Deliverable D8 Scenarios and wireless performance and coverage

Transcription

1 Deliverable D8 Scenarios and wireless performance and coverage Author(s): Jan Erik Håkegård (SINTEF) Bård Myhre (SINTEF) Per Hjalmar Lehne (Telenor) Terje Ormhaug (Telenor) Vidar Bjugan (Telenor) Marina Mondin (EuroConcepts) Muslim Elkotob (TUB) Frank Steuer (TUB) Partner(s): Telenor ASA EuroConcepts srl SINTEF Technische Universität Berlin Version: g Delivery Month: March 2005 Date: May 12, 2005 Workpackage, Activity: WP4, WP4-A2 Deliverable Type and Number: Report, D8 Distribution Type: Public Document Code: OBAN-WP4-SIN-049g-D Internet URL: IST 6FP Contract No Copyright by the OBAN Consortium.

2 OBAN-WP4-SIN-049g-D OBAN Consortium

3 Table of Contents 1. Introduction The role of WP4 in the OBAN project Description of the terms coverage and capacity Relations between WP4 and other WPs Organisation of WP Activity 1: Capacity and coverage scenarios Activity 2: Wireless capacity and coverage Activity 3: Emerging technologies within wireless Activity 4: Traffic analysis Contents of this deliverable Coverage analysis for single-cell networks Introduction Free-space propagation Indoor propagation Outdoor propagation Indoor-to-outdoor attenuation Coverage analysis Indoor coverage analysis Outdoor coverage analysis Indoor-to-outdoor communication Summary of coverage analysis Coverage analysis of multi-cell networks Introduction Access point distributions OBAN operators choice of policy Scenarios for access points in residential environments OBAN access points in small enterprises Assumptions about access points in medium and large enterprises Assumptions about access points in open spaces and vehicles Number of public hot spots Multi-cell interference Assumptions Statistical channel model Effect of uniform interference level Receiver input level Impact on coverage Statistical coverage analysis Residential area example Results, residential area Conclusions for multi-cell analysis Interference and coexistence issues Introduction Characteristics of interference sources Interference between WLAN and WPAN (Bluetooth) Performance evaluation of IEEE802.11b-IEEE interference Bluetooth IEEE g interference Microwave ovens OBAN-WP4-SIN-049g-D OBAN Consortium

4 Example of measured spectra Models of microwave oven interference Cordless phone interference Conclusions for interference analysis Capacity analysis Introduction MAC review DCF versus PCF Channel access schemes Fragmentation Throughput for one STA and one AP Time intervals of channel occupancy Saturation throughput The multiple STA case with perfect channel Estimated throughput Relation between transmission probability and packet collisions Cumulative saturation throughput Throughput per user Capture effect Capture probability Total throughput Throughput per user Loss of packets due to non perfect channel Throughput per user STAs transmitting at different data rate The performance anomaly of Normalised throughput Mixed g/b networks Unfairness between uplink and downlink traffic IEEE802.11e MAC EDCA HCCA ACK policies Throughput results Capacity of OBAN networks IEEE802.11b IEEE802.11a/g Conclusions for capacity analysis Measurements Introduction Telenor measurements Introduction WLAN antennas and measurement equipment Signal strength measurements and subjective test inside office building Outdoor test with 14.5 dbi beam antennas in the boarder area between LOS and NLOS Outdoor test with 14.5 dbi beam antennas and LOS Outdoor test with 14.5 dbi beam antennas with optical LOS and pine trees in the path area Conclusions TUB measurements Introduction Hardware constrains Constraints and Scenarios Detailed Scenario Descriptions and Measurement Results Conclusions of the TUB measurements Conclusions for the measurement campaigns OBAN-WP4-SIN-049g-D OBAN Consortium

5 7. Conclusions Coverage of single-cell networks with no interference Coverage of multi-cell networks with no interference Effect of interference on coverage Throughput References Appendix A Multi-rate Communication A.1. Introduction A.2. Data rates and modulation methods A Reference Model A.4. MAC Frames (MPDU) A.4.1. Exchange of data rate information A.5. PHY Frames (PPDU) A.5.1. Frequency hopping [Clause 14] A.5.2. Direct Sequence Spread Spectrum (2.4 GHz) [Clause 15, 18 and 19] A.6. OFDM (5 GHz and 2.4 GHz) [Clause 17 and 19] Appendix B Scenario descriptions OBAN-WP4-SIN-049g-D OBAN Consortium

6 OBAN-WP4-SIN-049g-D OBAN Consortium

7 1. Introduction 1.1. The role of WP4 in the OBAN project The overall objective of the OBAN project is to investigate the possibility of providing open broadband wireless access using WLAN access points (APs) connected to privately owned fixed broadband access lines. A residential gateway (RGW) connected to the termination of the fixed access line separates the traffic generated by the home user and the traffic generated by OBAN users. The RGW also ensures authentication and security of both home users and OBAN users, enables mobility through handover mechanisms to other OBAN cells or to other wireless communication systems, and provides each user with a certain degree of Quality of Service (QoS). Technical challenges related to these functionalities, together with the overall OBAN architecture, are assessed in WP2, and partly demonstrated in WP3, of the OBAN project. There is however another condition that must be met as well in order to fulfil the OBAN vision: The technology must be able to deliver sufficiently high capacity and coverage to enable an OBAN network to support requested applications and services with their associated QoS requirements to OBAN users. Moreover, the technology must enable the OBAN network to provide such services to a sufficiently high number of users to allow sound and realistic business cases to be built on the OBAN concept. The objective of WP4 is to provide the OBAN project with quantitative numbers on the capacity and coverage one may expect to obtain in realistic scenarios and under realistic assumptions Description of the terms coverage and capacity The capacity and coverage of an OBAN network is illustrated in Figure 1-1. The left part of the figure illustrates how the total available capacity is shared between home and OBAN (or guest) users. What is of interest in the OBAN project is the public capacity available to OBAN users. It depends on the total capacity of the fixed access line, on the wireless access capacity, on the capacity required by the home users, and on the number of OBAN users contending for the publicly available capacity. The right part of the figure illustrates the coverage, which depends on the density and distribution of WLAN APs and on their range. It is important to realise that capacity and coverage can not be considered separately. The coverage of an AP is larger for low capacities than for high capacities, as will be apparent later in this report. Local traffic (in-house and external) GSM, UMTS,. Guest Figure 1-1 Illustration of the separation of capacity between home users and OBAN users. Right: Illustration of OBAN coverage. OBAN-WP4-SIN-049g-D Page 1 of 177 OBAN Consortium

8 Relations between WP4 and other WPs WP4 takes input from WP1, WP2 and WP5: In WP1, a number of high level OBAN scenarios were defined. The parts of these scenarios that are of relevance for the WP4 issues are selected and defined in detail in WP4. In WP2, QoS functionalities of the RGW are assessed. These functionalities and their performance will be of importance when modelling and simulating the data traffic in WP4. If there are other functionalities related to security and mobility that have impact on capacity and coverage, these may be included in the traffic modelling as well. However, WP4 will not include any evaluation of mobility mechanisms such as handover. Mobility is solely considered in WP2 and WP3. In WP5, society aspects are considered. WP4 will take input related to placement, density and distribution of WLAN APs in various environments. WP4 provides output to WP2 and WP5: WP4 provides quantitative results for what throughput users may get access to under realistic conditions. These results will be of importance for the QoS activity of WP2. The WP4 results on coverage may be used as input for the mobility activity of WP2. WP5 will perform socio-economical analysis of OBAN business cases, and evaluate whether it is possible to build sound and realistic business cases on the OBAN concept. WP4 will provide valuable input to this activity, such as how many users may be supported simultaneously, what kind of services they may receive and the degree of contiguous coverage that may be obtained 1.2. Organisation of WP4 WP4 is divided into four activities. The relations between the activities are illustrated in Figure 1-2. A1 Coverage and Capacity scenarios A2 Wireless coverage and capacity A3 Emerging wireless technologies A4 Traffic analysis Figure 1-2 Relation between the activities in WP4 OBAN-WP4-SIN-049g-D Page 2 of 177 OBAN Consortium

9 Activity 1: Capacity and coverage scenarios Activity 1 consists in defining a number of capacity and coverage scenarios. These scenarios are loosely connected to high level OBAN scenarios developed in WP1. The capacity and coverage scenarios are however more detailed when it comes to e.g. applications, for which parameters such as data rate, packet lengths etc are defined, and to environments, for which propagation models are defined. The rationale behind this activity is the need to set some boundaries and limit the degrees of freedom before starting the analysis and traffic modelling. The objective is that these boundaries will make the work more focused and simplify the understanding and interpretation of the results. The result of this activity is reported in an internal report D8.1, and is included in this report as Appendix B Activity 2: Wireless capacity and coverage In activity 2 we consider the capacity and coverage that state-of-the-art WLAN technology offers to an OBAN type of network. Today, the bottleneck for the capacity is in most cases the fixed access line. This is in particular the case when the access line is ADSL, which has a throughput somewhere in the area of 4 Mbps to 8 Mbps, depending on how far the subscriber is located from the DSLAM. This capacity is far below that of WLAN technologies that offer MAC throughput in the order of 30 Mbps. The situation of the access line being the bottleneck is however in the process of changing. For one, new emerging ADSL technologies like ADSL2+ increase the downlink capacity to 25 Mbps, and VDSL and above all fibre-to-the-home (FTTH) increase the capacity even more. As people get used to more and more fixed broadband services, operators offer triple services (TV, phone, Internet access) etc, the penetration of higher capacity lines will increase. Secondly, an increasing number of local services will lead to increased local wireless traffic. Hence, the wireless network will need higher capacity than the fixed network to handle the local traffic in addition to the external traffic passing through the fixed access line. This activity considers current state-of-the-art WLAN equipment, i.e. IEEE802.11a devices that transmit in the 5 GHz band and IEEE802.11g devices that transmit in the 2.4 GHz band. Included in the work is also IEEE802.11b, as most current WLAN devices still uses this standard. Backward compatibility to IEEE802.11b is also an important part of the IEEE802.11g standard. When evaluating coverage, distribution of APs within an environment must be included. Assumptions must be made based on parameter related to the environment and on hypothesis of the penetration of OBAN APs in residents. Input from WP5 will be of importance to obtain realistic numbers, and also to align the results from this activity with results from WP5. The methodology applied is as follows. First we develop analytical models of the physical layer and of the MAC layer. These models are used to obtain an understanding of the mechanisms of these lower layers that form the basis of the capacity and coverage one obtains. The analytical and simulation results are compared with real measurements. Higher protocol layers (TCP/IP) are not considered in this activity. They will however be included in activity 4 as part of the traffic modelling and simulation. The result of this activity is presented in this report Activity 3: Emerging technologies within wireless If the fixed access line technologies are evolving, this is even more the case for WLAN technologies. There are currently a large number of working groups (WGs) within IEEE working on developing new amendments to the standards. In addition, other standardisation group as IEEE and IEEE develop standards that are of interest for OBAN. Within , WGn is of particular interest for the OBAN project. WGn is developing a new high rate amendment of The goal for this standard is to provide a data rate above 100 Mbps. This standardisation work is still early in the development, and will probably not be finalised before year Still, it is clear which strategies that will be applied to achieved the spectral efficiency needed for such high data rates using the allocated bandwidths. These strategies are the use of multiple antenna techniques and also more efficient coding and modulation techniques. OBAN-WP4-SIN-049g-D Page 3 of 177 OBAN Consortium

10 This activity consists of three tasks. The first task is to keep an eye on the development of the IEEE802.11n amendment and other standards and to do analytical work around these standards as they progress. We do this in order to obtain an understanding of the capacity and coverage that will be obtained under typical OBAN scenarios using new technology. The second task considers analytically the use of multiple antenna techniques. The system performance may profit from using multiple antennas in different ways depending on the environment, and which technique that is optimal will vary accordingly. This second task aims at increasing the understanding of how we can increase capacity and coverage in typical OBAN scenarios. The third task considers adaptive coding and modulation. As for the multiple antenna task, the goal is to increase the understanding of how we can enhance the capacity and coverage in typical OBAN scenarios. The results of this activity will be presented in the internal report D21.1, and will be included as appendix to the deliverable D Activity 4: Traffic analysis In activity 2 and 3, we do analytical evaluation of the capacity and coverage at the MAC layer and compared these results with simple simulations and measurements. In activity 4 we use the understanding obtained in these activities together with the scenarios we developed in activity 1 to develop traffic models. These models will contain the fixed network, functionalities of the RGW as well as the wireless network. It will also include differentiation between home users and OBAN users and differentiation between QoS classes. Finally, it introduces statistics regarding number of users, usages etc. Two simulators will be developed. One is a bit level simulator, and will provide results related to the use of multiple antennas. The second one is a network simulator developed using OPNET. The results from this activity constitute the main results from this WP. That is, they include quantitative results regarding the throughput available to OBAN users, number of OBAN users that can be supported, and coverage of an OBAN network. The result of this activity will be presented in deliverable D Contents of this deliverable This report contains the results of activity 2. The results of activity 1 are included in Appendix B. The report consists of a total of eight chapters, including this introduction, and two appendixes. Chapter 2 consists of an assessment of the propagation channels and of the WLAN physical layer. The IEEE802.11a, IEEE802.11b and IEEE802.11g physical layers are included in the assessment. The results of this Chapter include the relation between distance and channel bit rate for the various environments. Indoor, outdoor and indoor-to-outdoor propagation models are included. Chapter 3 contains a performance analysis of multi-cell networks. This Chapter builds on the results from Chapter 2, and assumes no coordination between APs. Interference from other OBAN APs then has the same effect as interference from non-oban WLAN systems. The Chapter also includes an assessment of the distribution and density of APs in an OBAN network in different environments. In Chapter 4 we include interference into our model. Overcrowded frequency bands and interference may prove to be the main threat to OBAN type of systems. Sources of interference may be Bluetooth equipment, microwave ovens and others. An assessment of different interference sources is performed and their impact on the performance of OBAN networks investigated. Chapter 5 consists of an assessment of the IEEE802.11MAC layer. An analytical model is used to model the CSMA access scheme applied by the standard. The results of this Chapter include the number of users that can be supported by an AP as function of the distance between the AP and the users, and of the data rate required by each user. Results are obtained for the three IEEE802.11a/b/g standards. OBAN-WP4-SIN-049g-D Page 4 of 177 OBAN Consortium

11 In Chapter 6 results from measurement campaigns are reported. Measurements have been done by Telenor and by TUB. Conclusions are given in Chapter 7. Appendix A contains an overview of the different IEEE physical layers and modes. Appendix B (D8.1) contains the results from activity 1. Seven scenarios have been defined, of which two represent the core residential OBAN scenario. Each scenario consists of use cases, applications and environments. OBAN-WP4-SIN-049g-D Page 5 of 177 OBAN Consortium

12 OBAN-WP4-SIN-049g-D Page 6 of 177 OBAN Consortium

13 2. Coverage analysis for single-cell networks 2.1. Introduction In this section we recall some basic Single-Input Single-Output (SISO) WLAN channel models [19] for outdoor, indoor and outdoor-to-indoor propagation. In particular, the free-space propagation attenuation and the modifications required to consider the presence of shadowing are recalled. These results will be used in Chapter 4 for the evaluation of the single cell coverage Free-space propagation Link budget calculations in case of Single-Input Single-Output (SISO) free-space propagation require an estimate of the received signal power. In case of free-space propagation the usual propagation equation is used: P R = P T G T G 2 R λ ( 4 π d ) 2 where P T and P R are the transmitted and the received power, respectively, G T and G R are the gains of the transmitting and receiving antennas, respectively, λ is the signal wavelength, and d is the distance between the transmitting and the receiving antennas. The previous equation can be expressed in decibels as: P [ db] = EIRP[ db] + G [ db] PL ( f, d)[ db] R R + whereby the parameter EIRP represents the effective isotropic radiated power, i.e.: EIRP[ db] = P [ db] G [ db] T + and PL fs (f c,d) is the free space attenuation (or Propagation Loss) expressed by: T fs c where PL fs ( f c, d)[ db] = 10 log 10 f c c 4 π d 8 m c = 3 10, f c is the carrier frequency, evaluated in Hz, and d is the distance measured in s meters. The free space attenuation factor can also be expressed as: PL ( f, d )[ db] = log f + 20 d fs c c log where f c is expressed in GHz and d in meters. By the previous setup, the link budget equation (in db) can be written as: P [ db] EIRP [ db] + G [ db] PL( f d ) [ db] = (3.1) R R c, OBAN-WP4-SIN-049g-D Page 7 of 177 OBAN Consortium

14 with PL=PL fs. In case of no-free space propagation (indoor or indoor-outdoor propagation), the previous equation is still valid, but the attenuation factor PL(f c,d) will assume different values depending on the type of propagation channel, and in particular PL(f c,,d) will account for shadowing, if present Indoor propagation The mathematical models we will refer to for the indoor environment have been proposed by Medbo et al. in [6][7] plus some additional models suitable for office environments. In summary, the characteristics of the models suggested for different environments [19][25] are shown in the table below: Model Environment Model [19] Conditions K Factor * Rms Delay Spread [ns] Number of Filter Taps A Flat fading LOS/ NLOS LOS/NLOS 0/ (no multipath) B Residential B LOS/B- LOS/NLOS 0 / NLOS C Residential / B LOS/C- LOS/NLOS 0 / Small Office NLOS D Typical C LOS/D- LOS/NLOS 3 / Office NLOS E Large Office D LOS/E- LOS/NLOS 6 / NLOS F Large Space (indoors / outdoors) E LOS/F- NLOS LOS/NLOS 6 / Table 2-1 Channel models * K factor for LOS conditions is to be applied just to the first channel tap, while for the other taps it is K= - db. In case of indoor propagation, up to the so called breakpoint distance, d BP, the free space signal attenuation model applies (where the attenuation is proportional with d 2 ), while for higher distances the additional attenuation increases with d 3.5.The shadowing must also be accounted for through a random variable s with proper statistical characteristics. The signal attenuation can therefore be expressed as: PL[ db] = PL [ db] + s with d fs d BP d PL [ db ] = PL fs ( f c, d BP )[db ]+35 log + s with d > d d 10 BP BP where s is a zero mean Gaussian random variable. The main parameters of the previous equations are summarized in Table 2-2. Model d BP (m) Slope for d<d BP Slope for d>d BP Standard deviation of shadowing variable s for d<d BP (LOS) Standard deviation of shadowing variable s for d>d BP (NLOS) A B C D E F Table 2-2 Path loss parameters of the channel models OBAN-WP4-SIN-049g-D Page 8 of 177 OBAN Consortium

15 The resulting path loss as function of distance for the various channel models is shown in Figure Attenuation [db] C1, C2 C3 50 C4 C Distance [m] Figure 2-1 Attenuation as function of distance for the different channel models Outdoor propagation In lack of good outdoor WLAN propagation models, we use models developed for typical cellular systems. In these models, the base stations are mounted high above ground to extend the coverage, while the mobile terminals are handheld and consequently located close to ground Propagation in suburban environment Signal attenuation in a suburban environment can be evaluated as: PL[dB]=PL fs (f c,d 0 ) + 10γ log 10 ( d / d 0 ) + s with d d 0 where λ is the wavelength, and d 0 is experimentally set to 100 m. The parameter γ, i.e., the attenuation exponent, is a Gaussian random variable that can be expressed as: γ = ( a b hb + c / hb ) + x σ γ whereby h b is the height of the base station antenna (usually 10m h b 80m), σ γ is the standard deviation of γ and it depends on the type of ground (see Table 2-3), x is a Gaussian zero mean random variable with distribution N[0,1], a, b, c are constants that depend on the type of ground (see Table 2-3). Finally, s represents the shadowing and can be expressed as: s = z σ where z is a Gaussian zero mean random variable with distribution N[0,1], while σ is a Gaussian random variable with distribution N[μ σ,σ σ ], where μ σ and σ σ are constants depending on the ground characteristics (see Table 3.3). The numerical values of the constants that depend on the type of ground are shown in Table 3.3. OBAN-WP4-SIN-049g-D Page 9 of 177 OBAN Consortium

16 Model parameters Type of ground A (Hilly/medium-high tree B (Hilly/small tree density C (Flat/small tree density) density) or Flat/medium-high tree density) a b (m -1 ) c (m) σ γ μ σ σ σ Table 2-3 Suburban environment parameters Mean attenuation exponent E[γ] Model A Model B Model C Base station height h [m] Figure 2-2 Mean attenuation exponent as function of base station height The mean value of the attenuation coefficient as function of the base station height is illustrated in Figure 2-2. For low base station heights, the model provides large values for the attenuation exponent. The variable terms can be simplified by computing the average attenuation. A good approximation of these terms is given by: σ [ ] 1/ 2 υ = 100 σ γ (log10 ( d / d 0 )) + μσ + σ σ where σ υ represents the standard deviation of a Gaussian random variable. The total attenuation can therefore be written as: PL[dB]=PL fs (f c,d 0 ) + 10 ( a b h b + c / h b ) log ( d / d 10 0 )] + υ σ υ with d d 0 where υ is a zero-mean Gaussian random variable with standard deviation σ υ. OBAN-WP4-SIN-049g-D Page 10 of 177 OBAN Consortium

17 Attenuation in pedestrian outdoor environments Two different attenuation models are typically considered, the Standard model ITU-R (M. 1225), and the Standard IEEE model (Stanford B). Standard model ITU-R (M.1225) In this kind of environment the attenuation model, used by the ITU R (M. 1225) [8], is expressed as: PL db ] = 40 log ( d) + 30 log ( ) s [ f c where d is the distance, expressed in km, f c is the frequency expressed in MHz, and s is the log-normal distribution of the random variable (with standard deviation of 10 db) representing the shadowing effect. Standard IEEE Model (Standard B) For this type of environment the model adopted by IEEE (Stanford B) [9] is given by: 4 π d 0 c d PL[ db] = 20 log + a b h + + s k b h b d log10 λ 0 (3.2) k = log f 10.8 log h u 2 (3.3) where h b and h u are, respectively, the base station and user heights (typically their values are 15m and 1.5m), d 0 is breakpoint distance selected as 100m, s is the log-normal distribution of the random variable due to the shadowing with 10 db standard deviation; a, b, c are constants depending on the ground (type B: a = 4, b = , c = 17.1), and f c is the frequency. Figure 2-3 Attenuation as function of distance for the outdoor model ITU-R (M.1225) in absence of shadowing effect. OBAN-WP4-SIN-049g-D Page 11 of 177 OBAN Consortium

18 Attenuation in a vehicular outdoor environment Two different attenuation models are typically considered, the Standard IEEE model (Stanford B) given above [9], and the Standard model ITU-R (M. 1225) [8], which is expressed as: [ Δh ] log ( d) 18 log ( Δh ) + 21 log ( f ) + s PL db] = [ b b 10 + where d is the distance in km, f is the frequency expressed in MHz, Δh b is the height of the base station with respect to the medium level of the buildings height, and s is the log-normal distribution of the random variable (with 10 db standard deviation) due to shadowing Indoor-to-outdoor attenuation Two different attenuation models are typically considered, the Standard model ITU-R (M. 1225), and the Standard IEEE model (Stanford B). ITU R (M. 1225) model [8] specifies the following attenuation model: PL[ db] = 40 log10 ( d) + 30 log10 ( f ) PLindooor loss + s where d is the distance in Km, f is the frequency, expressed in MHz, PL indoor-loss is the attenuation taking into account the effects of the transition outdoor-indoor (from 2 to 12 db), and s is the log-normal distribution of the random variable (with 14.4 db standard deviation) which accounts for shadowing. Material Attenuation Window in brick wall 2 db Metal frame glass wall into 6 db building Office wall 6 db Metal door in office wall 6 db Cinder block wall 4 db Metal door in brick wall 12.4 db Brick wall nexto to metal door 3 db Table 2-4 Attenuation at 2.4 GHz for various building materials Material Attenuation Interior hollow wall (10 cm) 3 db Interior hollow wall (15 cm) 4 db Interior office window 6 db Solid wood door (4.4 cm) 10 db Brick (9 cm) 4 db Exterior single pane window (1.25 cm) 6 db Exterior double pane coated glass (2.5 cm) 20 db Table 2-5 Attenuation at 5 GHz for various building materials A list of typical attenuations PL indoor-loss due to various building materials at 2.4 GHz [43] is given in table 2-4, while Table 2-5 [44] shows some attenuations in the WLAN 5 GHz band. The attenuation depends on frequency, material and polarization, and it is difficult to give final values. In general the attenuation for a given material tends to be higher in the 5 GHz band (with respect to the 2.4 GHz band), and much higher values than the one listed in Table 2-4 and Table 2-5 can be observed in thicker materials, like exterior concrete walls (in the order of 40 to 50 db for both the 2 GHz and the 5 GHz bands). OBAN-WP4-SIN-049g-D Page 12 of 177 OBAN Consortium

19 The model adopted from the IEEE (Stanford B) [9] is defined by equations (3.3) and (3.4): f 2000 h 2 u k = 6 log log10 + PL Indoor (3.4) where h b and h u are, respectively, the base station and the user heights (typical values for outdoor environments are 15m and 1.5m); d 0 is a normalization factor (100m); s is the log-normal distribution of the random variable (with 14.4 db standard deviation) which accounts for the shadowing effect; a, b, c are constants depending on the type of ground (type B : a = 4, b = , c = 17.1); f is the frequency. Finally, PL Indoor is the attenuation due to the transition outdoor-to-indoor environment (typically 20 db) Coverage analysis For each a/b/g standard the propagation models result in circular coverage zones. High data rates give zones with small radius, and the radius increases with decreasing data rate. In real environments, the coverage will of course not be circular due to obstructions and objects in the propagation environment. However, for the coverage analysis of multi-cell networks, the channel models provide a valuable tool Indoor coverage analysis In this section the coverage analysis of a single indoor cell scenario is described, utilizing the indoor channel models A trough F recalled in Section and considering the IEEE a/b/g standards. In the analysis, the minimum signal-to-noise ratio (SNR or E s / N0 ) allowing reliable communication is evaluated, obtaining an approximation for the transmitted bit rate as a function of the maximum coverage distance. The achievable bit rate is statistically averaged using Monte Carlo simulations by varying the statistical parameters of the propagation channel model, which have been previously described in Section 2.1. The received noise power is evaluated as: PN RX [ db] = TN[ db] + BW[ db] + NF[ db] whereby PN RX is the noise power at the input filter of the receiver, TN is the thermal noise (typical value is dbm for every WLAN standard considered in OBAN), BW is the channel bandwidth expressed in db, and NF is the noise factor assumed to be equal to 10 db. The receiver sensitivity (i.e. the minimum power allowed for the received signal) is then: PS MINRX ( Sensitivity) = PN RX + MI + 0 MIN where MI is the interference margin, and ( E s / N 0 ) min is the minimum SNR required for guaranteeing a bit error rate (BER) less than or equal to Finally, the maximum allowed attenuation (which depends on the minimum signal to noise ratio), can be derived as: AT [ db] = EIRP[ db] + G [ db] PS [ db] MAX whereby the EIRP parameter is the effective isotropic radiated power, G R is the gain of the receiving antenna, and PS MINRX is the minimum power of the received signal. The SNR as a function of the receiver distance is given as OBAN-WP4-SIN-049g-D Page 13 of 177 OBAN Consortium R Es N MINRX

20 Es N 0 = EIRP[ db] PL( f c, d)[ db] MI[ db] PN RX [ db] IEEE802.11a coverage The ECC has identified frequency bands that can be used by a equipment in Europe [42]: MHz. Only indoor use, mean EIRP limited to 200 mw, use of dynamic frequency selection (DFS) as well as transmitter power control (TPC) are required above MHz MHz. Indoor as well as outdoor use allowed, mean EIRP limited to 1 W, use of DFS and TPC required. As of March 2005, this decision is implemented in 6 countries, among them the Netherlands and Switzerland. In addition Italy, France and Sweden have committed to the decision. It is expected that most of the remaining countries in Europe will implement the decision in their National Frequency Allocation tables as well within short delay. Using the ECC regulations and combining the formulas seen above with the parameters of the IEEE a standard [28], the link budget for the two frequency bands can be derived (see Table 2-6 and Table 2-7). Current IEEE802.11a equipment does however have considerably lower transmit power than the maximum EIRP value allowed by ETSI. Typical values for maximum transmit power range from 12 dbm to 17 dbm. Using antennas with gain in the order of 2 dbi to 5 dbi, the resulting EIRP max is in the order of 14 dbm to 22 dbm. The range of conventional state-of-the-art equipment will consequently be lower than that indicated in the tables below. Center frequency [GHz] Bit rate [Mbps] Modulation BPSK BPSK QPSK QPSK 16QAM 16QAM 64QAM 64QAM Code rate 1/2 3/4 1/2 3/4 1/2 ¾ 2/3 ¾ EIRP max [dbm] RX antenna s gain [dbi] Thermal noise [dbm/hz] Channel band [MHz] Noise factor [db] Power of received noise [dbm] Margin [db] Minimum Es/N 0 [db] * Minimum power of received signal (Sensitivity) * [dbm] Maximum tolerable attenuation Table 2-6 Parameters of relevance for the standard IEEE a using the ETSI regulations for the MHz band. EIRP max = 23 dbm (200 mw) * Sensitivity and minimum Es/N 0 are those defined from the IEEE a standard, required to obtain BER OBAN-WP4-SIN-049g-D Page 14 of 177 OBAN Consortium

21 Center frequency [GHz] Bit rate [Mbps] Modulation BPSK BPSK QPSK QPSK 16QAM 16QAM 64QAM 64QAM Code rate 1/2 3/4 ½ 3/4 1/2 ¾ 2/3 ¾ EIRP max [dbm] RX antenna s gain [dbi] Thermal noise [dbm/hz] Channel band [MHz] Noise factor [db] Power of received noise [dbm] Margin [db] Minimum Es/N 0 [db] Minimum power of received signal (Sensitivity) [dbm] Maximum tolerable attenuation Table 2-7 Parameters of relevance for the standard IEEE a using the ETSI regulations for the MHz band. EIRPmax = 30 dbm (1 W) Using the maximum allowable path loss for the various data rate modes and the relation between attenuation and distance, the maximum distance for each channel model is easily obtained. The result is shown in Table 2-8 and Table 2-9 for the two frequency bands. Channel 6 Mbps 9 Mbps 12 Mbps 18 Mbps 24 Mbps 36 Mbps 48 Mbps 54 Mbps A-C D E F Table 2-8 Maximum distance for each mode and channel model for a. Applies for the frequency band MHz indoors (EIRPmax=200 mw) Channel 6 Mbps 9 Mbps 12 Mbps 18 Mbps 24 Mbps 36 Mbps 48 Mbps 54 Mbps A-C D E F Table 2-9 Maximum distance for each mode and channel model for a. Applies for the frequency band MHz (EIRPmax=1 W) OBAN-WP4-SIN-049g-D Page 15 of 177 OBAN Consortium

22 6 Mbps 18 Mbps 36 Mbps 54 Mbps A-C D 30 m F E Figure 2-4 Illustration of coverage zones for a using channel models A-F and the frequency band GHz (EIRP=200 mw) 6 Mbps 30 m 18 Mbps 36 Mbps 54 Mbps A-C D F E Figure 2-5 Illustration of coverage zones for a using channel models A-F and the frequency band GHz (EIRP=1 W) In Figure 2-6 and Figure 2-7 we show results from Monte Carlo simulations when we have added shadowing. Receiver sensitivity is -91 dbm for IEEE802.11a. OBAN-WP4-SIN-049g-D Page 16 of 177 OBAN Consortium

23 Figure 2-6 Indoor coverage for IEEE a using the ETSI regulations for the MHz band. EIRP max = 23 dbm (200 mw) Figure 2-7 Indoor coverage for IEEE a using the ETSI regulations for the MHz band. EIRPmax = 30 dbm (1 W) OBAN-WP4-SIN-049g-D Page 17 of 177 OBAN Consortium

24 IEEE802.11b coverage Parameters of relevance for this standard [29] are shown in Table Frequency[GHz] Bit rate [Mbps] Modulation CCK CCK CCK CCK EIRP max [dbm] RX antenna s gain[dbi] Thermal noise [dbm/hz] channel band [MHz] Noise factor [db] Power of received noise [dbm] Margin [db] minimum Es/N 0 [db] * Minimum power of received signal (Sensitivity) * [dbm] Maximum available attenuation Table 2-10 Parameters of relevance for the IEEE802.11b standard * Sensitivity and minimum Es/N 0 are those defined by the IEEE b standard, required to obtain BER Using the maximum allowable path loss for the various data rate modes and the relation between attenuation and distance, the maximum distance for each channel model is easily obtained. The results are given in Table 2-11 and shown graphically in Figure 2-8. Channel 1 Mbps 2 Mbps 5,5 Mbps 11 Mbps A-C D E F Table 2-11 Maximum distance for each mode and channel model for b (no shadowing) OBAN-WP4-SIN-049g-D Page 18 of 177 OBAN Consortium

25 1 Mbps 2 Mbps 5.5 Mbps 11 Mbps A-C 30 m D F E Figure 2-8 Illustration of coverage zones for b using channel models A-F (no shadowing). In Figure 2-9 we show results from Monte Carlo simulations where we have added shadowing. Figure 2-9 Indoor coverage for IEEE802.11b in LOS condition OBAN-WP4-SIN-049g-D Page 19 of 177 OBAN Consortium

26 IEEE802.11g coverage Parameters of relevance for this standard [30] are shown in Table 2-12 for the different considered modulations. Frequency [GHz] Bit rate [Mbps] Modulation BPSK BPSK QPSK QPSK 16QAM 16QAM 64QAM 64QAM Code rate 1/2 3/4 1/2 3/4 1/2 ¾ 2/3 3/4 EIRP max [dbm] RX antenna s gain[dbi] Thermal noise [dbm/hz] Channel band [MHz] Noise factor [db] Power of received noise [dbm] Margin [db] Minimum Es/N 0 [db] * Minimum power of received signal (Sensitivity) * [dbm] Maximum available attenuation Frequency [GHz] Bit rate [Mbps] Modulation CCK CCK CCK CCK EIRP max [dbm] RX antenna s gain [dbi] Thermal noise [dbm/hz] Channel band [MHz] Noise factor [db] Power of received noise [dbm] Margin [db] Minimum Es/N 0 [db] * Minimum power of received signal (Sensitivity) * [dbm] Maximum available attenuation Table 2-12 Parameters of relevance for the IEEE802.11g standard * Sensitivity and minimum Es/N 0 are those defined from the IEEE g standard, required to obtain BER The maximum distance for each data rate mode and channel model to obtain BER= 10 is listed in Table 2-13 and shown graphically in Figure Channel 6 Mbps 9 Mbps 12 Mbps 18 Mbps 24 Mbps 36 Mbps 48Mbps 54 Mbps A - C D E F Table 2-13 Maximum distance for each mode and channel model (no shadowing) OBAN-WP4-SIN-049g-D Page 20 of 177 OBAN Consortium

27 6 Mbps 30 m 18 Mbps 36 Mbps 54 Mbps A-C D F E Figure 2-10 Illustration of coverage zones for g using channel models A-F (no shadowing). Figure 2-11 shows the results of Monte Carlo simulations using the channel models A to F and taking into account the shadowing. Figure 2-11 Indoor coverage for IEEE802.11g in LOS conditions OBAN-WP4-SIN-049g-D Page 21 of 177 OBAN Consortium

28 Coverage comparisons among IEEE802.11a/b/g The comparative graph that follows shows theoretical performance of the three standards IEEE a/b/g, obtained with channel model A without fading and shadowing. Figure 2-12 Theoretical indoor performance of the analyzed standards. Figures 2.6 to 2.11 compare coverage results of WLAN standards a-b-g in terms of bit rate as a function of the distance, for the channel models A-F described in Table 2.2 above Coverage for channel model A IEEE a GHz IEEE b IEEE g IEEE a GHz 40 Rate [Mbps] Distance [m] Figure 2-13 Coverage for the examined standards and channel model A. OBAN-WP4-SIN-049g-D Page 22 of 177 OBAN Consortium

29 Figure 2-14 Coverage for the examined standards and channel model B Coverage for channel model C IEEE a GHz IEEE b IEEE g IEEE a GHz 40 Rate [Mbps] Distance [m] Figure 2-15 Coverage for the examined standards and channel model C. OBAN-WP4-SIN-049g-D Page 23 of 177 OBAN Consortium

30 60 50 Coverage for channel model D IEEE a GHz IEEE b IEEE g IEEE a GHz 40 Rate [Mbps] Distance [m] Figure 2-16 Coverage for the examined standards and channel model D Coverage for channel model E IEEE a GHz IEEE b IEEE g IEEE a GHz 40 Rate [Mbps] Distance [m] Figure 2-17 Coverage for the examined standards and channel model E. OBAN-WP4-SIN-049g-D Page 24 of 177 OBAN Consortium

31 60 50 Coverage for channel model F IEEE a GHz IEEE b IEEE g IEEE a GHz 40 Rate [Mbps] Distance [m] Figure 2-18 Coverage for the examined standards and channel model F Outdoor coverage analysis In this section the coverage analysis of a single outdoor cell scenario is described, considering in particular the pedestrian outdoor environment according to the Standard model ITU-R (M. 1225) and the indoor-to-outdoor case (outdoor propagation due to an indoor transmitter). The standards IEEE a/b/g have been considered, which are typical low-mobility standards (which is the reason why the outdoor vehicular environment is not considered. As in the indoor case, the minimum SNR allowing reliable communication is evaluated, obtaining an approximation for the transmitted bit rate as a function of the maximum coverage distance. The achievable bit rate is statistically averaged using Monte Carlo simulations by varying the statistical parameters of the considered propagation channel models, described in section Pedestrian outdoor environment The outdoor coverage for a pedestrian environment has bee evaluated according to the standard model ITU-R (M. 1225) in Section using the IEEE a/b/g parameters reported in Table 2-6 to Table The comparison between the results obtained for the three standards is shown in Figure It can be observed how the achievable bit rate rapidly decreases, because of the high attenuation (increasing ad d 4 ), and the high fading of the standard model ITU-R (M. 1225). OBAN-WP4-SIN-049g-D Page 25 of 177 OBAN Consortium

32 60 50 Outdoor coverage IEEE a GHz IEEE b IEEE g IEEE a GHz 40 Rate [Mbps] Distance [m] Figure 2-19 Outdoor coverage for a pedestrian environment (standard model ITU-R (M. 1225)) for the IEEE a/b/g standards Indoor-to-outdoor communication An estimate of the indoor-to-outdoor propagation is obtained by recalling formulas given in Section 2.1.4, the parameters of Table 2-2 describing models A through F, and adding a loss term PL indoor-loss of 6 db to account for the loss due to the walls. The path loss can be expressed as: PL [ db ] = PL [ ] + fs db s+pl indoor loss with d d BP d PL [ db ] = PL fs ( f c, d BP )[db ] +35 log + S+PL indoor loss with d > d d 10 BP BP In Table 2-14 the reduction in range is illustrated for different values of the additional loss due to external windows or walls when the shadowing is not taken into account (see Table 2-4 and Table 2-5 for examples of attenuation at the 2.4 GHz and 5 GHz band). The additional loss leads to a constant reduction in range for all standards and channel data rates. This reduction is shown in Table 2-15 together with the corresponding reduction in coverage area. Notice that the attenuations are somewhat optimistic, since they have been selected in the hypothesis of having a favorable positioning of the access point for outdoor propagation a (5.2 GHz) a (5.6 GHz) g b No wall db additional loss db additional loss db additional loss Table 2-14 Range as function of additional attenuation due to external walls or windows and no shadowing. Maximum channel data rate is considered for each standard. OBAN-WP4-SIN-049g-D Page 26 of 177 OBAN Consortium

33 Range Coverage area No wall 100 % 100 % 2 db additional loss 88 % 77 % 6 db additional loss 67 % 45 % 12 db additional loss 45 % 20 % Table 2-15 Reduction in range and coverage due to propagation through an external wall or window. The coverage when the shadowing is taken into account is shown in Figure 2-20 to Figure 2-23 for the three considered standards. Notice that this case is optimistic, since it is based on the hypothesis that the propagation in the so called outdoor environment follows the same rules as the so called indoor environment, and in particular it increases as d 3.5. The achievable bit rate decreases in fact less rapidly than in the outdoor case. It can be observed that in the given hypothesis the achievable bit rate is similar to the indoor case. The coverage characteristics depend however from the loss coefficient PL indoor loss, which may vary as shown section , and may assume very high values in case of concrete external walls, making indoor-to-outdoor propagation practically impossible. This result tells us that if an external coverage is sought after, like in the case of a garden of a terrace, the access point should be properly located in order to avoid attenuation due to walls Indoor-to-outdoor coverage for IEEE a (5.6 GHz band) model A model B model C model D model E model F 40 Rate [Mbps] Distance [m] Figure 2-20 Indoor-to-outdoor coverage for IEEE a (5.6 GHz band) in indoor models A-F. OBAN-WP4-SIN-049g-D Page 27 of 177 OBAN Consortium

34 60 50 Indoor-to-outdoor coverage for IEEE a (5.25 GHz band) model A model B model C model D model E model F 40 Rate [Mbps] Distance [m] Figure 2-21 Indoor-to-outdoor coverage for IEEE a (5.25 GHz band) in indoor models A-F Indoor-to-outdoor coverage for IEEE b model A model B model C model D model E model F 8 Rate [Mbps] Distance [m] Figure 2-22 Indoor-to-outdoor coverage for IEEE b and indoor models A through F. OBAN-WP4-SIN-049g-D Page 28 of 177 OBAN Consortium

35 60 50 Indoor-to-outdoor coverage for IEEE g model A model B model C model D model E model F 40 Rate [Mbps] Distance [m] Figure 2-23 Indoor-to-outdoor coverage for the IEEE g standard and indoor models A through F As a comparison, the indoor-to-outdoor coverage obtained by adding loss term PL indoor loss of 6 db to the previous pedestrian outdoor model (standard model ITU-R) is also shown, in the hypothesis that the access point is located inside a building but extremely close to the external wall, so that the propagation is substantially external, apart from the loss term due to the walls. The achievable rate as a function of the distance is shown in Figure It can be observed that also in this hypothesis the achievable bit rate is similar to the indoor case Indoor-to-outdoor coverage IEEE a GHz IEEE b IEEE g IEEE a GHz 40 Rate [Mbps] Distance [m] Figure 2-24 Indoor to outdoor coverage for IEEE a/b/g standards and outdoor pedestrian model. OBAN-WP4-SIN-049g-D Page 29 of 177 OBAN Consortium

36 2.3. Summary of coverage analysis In this section we have presented channel propagation models for different environments. The channel models are characterised by a deterministic attenuation as function of frequency and distance, and by an additive stochastic shadowing process. Further, we have extracted parameters from the IEEE WLAN standards to find the minimum received signal power necessary to obtain reliable communication which is defined as obtaining a BER better than or equal to Based on the channel models, the system parameters and the European regulations on maximum emitted energy in the relevant frequency bands, the range and coverage as function of channel data rate is found in various environments. The results of the coverage analysis are summarised in Table 2-16 to Table Due to the statistical nature of the channel models, the propagation attenuation at a certain range will vary. As the channel data rate depends on the propagation attenuation, this will be stochastic as well. This is why we show the coverage as function of the average channel data rate. The results are based on the following assumptions: The outdoor loss exponential is 4, while the indoor loss exponential is 2 at distances smaller than the breakpoint distance and 3.5 at distances larger than the breakpoint distance. The breakpoint distance depends on the model. Hence, the signal is attenuated much faster outdoors than indoors. In rural and suburban areas, this is not necessarily the case. In urban areas, where the distance of communication is larger than block sizes, it is more likely to be correct. The indoor-to-outdoor propagation loss is set to 6 db, in the hypothesis of placement of the access point close to a window or a thin wall. This loss will depend on the construction materials and the thickness of the external wall the signal is passing through, and may range from 2-3 db to 50 db. In the coverage calculations we have used the maximum permitted EIRP. Current IEEE802.11a stateof-the-art equipment is not capable of transmitting this high power levels. This is particularly the case for WLAN cards implemented in mobile terminals. Hence, without directive antennas, the coverage of current systems will be less than that estimated in this section. IEEE802.11a equipment transmitting in the MHz band may only be used indoors. We have not considered this limitation and estimate outdoor coverage also for this frequency band. From the tables below we can draw the following conclusions: The results clearly show that coverage is closely related to data rate. For instance, the range of IEEE802.11g equipment is about 3 times larger with average channel data rate 10 Mbps than it is with average channel data rate 50 Mbps. The difference in coverage between IEEE802.11g and IEEE802.11a is not significant. The path attenuation is larger for the a standard due to higher transmit frequency. However, the permitted EIRP is higher for the a standard, leading to comparable coverage. It is however important to note that directive antennas is necessary for a equipment to transmit at maximum EIRP. The indoor coverage for all models is larger than the typical sizes of the environments (e.g. residents, small offices). Hence, the practical coverage for an implementation is limited more by the actual surroundings, building materials and structures in the propagation environment than by the theoretical limits. The outdoor range is relatively short using our outdoor model. The model is developed for cellular systems with base stations at elevated locations and handheld terminals at street level. In typical OBAN environments it is likely that the outdoor coverage will be larger than that obtained with the outdoor model. Probably the Large space indoor/outdoor model F gives a better estimate of the actual outdoor OBAN coverage. In the results for indoor-to-outdoor coverage we have assumed an additional path loss of 6 db due to propagation through an external thin wall or window, and the same channel model on both sides of the wall. The range compared to the range without the additional loss is about 67 %. If the additional loss due to the propagation through a thin wall or a window increases to 12 db, the corresponding range is reduced to 45 % of the range with no additional loss. If an exterior concrete wall is present, no external propagation is possible. This clearly indicates that an access point should be placed on an outside thin wall or close to a window to provide large coverage for OBAN users located outdoors. OBAN-WP4-SIN-049g-D Page 30 of 177 OBAN Consortium

37 Environment (Channel model) 50 Mbps 40 Mbps 30 Mbps 20 Mbps 10 Mbps Residential (B) 21 m 26 m 33 m 46 m 63 m Residential/small office (C) 20 m 25 m 32 m 45 m 62 m Typical office (D) 27 m 35 m 43 m 54 m 70 m Large office (E) 33 m 46 m 54 m 74 m 103 m Large space indoor/outdoor (F) 40 m 55 m 67 m 88 m 122 m Outdoor 7 m 10 m 14 m 16 m 24 m Indoor-to-outdoor (percentage of indoor range) Table 2-16 Maximum range as function of average channel data rate using different channel models for IEEE802.11a equipment transmitting in the 5.25 GHz band. Environment (Channel model) 50 Mbps 40 Mbps 30 Mbps 20 Mbps 10 Mbps Residential (B) 30 m 43 m 51 m 65 m 95 m Residential/small office (C) 28 m 42 m 49 m 62 m 92 m Typical office (D) 42 m 55 m 62 m 82 m 122 m Large office (E) 50 m 72 m 85 m 112 m 158 m Large space indoor/outdoor (F) 60 m 85 m 102 m 130 m 183 m Outdoor 10 m 16 m 20 m 25 m 35 m Indoor-to-outdoor (percentage of indoor range) Table 2-17 Maximum range as function of average channel data rate using different channel models for IEEE802.11a equipment transmitting in the 5.6 GHz band. Environment (Channel model) 50 Mbps 40 Mbps 30 Mbps 20 Mbps 10 Mbps Residential (B) 28 m 35 m 43 m 56 m 78 m Residential/small office (C) 25 m 33 m 40 m 53 m 75 m Typical office (D) 34 m 45 m 54 m 70 m 100 m Large office (E) 42 m 58 m 70 m 90 m 130 m Large space indoor/outdoor (F) 50 m 69 m 83 m 108 m 154 m Outdoor 10 m 16 m 20 m 25 m 35 m Indoor-to-outdoor (percentage of indoor range) Table 2-18 Maximum range as function of average channel data rate using different channel models for IEEE802.11g equipment. Environment (Channel model) 10 Mbps 5 Mbps 2 Mbps Residential (B) 55 m 76 m 97 m Residential/small office (C) 50 m 74 m 95 m Typical office (D) 67 m 100 m 128 m Large office (E) 82 m 130 m 167 m Large space indoor/outdoor (F) 100 m 150 m 200 m Outdoor 18 m 34 m 42 m Indoor-to-outdoor (percentage of indoor range) Table 2-19 Maximum range as function of average channel data rate using different channel models for IEEE802.11b equipment. OBAN-WP4-SIN-049g-D Page 31 of 177 OBAN Consortium

38 OBAN-WP4-SIN-049g-D Page 32 of 177 OBAN Consortium

39 3. Coverage analysis of multi-cell networks 3.1. Introduction In this section, we look at how the performance is influenced by interference from uncoordinated OBAN APs, or from non-oban APs. A simplified approach is studied with the motivation of giving a first impression of the impact of interference from neighbouring access points. The major assumption is that we simply look at interference from other sources just like an additive noise contribution. By taking this approach, the capacity analysis will be similar to the mono-cell case, and is not repeated here. The impact is on the coverage from the OBAN AP, and estimations for this are done here. In order to do a complete analysis of the interference situation and impact on a WLAN scenario, only simulations where the MAC performance is simulated will give the correct result. This will be done in activity 4 of OBAN WP4, and is consequently not covered here. This section is structured as follows. In Section 3.2, the distribution of access points is addressed based on available statistics and forecasting. This information is used to set up a case for estimating the resulting impact on the interference level. In Sections 3.3, the statistical channel model is presented together with some general assumptions on the method. Section 3.4 shows, as an introduction, how much a specific level of interference influences the coverage and range from an OBAN AP when a uniform interference level is assumed. In Section 3.5 the actual expected interference levels for a residential scenario is calculated based on the penetration figures from Section 3.2. The assumption and method is described and can be used for other parameter sets covering other scenarios Access point distributions The distribution of access points in an environment is crucial for the coverage and capacity of OBAN services. Compared to e.g. GSM the OBAN concept is based heavily on random growth of broadband accesses and installations of WI-FI zones in the private and enterprise sectors. But to some extent it can also be determined by degree of active promotion done by OBAN operators. Operators may chose different strategies with respect to actively pick OBAN sites and support investments e.g. on WI-FI equipment. Operators can be access providers and choose to utilise the full capacity of a fixed access line, beyond the subscribed capacity. They may also choose to establish own access points to fill in the radio coverage, and operate or have roaming agreements with public hotspots. This section will describe scenarios for the future distribution of OBAN access points, in a time frame of some 3-5 years. For the private sectors the scenarios will be based on assumptions about growth and deployment of broadband accesses and WI-FI, and assumptions about how operators may influence the acceptance and growth of installing OBAN access points. The same kind of reasoning is applied for small enterprises. For the medium and large enterprises and public places it is anticipated that OBAN access points may be deployed to a larger degree as needed. A section on public hotspots is also included OBAN operators choice of policy The OBAN concept may be promoted more or less actively by OBAN operators. Some choices are: Establish OBAN with all interested hosts Follow a pick host strategy to obtain a best possible coverage OBAN-WP4-SIN-049g-D Page 33 of 177 OBAN Consortium

40 Utilise excess fixed line capacity or only rely on the hosts subscribed capacity Install WI-FI access points if this is not in place at chosen host sites For the scenarios the choices may be categorised as active or passive promotion policies Scenarios for access points in residential environments The environment description is: E1- Residential Intra room, room-to-room, indoor to outdoor, single and multi family dwelling Main factors that determine scenarios for access point distributions in this kind of environment are assumed to be installed number and capacities of access points and the proliferation of private WI-FI zones, along with the policy taken by OBAN operators Installed number of broadband access lines in households. In 2004 the installation of broadband access to private homes varied approximately between 10% and 30% across the European countries. The growth rates were about %. In a 3-5 years time frame this indicates a substantial coverage of broadband accesses in Europe. Some argue that broadband access will become as natural as tap water and electricity in a few years time. On the other hand we see in Sweden that the total number of Internet subscriptions seems to level out at about 60-70% of the total number of households. Based on these observations one could argue that for the near future the following assumptions could apply: High coverage 80% Low coverage 50% With the proliferation of networked equipment, e.g. lap tops, STB s, etc., and cost reductions of wireless equipment, it is reason to assume that the great majority of broadband users also will have wireless routers installed. A moderate guess could be 80%. In scenarios with a passive promotion policy by the operators, installed WI-FI will be a limiting factor. If an active promotion policy is assumed, WI-FI will be installed as needed Installed capacity per access line Broadband access will be carried on several access technologies. Today DSL is the dominant technology in Europe, but other technologies like cable modems and shared fibre access to apartment buildings are also growing fast in several countries. New standards enhance the capacity of DSL systems and the capacity of cable modems depends on network layout. Fibre systems can be upgraded almost with no limit. A clear trend is that operators offer more capacity for the same price as competition increases. Figure 3-1 shows a scenario for subscribed capacity per access line that may be typical in the Nordic countries in 5 years time. OBAN-WP4-SIN-049g-D Page 34 of 177 OBAN Consortium

41 120 % 100 % Installed cap>x 80 % 60 % 40 % 20 % 0 % 0, Mbps on established bb. access lines Figure 3-1 Anticipated relative profile of subscribed capacity for broadband access lines. For the OBAN concept it may also be relevant to assume that access providers and OBAN operators utilise excess installed capacity for connecting access points. E.g. an ADSL customer may subscribe for 2 Mbps, while the service is delivered on a modem capable of 8 Mbps. In Figure 3-2 it is made an assumption that all existing DSL customers have modems capable of 8 Mbps, and all new DSL customers in the coming years will get their services on VDSL or ADSL2+ modems. 120 % 100 % 80 % Installed cap>x 60 % 40 % 20 % 0 % 0, Mbps on established bb. access lines Figure 3-2 : Anticipated relative capacity profile of installed broadband access lines. It may be argued that only households with subscribed capacity of more than 4-8 Mbps are potential OBAN hosts, if only subscribed capacity is utilised. On the assumption that OBAN operators will utilise installed capacity, the potential will be practically all installed broadband access subscribers. These assumptions must OBAN-WP4-SIN-049g-D Page 35 of 177 OBAN Consortium

42 however be tested by the traffic simulations that shall be performed in activity 4 of WP4, to verify what is the needed capacity on the fixed line. An optimistic scenario for the number of OBAN hosts could be 50% of the above totals. A moderate scenario could be 10%. Such figures will however depend heavily on strategies taken by OBAN operators, and how the OBAN concept is promoted towards broadband site owners Scenarios for number of OBAN access points By combining alternative options outlined in the sections above, scenarios A-H as shown in Table 3-1 can be constructed. Scenarios: A B C D E F G H 1 Operator policy Act/pass Active Active Active Active Pass. Pass. Pass. Pass. 2 Bb connected households % of hh WI-FI installed i bb housh.*) % of Potential OBAN hosts % of OBAN installed, of potential % of Total WI-FI in households % of hh Private OBAN access points % of hh ,8 Table 3-1 Scenarios for % of households (hh) that may have installed WI-FI and be OBAN hosts, in 3-5 years. *) Effectively 100% when calculating 7 under the 1/active assumption. For 1/passive assumption calculate with given entry Spatial distribution of OBAN access point In the studies several types of dwellings and location in urban and rural environments will be taken into account. On the condition of a passive policy by the operators it is a reasonable assumption that the deployment of access points will be independent of type of dwelling, and independent of whether adjacent dwellings are OBAN sites or not. Considering a case with given placement of houses the number and placement of access points should be anticipated by independent random drawings for each dwelling, based on the figures given in row 7 of Table 3-1. For the case of an active policy one could follow a similar procedure, but adjust the result if needed to improve the coverage. In any case it is reasonable to assume that the OBAN access points will have an optimal placement also for outdoor coverage, and in the active policy case, also may be supplemented by outdoor antennas under certain conditions OBAN access points in small enterprises The environment definition is: E2 - Small Enterprise Enclosed offices, classroom, meeting/conference room, coffee shop Small enterprises in OBAN context will be units with less than 20 employees encompassing production, public and private services and commercial activities. The reason for this distinction is that enterprises of this size may be located in similar environments as dwellings. OBAN-WP4-SIN-049g-D Page 36 of 177 OBAN Consortium

43 Statistics about the number of this size of companies are not readily available. Statistics from Norway, Denmark and UK, indicates that they amount to about 5% of the number of households. It is reason to assume that in 3-5 years they will have a high coverage of broadband access, as well as installed WLAN. Table 3-2 summarises assumptions about access points in small enterprises. The figures are given as relative to number of households. For small enterprises the take-rate for OBAN is assumed to depend on active/passive policy adopted by OBAN operators, with same rates as for residential households. Scenarios: A B 1 Operator policy Act/pass Active Passive 2 Number of small enterprises % rel hh WI-FI installed in small enterprises % of Potential OBAN hosts % of OBAN installed, of potential % of Total WI_FI in small enterprises % rel hh OBAN access points in small enterpr. % rel hh Table 3-2 : Scenarios for % of established OBAN access points in small enterprises in 3-5 years, relative to number of households Assumptions about access points in medium and large enterprises The environment definitions are: E3 - Medium enterprise Enclosed offices, classroom, meeting/conference room, office landscape multi storey office, campus E4 - Large enterprise Large factory floor, hospital, warehouse, library concert hall, movie theatre Medium and large enterprises will normally be located away from dwelling areas. In OBAN context the most interesting situation would be to consider access for customers and visitors being at the enterprise premises. Today that is prohibited in many companies for security reasons. Other companies have solved that problem and allow access to Internet by their WLAN zones. In a few years time it may be assumed that more companies will allow this, and that coverage in such surroundings will be the result of planned initiatives to offer good access services Assumptions about access points in open spaces and vehicles Definitions of these environments are: E5 - Large open space Airports, railway stations, city square, high street E6- Mobile Bus, train, plane These are environments that mainly will be the arena for planned deployment of WLAN zones by OBAN operators or actors they have roaming agreements with. Assumptions about OBAN access points should therefore be as needed Number of public hot spots The significance of public hot spots in an OBAN environment is partly that they may be owned by or have roaming agreements with OBAN network operators. But they may also be sources of interference with OBAN access points. OBAN-WP4-SIN-049g-D Page 37 of 177 OBAN Consortium

44 The number of public hotspots is growing fast. Popular sites for locating hotspots are communication centres, hotels and conference sites, restaurants and coffee shops, and road service stations. One of the better tools for users to locate wi-fi hot spots is Intel s Hotspot Finder, Figure 3-3 shows two examples of spatial locations in urban areas in Figure 3-3 Locations of hotspots in central Newcastle upon Tyne (left) and Paris (right) 2004 (blue and red only distinguishes relation between Hotspot owner and Intel Hotspot Finder) The information in Intel s Hotspot Locator underestimates the real number of hot spots quite a lot, but it still can give some clues about possible future evolutions. Compared to population numbers the number of public hotspots in Europe are substantially less than 1 pr inhabitants, even in city areas. The spatial densities in the two cases shown in Figure 3-3 are 17/sqkm and 34/sqkm respectively. It is reason to believe that sites like hotels, conference centres and libraries will to a large degree have installed hotspots in a few years time. With the assumption of a high growth rate of 200% per year the number of hot spots compared to population will still be far less than 1% pr inhabitant, or 2% per household, but this will give a number of public hotspots comparable to a low scenario for OBAN access points, as proposed above. In central areas like the Paris example, such a growth would lead to a density of one hot spot per meters, which is likely to be a saturation level Multi-cell interference In Chapter 4, the case of single OBAN cells was analysed with respect to capacity and coverage. We will now take the interference from neighbouring OBAN access points into account. First we will develop a simplified model for the reduction in signal-to interference and noise ratio, SINR, or C/I, due to other APs. Further, we map this onto the transmission modes of the most common WLAN standards, IEEE b, and g. In this initial study, the IEEE a standard is not treated, however the case is similar and will be handled in the WP4 activity 4 on simulations Assumptions Here, we will only handle the case of WLANs operating in the 2.45 GHz ISM 1 band, i.e. the IEEE b and g versions. The case of IEEE a (or HIPERLAN/2), operating in the 5 GHz band is similar, with changes occurring in the values for maximum transmitter power and the resulting path loss. The regulatory constraints are given in Table ISM: Industrial, Scientific and Medical OBAN-WP4-SIN-049g-D Page 38 of 177 OBAN Consortium

45 Frequency band Standard Max EIRP [mw], [dbm] IEEE b, IEEE g 100 mw (20 dbm) GHz (ISM) GHz, Band A IEEE a with no transmitter power control (TPC) and with/without dynamic 60 mw (18 dbm) Indoor only frequency selection (DFS) IEEE a + IEEE h, employing TPC without DFS 120 mw (21 dbm) Indoor only IEEE a + IEEE h, employing 200 mw (23 dbm) GHz, Band B both TPC and DFS IEEE a + IEEE h, employing both TPC and DFS Indoor only 1 W (30 dbm) Indoor and outdoor use Table 3-3 European regulations for the WLAN frequency bands regarding transmitter power. Other aspects that are considered are: The channels in use by the interfering APs determine the interference level experienced by the serving OBAN AP and its users. The experienced interference decreases when the interfering AP is using another channel than the serving OBAN AP. The amount of overlap decides the resulting impact. Transmitter power levels of both the serving OBAN AP and the interfering APs are assumed to be the maximum allowable according to the European regulations. The resulting interference level can be modelled in two ways: o An average and uniform interference level is applied over the area. The case is treated briefly in Section 3.4. o An actual case with a number of I-APs gives an interference contour, which is used when calculating the actual impact on the serving OBAN AP and its users. This is treated in Section Statistical channel model Based on the propagation models presented in section 2.1 the following simplified model for the received power level is adopted: Where: 20 log( d) for d d λ PR( d) = EIRP + 20 log( ) + GR X d 4π 20 log( dbr ) 10 n log for d > d dbr EIRP Equivalent Isotropic Radiated Power [dbm] d distance [m] d BR breakpoint distance [m] (free space propagation for d < d BR ) P R (d) received power [dbm] λ wavelength [m] n path loss exponent G R antenna gain receiver antenna [dbi] X shadowing attenuation [db] BR BR OBAN-WP4-SIN-049g-D Page 39 of 177 OBAN Consortium

46 To estimate the statistical properties of the received power level in the scenarios, the following three parameters in the model are modelled as random variables: The breakpoint distance, d BR, is assumed to be log normally distributed with mean, μ BR and standard deviation, σ BR. The path loss exponent minus one, ν = n-1, is assumed to be log-normally distributed with mean, μ ν, and standard deviation, σ ν. The path loss exponent is given the bias one to prevent that values smaller than one should give a positive propagation gain increasing with increasing distance. The shadowing attenuation, X, is assumed to be normally (Gaussian) distributed with mean, μ X and standard deviation, σ X. The normal distribution (Gaussian) is given by: ( ) f T 1 = e σ 2π T 2 1 T μt 2 σt A random variable is log normally distributed if the logarithm of the random variable is normally distributed. For the breakpoint distance it follows that, T = ln(d BR ) in the expression above. In addition the following substitutions for mean and standard deviation are necessary: and 2 1 σ BR μt = ln ( μbr) ln μbr σ T 2 σ BR = ln μbr A similar substitution is necessary for the path loss exponent parameters, ν, μ ν and σ ν. Figure 3-4 shows the lognormal distributions for the breakpoint distance d BR and the path loss exponent n. In this example the mean breakpoint distance is μ BR = 5 m and the standard deviation, σ BR = 2. For the path loss exponent the mean value is μ n = 3.5 m and consequently μ ν = = 2.5 m. The standard deviation is, σ ν = 2, in this example. OBAN-WP4-SIN-049g-D Page 40 of 177 OBAN Consortium

47 Statistical channel parameters d n probability break point distance, d[m] and path loss exponent, n Figure 3-4 Lognormal distributions for the breakpoint distance, dbr, and the path loss exponent Effect of uniform interference level In this section, we apply a simplified approach to analyse the interference impact on the coverage of an OBAN AP. To see this, we assume that the interference level is uniform throughout the area. This makes it simple to analyse, because it has the same effect as raising the noise floor, or decreasing the transmitter output power Receiver input level Based on Table 2-6 and Table 2-6 in Section we can provide a mapping between experienced C/I to the transmission mode. The numbers are repeated for convenience in Table 3-4. IEEE b and g IEEE a and g Transmission mode (MCS) bit rate [Mb/s] Sensitivity (no interference) [dbm] Required BER 10-5 [db] Table 3-4 Sensitivity and Signal-to-noise-and-interference-ratio for the standards IEEE a/b/g. When the interference level rises above a certain level, it dominates over the thermal noise and we can say that the system is interference limited. In the further calculations, we will define interference as the sum of the thermal noise and interference: OBAN-WP4-SIN-049g-D Page 41 of 177 OBAN Consortium

48 I tot = I + N (NB! Linear scale) N is the total noise power in the channel under consideration, N = N 0 B. In this case B = 22 MHz and N 0 = -174 dbm/hz, giving N = dbm. Adding the estimated receiver noise figure of 10 db, giving a total thermal noise in the receiver front end of N = dbm. Figure 3-5 shows how the total noise and interference behaves and shows the transition between the noise limited and interference-limited case Thermal noise Interference Noise+interference -75 Resulting level [db] Added interference level, I [db] Figure 3-5 Total noise and interference level when interference is treated as an additive noise contribution. Thus, the resulting minimum received level in order to maintain the required C/I is given by: C P = + MINRX I tot I min [db] This is shown in Figure 3-6 for the different transmission modes of IEEE b/g for an interference level from 100 dbm (noise limited system) up to 70 dbm (interference limited system). OBAN-WP4-SIN-049g-D Page 42 of 177 OBAN Consortium

49 Rx level [db] Mb/s 2 Mb/s 5,5 Mb/s 11 Mb/s 6 Mb/s 9 Mb/s 12 Mb/s 18 Mb/s 24 Mb/s 36 Mb/s 48 Mb/s 54 Mb/s Minimum Rx level degradation due to interference for IEEE b/g I [db] Figure 3-6 Increase in minimum receiver level for IEEE b and g in order to maintain the required minimum C/I for different interference levels. Correspondingly, the maximum allowable path loss assuming the maximum allowable transmitter power level can be calculated and is shown in Figure Maximum path loss reduction due to interference for IEEE b/g Max allowable path loss [db] Mb/s 2 Mb/s 75 5,5 Mb/s 11 Mb/s 6 Mb/s 70 9 Mb/s 12 Mb/s 18 Mb/s 24 Mb/s Mb/s 48 Mb/s 54 Mb/s I [db] Figure 3-7 Maximum permissible path loss degradation due to interference for IEEE b and g Impact on coverage The theoretical coverage will be smaller when interference level increases, because the maximum permissible attenuation decreases. Using the channel models A F, ref. Section 2.1, the new coverage can be calculated for different interference levels. The so-called fall-back points between the transmission modes will get closer to the AP. OBAN-WP4-SIN-049g-D Page 43 of 177 OBAN Consortium

50 For the assumption of a uniform interference level, the maximum allowable path loss is found directly when the required receiver level is known, as: AT MAX = P TX P MINRX [db] Mapping AT MAX directly to the path loss models from section 2.1, the maximum range or coverage for each mode is given. This is shown in Figure 3-8 for IEEE b and Figure 3-9 for the additional modes of IEEE g Maximum distance reduction due to interference for IEEE b, propagation model A-C 1 Mb/s 2 Mb/s 5,5 Mb/s 11 Mb/s Maximum distance reduction due to interference for IEEE b, propagation model D 1 Mb/s 2 Mb/s 5,5 Mb/s 11 Mb/s Max distance [m] 80 Max distance [m] Added interference at receiver, I [db] Added interference at receiver, I [db] Maximum distance reduction due to interference for IEEE b, propagation model E 1 Mb/s 2 Mb/s 5,5 Mb/s 11 Mb/s Maximum distance reduction due to interference for IEEE b, propagation model F 1 Mb/s 2 Mb/s 5,5 Mb/s 11 Mb/s 200 Max distance [m] Max distance [m] Added interference at receiver, I [db] Added interference at receiver, I [db] Figure 3-8 Estimated maximum range for the transmission modes of IEEE b as a function of additive interference level Maximum distance reduction due to interference for IEEE g, propagation model A-C 6 Mb/s 9 Mb/s 12 Mb/s 18 Mb/s 24 Mb/s 36 Mb/s 48 Mb/s 54 Mb/s Maximum distance reduction due to interference for IEEE g, propagation model D 6 Mb/s 9 Mb/s 12 Mb/s 18 Mb/s 24 Mb/s 36 Mb/s 48 Mb/s 54 Mb/s Max distance [m] 60 Max distance [m] Added interference at receiver, I [db] Added interference at receiver, I [db] OBAN-WP4-SIN-049g-D Page 44 of 177 OBAN Consortium

51 Maximum distance reduction due to interference for IEEE g, propagation model E 6 Mb/s 9 Mb/s 12 Mb/s 18 Mb/s 24 Mb/s 36 Mb/s 48 Mb/s 54 Mb/s Maximum distance reduction due to interference for IEEE g, propagation model F 6 Mb/s 9 Mb/s 12 Mb/s 18 Mb/s 24 Mb/s 36 Mb/s 48 Mb/s 54 Mb/s Max distance [m] Max distance [m] Added interference at receiver, I [db] Added interference at receiver, I [db] Figure 3-9 Estimated maximum range for the transmission modes of IEEE g as a function of additive interference level. Figure 3-10 shows an example of how the covered area decreases when interference increases. Figure 3-10 Coverage reductions due to uniform interference. From left to right: No interference, -85 dbm interference level, -75 dbm interference level. Example for IEEE b. The above examples are for the assumptions that a uniform carpet of interference is present and that propagation is equal in all directions Statistical coverage analysis We go a step further and put a finite number of interference sources modelled as point sources with a certain distance to the serving AP. In this case, the interference level is not uniform, but is dependent on the distance from each of the sources to the point of study, and the fallback points will occur closer to each other Residential area example Based on the figures in Section 3.2, we can set up an example, corresponding to an environment defined in milestone D8.1. For the analysis example a residential environment is chosen together with the parameters for scenario E in Table 3-1. Environmental definitions: The household density is calculated for a typical residential area in the Nordic countries consisting of villas. The average building site is chosen to be 600 m 2, which corresponds to approx households per km 2. OBAN-WP4-SIN-049g-D Page 45 of 177 OBAN Consortium

52 The calculation was done over a square area of 100 m x 100 m ( m 2 ) with a single OBAN AP placed in the centre of the area. It is assumed that interfering APs longer away than 50 m has minor impact on the performance. The WLAN (OBAN and non-oban) AP penetration is according to scenario E from the table in section : o 56 % of households have WiFi installed o 22 % of households have OBAN o This results in 34 % of households have non-oban WiFi APs The average number of non-oban (interfering) APs (I-AP) is given from the penetration figures above. With 34 % of households equipped with non-oban WiFi APs (I-APs) and a household density of per km 2, there is an average of 5.7 I-APs within a m 2 square The number of I-APs are follows the Gaussian distribution with mean, μ niap = 5.7 and σ niap =3 Propagation characteristics channel model The propagation from the OBAN AP follows the channel model A-C with fixed values. The breakpoint distance d BR = 5 m and the path loss exponent is 3.5. The propagation from the I-APs is randomly selected according to the statistical model described in Section The break point distance, d BR, is log normally distributed with mean μ BR = 5 m and standard deviation σ BR = 2. The path loss exponent (d > d BR ) is also log normally distributed with μ n = 3.5 (μ v = 2.5) and σ v = 2. Frequency channel selections and power The OBAN AP always use channel 7 (2.442 GHz) The channels of the I-APs are randomly chosen between 1 and 13. All channels are assumed to be equally likely 2. This affects the resulting interference level. The interference impact of each I-AP is filtered according to the frequency separation and degree of overlap. We have assumed that the transmitted spectrum, as well as the receiver bandwidth, follows a sin(x)/x function multiplied by a spectral mask to ensure reduced side lobe power. π sin ( f fc ) 11 MHz PTX ( f ) = Txmask( f ) π ( f fc ) 11 MHz Where the spectral mask is defined as: 0 dbr, f fc < 11 MHz Txmask( f ) = 30 dbr, 11 MHz f fc < 22 MHz 50 dbr, f fc 22 MHz All APs are IEEE b with maximum transmitter power of 20 dbm Calculations have been done for the example described above. A number of 300 calculations have been done with the randomness described above. The coverage areas for each of the PHY modes for IEEE b has been calculated each time and processed to find the mean, median, 10%- and 90%-percentiles and the standard deviation. Only interference from neighbouring APs is considered, however STAs associated to these have the same effect. 2 Off-the-shelf APs and wireless routers are usually set up with default channels from the factory. Currently, most IEEE b and g APs are set up from the US-based non-overlapping set, channel 1, 6 or 11, with a majority on channel 6. OBAN-WP4-SIN-049g-D Page 46 of 177 OBAN Consortium

53 Results, residential area Table 3-5 gives the results for the residential example above with 100 calculations. It shows that the average reduction in the covered area is around 90 % for all transmission modes. The equivalent range reduction is shown in Table 3-6 and shows a degradation of around -70 % for all modes. Mode Maximum coverage, propagation model A-C [m 2 ] 10 percentile coverage [m 2 ] Median coverage [m 2 ] Standard deviation [m 2 ] Mean coverage [m 2 ] 90 % percentile coverage [m 2 ] Average coverage reductions [%] 1 Mb/s Mb/s Mb/s Mb/s Table 3-5 Estimated coverage reductions for IEEE b modes in a residential area. Average number of non-oban interferers is 6 in an area of 100 m x 100 m. Number of iterations is 100. As the example diagram shows, the coverage area is no longer circular, but takes an irregular shape. However, based on numbers for the coverage area, an equivalent radius can be computed. Mode 1 Mb/s 2 Mb/s 5.5 Mb/s 11 Mb/s Maximum distance, no interference [m] Maximum distance, with interference [m] Range reduction [%] Table 3-6 Maximum distance with and without interference for the residential scenario case. IEEE b modes, channel model A-C. A sample from the above calculation is shown in Figure 3-11 and Figure 3-12 with 9 I-APs. The frequencies are shown in the C/I plot. The sample also shows that the effect of the interference is dependent on the frequency distance between the OBAN AP and the I-APs. 100 Interference level [db] C/I [db] [m] [m] [m] [m] Figure 3-11 Interference contour and C/I contour for an example of 9 neighbouring I-APs. The frequencies of the I-APs are shown in the C/I-contour and are 2.442, 2.447, (2x), (4x) and GHz. The OBAN AP is on channel 7, GHz OBAN-WP4-SIN-049g-D Page 47 of 177 OBAN Consortium

54 100 C/I contours [m] [m] Figure 3-12 Resulting coverage for the OBAN access point in the sample above with 9 neighbouring I- APs. It shows clearly that the coverage area decreases, in this case to a radius of approx. 12 m for the 11 Mb/s PHY mode, which is 85 % reduction compared to the non-interference case Conclusions for multi-cell analysis In this chapter, we have addressed the performance in multi-cell networks. We have suggested models for the distribution and density of the access points in order to model realistic scenarios. Further we have performed simplified interference analysis based on the assumption that interference can be modelled as an additive noise contribution. An example has been calculated for a typical Nordic residential area, and this shows that a coverage reduction of 90 % can be expected given certain assumptions about WiFi penetration and propagation properties. This method does not take WLANs MAC-layer into account, which will be studied in the simulation work in activity 4. OBAN-WP4-SIN-049g-D Page 48 of 177 OBAN Consortium

55 4. Interference and coexistence issues 4.1. Introduction In this section the most important sources of interference are described, with particular attention to the interference effects due to Bluetooth devices (Section 4.3) and to microwave ovens (Section 4.4). As far as Bluetooth interference is concerned, the main methods for performance evaluation in an interfered environment are summarized, and recommended practice on coexistence mechanisms operating at physical layer are recalled. As far as interference due to microwave ovens is concerned, some examples of measured spectra of interfering signals are presented, and analytical and simulative interference models are described Characteristics of interference sources Interference can be classified as intra-system or inter-system interference depending on licensed or unlicensed frequency band exploited by the system. Intra-system (intra-cell and inter-cell) interference is one of the most important impairments for wireless communication. It depends on the particular wireless technology and will be specified at later stages of the OBAN project. According to the OBAN WP1 Scenarios and Requirements [10] inter-system interference is of special interest because of the importance of the unlicensed spectrum for the OBAN project. As an example, possible co-existing technologies in 2.4 GHz and 5.0 GHz frequency bands may include [11]: IEEE a/b/g/n and proprietary extensions IEEE (Bluetooth) IEEE IEEE a IEEE IEEE a 2.4 GHz Cordless Phone 5.0 GHz Cordless Phone 2.4 GHz Video Transmitter 5.0 GHz non-ofdm Video Transmitter 2.4 GHz Microwave Oven It is worth emphasizing the special importance of the coexistence problem for an OBAN system. In the conventional WLAN environment without Visiting User Terminals, all the wireless equipment such as Indoor AP, Host User Terminal and Interferer belong to the host user. Thus, coexistence is a local problem of the particular host user in this example. That means that even if the WLAN throughput degrades significantly because of the Interferer, only the Host User suffers from his own interference. In the OBAN environment with Visiting User Terminals, the coexistence problem becomes a common OBAN problem. Indeed, in this case the Indoor AP becomes available for visiting OBAN users and if the throughput drops below some level, the OBAN coverage may be violated. Taking into account that the underlying idea for an OBAN system is that a host user must be unaffected when he becomes an OBAN subscriber, e.g. a host user cannot be restricted in terms of using coexisting technologies from the list presented above, it becomes clear that OBAN APs may require a special protection from coexisting wireless technologies compared to the conventional wireless systems. The main feature of the inter-system interference is that there may be no co-operation between co-existing systems. At the MAC layer that means that interference avoidance protocols may fail and cause uncontrollable throughput degradation. At the PHY layer that means that the interference may be completely asynchronous (sparse) affecting only some parts of the frequency band and/or time interval. This may significantly decrease the efficiency of interference mitigation schemes because of the poor overlap between the interference and the training data of the signal of interest [12]. OBAN-WP4-SIN-049g-D Page 49 of 177 OBAN Consortium

56 Different types of interference sources are included in the list above. Some of them are extensively studied and well understood as interference sources for some of the current WLAN technologies (mainly Wi-Fi). Bluetooth devices and microwave ovens in the Industrial-Scientific-Medical (ISM) band (2.4 GHz) are among these. Other interference sources especially in the 5.0 GHz band have not attracted significant attention yet. A few examples of this kind of interference sources are video transmission devices, plasma lighting systems [13], ultra wide band (UWB) devices [14] and radar systems [15]. Some details on the main interference sources are described in [16-18], while the simulation models of microwave ovens, and in general of non-oban interference sources, are described in section 6. Due to the high cost of licensed spectrum, the unlicensed frequency band offers a free and attractive alternative. Currently the most popular unlicensed frequency band is the 2.4 GHz ISM band that is available in many parts of the world. For most countries, the 2.4 GHz ISM band ranges fro 2.4 to GHz. As a consequence of popularity, this band is witnessing a rapid growth in both the number of systems as well as the diverse technologies being used within the band. This latitude comes at a price in the form of susceptibility to excessive mutual interference. Because the 2.4 GHz ISM band is available for unlicensed use and all systems operate independently, mutual interference is inevitable. Due to the differences between these communication technologies, the mutual interference has different characteristics as well. Sources of such interference include unintentional radiators such as microwave ovens and in general non communications devices, or intentional radiators like cordless phones and Bluetooth devices Interference between WLAN and WPAN (Bluetooth) Bluetooth [31] is a widely spread technology. The Bluetooth specification allows for three levels of radio output power: Class I 100 mw (20 dbm) Class II 2.5 mw (4 dbm) Class III 1 mw (0 dbm) Class III devices (computers, printers, PDAs, mobile phones) are designed to operate over a range of approximately 10 m, while Class I devises (access points for enabling wide coverage networks) are designed to operate with a range of approximately 100 m. A fast frequency hopping scheme is used at the PHY layer with a hop rate of 1600 hops/s. Hopping frequencies range over 79 frequency channels in the ISM band, each of the channels being 1 MHz wide. A TDD technique is used to transmit and received data. Each data packet transmitted in a slot occupies 366 μs. Slots are centrally allocated by the master device. One can see that Bluetooth technology generates interference bursts, which may randomly overlap with WLAN data slots in the frequency and time domains creating asynchronous (sparse) interference similar to that introduced in [12]. The collision probability depends on the WLAN and Bluetooth packet duration and also on the frequencies occupied by both systems. This section describes the problem of interference of wireless personal area networks (WPAN) and wireless local area networks (WLAN), with particular attention at single antenna systems and at physical and MAC layer behaviour. Notice that we will not analyze the oldest physical layers of IEEE , namely FH (Frequency hopping) and DSSS (Direct Sequence Spread Spectrum), since they are of limited interest today Performance evaluation of IEEE802.11b-IEEE interference Both the IEEE coexistence Task Group and the Bluetooth Special Interest Group (SIG) are devoted to the development of techniques to alleviate the impact of interference. The attention is focused on the coexistence of two standards that work in the ISM frequency band: Bluetooth (IEEE ) for WPANs and the IEEE family for WLANs. In this section the analytic/simulation framework for the evaluation of OBAN-WP4-SIN-049g-D Page 50 of 177 OBAN Consortium

57 SIR and BER performance at PHY layer in presence of WLAN/WPAN interference suggested in [24] is described. There are many factors that affect the level of interference, namely, the separation between the WLAN and the WPAN devices, the amount of data traffic flowing over each of the two wireless networks, the power levels of the various devices, and the data rate of the WLAN. Also different types of information being sent over the wireless networks have different levels of sensitivity to the interference. For example, a voice link may be more sensitive to interference than a data link being used to transfer a data file Analytic BER and SIR evaluation Figure 4-1 Model used for BER calculation The basic model for analytical performance evaluation in an interfered environment is shown in Figure 4-1. Supposing that the device positions and transmission parameters are known, the model analytically evaluates the BER performance from those parameters during the periods of stationarity, which are the periods of time over which the transmission parameters do not change. The empirical indoor path loss model shown in Table 4-1 is considered and fed to the scheme of Figure 4-1. The model does not apply below about 0.5 meter due to near field and implementation effects. Equation Path Loss = log 10 (d) Path Loss = log 10 (d/8) Condition 0.5 m d 8 m d > 8 m Table 4-1 Equations for path loss (db) at 2.4 GHz versus distance (m). The power levels are derived from the transmission model and the system geometry, using the so called spectrum factors, which represent the combined effect of transmitter and receiver masks [24], and frequency offset (the difference between the interference and the information signal centre frequency). Example of worst case spectrum factor values are shown as an example in Table 4-2. OBAN-WP4-SIN-049g-D Page 51 of 177 OBAN Consortium

58 Frequency offset (MHz) Spectrum Factor (db) IEEE to IEEE b IEEE b to IEEE Table 4-2 Spectrum factor values for IEEE and IEEE b. The SIR (Signal to Interference Ratio) is given by the ratio of the received signal power to the total received interferer power. The powers are calculated after the spectrum factor has been applied, and so this ratio corresponds to the value after the receiver filter. Receiver noise is not considered in this model. The BER calculation is based on SIR. It is assumed that above a certain SIR the BER is effectively zero and below a certain SIR the BER is effectively 0.5., according to the limits shown in Table 4-3. Receiver Upper limit on SIR Lower limit on SIR b 10 db -3 db db 1 db Table 4-3 Assumed limits on SIR. Figure 4-2 shows the results of analytical calculating BER for SIR values in the range 2 to 10 db for each IEEE b modulation type. OBAN-WP4-SIN-049g-D Page 52 of 177 OBAN Consortium

59 Figure 4-2 BER versus SIR for IEEE b modulation types. While the analytical model uses transmitter power and distance as input parameters, a simulation model which uses the signal-to-noise ratio (SNR) and the SIR can also be implemented. In both cases, the output is BER. In this section simulation models for IEEE and IEEE b, and their interference is described [24]. The results of an examination of the BER performance in interference-limited environments are presented. The performance is determined using Monte Carlo simulation methods Monte Carlo simulation of BER and SIR performance In [24] also the simulated performance of IEEE b interfered by IEEE are reported. The various standards are simulated through their transmit and receive schemes. The BER performances of the 1 Mbps IEEE b system, in an interference-limited environment with SNR=35 db, obtained by Monte Carlo simulations, are shown in Figure 4-3. Both SNR and SIR are measured at the input to the receiver band pass filter (BPF). The most disturbing interference is located at f d =1 MHz, which needs a minimum SIR of -5 db. For frequency offset greater than 8 MHz, the SIR value has to be very low in order to get a high BER. This fact is due to the BPF in the IEEE b receiver having high attenuation at frequencies near 11 MHz. Figure 4-4 shows the performance of the 5.5 Mbps IEEE b receiver with IEEE interference in AWGN channel. Unlike the case of 1 Mbps IEEE b system, co-channel interference from IEEE (f d =0 MHz) significantly degrades the performance of the 5.5 Mbps IEEE b receiver. For frequency offset greater than 2 MHz, an SIR of 1 db is required to achieve the BER of In this case, the 5.5 Mbps IEEE b receiver achieves almost 3 db improvement over the 11 Mbps IEEE b system. OBAN-WP4-SIN-049g-D Page 53 of 177 OBAN Consortium

60 Figure Mbps IEEE b BER performance as a function of the signal over interference ratio (SIR) with IEEE interference with frequency offset f d. AWGN channel, SNR=35 db. Figure Mbps IEEE b BER performance as a function of the signal over interference ratio (SIR) with IEEE interference with frequency offser f d. AWGN channel, SNR=35 db. Figure 4-5 illustrates the performance of the 11 Mbps IEEE b receiver with IEEE interference. This figure indicates that the CCK modulation is more vulnerable to the interference signal than the 1 Mbps DSSS. A minimum SIR of 3 db has to be achieved to get BER = 10-2 for all frequency offsets. This result is no surprising, because the CCK provides a higher bit rate but occupies the same 22 MHz bandwidth, thereby having a smaller coding gain. OBAN-WP4-SIN-049g-D Page 54 of 177 OBAN Consortium

61 Figure Mbps IEEE b BER performance as a function of the signal over interference ratio (SIR) with IEEE interference with frequency offser f d. AWGN channel, SNR=35 db Recommended practice on coexistence mechanisms operating at physical layer This section summarizes some coexistence mechanisms that can be used to facilitate coexistence of WPANs and WLANs. Both the IEEE coexistence Task Group and the Bluetooth Special Interest Group (SIG) are devoted to the development of techniques to alleviate the impact of interference [24]. Considering, as we are in this section, single antenna systems and focusing on physical layer solutions, two main categories of coexistence mechanisms must be looked at: collaborative and non collaborative. Collaborative coexistence mechanisms exchange information between two wireless networks, that is a collaborative coexistence mechanisms requires communication between the WLAN and the WPAN. Non collaborative mechanisms do not exchange information between two wireless networks. These mechanisms are only applicable after a WLAN or a WPAN are established and user data is to be sent. Both types of coexistence mechanisms are designed to mitigate interference. Collaborative coexistence mechanisms are intended to be used when at least one WLAN station and WPAN device are collocated within the same physical unit. When collocated, there needs to be a communication link between the WLAN and WPAN devices within the physical unit, which could be a wired connection between these devices or an integrated solution. Non collaborative coexistence mechanisms are intended to be used when there is no communication link between the WLAN and WPAN. Collaborative coexistence mechanisms The main collaborative mechanisms between IEEE b and IEEE proposed in the literature involve coordinated scheduling of packet transmission at MAC sub-layer or programmable notch filtering of narrowband interferers at PHY layer. These collaborative mechanisms may be used separately or combined with others, and are here briefly summarized. A deeper discussion can be found in [24]. The Alternating Wireless Medium Access (AWMA), which can be used only if the BT master is dual mode (WLAN/BT) and connected with the WLAN AP, utilizes a portion of the wireless IEEE beacon interval OBAN-WP4-SIN-049g-D Page 55 of 177 OBAN Consortium

62 for wireless operations. From a timing perspective, the medium assignment alternates between usage following IEEE procedures and usage following IEEE procedures. Each wireless network restricts their transmission to the appropriate time segment, which prevents interference between the two wireless networks. In AWMA, a WLAN radio and a WPAN radio are collocated in the same physical unit. This allows for a wired connection between the WLAN radio and the WPAN radio. This wired communication link is used by the collaborative coexistence mechanism to coordinate access to the wireless medium, between the WLAN and WPAN. The AWMA mechanism uses the shared clock within all the WLAN-enabled devices and thus all WLAN devices connected to the same WLAN AP share common WLAN and WPAN time intervals. Therefore, all devices connected to the same AP restrict their WLAN traffic and WPAN traffic to non overlapping time intervals. As such, there will be no WLAN/WPAN interference for any devices connected to the same WLAN AP. In the case of multiple APs, typically the APs are not synchronized. In that case there will be some residual interference between WPAN devices synchronized with one WLAN AP and WLAN devices synchronized with another AP. If the WLAN APs are synchronized then this residual interference can also be eliminated. The IEEE WLAN AP sends out a beacon with a periodic interval of duration T B. AWMA subdivides this interval into two subintervals: one for WLAN traffic and one for WPAN traffic. Figure 4-6 illustrates the separation of the WLAN beacon interval into two subintervals. Figure 4-6 Timing of the WLAN and WPAN subintervals. It is recommended to use AWMA whenever there is a high density of devices with collocated WLAN/WPAN radios. The AWMA mechanism is also to be used when the WLAN and/or WPAN network bandwidth allocation needs to be deterministically controlled and not dependent on the traffic load of either WLAN or WPAN. The Packet Traffic Arbitration (PTA) control entity provides per packet authorization of all transmissions. In the PTA mechanism, the IEEE b station and IEEE node are collocated. Each attempt to transmit by either the b or is submitted to PTA for approval. PTA may deny a transmit request that would result in collision. The PTA mechanism coordinates sharing of the medium dynamically, based on the traffic load of the two wireless networks. PTA uses its knowledge of the duration of IEEE b activity and future IEEE activity of a number of slots into the future to predict collisions. When a collision would occur, PTA prioritizes transmissions based on simple rules that depend on the priorities of the various packets. It is recommended to use PTA whenever there is a high variability in the WLAN and WPAN traffic load or whenever an IEEE SCO link needs to be supported. The PTA mechanism uses a dynamic packet scheduling mechanism that automatically adapts to changes in traffic loads over the WLAN and WPAN networks. The PTA mechanism supports IEEE SCO links while the AWMA mechanism does not. The Deterministic Interference Suppression is designed to mitigate the effect of IEEE interference on IEEE b. The basic idea of the interference suppression mechanism is to put a null in the IEEE b receiver at the frequency of the IEEE signal. However, because IEEE is hopping to a new frequency for each packet transmission, the IEEE b receiver needs to know the FH pattern, as well as timing, of the IEEE transmitter. This knowledge is obtained by employing an IEEE receiver as a part of the IEEE b receiver. On account of this being primarily a physical layer solution, it may be OBAN-WP4-SIN-049g-D Page 56 of 177 OBAN Consortium

63 integrated with the PTA MAC sub-layer solution. This procedure is applicable to all basic rates in IEEE b (1, 2, 5.5, and 11 Mbps). Recommendations on the utilization of collaborative mechanisms It is recommended that when it is possible, or necessary, to collocate a WLAN and WPAN device within the same physical unit, that either the AWMA collaborative coexistence mechanism or the PTA mechanism be used. If the PTA mechanism is used it is also recommended that the deterministic interference suppression mechanism be used in concert with the PTA mechanism. While PTA can be used without deterministic interference suppression, the combination of the two mechanisms leads to increased WLAN/WPAN coexistence. If there is a high density of physical units incorporating both WLAN and WPAN device in a common area (greater than or equal to three units in a circle of radius 10 meters) and WPAN SCO link (voice link) is not being utilized, then it is recommended that the AWMA mechanism be used. If the density of units is low (less than three units in a circle with a radius of 10 meters), or the WPAN SCO link is used, then it is recommended that the PTA mechanism be used in concert with deterministic interference suppression mechanism. Non Collaborative coexistence mechanisms The non collaborative coexistence mechanisms proposed by the IEEE Task Group [24] and involving IEEE b WLANs are based on suppressing unwanted interference or adapting the transmission according to channel conditions. The foundation for the effectiveness of these types of methods is to be able to figure out the current channel conditions accurately and in a timely manner. Channel estimation may be done in a variety of ways: received signal strength indication (RSSI), header error check (HEC) decoding profile, bit error rate (BER) and PER profile, and an intelligent combination of all the above. The adaptive interference suppression method, which may be of interest for the OBAN scenario, is the so called Adaptive Interference Suppression. This technique is based solely on signal processing in the physical layer of the WLAN and is recommended in order to cancel the IEEE interference. In this method the WLAN has no a priori knowledge of the timing and frequency used by the system, and it uses an adaptive filter to estimate and cancel the interfering signal. A more in-depth analysis of the method can be found in [24]. Recommendations on the utilization of non collaborative coexistence mechanisms There are possible range limitations under which a non collaborative mechanism may not be sufficient. For example, when an IEEE b system and an IEEE system (Class 3) are operated 30 centimetre apart, the IEEE signal will be considerably above the detection threshold of the WLAN system, even when out of band; thus, non collaborative schemes relying on channel estimation and interference detection will be unable to prevent interference in these short range situations. It is in general recommended that adaptive interference suppression be used also in presence of other (scheduling) mechanisms, because it operates at the physical layer; for this reason it can also be used by itself. It is recommended that the adaptive interference suppression filter be used when there is sufficient IEEE interference to noticeably degrade performance and delaying the IEEE traffic is not sufficient. Specifically, delay sensitive traffic such as streaming media will benefit from the use of this mechanism Bluetooth IEEE g interference For a BT transmission to disrupt the IEEE g packets, there must be an overlap in frequency as well as time, as illustrated in Figure 4-7. OBAN-WP4-SIN-049g-D Page 57 of 177 OBAN Consortium

64 Figure 4-7 Interference of BT signal on g signal (a) in frequency (b) in time The probability that both the BT and g signals overlap in frequency is roughly 20/79 25%. In the time domain, long g long packets stand a higher chance of overlapping a number of BT time slots than short packets. The WLAN receiver sees the BT signal as a narrowband interference mainly affecting a small number of sub-carriers. Based on this behaviour, many authors have suggested mitigation techniques for the interference generated by BT on OFDM based IEEE g WLANs [37-42]. We will summarize here the main results Channel estimation in presence of BT interference on IEEE g In this section a possible PHY layer solution to combat the effects of Bluetooth interference on the IEEE g standard [37] is described. This solution consists in a PHY layer algorithm for simultaneously estimating the multipath channel and interference characteristics and using these estimates in the convolutional decoder to improve the system performance in the presence of Bluetooth interference. This algorithm allows the packet error rate (PER) of the g WLAN system to be maintained even if the WLAN packet collides with a Bluetooth packet. Such an algorithm will allow the WLAN system to be more robust in the presence of Bluetooth interference. The modulation format of IEEE g is OFDM based on 64 point FFT. The data bits are first scrambled, then convolutionally encoded, interleaved and mapped onto a signal constellation before being OFDM modulated. The convolutional code used is the industry-standard 64 state rate 1/2 convolutional encoder. Punctured patterns are defined in the standard to give rate 2/3 and 3/4 codes from basic code. Assuming perfect channel estimates for the moment, the received signal at frequency bin k can be written as: r = H a + n, k=1,,n k k k k where H k is the channel value, a k the transmitted symbol, n k the received noise with variance σ k 2 and N the FFT size, which is 64 in the g system. a k can be a symbol from either a BPSK, QPSK, 16QAM or 64QAM constellation representing 1, 2, 4 or 6 bits respectively. Let these bits be denoted as b 0, b 1, b 2, b 3, b 4 and b 5. The Viterbi decoder requires a soft metric for p each of the bits in the symbol a k, denoted by m k (b i ). Let us define the subset of constellation point Ci as the set of symbols from the defined constellation such that b i =p where p is either 0 or 1. The first step is to find two symbols a 0, i and a 1, i for each bit b i is: a a 2 0, i = / k i 2 arg min 0 [ ] rk H a c k ak σ k 2 1, i = / k i 2 arg min 1 [ ] rk H a c k ak σ k (6.5) (6.6) Then m k (b i ) can be defined as: OBAN-WP4-SIN-049g-D Page 58 of 177 OBAN Consortium

65 m k k k 0, i ( bi ) 2 2 k 2 r H a rk H k a1, i = (6.7) σ This soft metrics are then deinterleaved and used in a soft decision Viterbi decoder to decode the transmitted bits. The g has 2 OFDM frames in the preamble that are used for channel estimation. Assuming that the sequence a k is known, the conventional channel estimate per frequency bin is: ) H = r / a (3.5) k k k The above relation treats the channel response at each frequency bin to be independent of the response at other frequency bins. However, assuming that the multipath channel has an impulse response that is limited to the cyclic prefix length, N c, it is clear that the frequency response parameters have only N c independent variables and the remaining N-N c parameters are dependent. Moreover, by estimating the channel from equation above, no information about the noise variance can be obtained. Let the sample spaced channel impulse response be denoted by h i, i=0,,n c -1, then the received vector can be expressed in matrix form as r = AF h + n where A is a N N diagonal matrix composed of the known transmitted symbols a k and F is a N N c truncated Fourier matrix. Defining R n to be the correlation matrix of the noise vector n, and denoting G=AF, we can then write the least-squares estimate of the channel frequency response vector as: Hˆ LS = F H 1 1 H 1 ( G R G) G R r n n In the situation with Bluetooth interference, the noise correlation matrix R n cannot be assumed to be white. However, it is reasonable to assume that R n is a diagonal matrix with entries as follows: R n 2 σ w M 0 = M 0 0 L M L M L L 2 b 0 M 2 σ + σ M L L w 2 σ + σ 2 b L M L M L w 0 M 0 M 0 2 σ w where σ w 2 is the noise variance of the AWGN and σ b 2 is the noise variance of the Bluetooth interference. Since the Bluetooth is narrowband compared to g, only about 3 consecutive carriers will have a higher noise variance than the rest. Once the frequency response has been estimated, the noise variance at each frequency can be estimated as follows: let e be the error vector defined as e = r ˆ. Then, AH LS ˆ σ 2 k = e k 2 OBAN-WP4-SIN-049g-D Page 59 of 177 OBAN Consortium

66 This noise variance estimate is then averaged over the two training frames for each frequency bin k and the i-th OFDM frame, according to the following procedure: 1. Estimate Ĥ LS 2 ˆ k,0 from equation (3.6), andσ from equation (3.7) on the 2 training frames. 2. During the ith OFDM data frame, use the channel estimate Ĥ LS from Step 1 to estimate the transmitted symbol at frequency k and time i as follows: a~ k i rk i / Hˆ, =, k ˆ = a~ sliced to nearest constellation point a k, i k, i 3. Estimate the noise variance at frequency bin k for the ith OFDM frame as : ˆ σ 2 ˆ k, i = rk, i H k ak, i ˆ 2 Average the variance estimates over N f OFDM data frames: N f ˆ σ = ˆ σ k=1,,n k N f 1 i= 0 k, i The channel and noise variance estimates obtained above can now be used to determine the soft metrics for use in the Viterbi decoder. The simulated performance is shown in Figure 4-8 for the case when Bluetooth interference at a certain frequency corrupts the entire g packet. SIR is defined only over the frequency bins that are affected by interference, whereas SNR is defined for the full bandwidth of the g signal. The simulation results have been obtained with an SIR of 3 db using the traditional channel estimator and no noise variance estimation. Even if the receiver has perfect channel state information (CSI), i.e. knows the multipath channel perfectly but not the interference, the performance is only marginally improved. The difference in performance between multipath channel alone and multipath with Bluetooth interference is large. Figure 4-9 shows the performance of the proposed algorithm for various values of N f. There is not much performance difference between estimating the noise over 10 frames versus 60 frames. In both cases the performance at a PER of 0.01 is about 4 db worse than the ideal receiver that knows the channel and noise variances perfectly. With noise variance estimation (3.6) over only 2 OFDM training frames, i.e. N f =0, the performance is worse by about 7 db. Comparing Figures 3.8 and 3.9 we see that even estimating the noise variance over the 2 training frames alone gives significant performance improvement. The gap in performance of about 4 db between perfect channel (multipath + interference) information and estimated values is primarily due to the errors in step 2 incurred by the symbol slicer since this slicing to the nearest constellation point is being done before the convolutional decoding. OBAN-WP4-SIN-049g-D Page 60 of 177 OBAN Consortium

67 Figure 4-8 Performance of conventional channel estimator (3.5) with multipath (MP) and Bluetooth (BT) interference for SIR=30 db Figure 4-9 Performance of new channel and noise estimators (3.6 and 3.8) with multipath (MP) and Bluetooth (BT) interference for SIR=3 db Erasures management in BT-interfered IEEE g In this section is illustrated a study conducted by K.K. Wong and T. O Farrell [38]. They evaluate the impact of the BT interference on an g OFDM PHY and study the insertion of erasures as a method to mitigate the effects of such interference. As seen before, the WLAN receiver sees the BT signal as a narrowband interference mainly affecting a small number of sub-carriers. The SIR level for each independent sub-carrier is determined instantaneously by the power of both the OFDM symbol and the BT signal transmitted over the bandwidth corresponding to that subcarrier. Those data symbols corresponding to sub-carriers with low SIR will be replaced by erasures in input to the IEEE g Viterbi decoder. This avoids a large bias to the path metrics in the Viterbi algorithm OBAN-WP4-SIN-049g-D Page 61 of 177 OBAN Consortium

68 introduced by the corrupted data symbols. Besides that, the use of bit interleaving across the OFDM symbol helps to reduce the burst of errors generated by several adjacent corrupted data symbols. Since the power of the BT signal mostly concentrates around its centre frequency, the erasures would be inserted over those OFDM sub-carriers closest to the BT centre frequency. To simplify the analysis, it is assumed that the receiver has knowledge of the BT interferer centre frequency. This could be learnt through channel measurements. The proximity of the BT device to the WLAN receiver has a tremendous effect on the degree of interference. Even though the WLAN has a higher transmit power (20 dbm) than the BT (0 dbm), the WLAN signal power drops considerably due to the path loss during transmission to the receiver at a sufficient distance from the access point, and is likely to be drowned out by the BT signal. For a WLAN receiver at 10 m from the access point with interference from a 1 m apart BT device, the SIR is 0 db. Figure 4-10 shows the simulation results for an g WLANs transmitting 100 byte packets at 24, 36, 48 and 54 Mbps with the number of erasures as a parameter. Without erasures, the BT interference completely corrupts the WLAN signal at an SIR=0 db, giving error floors at above 10% PER. To achieve the target PER, the SIR has to be raised to 15 db for 24 Mbps and above 20 db for higher data rates. This translates to limiting the WLANs coverage in order to maintain high SIR levels. The insertion of erasures significantly improves performance. For 24 Mbps, the SIR required is reduced from 15 db to 7 db when 9 erasures are used; for 36, 48 and 54 Mbps, the SIR needed is lowered by as much as 20 db when up to 5 erasures are used. For the same number of erasures, the SIR required for 1% PER rises as the data rate increases. Figure 4-10 Plots of PER vs. E b /N 0 showing SIR required by g transmitting 100 byte packets in multipath channel with and without erasures to achieve target 1% PER at (a) 24 Mbps (b) 36 Mbps (c) 48 Mbps and (d) 54 Mbps. OBAN-WP4-SIN-049g-D Page 62 of 177 OBAN Consortium

69 It has to be noted that too many erasures can cause negative effects when punctured convolutional codes and higher modulation orders (i.e. more compact signal constellation) are used. This effect is evident at 36, 48 and 54 Mbps modes where the convolutional code has been punctured to rates ¾, 2/3 and ¾, respectively, and the modulations are 16-QAM, 64-QAM and 64-QAM respectively. For these three data rates, a maximum of 5 erasures gives sufficient amelioration as 7 erasures give similar performance, while 9 erasures result in a worse performance although with higher SIR levels. Since punctures are equivalent to erasure insertion, such a large number of erasures tend to leave paths unresolved within the Viterbi algorithm. A similar mitigation method, based on the insertion of erasures and on decoder metrics modification, is described in [39], while an assessment of the IEEE g performance in a BT-interfered environment is presented in [40,41] Microwave ovens Microwave ovens are possible interference sources. A microwave oven is composed of a magnetron (typically one and two of them in the residential and commercial cases correspondingly) powered by the AC mains, and producing radiation. Emissions have a 50% duty cycle based on the mains period (50 Hz in Europe). Each transmission starts by a transition phase (about 1 ms duration), where the emissions swipe across the band, before the magnetron settles on its main frequency phase (about 8 ms duration out of the 20 ms mains power cycle at 50 Hz). A similar pulsed transition phase appears in the end of a burst. The centre burst frequency is mostly around MHz. Commercial microwaves generate more complicated interference compared to the residential ones and cover a wider frequency band with larger power variations. This interference may have a significant impact on WLAN systems especially in the case of a number closely spaced commercial microwave ovens working on different mains phases. Possible simulation models able to describe MicroWave Oven (MWO) interference are described in section 6.3. The most common unintentional radiator in the 2.4 GHz ISM band is the microwave oven. The source of the radiation in a microwave oven is a magnetron tube that is used to convert electric current into electromagnetic radiation energy. The domestic kitchen or residential microwave ovens usually have only one magnetron tube. The commercial microwave ovens may have two or more magnetron tubes. During the active period, the emitted signal is a continuous wave (CW) that has a more or less stable frequency near 2.45 GHz, exactly in the middle of the ISM band, occurring once per mains power cycle. In practice, the output spectrum of a microwave oven varies in frequency and can show multiple interfering tones at other frequencies in the ISM band. For a microwave oven with one magnetron tube, the duty cycle is close to 50%, i.e., 10 ms ON and 10 ms OFF for AC 220V/ 50 Hz input power (1/50 x 0.5s =10 ms) Example of measured spectra A commercial microwave oven has two magnetron tubes that are active during one half of the power cycle. During the ON period the magnetron tube flickers on and off rapidly like a fluorescent tube and the rapid flickering which can have a fast rise-and-fall time causes the wideband noise. The frequency sweep usually covers 2 to 15 MHz for a residential microwave oven to over tens of MHz for a commercial microwave oven. The performance impact by microwave oven interference is mainly dependent upon the distance and frequency separation between the microwave oven and the WLANs devices. Measurements of the microwave ovens emissions have been performed by the NTIA (National Telecommunications and Information Administration, in USA). The NTIA describe measurement results for residential microwave ovens that have an effective isotropic radiated power (EIRP) which can range from 16 dbm to 33 dbm, with power consumption in the Watt range. Figure 6.1 and Figure 6.2 give respectively the max-hold spectrum and the active time of a typical residential oven, according to the measurements performed by A. Kamerman and N. Erkoçevic [17]. These results are in line with the NTIA measurements. During the active period the emitted signal is a CW with a frequency that moves over few MHz. OBAN-WP4-SIN-049g-D Page 63 of 177 OBAN Consortium

70 Although there are many differences between the emissions from ovens of different manufactures, the centre burst frequency is mostly somewhere around MHz, and the sweep goes over 2-6 MHz. Likewise, the total active period is about 8 ms (out of 20 ms mains power cycle at 50 Hz, or 16 ms at 60 Hz). Figure 4-11 Example of spectrum for residential microwave oven. Figure 4-12 Active period for residential microwave oven. Microwave ovens which are used for commercial applications are based on two magnetron tubes which are alternately active during one half of the mains power cycle of 20 ms (or 16 ms). This type of ovens has power consumption in the Watt range. Measurements made by A. Kamerman and N. Erkoçevic [17] show that a commercial microwave ovens occupies a much wider spectrum than the one found for residential oven shown, for example, in Figure Figure 4-13 and Figure 4-14 show the spectrum of two different type of commercial microwave oven, a model A with a rotating mirror plates to illuminate the food and a model B with reflector in bottom and ceiling. The commercial ovens show a random variation in frequency over tens of MHz. OBAN-WP4-SIN-049g-D Page 64 of 177 OBAN Consortium

71 Figure 4-13 Example of spectrum for commercial microwave oven (model A). Figure 4-14 Example of spectrum for commercial microwave oven (model B) Models of microwave oven interference When we analyze the performance of digital radio communication systems in the presence of micro wave oven (MWO) interference it is very useful to acquire the interference data by simulation. Two main models have been proposed in the literature, a mathematical model which represents the first order statistic of microwave oven interference by using the Middleton s class-a noise model [20,26], and the time domain model proposed by Y. Matsumoto et al [23,27], which takes into account the noise generation mechanism and its characteristics, and basically suggests a periodic frequency modulation. Additional work on the topic of MWO noise modelling and measurement can be found in [32-35] Class-A impulsive noise model In the case of class-a impulsive noise model, the high level interferences are emitted randomly. On the other hand, the interference emitted from microwave oven is a burst-type interference. As shown in Figure 6.5, the OBAN-WP4-SIN-049g-D Page 65 of 177 OBAN Consortium

72 first order statistic such as amplitude probability distribution (APD) does not depend on the burst-ness of interference. Figure 4-15 APDs of random and burst interferences. Figure 4-15 is an example of the two waveforms of random interference and burst interference, which have the same probability distribution of amplitude. Results obtained in [21] show that the first order statistic of microwave oven interference can be modelled as a class-a impulsive noise model. In the Middleton class-a noise model the impulsive random noise is represented as the sum of various Gaussian noises with different variances, and the emissions of these Gaussian noise sources obey Poisson distribution; the high amplitude impulse interferences are emitted randomly. Generally, narrow band noise n(t) can be expressed as the in-phase component x(t) and quadrature phase component y(t) as n( t) = x( t)cos 2πf t y( t)cos 2πf t where f c is a central frequency of the noise. In the class-a impulsive radio noise model the in-phase component and quadrature component of the carrier wave are normalized to the total mean noise power to be unity. The probability density functions of each component are given as where = p( x) exp( A) = m 0 = p( y) exp( A) = σ m 0 c m! m! A m 2πσ A m 2πσ m + Γ = 1 + Γ 2 A m exp c 2 2 m exp 2 2 m In the above equations, Γ is the mean power ratio of the Gaussian noise component to the non Gaussian noise component and A is the impulse index (product of the average number of impulsive noise events at the receiver in a unit time and their average duration). A smaller value of A corresponds to strongly impulsive noise and a larger value of A implies greater temporal continuity of the impulses. If represents the experimentally measured envelope of interference, the parameters A and Γ of class-a impulsive noise can be estimated as follows: 2 x σ 2 m 2 y σ 2 m OBAN-WP4-SIN-049g-D Page 66 of 177 OBAN Consortium

73 where e 2 =E[ 2 ], e 4 =E[ 4 ], e 6 =E[ 6 ]. 9( e4 2e ) A = 2( e + 12e 2e Γ = 2 6 ( e 6 3( e + 12e e2e e2e4 2 2 e2 ) ) 2 ) 1 N. Mingxin and L. Ling [22] provide an improved simulation method considering both the amplitude distributions and impulsive burst characteristics. To do so requires two main steps including generating a series of class-a impulsive noise with the same APD as that of MWO interference, and then rearranging the series according to impulsive burst characteristics, remaining the APD at the same time, in order to create a periodical and burst high level interference similar to MWO interference Time-domain MWO noise model The microwave oven noise model proposed by Y. Matsumoto, M. Takeuchi, K. Fujii, A. Sugiura and Y. Yamanaka [23], is a time domain model that takes into account the noise generation mechanism and its characteristics. The model is given by a simple function in the time domain. In a microwave oven a magnetron behaves like a diode and generates RF energy only in the time interval while the driving voltage, produced from the ac mains voltage supplying it to a transformer directly or through an inverter, exceeds the threshold for the oscillation. As a result, microwave ovens generate RF pulses in the 2.4 GHz band at the frequency of the ac mains (50 to 60 Hz) or at the switching frequency of the inverter (typically 30 to 50 khz). Commercially available ovens are categorized into two groups with respect to the high voltage generation: the transformer type and the inverter type. The RF pulse is a pulsed sinusoidal wave whose instantaneous amplitude and frequency vary widely with the instantaneous magnetron driving voltage. Figure 4-16 shows examples of the envelope of RF pulses emitted from the two different types of oven. The duration of the RF pulses are around 8 ms and 15 s for the transformer type and the inverter type respectively. OBAN-WP4-SIN-049g-D Page 67 of 177 OBAN Consortium

74 Figure 4-16 Noise waveform observed at different frequencies. (a) Transformer type oven fc=(1)2.475 GHz,(2)2.462 GHz,(3)2.452 GHz,(4)2.442 GHz,(5)2.432 GHz. (b) Inverter type oven, f c =(1)2.462 GHz,(2)2.452 GHz,(3)2.442 GHz. Considering the mechanism of RF pulse generation, the following assumptions on the RF pulse characteristics have been made in developing the noise model: 1) A noise pulse has a width equal to the time interval during which the driving voltage V(t) exceeds the threshold voltage V 0. 2) The instantaneous amplitude of the pulse envelope varies linearly with the driving voltage (AM). 3) The instantaneous frequency also changes linearly with the driving voltage (FM). As a consequence, the following time domain model has been suggested: t I( t) = I 0 U[ V ( t) ] exp 2π j f 0t + f max V ( ξ ) dξ (6.7) Note that f 0 and f max are the carrier frequency (about 2.45 GHz) and the maximum frequency deviation for the FM, respectively, and that the driving voltage V(t) is normalized by its maximum value. The maximum amplitude of the envelope is given by I 0, and the phase of I 0 is assumed uniformly distributed within [0, 2π]. The AM with a cut-off threshold voltage V 0 is given by V, forv V U ( V ) = 0, forv < V 0 0 OBAN-WP4-SIN-049g-D Page 68 of 177 OBAN Consortium

75 This model can be realized by combining an FM modulator and an AM modulator as schematically shown in Figure Figure 4-17 Model of the microwave oven noise. The driving voltage applied to the magnetron is represented in a normalized form as cos(2πf v ) t V ( t) = cos(2πf vt) cos(2πf st) cos(2πf vt) cos(2πf st) for transformer type for inverter type I for inverter type II In the equation above, the ac mains frequency and the inverter switching frequency are denoted by f v and f s, respectively. The voltage model for inverter type II is applicable to an inverter type oven having a full wave rectifying circuit following a transformer. Furthermore, the noise parameter is assumed to be constant. In the receiver of a wireless system, interfering oven noises are band limited by RF and IF filters, which can be modelled with a convolution operation [23]. The proposed noise model has six independent parameters, f v, f s, f 0, f max, I 0, and V 0. All the parameters can be determined by measuring noise characteristics. Table 6.1 summarizes typical noise parameters estimated from measured waveforms and spectra. Microwave oven type Transformer type Inverter type II AC mains frequency: f v 50 Hz 50 Hz Switching frequency: f s - 30 khz Threshold voltage: V Frequency: f GHz GHz Frequency deviation: f max 43 MHz 40 MHz Amplitude I 0 89 dbμv 101 dbμv Table 4-4 Noise parameters estimated from Figures 6.6 and 6.9 (a). OBAN-WP4-SIN-049g-D Page 69 of 177 OBAN Consortium

76 The simulation scheme shown in Figure 4-17 models accurately actual measurements, as shown in Figure 4-18 and Figure 4-19, where measured and simulated noise spectra and time waveforms are compared. The results prove that the proposed model is effective both in the time and in the frequency domain. Figure 4-18 Measured and simulated noise spectra (BW=1 MHz). (a) Measured. (b) Simulated OBAN-WP4-SIN-049g-D Page 70 of 177 OBAN Consortium

77 Figure 4-19 Simulated noise waveforms (BW=10 MHz). (a) Transformer type oven, fc=(1)2.472 GHz,(2)2.462 GHz, (3)2.452 GHz, (4)2.442 GHz, (5)2.432 GHz. (b) Inverter type II oven, fc=(1)2.462 GHz, (2)2.452 GHz, (3)2.442 GHz Cordless phone interference The main characteristics of most cordless phones are low power and narrow bandwidth. The transmitted power can be 10 dbm or lower. For FHSS cordless phones, the bandwidth is 1 MHz. For most DSSS cordless phones, the 6-dB bandwidth is usually less than 2 MHz. A series of tests were performed to investigate the performance impact of several different types of cordless phones on the DSSS WLAN devices. The performance degradation depends on the cordless phone signal strength, occupied frequency bandwidth, frequency separation and distance between the WLAN device and the cordless phone. A noted design feature of several cordless phones is the ability to re-tune their channels (channel selection) after detecting the presence of stronger DSSS signals. The DSSS cordless phones usually have high output power (>20 dbm) and wider bandwidth (6-dB bandwidth >3 MHz). The IEEE DSSS devices carrier frequency should have at least 20 MHz of frequency separation from these cordless phones in order to maintain acceptable link quality Conclusions for interference analysis In this section the main interference sources affecting WLAN performance have been recalled, and their characteristics summarized, with particular attention to Bluetooth (BT) based devices and microwave ovens OBAN-WP4-SIN-049g-D Page 71 of 177 OBAN Consortium

78 (MWO). The main analytical and simulative interference models for MWO and BT devices have been summarized, together with performance evaluation methods and physical layer mitigation techniques. The main conclusion that can be drawn from the analysis is the fact that the proximity of the interfering device to the WLAN receiver has a tremendous effect on the degree of interference. While microwave ovens, however, are often located far from WLAN receivers, so that their effect can be neglected in most working environments, this is not always true for Bluetooth-based devices. Even though WLANs have higher transmit power than BT devices, the WLAN signal power may drop considerably due to the path loss during transmission to the receiver at a sufficient distance from the access point, making the system vulnerable to BT interference. For this reason, in absence of robust interference mitigation techniques, a proper placement of BT and WLAN devices (as well as MWO) must be selected in order to ensure acceptable WLAN link quality. OBAN-WP4-SIN-049g-D Page 72 of 177 OBAN Consortium

79 5. Capacity analysis 5.1. Introduction In this section we do an analysis of the performance of WLAN systems, starting with simple cases and gradually making the system model more complex by including elements present in realistic implementations. This approach is selected to obtain a thorough understanding of the mechanisms of the MAC and their effect on the performance. The complexity in the model is increased in the following manner: First we consider the case of one STA and one AP. The throughput is then given by the number of information bits in a frame divided by the time the channel is sensed busy due to that transmission. Then we include several STAs contending for the channel. In this case collisions will occur, reducing the throughput. Next we include the fact that received frames from various STAs at different locations will have varying power levels, leading to the capture effect. Above, the channel is assumed to be perfect. In real systems, frames may be lost due to non-perfect channel as well as due to collisions. This must be taken into account. The next issue included is that STAs may transmit at different data rates. This gives rise to what is called the performance anomaly of Further, some issues are mentioned related to mixed uplink and downlink traffic. The results are applied to OBAN scenarios considering packet lengths, QoS parameters etc. for the defined applications. Results are shown for b and a/g systems. For g systems, the ERP-OFDM PHY, which for our purposes is similar to the OFDM PHY of the a amendment except for the frequency band, is considered. The various modulation modes are described in Appendix A MAC review DCF versus PCF The MAC contains two medium access protocols: the contention-based DCF (distributed coordination function) and the PCF (point coordination protocol). When PCF is enabled, the time is divided into superframes. Each superframe consists of a contention-free period for PCF and a contention period for DCF. At the beginning of the contention-free period, the access point contents for the channel. When it acquires the channel, it cyclically polls high-priority STAs and grants them the channel. There are several disadvantages related to the use of PCF: It experiences substantial delay at low load. The effective contention-free period may vary, as the AP needs to contend for the channel. Handling a large number of interactive streams in PCF harms the applications using DCF contention. PCF has not drawn much attention from either the research community or the industry. Therefore we focus on DCF in this document. OBAN-WP4-SIN-049g-D Page 73 of 177 OBAN Consortium

80 Channel access schemes DIFS SIFS DIFS Source Data Destination ACK Contention Window Other Defer Access Backoff after defer Figure 5-1 Basic access scheme DIFS SIFS SIFS SIFS DIFS Source RTS Data Destination CTS ACK CW Other NAV (RTS) NAV (CTS) Figure 5-2 RTS/CTS access scheme. The DCF contains two ways for a STA to access the channel: The default scheme is the basic access mechanism which is a two-way handshaking technique. A STA having a packet to transmit may do so if it has observed that the medium is idle for duration equal to the DCF interframe space (DIFS). The receiving STA responds upon successful reception by OBAN-WP4-SIN-049g-D Page 74 of 177 OBAN Consortium

81 transmitting a positive acknowledgement (ACK) after a Short IFS (SIFS). A SIFS is shorter than a DIFS. The basic access scheme is illustrated in Figure 5-1. The RTS/CTS mechanism is a four-way handshaking technique. The transmitting STA then reserves the channel by sending a Request-to-send (RTS) short frame. The destination STA acknowledges the receipt of an RTS frame by returning a Clear-to-send (CTS) frame after an interval SIFS. Then normal packet transmission and ACK response occur. The RTS/CTS frames contain a Duration/ID field that defines the period of time that the channel is to be reserved to transmit the data frame and the ACK. All STAs within reception range of both transmitter and receiver shall learn of the channel reservation, hence eliminating the effect of the hidden terminal problem. The RTS/CTS access scheme is illustrated in Figure 5-2. An advantage of RTS/CTS mechanism occurs in the case where the coverage of several WLAN cells using the same frequencies overlap, as the mechanism works across net boundaries. The use of RTS/CTS handshake is controlled by the artsthreshold attribute, which may be set on a per-sta basis. RTS/CTS handshake is used for frame lengths larger than artstheshold. Setting this attribute to be larger than the maximum MSDU 3 size has the effect of turning off the RTS/CTS mechanism. Setting this attribute to zero has the effect of turning it on for all frames Random backoff As mentioned above, a STA having a packet to transmit may do so if it has observed that the medium is idle for duration equal to the DIFS. If the medium is busy, the STA defers until the medium is determined to be idle without interruption for a period of time equal to DIFS plus a backoff time. The backoff time is measured in discrete time units called backoff slots. The backoff time is equal to: Backoff Time = Random() x aslottime, where Random() is a pseudorandom integer from a uniform distribution of the interval [0,CW] (CW is the contention window). After observing an idle medium for DIFS, the STA decrements its backoff counter at the end of each observed backoff slot. When the backoff counter reaches zero, the STA transmits the MSDU. If the channel becomes busy during a backoff slot, the STA suspends the backoff procedure until it again detects that the medium is idle for DIFS. CW is within the interval acwmin and acwmax, which are defined for each PHY. The initial CW value is acwmin. For each time a MSDU collides, CW is increased until it reaches acwmax. When a transmission has collided i times, the CW can be expressed as: CW i = min i [( acwmin + 1) 2 1, acwmax] CW reaches acwmax after m retransmissions. After every successful transmission of a MDSU, CW is reset to acwmin Fragmentation In the transmitter, MSDU packets and MMPDU 4 packets may be partitioned into a number of smaller MAC level packets (MPDUs). The reason for this fragmentation is to increase the probability of successful transmission in cases where channel characteristics limit reception reliability for longer frames. 3 MSDU: MAC Service Data Unit, a unit of data arriving at the MAC from the higher level in the protocol stack. 4 MMPDU: MAC Management Protocol Data Unit, a management packet arriving at the MAC from the higher lavel. OBAN-WP4-SIN-049g-D Page 75 of 177 OBAN Consortium

82 The threshold for partitioning a MSDU or MMPDU packet is set by the afragmentationthreshold parameter. Each fragment is a frame no larger than afragmentationthreshold. The fragmentation frame contains a MAC header and CRC. Each MPDU resulting from the fragmentation is sent as an independent transmission acknowledged separately. Hence, retransmissions occur per fragment. Unless interrupted by occupancy limitations of the channel, the fragments of a single MSDU/MMPDU are sent as a burst, using a single invocation of the channel access procedure Throughput for one STA and one AP Time intervals of channel occupancy When there is only one STA contending for the channel, collisions will not occur. The throughput is therefore given by the relation between the number of information bits in a frame, and the time the channel is busy because of the transmission of that frame. We consider the case where the STA always has a packet to transmit, i.e. the transmission queue of each STA is assumed to be always nonempty. In other words, we operate in saturation conditions 5. After the completion of each successful transmission, the STA must wait a DIFS plus the random backoff time before transmitting the next packet. For the basic and RTS/CTS schemes, the time to transmit a frame is then given by: T = T + T + T + T + T basic DIFS backoff data SIFS ACK T = T + T + T + T + T + T + T + T + T RTS / CTS DIFS backoff RTS SIFS CTS SIFS data SIFS ACK The inter frame space times T DIFS and T SIFS are defined for the different PHYs. The average backoff time T backoff is equal to half the minimum size of the contention window. The length of the control frames T RTS, T CTS and T ACK are also given for each PHY and data rate. T data depends on the length of the payload b The relevant parameters of the b amendment are listed in Table 5-1. The PHY preamble is transmitted using 1 Mbps data rate and the PHY header using 2 Mbps data rate. The duration of the PHY preamble and header then totals to 96 μ s. The duration of the different packets are given by: T data T CTS = P ( T + ( P) / R ) μs = μs = T preamble _ header b + ACK = 11 μs 152 ( T + / R ) μs preamble _ header 112 b. contr = T RTS = ( T + / R ) μs 176 μs preamble _ header 160 b. contr = 5 Saturation throughput is defined as the limit reached by the system throughput as the offered load increases, and represents the maximum load that the system can carry in stable conditions. This performance figure is smaller than the maximum throughput, but gives more meaning as it in practical cases will be very difficult to operate at maximum throughput for a long period of time. OBAN-WP4-SIN-049g-D Page 76 of 177 OBAN Consortium

83 where R b is the data bit rate in Mbps, and R b, contr is the basic bit rate in Mbps used by control frames. For b, the supported bit rates are: 1 Mbps, 2 Mbps, 5.5 Mbps and 11 Mbps. The basic bit rate set is a subset of the supported bit rates and must be supported by all STAs in order to associate with a BSS. In the calculations above we have selected R b = 11 and R b, contr = 2. P denotes the payload length and takes on values from 0 to 2312 bytes. It is then straightforward to calculate the time it takes to transmit a data frame for the two access schemes: T RTS / CTS P P T basic = μs = μs P P = μs = μs PHY preamble 72 bits at 1 Mbps PHY header 48 bits at 2 Mbps PHY Short preamble+header (802.11b) 96 μ s MAC header + FCS 272 bits T 10 μ s SIFS T 50 μ s DIFS aslottime 20 μ s acwmin 31 acwmax 1023 ACK 112 bits + PHY header/preamble RTS 160 bits + PHY header/preamble CTS 112 bits + PHY header/preamble Supported data rates 1, 2, 5.5, 11 Mbps Basic data rates Specified for each BSS. Subset of supported data rate set. Payload lengths bytes Table 5-1Relevant parameters for IEEE802.11b a/g The relevant parameters of the a standard and the ERP-OFDM PHY of the g standard are listed in Table 5-2. Using these parameters we obtain: T T cts rts T data = T = T = T ack PHYpreamble = T PHYpreamble + T PHYpreamble + T 22 + MAC_header + P P ceil = + ceil R 20 4 b R 4 4 b 22 + CTS_packet ceil ceil 134 μs R = b contr R 4. 4 b, contr ceil 22 + RTS_packet ceil 182 = + [ μs] R 20 4 b contr R 4, 4 b, contr OFDMsymb + T OFDMsymb OFDMsymb The relevant parameters are listed in Table 5-2. [ μs] [ ] OBAN-WP4-SIN-049g-D Page 77 of 177 OBAN Consortium

84 T 20 μ s PHYpreamble T 4 μ s OFDMsymb MAC header + FCS 272 bits Number of data bits per OFDM symbol 4 R b T 16 μ s SIFS T 34 μ s DIFS aslottime 9 μ s acwmin 15 acwmax 1023 ACK 112 bits + PHY preamble RTS 160 bits + PHY preamble CTS 112 bits + PHY preamble Supported data rates 6, 9, 18, 24, 36, 48, 54 Mbps Basic bit rates 6, 12, 24 Mbps Payload length bytes Table 5-2 Relevant parameters (IEEE a/g). The data rate R b and the control rate R b, contr used for the ACK, RTS and CTS packets may be different, as the control packets are transmitted at a low rate to assure long reception distance. The data rate R b can take on the values 6, 9, 18, 24, 36, 48 and 54 Mbps. The control rate R b, contr can take on the rates 6, 12 and 24 Mbps. The payload length P takes on values from 0 to 4061 bytes Saturation throughput The saturation throughput for the two access schemes are given by the number of payload bits divided by the period of the time the channel is busy due to that transmission: S S = basic RTS / CTS = P Tbasic P T RTS / CTS The saturation throughput of the two access schemes as function of frame length is depicted for b and a/g devices in Figure 5-3. OBAN-WP4-SIN-049g-D Page 78 of 177 OBAN Consortium

85 Saturation throughput S [Mbps] Basic RTS/CTS Payload length P [bytes] Saturation throughput S [Mbps] Basic RTS/CTS Payload length P [bytes] Figure 5-3 Saturation throughput for one STA contending for the channel. Left: b (Rb=11 Mbps, Rb,contr = 2 Mbps). Right: a (Rb = 54 Mbps, Rb,contr = 24 Mbps). In Table 5-3 we have listed some numbers for the saturation throughput. From the numbers we see that the maximum saturation throughput is 7.92 Mbps in an 11 Mbps b system and 41.7 Mbps in a 54 Mbps a/g system. If we have shorter than maximum length frames, e.g bytes, or if we employ the RTS/CTS access scheme, the throughput is considerably lower. Saturation throughput Saturation throughput b basic access scheme a/g basic access scheme P=1500 bytes P=2312 bytes P=1500 bytes P=4061 bytes 6.89 Mbps 7.92 Mbps 30.2 Mbps 41.7 Mbps b RTS/CTS access scheme a RTS/CTS access scheme P=1500 bytes P=2312 bytes P=1500 bytes P=4061 bytes 5.74 Mbps 6.9 Mbps 24.8 Mbps 37.8 Mbps Table 5-3 Saturation throughput for IEEE b and IEEE a/g The multiple STA case with perfect channel In the previous section we considered the case where there was only one STA contending for the channel. In this section we consider the case where there are a number N STAs contending for the channel and where packets only are lost due to collisions. In the one STA case, there was no collision, meaning that the average backoff time was equal to half of the minimum contention window, as the backoff time after a transmission is drawn from a uniform distribution in the interval [ 0, CW min ]. The probability that the STA was transmitting in an arbitrary time slot (i.e. that the backoff timer is zero) was then 2 /( CW + min 1). This is no longer the case when we have several STAs contending for the channel. The probability that a STA is transmitting in an arbitrary time slot now depends on how many times the previous packet has been retransmitted. In [1], a model is developed to analyze the probability that the STA is transmitting in an arbitrary time slot. This is then used to analyze the throughput Estimated throughput An alternative definition of the normalised throughput is as follows: OBAN-WP4-SIN-049g-D Page 79 of 177 OBAN Consortium

86 S = E [ Information bits transmitted in a time slot] E[ Length of a slot time] When no STA is transmitting, the slot length is given by the aslottime parameter in the standard. When a STA is transmitting, the slot length lasts from a STA captures the channel until the transmission has finished and the channel has been idle for a period DIFS. There are three situations that can occur that affect the length of the slot time, no STA is transmitting, one STA transmits and the transmission is successful, and more than one STA transmit and collisions occur. Information bits are only successfully transmitted in the case where only one STA transmits. We define: P s : The probability that a transmission occurring on the channel is successful P tr : The probability that at least one STA is transmitting P : The probability that a STA transmits in an arbitrary chosen time slot ts The relations between these three probabilities are: P s P = NP The normalised throughput can then be expressed as: S = ts ( 1 P ) P = 1 1 tr ts N 1 ( ) N tr P ts Ps Ptr E[ P] ( 1 P tr ) Tσ + P s P tr T s + P tr ( 1 P s ) T c N STAs don t transmit times the average length of the payload. The first part of the denominator corresponds to the probability that no STAs transmit times the length of a backoff slot T σ. The second term corresponds to the probability than one STA The nominator corresponds to the probability that one STA transmits while ( 1) transmits and ( N 1) STAs don t transmit times the average time the channel is sensed busy because of a successful transmission T s. The third term corresponds to the probability that more than one STA transmits in the slot times the average time the channel is sensed busy because of an unsuccessful transmission T c. For the basic and RTS/CTS access mechanisms, we have the following expressions for T = T + T + T + TSIFS + TACK T T + TSIFS + TCTS + TSIFS + Tdata + TSIFS + T T + bas s DIFS backoff data bas c = TDIFS + Tbackoff + Tdata rts s = TDIFS + Tbackoff + TRTS rts c = TDIFS + Tbackoff TRTS ACK T s and T c : T data is the duration of the longest packet involved in a collision. In this section we assume that all STAs transmit packets with the same length. Hence, T = T. data data OBAN-WP4-SIN-049g-D Page 80 of 177 OBAN Consortium

87 Relation between transmission probability and packet collisions In [1] a Markov chain model is used to estimate the probability P ts that a STA transmits in an arbitrary time slot. The analysis makes one key approximation. Each packet collides with constant and independent probability p, regardless of the number of retransmissions suffered. If only two STAs are contending for the c channel, this is obviously far from accurate. When packets from the two STAs collide, they both increase their contention window, and the probability that the next packet that is transmitted collides is reduced to half. If many STAs contend for the channel on the other hand, then the contention window of most of them will be unaffected by a collision and the collision probability of a retransmitted packet will be close to the collision probability of the first transmission. Consider a network consisting of N STAs. All of them have packets to transmit. It is then possible to derive a relationship between the probability that a STA transmits a packet in a time slot, given the backoff and contention window parameters, and the probability p c that the packet collides with a packet from another STA. The probability P ts is easily derived from the Markov chain model defined in [1]: P ts = 2( 1 2 pc ) ( 1 2 p )( W + 1) + p W 1 ( 2 p ) c c m ( ) The backoff time parameters are given by the maximum and minimum contention window sizes: W = acwmin +1 2 m W = acwmax +1 c The parameter m represents the maximum backoff stage. On the other hand, the probability that a packet encounters a collision is equal to the probability that, in a time N 1 remaining STAs transmits: slot, at least one out of the ( ) p = 1 N 1 ( 1 ) c P ts The two equations above represent a nonlinear system in the two unknowns P ts and p c. The system has always a unique solution that may be found using numerical methods. The solution for P ts is then inserted into the expression of the saturation throughput Cumulative saturation throughput In Figure 5-4, the cumulative saturation throughput as function of STAs is illustrated for b and a. The frame length is set to 1500 bytes, which corresponds to the MSDU size of some streaming applications and file download applications. We see that the basic access scheme is much more sensitive to increasing number of STAs than the RTS/CTS scheme. But even for the basic access scheme, the total throughput for b only decrease from 7.4 Mbps for 3 STAs (which is the number of STAs that maximize the total throughput) to 7 Mbps for 10 STAs and to 6.5 Mbps for 20 STAs. For a/g, the cumulative throughput is 31.5 Mbps with 2 STAs, 28.3 Mbps for 10 STAs and 26.3 Mbps for 20 STAs. This shows that although the MAC protocol does add some overhead, the efficiency remains relatively constant as the number of STAs increase. OBAN-WP4-SIN-049g-D Page 81 of 177 OBAN Consortium

88 Basic RTS/CTS Basic RTS/CTS Saturation throughput [Mbps] Saturation throughput [Mbps] Number of STAs Number of STAs Figure 5-4 Saturation throughput as function of number of STAs. Left: b (Bit rate=11 Mbps, L=1500 bytes). Right: a (Bit rate=54 Mbps, L=1500 bytes). Saturation throughput S [Mbps] N=1, Basic 2 N=5, Basic 1 N=20, Basic Payload length P [bytes] Saturation throughput S [Mbps] N=1, RTS/CTS 2 N=5, RTS/CTS 1 N=20, RTS/CST Payload length P [bytes] Figure 5-5 Saturation throughput as function of payload length for b. Left: Basic access scheme. Right: RTS/CTS access scheme. OBAN-WP4-SIN-049g-D Page 82 of 177 OBAN Consortium

89 50 50 Saturation throughput S [Mbps] N=1, Basic N=5, Basic N=20, Basic Payload length P [bytes] Saturation throughput S [Mbps] N=1, RTS/CTS N=5, RTS/CTS N=20, RTS/CST Payload length P [bytes] Figure 5-6 Saturation throughput as function of payload length for a/g. Left: Basic access scheme. Right: RTS/CTS access scheme. In Figure 5-5 and Figure 5-6, the normalised throughput is illustrated as function of the frame length for 1, 5 and 20 contending STAs. When the RTS/CTS scheme is used, the throughput is practically independent of number of STAs, while the dependency is clearer for the basic scheme. For a few STAs and small packets, it is more efficient to use the basic scheme, assuming that all STAs can hear eachother. As the number STAs and/or packet lengths increase, the RTS/CTS scheme performs relatively better, and at a certain point outperforms the basic scheme. This is due to the fact that the number of collisions increases with the number of STAs. This is also illustrated in Figure 5-7. As the RTS frame is much shorter than a data frame (in general), it is better to loose a RTS frame in a collision than a data frame. The more STAs that are competing for the channel, the more collisions will occur. And the longer data frames, the more time is wasted in each retransmission. The a/g standard has a higher average number of transmission due to the smaller minimum contention window size. Average number of transmissions b a Number of STAs Figure 5-7 Average number of transmissions per packet as function of number of STAs. OBAN-WP4-SIN-049g-D Page 83 of 177 OBAN Consortium

90 Throughput per user The MAC protocol gives all STAs that have a packet ready for transmission and contending for the channel equal chance to capture the channel. If the packet lengths and the information bit rate is the same for all STAs, the capacity is equally distributed. In this section we assume that this is the case. Figure 5-8 shows the estimated throughput per user as function of number of STAs contending for the channel. The numbers are also shown in Table Basic RTS/CTS Basic RTS/CTS Throughput per STA [Mbps] Throughput per STA [Mbps] Number of STAs Number of STAs Figure 5-8 Throughput per user as function of number of STAs. Frame length: 1500 bytes. Left: b (11 Mbps), Right: a (54 Mbps). Number of STAs b (11 Mbps) a (54 Mbps) N Basic RTS/CTS Basic RTS/CTS Table 5-4 Estimated throughput per STA in Mbps as function of number of STAs. Frame length 1500 bytes. The results show that the throughput per user is quite similar for the two access schemes. Hence, the loss in throughput one has to pay for eliminating the hidden terminal problem is relative small for payload lengths of 1500 bytes. For small packets, however, the difference in throughput between the two access schemes is larger. This is illustrated in Figure 5-9. With a payload length of only 20 bytes, it is very inefficient to use the RTS/CTS scheme. OBAN-WP4-SIN-049g-D Page 84 of 177 OBAN Consortium

91 Throughput per STA [Mbps] Basic RTS/CTS Throughput per STA [Mbps] Basic RTS/CTS Number of STAs Number of STAs Figure 5-9 Throughput per user as function of number of STAs. Frame length: 20 bytes. Left: b (11 Mbps), Right: a (54 Mbps) Capture effect In the previous sections we have assumed that whenever more than one frame is transmitted at the same time, the information contained in all the colliding frames is lost. This model may be too pessimistic if the signals transmitted by different STAs arrive at the AP with different power levels. The AP may capture a frame if its power exceeds the joint power of the n interfering signals by a certain threshold factor for the duration of a certain time period. This is commonly referred to as the capture effect. The predefined threshold is called the capture ratio. From Chapter 2 we have that the minimum received E s / N0 is about 10 db for 11 Mbps b and about 20 db for 54 Mbps a/g. We assume that the capture ratio equals these values. STA i r i 1 AP Figure 5-10 A generic WPAN cell with radius 1 and with an AP in the centre. OBAN-WP4-SIN-049g-D Page 85 of 177 OBAN Consortium

92 Capture probability To analyze the capture effect, we consider a WLAN cell with radius 1, and with an AP in the center (see Figure 5-10). The STAs are uniformly distributed within the coverage area of the AP. The pdf of the distance between the AP and the STAs is then given by: h ( r) = 2r, r < 0 1 Denoting the received power of a frame from the i th STA p cap 0 ( zo, n) = Pr γ = > z = n 0 P i, the capture probability is given by [3]: P n 1 i= 1 P ri = i 1 z 0 1 r0 i 4 where P 0 is the received power of the useful signal, P1, KPn are the received power levels of the n interfering signals, γ is the signal-to-interference ratio and z 0 is the capture ratio. The free space loss exponential is set to -4 and the fading of the signal envelope is Rayleigh distributed. Statistically, all the factors of the expression above are equal. The average capture can then be derived by integrating over the geometrical distribution of the STAs [3]: where: p cap ( ) ( ( ) ) n z ( ) 0 n I r0, z0 h r0, dr = 0 I ( r, z ) 0 0 ( r ) h i dri = ri 1+ z o ro 4 = 1 r 2 o z 0 arctan r z 0 The average capture probability is plotted in Figure 5-11 as function of the capture ratio and the number of STAs. As expected, the capture probability decreases as the number of STAs and capture ratio increase. The overall probability of frame capture P cap depends on the probability that a STA is transmitting in an arbitrary time slot P ts. The probability that i interfering frames are being generated in the observed time slot is given by: R i N = P i + 1 ts N i 1 ( 1 P ) ts The overall capture probability then becomes: P cap = N 1 i= 1 R i p cap ( z i) 0, OBAN-WP4-SIN-049g-D Page 86 of 177 OBAN Consortium

93 n = 3 n = 6 n = 9 p cap (z 0,n) z 0 [db] Figure 5-11 Average conditional capture probability as function of the capture ratio and number of STAs Total throughput The probability of a successful transmission taking into account the capture effect is given by: P s = NP ts ( 1 P ) ts P tr N 1 + P cap The expression for the normalised throughput is then the same as when capture was not included in the model: S = Ps Ptr E[ P] ( P tr ) T + P s P tr T s + P tr ( 1 P s ) T c 1 σ In Figure 5-12, the impact of the capture effect is illustrated. Infinite capture ratio corresponds to no capture effect. As expected, the capture effect has an impact when the number of STAs is small. When many STAs are contending for the channel, the effect is very limited. Also, the impact is larger using the basic access scheme than using the RTS/CTS scheme. OBAN-WP4-SIN-049g-D Page 87 of 177 OBAN Consortium

94 Saturation throughput S [Mbps] Basic, z=10 db Basic, z= RTS/CTS, z=10 db RTS/CTS, z= Normalised throughput Basic, z=20 db Basic, z= RTS/CTS, z=20 db RTS/CTS, z= Number of STAs Number of STAs Figure 5-12 Normalised throughput as function of number of STAs with and without taking into account the capture effect. Capture ratio is 10 db. Frame length 1500 bytes. Left: b. Right: a/g Throughput per user The throughput per user with a capture ratio of 10 db is illustrated in Figure The STAs are uniformly distributed within the circular coverage area. The curves show that the impact of the capture effect is very small. With our assumptions on STA distribution and capture ratio, it is therefore little to gain in including the capture effect in our model. For other STA distributions the situation may be different. For example, if the AP is placed indoors, and one STA is located indoors close to the AP and the rest of the STAs are located outdoors, then the indoor STA may capture the channel due to its better SNR. Throughput per user [Mbps] Basic, z=10 db Basic, z= RTS/CTS, z=10 db RTS/CTS, z= Throughput per user [Mbps] Basic, z=20 db Basic, z= RTS/CTS, z=20 db RTS/CTS, z= Number of STAs Number of STAs Figure 5-13 Throughput per user with and without the capture effect taken into account. Frame length 1500 bytes. Left: b (11 Mbps). Right: a (54 Mbps) Loss of packets due to non perfect channel In [2], the model in [1] is expanded to include frame-error probability P f, and also maximal allowable number of transmission attempts. The result is a generalization of the results in [1]. OBAN-WP4-SIN-049g-D Page 88 of 177 OBAN Consortium

95 An unsuccessful transmission attempt can now be caused by either packet collision and/or reception of an erroneous frame. These are independent happenings, and the probability for unsuccessful transmission p can be expressed as: p = p c + P f p c P f The probability P ts can be expressed as [2]: P ts = m+ f + 1 2( 1 2 p)( 1 p ) m+ f + 1 m f 1+ f ( 1 2 p)( 1 p ) + W ( 1 p p( 2 p) )( 1 p 2 p ) where m + f corresponds to the maximum number of retransmission attempts. After m + f + 1 attempts to transmit a packet, it is either successfully received or dropped. The throughput is given by: S = Ps Ptr ( 1 Pf ) E[ P] ( 1 Ptr ) Tσ + Ptr Ps ( 1 Pf ) Ts + Ptr ( 1 Ps ) Tc + Ptr Ps Pf Te where T e is the average time the channel is sensed busy from a frame that suffers transmission errors. One or more bit errors in a data packet lead to packet loss. Both data packets and control packets may contain errors. The MAC header in all packets contains a Duration/ID field. This field is used by all STAs receiving it to predict the time interval the channel is busy. Hence, as long as the packet header is received correctly, the channel is considered as busy for a time interval equal to the time interval of a successful transmission. Assuming that most errors occur in the payload part of a frame, we have approximately that T = T. Figure 5-14 illustrates how the throughput decreases with the frame error rate P f. The number of STAs contending for the channel is 5, and the maximum number of retransmissions f is set to 100 which approximates infinity. The curves show that a frame error rate below 10 % gives a relatively moderate decrease in throughput. e s 8 7 Basic RTS/CTS Basic RTS/CTS Saturation throughput [Mbps] Saturation throughput [Mbps] Frame error rate P f Frame error rate P f Figure 5-14 Throughput as function of frame error rate (N=5, f=100, P=1500 bytes). Left: b, Right: a. OBAN-WP4-SIN-049g-D Page 89 of 177 OBAN Consortium

96 Throughput per user The throughput per user for various frame error rates and number of users is illustrated in Figure The curves show that the throughput per user decreases with close to the same percentage as the packet loss rate. Throughput per user [Mbps] P f =0 P f =0.1 P f =0.2 P f =0.4 Throughput per user [Mbps] P f =0 P f =0.1 P =0.2 f P =0.4 f Number of STAs Number of STAs Figure 5-15 Throughput per user as function of number of STAs for different packet loss probabilities P f. Frame length 1500 bytes. Left: b. Right: a/g STAs transmitting at different data rate Both the b and a/g standards have a number of data rates that can be used for data packets (see Table 5-1 and Table 5-2. STAs operating in the same BSS may transmit data using different rates. STAs operating under low SNR conditions may for instance be forced to reduce the spectral efficiency and hence use a lower data rate mode than STAs operating under high SNR conditions to be able to transmit data at all. When one STA reduces its data rate, however, it will have serious consequences on the throughput for other STAs in the BSS. This effect is sometimes called the performance anomaly of The performance anomaly of STAs in a BSS have the same probability of accessing the channel independently of the data rates. Hence, a STA transmitting at a low data rate will access the channel just as often as a STA transmitting at a high data rate and obtain the same packet rate. As the packet lengths are not related to the data rate, all STAs will obtain the same (long term) throughput, irrespective of their data rate, as long as they use the same packet lengths. This means that if one STA transmits 6 Mbps in an a network, another STA transmitting at 54 Mbps will experience a drop in throughput to below 6 Mbps Normalised throughput We consider the case where there are two types of STAs in a BSS, N slow STAs transmitting at a low data rate and N fast STAs transmitting at a high data rate. We assume that collisions involve only two packets. This is an approximation that is closer to reality for networks containing a few STAs than networks containing many STAs. The average time it takes to successfully transmit a packet is given by: OBAN-WP4-SIN-049g-D Page 90 of 177 OBAN Consortium

97 N N slow fast T s = Ts, slow + Ts, fast N When a collision occurs, the channel is busy for a period of time determined by the longest (in time) data packet. Hence, collisions involving one or more low data rate packets will occupy the channel for a longer period of time and collisions involving only high data rate packets. The average time the channel is occupied due to a collision between two packets is given by: N ( N 1) ( N 1) N ( N 1) fast fast fast fast T c = 1 Tc, slow + T, N N N c fast ( N 1) These expressions can be inserted into the equation for the saturation throughput. In Figure 5-16, the resulting throughput per user is shown for b and a/g. The curves show how one slow user severely degrades the performance of all users. As the proportion of slow users increases, the throughput of each user degrades further. For a few STAs however, the main degradation occurs from zero to one slow user. Throughput per user[mbps] All fast One slow All slow Throughput per user [Mbps] All fast One slow All slow Number of STAs Number of STAs Figure 5-16 Throughput per user as function of number of STAs for mixed data rate case. Frame length 1500 bytes, basic access scheme. Left: b (Data rate slow user 2 Mbps, fast users 11 Mbps). Right: a/g (Data rate slow user 6 Mbps, fast users 54 Mbps) Mixed g/b networks A network may contain both b and g radios. However, b radios do not hear when the channel is busy with g OFDM signals. Therefore, the g standard provides protection mechanisms for managing communication in a mixed b/g environment. The protection mechanisms prevent b clients from transmitting after improperly assessing that the airspace is empty while g OFDM signals are being transmitted. The g products still communicate at the same g OFDM data rates when protection is in use, but a short b rate message signals to b products to not transmit for a specified duration because an g OFDM message is immediately following. The b protection messages cause signalling overhead and result in reduced throughput. The g standard contains two protection mechanisms. The first one is a standard RTS/CTS protocol, using b rates. The other mechanism is called CTS-to-self. The g radio then transmit a CTS message using an b rate to clear the air, and then immediately follows with data using an g data rate. There are then three situations that can occur related to mixed g/b environments: OBAN-WP4-SIN-049g-D Page 91 of 177 OBAN Consortium

98 802.11g only: The g AP detects that all of the clients are g and instructs the network not to use any protection method g AP, mixed STAs: The AP senses both technologies on the network. It instructs g STAs to use a protection mechanism. The g STAs then transmit at a lower throughput then in g only networks, but at higher throughput than b STAs b AP, g STAs: Communications use CCK modulation and achieve typical b throughput. An g STA can always function as an b STA. The performance of mixed b/g networks will be evaluated in activity A4 and the results will be presented in delivery D Unfairness between uplink and downlink traffic There is unfairness between uplink and downlink flows in WLANs. The reason is that there are multiple uplink CSMA instances contending with only one downlink CSMA instance (the one at the AP). This is particularly noticeable when the main traffic is downstream. This phenomenon mainly affects UDP traffic flows, while it does not have the same impact on TCP traffic, since TCP is a closed-loop protocol. The effect of this unfairness will be assessed in activity A4 and the results will be presented in delivery D IEEE802.11e MAC The introduction of a and g equipment with their increased data rate capabilities makes bandwidth intensive applications such as video/audio streaming, interactive gaming etc. feasible even with multiple STA connected to an AP. Such applications do however need provision of QoS. This was the rationale behind the creation of the e task group. The e MAC is called the Hybrid Coordination Function (HCF). It is called hybrid as it combines a contention channel access mechanism, referred to as the Enhanced Distributed Channel Access (EDCA), and a polling-based channel access mechanism, referred to as the HCF Controlled Channel Access (HCCA). EDCA is an enhanced version of the legacy DCF and is used to provide prioritized QoS service. HCCA is used to provide parameterized QoS service, where a negotiation of QoS requirements between the STA and the Hybrid Coordinator (HC) takes place before the communication commences. Both mechanisms operate simultaneously and continuously within the BSS. The time is divided into superframes. Each superframe starts with a beacon frame, and the remaining time is divided into a Contention Free Period (CFP) and a Contention Period (CP). The EDCA works during the CP, and the HCCA works during the CFP. The beacon frame contains the QoS parameter set elements (QPSE) that provides information needed for the STAs for proper operation of the QoS facility during the CP EDCA The EDCA mechanism is designed primarily for traffic that has no hard QoS requirement. It is well suited for TCP traffic that has no sustained bandwidth or delay requirements, but still requires a certain level of reliability. With EDCA, a single MAC can have multiple queues that work independently, in parallel, and with different priority levels. Each frame arriving at the MAC from higher layers carries a specific priority value. Up to 8 priority levels are allowed, which are in turn mapped to 4 access categories (ACs) or physical MAC queues. This is illustrated in Figure Each queue has its own Adaptive Interframe space (AIFS) and maintains its own Backoff Counter (BC). The contention window and backoff required for each packet are parameterized with AC specific parameters (see Table 5-5). Each QoS data frame carries its priority level in the MAC frame header. OBAN-WP4-SIN-049g-D Page 92 of 177 OBAN Consortium

99 A virtual collision handler is used whenever more than one AC finishes its backoff at the same time within one STA. The highest priority frame among the colliding frames is chosen and transmitted, and the others increase their backoff counters with CWmin[i]. AC0 AC1 AC2 AC3 AIFS[0] BC[0] AIFS[1] BC[1] AIFS[2] BC[2] AIFS[3] BC[3] Virtual Collision Handler Transmission attempt Figure 5-17 Illustration of Access categories (ACs) for EDCA. AC # Designation (Informative) Cwmin CWmax AIFSN TXOP limit b TXOP limit a/g 0 Background acwmin acwmax Best Effort acwmin acwmax Video (acwmin+1)/2-1 acwmin 2 6,016 ms 3,008 ms 3 Voice (acwmin+1)/4-1 (acwmin+1)/ ,264 ms 1,504 ms Table 5-5 Default EDCA parameter set Different levels of priority are provided to each AC through a combination of three priority differentiation mechanisms as follows: Arbitrary interframe space (AIFS). AIFS is the constant duration (for an AC) for which the medium must be silent before commencing random backoff (see Figure 5-18). Contention window sizes ( CW min and CW max ) Transmission Opportunity (TXOP) limit, which is the maximum duration for which a STA can use the medium for transmission, having won the contention. Table 5-5 contains the default EDCA parameters. However, the AP can adapt these parameters dynamically depending on network conditions. For 0 i j 3 we have CWmin[i] CWmin[j], OBAN-WP4-SIN-049g-D Page 93 of 177 OBAN Consortium

100 CWmax[i] CWmax[j] and AIFS[i] AIFS[j], and at least on of the above inequalities must be not equal to. The parameters are announced by the AP via beacon frames. AIFSD[AC] AIFSD[AC] + SlotTime PIFS SIFS Busy medium Contention Window [1,CWmin[AC]+1] Figure 5-18 IEEE e EDCA channel access AIFS The AIFS[AC] is determined by: AIFS[AC] = AIFSN[AC] aslottime + asifstime where AIFSN[AC] is an integer greater than zero (see Table 5-5). The smaller AIFS[AC], the higher priority. The use of AIFS gives high priority STAs better service in two ways: High priority STAs may transmit in backoff slots that lower priority STAs still waiting in AIFS cannot. Hence, the probability of collision is reduced. High priority STAs will progress through backoff slots relatively faster since they may decrement their backoff counter while low priority STAs wait for the end of their AIFS. For instance, if two transmissions collide and by chance choose identical backoff counter value, then the STA with the smaller AIFS will transmit sooner Contention Window In legacy DCF, initial values for backoff counters are randomly selected from the interval [0,CW], where CW is a function of the PHY-specific acwmin and acwmax attributes. In contrast, backoff counters in EDCF are selected from the interval [1, CW+1], and CW is a function of the AC specific acwmin[i] and acwmax[i] attributes. The lower bound of the interval is increased to one to accommodate AIFS values of one. If the interval were [0,CW], it would be possible for a AC with AIFS[i] = 0 to preempt STAs seeking coordinate use of the medium after the priority IFS (PIFS) (see Sec ). High priority STAs have small acwmin and acwmax values, which corresponds to fewer backoff slots being traversed per transmission on average TXOP During an EDCA TXOP, a STA is allowed to transmit multiple MPDUs from the same AC with a SIFS time gap between the ACK and the subsequent frame transmission. The TXOP is determined by the TXOPLimit[i] parameter. Such multiple MPDU transmissions are sometime referred to as Contention-Free Bursts (CFB) or TXOP bursting. STAs with longer TXOP limits have to contend for the medium access less often than STAs with similar traffic arrival rate but shorter TXOP limits. Less frequent contention means lower transmission overhead per unit of payload, fewer collisions and, thus, superior service. On the other hand, a STA cannot transmit a frame that extends beyond a TXOP. If a frame is too large to be transmitted in a TXOP, it should be fragmented into smaller frames. OBAN-WP4-SIN-049g-D Page 94 of 177 OBAN Consortium

101 The QPSE transmitted in the beacon frames also contains the parameters TXOPBudget[i], Load[i] and SurplusFactor[i]. For the AC[i], TXOPBudget[i] specifies the additional amount of time available during the next beacon interval, Load[i] specifies the amount of time used during the previous beacon interval, and SurplusFactor[i] represents the ratio of over-the-air bandwidth reserved to the bandwidth of the transported frame required for successful transmission. These parameters are calculated by the AP for each beacon interval and embedded into the next beacon frame transmitted to each STA HCCA The HCCA mechanism is designed primarily for applications that have a sustained bandwidth and/or delay requirements. Such requirements are typical for audio/video and multimedia streaming, and interactive applications. The HCCA guarantees that the QoS requirements are met once a stream has been admitted into the network. A STA requesting parameterized services uses the Traffic Specification (TSPEC) element, which contains the set of parameters that characterize the traffic stream that the STA wishes to establish with the HC. The HC analyses the parameters and decides whether to admit the stream into the network. This is known as the admission control process. Once a stream for a STA is established, the HC allocates TXOPs via polling to the STA, in order to guarantee its QoS requirements. The HC enjoys free access to the channel during both the CFP and during the CP. During the CP, it uses the highest EDCA priority and its access to the channel is guaranteed once it becomes idle. I.e., it makes use of the PCF Interframe Space (PIFS), which is the shortest AIFS value, and sets CWmin = CWmax = 0 to seize and maintain control of the channel. Once the HC has control of the medium, it starts to deliver parameterized downlink traffic to STAs and to issue QoS contention-free polls (QoS CF-polls) frames to those STAs that have requested uplink parameterized services. If the STA being polled has traffic to send, it may transmit several frames for each QoS CF-poll received, respecting the TXOP limit specified in the poll frame. In addition, in order to utilize the channel more efficiently, the STAs are allowed to piggyback both the ACK-packet and the CF-Poll onto data frames. Several scheduling disciplines can be used in the HC, and the performance of the HCCA depends on the choice of admission control and scheduling algorithm ACK policies Two ACK policies are defined in addition to the ARQ mechanism used by legacy DCF: the NO ACK policy and the Block ACK policy. When NO ACK policy is used, the transmission of a packet is not accompanied by an ACK. This policy increases the throughput at the expense of potential performance degradation in high error rate conditions. The Block ACK policy aggregates several acknowledgements into one frame. There are two types of Block ACK mechanisms: immediate and delayed. Immediate Block ACK is suitable for high-bandwidth, low latency traffic while the delayed Block ACK is suitable for applications that tolerate moderate latency. When the immediate Block ACK policy is used, the recipient shall respond to a BlockAckReq, with a BlockAck frame. If the recipient sends the BlockAck frame, the originator updates its own record and retries any frames that are not acknowledged in the BlockAck frame, either in another block or individually. When the delayed Block ACK policy is used, the recipient shall respond to BlockAckReq with an ACK frame. The recipient shall then send its BlockAck response in a subsequently obtained TXOP. Once the contents of the BlockAck frame have been prepared, the recipient shall send this frame in the earliest possible TXOP using the highest priority AC. The originator shall respond with an ACK frame upon receipt of the BlockAck frame. OBAN-WP4-SIN-049g-D Page 95 of 177 OBAN Consortium

102 Throughput results There are a number of recent papers investigating the performance and throughput of the e protocol. Below we describe some of the reported results Fairness in time In [4] it is considered how the TXOP can be used to provide fairness in time. The scenario is that a STA located far from the AP may be forced to use a coding and modulation scheme that provides low data rate due the low SNR. As seen in Sec. 5.7, this would seriously degrade the throughput for other high data rate STAs using the legacy DCF. Figure 5-19 is taken from [4] and shows results from OPNET simulations. The figure illustrates how the throughput of the fast STA decreases when the data rate of the slow STA decreases using DCF. The figure also illustrates that this can be avoided if the e standard is implemented. Using both EDCA and HCCA, the throughput of the 11 Mbps fast STA remains practically constant when the data rate of the slow STA decreases to 5.5 Mbps, 2 Mbps and 1 Mbps with. This is because the channel is allocated to a STA for a certain period of time, and not for one packet. The low data rate STA must use fragmentation to fit the frame size within the TXOP limit. Figure 5-19 Comparison of throughput with one fast and one slow STA for legacy DCF, EDCA and HCCA using b [4] Performance of priority mechanisms In [5] the Markov chain model developed in [1] is extended to include the EDCA. It incorporates the AIFS and contention windows associated with different ACs, and enables an analytical evaluation of the throughput for two STAs with different priority levels. The model does not incorporate the TXOP mechanism. The conclusions of the paper are as follows. Contention window sizes provide efficient differentiation under low load conditions, but as loads increase low priority stations may be starved of bandwidth. This is particularly the case when the maximum contention window size is much larger for the low priority STAs. Also high priority STAs suffer performance degradation due to lower priority STAs. This is in contrast to AIFS differentiation which does not sacrifice service provided to high priority traffic under heavy low priority traffic loads. Low priority traffic is however susceptible to being starved using AIFS differentiation. OBAN-WP4-SIN-049g-D Page 96 of 177 OBAN Consortium

103 5.11. Capacity of OBAN networks In this section we use results given earlier in this Chapter to assess the capacity of an OBAN cell. The capacity is measured as throughput per user as function of number of users, and as the number of users that can be supported as function of throughput per user. Results are provided for long frames (1500 bytes) and for short frames (120 bytes), and for both basic and RTS/CTS access schemes. The analysis is limited to the saturation throughput of one way data streams. A number of issues are left for the traffic simulations of activity 4. Of particular importance are the following issues: In this section the MAC throughput is considered. This throughput is similar to the UDP throughput, with the only difference that IP and UDP headers must be included as part of the MAC data frame. For TCP traffic, the throughput may be significantly different as TCP is a close loop protocol. After a user transmits a TCP packet it must wait for an end-to-end TCP-ACK packet. The TCP-ACK packet is from the MAC layers point of view treated as any other data packet, although a short one. The length of the window, i.e. the number of TCP packets that is transmitted between each ACK packet, also varies. It is therefore considered as more adequate to simulate TCP traffic using the OPNET simulator developed in activity 4. The AP contends for the channel on equal term with the STAs. Hence, if there are multiple downstream communications, they share a throughput equal to the throughput available for one upstream connection. This is of particular importance for two-way traffic such as IP telephony. An IP telephony connection requires 96 kbps in each direction. As the number of parallel IP telephony sessions handled by an AP increases, the downlink capacity will suffer before the uplink capacity does. This asymmetry is not considered in this section. One realistic case is that a mix of g and b STAs connect to the same AP. As explained earlier in this chapter, there are protection mechanisms defined in the g standard that handle this situation. These mechanisms are not considered in this section. In the residential gateway there will be functionalities to give different priorities to different users, e.g. to home users and OBAN users. Such functionalities are not considered in this section. Here it is assumed that all users have the same priority, and that the throughput is shared equally among all users. The results are based on a frame error rate of For frame length 1500 bytes, this corresponds to a BER equal to For frame lengths of 120 bytes, this corresponds to a BER equal to We have then made the approximation than one frame error is caused by one bit error. Below we provide the results of the throughput per user as function of number of users, and the number of users that can be supported as function of throughput per user for IEEE802.11b and IEEE802.11a/g networks. Results are provided for 1500 byte frames, which corresponds to the frame length of video streaming, and for 120 byte frames, which corresponds to the frame length of IP telephony. Results for both basic and RTS/CTS access schemes are included. Finally, results for all users transmitting at the same data rate and results for both fast and slow users are included. OBAN-WP4-SIN-049g-D Page 97 of 177 OBAN Consortium

104 IEEE802.11b All STAs transmitting at the same rate Throughput per user [Mbps] R =11 Mbps b R b =5.5 Mbps R b =2 Mbps R b =1 Mbps Number of STAs Figure 5-20 Throughput as function of number of STAs and data rate for b. Basis access scheme, frame length 1500 bytes Throughput per user [Mbps] Data rate [Mbps] Table 5-6 Number of users that can be supported by an b AP as function of throughput per user. Basis access scheme, frame length 1500 bytes OBAN-WP4-SIN-049g-D Page 98 of 177 OBAN Consortium

105 Throughput per user [Mbps] R b =11 Mbps R b =5.5 Mbps R =2 Mbps b R =1 Mbps b Number of STAs Figure 5-21 Throughput as function of number of STAs and data rate for b. RTS/CTS access scheme, frame length 1500 bytes Throughput per user [Mbps] Data rate [Mbps] Table 5-7 Number of users that can be supported by an b AP as function of throughput per user. RTS/CTS access scheme, frame length 1500 bytes. OBAN-WP4-SIN-049g-D Page 99 of 177 OBAN Consortium

106 Throughput per user [Mbps] R b =11 Mbps R b =5.5 Mbps R =2 Mbps b R =1 Mbps b Number of STAs Figure 5-22 Throughput as function of number of STAs and data rate for b. Basis access scheme, frame length 120 bytes. Throughput per user [Mbps] R =11 Mbps b R b =5.5 Mbps R =2 Mbps b R b =1 Mbps Number of STAs Figure 5-23 Throughput as function of number of STAs and data rate for b RTS/CTS access scheme, frame length 120 bytes Data rate 1 Mbps 2 Mbps 5.5 Mbps 11 Mbps Basic access scheme RTS/CTS access scheme Table 5-8 Number of 96 kbps streams supported by an b AP. Frame length 120 bytes OBAN-WP4-SIN-049g-D Page 100 of 177 OBAN Consortium

107 STAs transmitting at different data rates Throughput per user[mbps] All fast One slow All slow Number of STAs Figure 5-24 Throughput per user as function of number of STAs for users transmitting at different data rate modes. Fast users transmit at 11 Mbps, slow users transmit at 2 Mbps. Basic access scheme. Frame length 1500 bytes. Throughput per user [Mbps] All fast One slow All slow Table 5-9 Number of users that can be supported by a b AP as function of throughput per user. Basic access scheme. Frame length 1500 bytes. OBAN-WP4-SIN-049g-D Page 101 of 177 OBAN Consortium

108 Throughput per user[mbps] All fast One slow All slow Number of STAs Figure 5-25 Throughput per user as function of number of STAs for users transmitting at different data rate modes. Fast users transmit at 11 Mbps, slow users transmit at 2 Mbps. RTS/CTS access scheme. Frame length 1500 bytes. Throughput per user [Mbps] All fast One slow All slow Table 5-10 Number of users that can be supported by an b AP as function of throughput per user. RTS/CTS access scheme. Frame length 1500 bytes. OBAN-WP4-SIN-049g-D Page 102 of 177 OBAN Consortium

109 Throughput per user[mbps] All fast One slow All slow Number of STAs Figure 5-26 Throughput per user as function of number of STAs for users transmitting at different data rate modes. Fast users transmit at 11 Mbps, slow users transmit at 2 Mbps. Basic access scheme. Frame length 120 bytes. Throughput per user[mbps] All fast One slow All slow Number of STAs Figure 5-27 Throughput per user as function of number of STAs for users transmitting at different data rate modes. Fast users transmit at 11 Mbps, slow users transmit at 2 Mbps. RTS/CTS access scheme. Frame length 120 bytes. Data rate All fast users One slow user All slow users Basic access scheme RTS/CTS access scheme Table 5-11 Number of 96 kbps streams supported by an b AP. Frame length 120 bytes. OBAN-WP4-SIN-049g-D Page 103 of 177 OBAN Consortium

110 IEEE802.11a/g All STAs transmitting at the same data rate 15 R b =54 Mbps Throughput per STA [Mbps] 10 5 R b =36 Mbps R b =18 Mbps R b =6 Mbps Number of STAs Figure 5-28 Throughput as function of number of STAs and data rate for g. Basic access scheme, frame length 1500 bytes. Throughput per user [Mbps] Data rate [Mbps] Table 5-12 Number of users that can be supported by an g/a AP as function of throughput per user. Basic access scheme. Frame length 1500 bytes. OBAN-WP4-SIN-049g-D Page 104 of 177 OBAN Consortium

111 Throughput per STA [Mbps] R b =54 Mbps R =36 Mbps b R =18 Mbps b R b =6 Mbps Number of STAs Figure 5-29 Throughput as function of number of STAs and data rate for a/g. RTS/CTS access scheme, frame length 1500 bytes. Throughput per user [Mbps] Data rate [Mbps] Table 5-13 Number of users that can be supported by an g/a AP as function of throughput per user. RTS/CTS access scheme, frame length 1500 bytes. OBAN-WP4-SIN-049g-D Page 105 of 177 OBAN Consortium

112 Throughput per STA [Mbps] R b =54 Mbps R b =36 Mbps R =18 Mbps b R =6 Mbps b Number of STAs Figure 5-30 Throughput as function of number of STAs and data rate for a/g. Basic access scheme, frame length 120 bytes. Throughput per STA [Mbps] R b =54 Mbps R =36 Mbps b R b =18 Mbps R =6 Mbps b Number of STAs Figure 5-31 Throughput as function of number of STAs and data rate for a/g. RTS/CTS access scheme, frame length 120 bytes. Data rate 6 Mbps 18 Mbps 36 Mbps 54 Mbps Basic access scheme RTS/CTS access scheme Table 5-14 Number of 96 kbps streams supported by an a/g AP. Frame length 120 bytes. OBAN-WP4-SIN-049g-D Page 106 of 177 OBAN Consortium

113 STAs transmitting at different data rates Throughput per user [Mbps] All fast One slow All slow Number of STAs Figure 5-32 Throughput per user as function of number of STAs for users transmitting at different data rate modes. Fast users transmit at 11 Mbps, slow users transmit at 2 Mbps. Basic access scheme. Frame length 1500 bytes. Throughput per user [Mbps] All fast One slow All slow Table 5-15 Number of users that can be supported by an g/a AP as function of throughput per user. Basic access scheme, frame length 1500 bytes. OBAN-WP4-SIN-049g-D Page 107 of 177 OBAN Consortium

114 Throughput per user [Mbps] All fast One slow All slow Number of STAs Figure 5-33 Throughput per user as function of number of STAs for users transmitting at different data rate modes. Fast users transmit at 54 Mbps, slow users transmit at 6 Mbps. RTS/CTS access scheme, frame length 1500 bytes. Throughput per user [Mbps] All fast One slow All slow Table 5-16 Number of users that can be supported by an g/a AP as function of throughput per user. RTS/CTS access scheme, frame length 1500 bytes. OBAN-WP4-SIN-049g-D Page 108 of 177 OBAN Consortium

115 Throughput per user [Mbps] All fast One slow All slow Number of STAs Figure 5-34 Throughput per user as function of number of STAs for users transmitting at different data rate modes. Fast users transmit at 54 Mbps, slow users transmit at 6 Mbps. Basic access scheme. Frame length 120 bytes. Throughput per user [Mbps] All fast One slow All slow Number of STAs Figure 5-35 Throughput per user as function of number of STAs for users transmitting at different data rate modes. Fast users transmit at 54 Mbps, slow users transmit at 6 Mbps. RTS/CTS access scheme. Frame length 120 bytes. Data rate All fast users One slow user All slow users Basic access scheme RTS/CTS access scheme OBAN-WP4-SIN-049g-D Page 109 of 177 OBAN Consortium

116 Table 5-17 Number of 96 kbps streams supported by an a AP. Frame length 120 bytes Conclusions for capacity analysis Several conclusions can be drawn from the capacity analysis: The importance of the frame length on the throughput is very apparent. For b, 10 users transmitting 1500 byte packets may each obtain a throughput of 600 kbps using the basic access scheme. If they transmit 120 byte packets, they may only obtain 200 kbps. For the a/g standards, the corresponding numbers are 2.5 Mbps and 550 kbps. Hence, the services and traffic types applied, and the corresponding protocols defining the frame lengths, are very decisive for the throughput capacity of an AP. It will often be necessary to use the RTS/CTS scheme to avoid the hidden terminal problem. When the number of users is small, this will lead to a significant degradation in throughput per user and/or in number of users that can be supported. When the number of users is high, however, the probability of collision increases. For long frames it is therefore advantageous to use the RTS/CTS scheme, as less time is lost due to collisions of short control packets. For short data packets of similar lengths as the control packets on the other hand, capacity is always lost when using the RTS/CTS scheme. For 11 Mbps b and 1500 bytes frames, it is more efficient to use RTS/CTS when 24 or more users are contending for the channel. For 54 Mbps a/g and 1500 bytes frames it is more efficient to use the RTS/CTS scheme when more than 18 users are content. For 120 byte packets, it will always be more efficient to use the basic access scheme. If a user transmitting at a low data rate enters a network of users transmitting at high data rates, the throughput of all users suffers. This is very apparent when the total number of users is relatively low. The throughput of each user is then close to that of a network of only slow users. When the number of fast users increases, however, the relative impact of the slow user is reduced. This is because the percentage of time the slow user gets access to the channel decreases. OBAN-WP4-SIN-049g-D Page 110 of 177 OBAN Consortium

117 6. Measurements 6.1. Introduction Measurements were conducted by Telenor and by TUB to complement the analytical results. The Telenor campaign contains signal strength measurements inside an office building and outdoors with both LOS and NLOS using b equipment. For the outdoor measurements, a directive antenna was used. The TUB campaign contains both signal strength and throughput measurements using IEEE802.11b equipment. Measurements were conducted indoors in a multi office building. The results include both the single-user case and the multi-user case, and the case with no interference and with interference from both microwave ovens and from Bluetooth communications. The two measurement campaigns and their results are presented in the two following sections, while conclusions are summarised in Sec Telenor measurements Introduction In connection with OBAN evaluation it s a challenge to obtain realistic knowledge about WLAN coverage and interference both indoors and outdoors. The consequences of coexistence of outdoor and indoor systems WLAN do it more defiant. The following four tests give some knowledge about the relationship between propagation models and the real conditions in different surroundings: 1) Signal strength measurements and subjective test inside an office building 2) Outdoor test with 14.5 dbi beam antennas in the boarder area between LOS and NLOS 3) Outdoor test with 14.5 dbi beam antennas and LOS 4) Outdoor test with 14.5 dbi beam antennas with optical LOS and some pine trees in the path area. All the test are done with IEEE b WLAN equipment in the GHz band WLAN antennas and measurement equipment In this part a summary of the WLAN and measurement equipment used in the tests are given. Access point: SMC2682W - 11Mbps Wireless Bridge Specifications: Output Power: +13dBm (minimum) Sensitivity: Min.-76dBm for 10E-5; Min.-80dBm for 10E-5 Mobile terminal: SMC2662W V.3-11Mbps Wireless USB Adapter OBAN-WP4-SIN-049g-D Page 111 of 177 OBAN Consortium

118 Specifications: Output Power: +14dBm Sensitivity: 1/2/5.5/11 Mbps:-90/-88/-83/-80 dbm (min) External dipole antenna (standard) Gain: 2.2 dbi Narrow beam antennas: SMCANT-DI145 Figure 6-1 Narrow beam antennas SMCANT-DI145 with 14.5 dbi gain Specifications: Freq Range: GHz Gain: 14.5 dbi Polarization: Elliptical Connector: N Female Beam width (-3dB): 26 Degrees VSWR: <1.22 F/B Ratio: 15 db Impedance: 50 Ohm Length: Max Range Point-to-Point: 14 km 20º 6.0 km 90º 5.0 km 134º 3.0 km 180º 1.5 km Reference signal source: HP 8350 Sweep Oscillator with dipole antenna Signal strength measurement: Advantest R3261A Spectrum Analyzer with dipole antenna As a reference for the tests presented in the following, the manufacturers coverage data are quoted. OBAN-WP4-SIN-049g-D Page 112 of 177 OBAN Consortium

119 Environmental Outdoors: A line-ofsight environment with no interference or Obstruction between the Access Point and users. Indoors: A typical office or home environment with floor to ceiling obstructions between the Access Point and users b Wireless Distance Table Condition Speed and Distance antenna 11 Mbps 5.5 Mbps 2 Mbps 1 Mbps 160 m 195 m 255 m 350 m 72 m 73 m 73 m 75 m Table 6-1 Speed and coverage data for SMC2682W from equipment the specifications Signal strength measurements and subjective test inside office building Description of the actual office building The building consists of lower floor, first floor and second floor. The measurements and subjective tests are mainly carried out at the first floor. Figure 6-2 shows the actual area at the first floor. Outside walls consist of reinforced concrete and bricks. Inside the building there is reinforced concrete pillars and mainly metallic building framework with plaster board walls. The plaster boards are coated with painted woven glass cloth. The panelled ceilings in the corridors consist of separate metallic profiles hanging from the concrete floor. In the other rooms mineral wool plates fasten in a metallic grid hanging from the concrete floor. Between the first floor and second floor there is a 230 mm reinforced concrete floor. The areas at the second floor above the actual area at the first floor are mainly similar to the first floor areas. OBAN-WP4-SIN-049g-D Page 113 of 177 OBAN Consortium

120 Figure 6-2 Location 1) is access point/reference transmitter. Mobile terminal locations 2) -15 OBAN-WP4-SIN-049g-D Page 114 of 177 OBAN Consortium

121 Figure 6-3 The picture is taken from the doorway to the room where the access point is located at the first floor. The end of the corridor is the border of coverage Measurements and subjective tests at the first floor The distances are given as the sum of movements in x and y direction in accordance with table A2. Based on the measurements, the mean received power is estimated to be 7 db below Prmax. The measured signal fluctuation between Prmax and Prmin is described as a combination of Log Normal and Rice Fading. The fluctuations are the effect of small movements of the mobile terminal antenna (a few decimetres), people who moves in the area (at least 0.5 meter from the antennas) and for example doors in the area that are opened and closed. The received signal power is for example given as: λ P = EIRP 10 n log ( d) + G + 20 log L 4π r 10 R 10 ADD where: P r = receiver input signal power in dbm EIRP = equivalent isotropic radiated power in dbm n = attenuation factor (for free space n = 2) d = distance in meter G R = receiver gain in dbi λ = wave length L ADD = additive attenuation caused by obstacles in db Assuming free space propagation he received power in distance one meter from the access point should be: P r = = -24 dbm, OBAN-WP4-SIN-049g-D Page 115 of 177 OBAN Consortium

122 The mean value of the measured signal level one meter from the access point is 10 db below the calculated value for free space case. The main reason for this difference is the multipath conditions in the room. In addition there may be some unforeseen attenuation in the equipment Great changes in the level was observed when people passing a couple of meters from the antennas. Position, ref Figure 6-2 Distance (x,y) [m] Straight distance [m] Prmax [dbm] Prmin [dbm] PRmean, μ [dbm] Subjective test with b AP 2) ) ) ) 5, ) 5, Stable connection 7) 5, < ) (5-2), < ) 5, < (Lab door open mirror) 10) 5, < (Lab door closed) 11) (5-2), < Unstable connection 12) 5, < ) 5, < Unstable connection 14) 5, < ) (5+2), < No connection Table 6-2 WLAN coverage first floor, Measured max, min and mean signal strength. The positions are in agreement with the positions in Figure 6-2. The AP is at position 1) with EIRP = 14 dbm. Figure 6-4The plot shows the received signal strength in dbm in the corridor as function of distance in meters. XX measured signal strength and dotted line calculated values OBAN-WP4-SIN-049g-D Page 116 of 177 OBAN Consortium

123 Along the corridor the signal level is -74 dbm at distance 10 meters from the access point and -81 dbm at distance 30 meters. I.e. the attenuation factor is about n = 1.4, which is less than in free space. The corridor behaves like a lossy waveguide. The equation on page 115 describes approximately the signal level as function of distance at the first floor with following parameters: EIRP = 14 dbm n = 1.4 G R = 2 dbi λ = m = 36 db L ADD The estimated received power is plotted as a dotted line in Figure 6-4. The subjective observations are in good agreement with measurements and specifications for the actual WLAN equipment from SMC. The max distance in the coverage area is less than the half the distance specified in table 1, however. Out of coverage is where connection on all data rates fails. Inside the coverage area the transmitted gross data rate will automatically change in the range 1 11 Mbps Measurements and subjective tests at the second floor Position, ref Figure 6-2 Distance (v,x,y) [m] Straight distance (3D) [m] PRmean, μ [dbm] Subjective test with b AP 2) 3,0, ) 3,4, ) 3,4, Stable connection 5) 3,4, ) 3,4, ) 3,4, No connection 8) 3,0, Table 6-3 WLAN coverage second floor, Measured mean signal strength and subjective test as function of the position in relation to the access point at the first floor. AP is at position 1) EIRP = 14 dbm OBAN-WP4-SIN-049g-D Page 117 of 177 OBAN Consortium

124 Figure 6-5The plot shows the signal strength in dbm at the second floor as function of distance in meters. XX measured signal strength and dotted line calculated values At the second floor three meters straight above the access point antenna at the first floor the signal strength is 59 dbm, which is about 16 db below the level at distance three meters from the access point at the first floor. This difference in signal level is a result of penetration loss through the floor, but also the fact that the antennas have max radiated power in the horizontal plane. The signal level at the second floor is mostly a function of distance, and seems to be less sensitive to either the mobile terminal is in the corridor or inside rooms near by. In this test environments the received level seems to be in good agreement with an n = 4.1 law all over the area. The equation on page 115 describes approximately the signal level as function of distance at the second floor with following parameters: EIRP = 14 dbm n = 4.1 G R = 2 dbi λ = m = 6 db L ADD The estimated received power is plotted as a dotted line in Figure 6-5. OBAN-WP4-SIN-049g-D Page 118 of 177 OBAN Consortium

125 Outdoor test with 14.5 dbi beam antennas in the boarder area between LOS and NLOS Figure 6-6 The radio path between Teleskolen radio tower at Grimstad and a building in the central are of the town Wireless Bridge Master and Slave with antenna gain 14.5 dbi. During the test the antennas and terminal equipment in each end of the link are similar. EIRP in this system is about 13 dbm db = 27.5 dbm assuming zero cable attenuation. This EIRP level is not in agreement with ECC, however, that specifying legal EIRP < 20 dbm. The distance between the terminals is 1330 meter. A big birch touches the optical line of sight near the Wireless Bridge Slave (WB-S) as shown on the map in Figure 6-6. Without leaves on the tree the connection is excellent. With leaves on the tree, however, the bit rate is often reduced and the connection is instable in windy weather. As the photo shows, the clearance between the line of sight and the surrounding terrain is large most of the distance. The parameters in the actual system and as a first approximation assuming free space and L ADD = 0 db: EIRP = 27.5 dbm d = m n = 2 G R = 14.5 dbi λ = m = 0 db L ADD In agreement with equation on page 115, the received signal power is then: OBAN-WP4-SIN-049g-D Page 119 of 177 OBAN Consortium

126 In free space the received power in this situation should be: P r = = dbm With leaves on the tree the terminal is at the boarder of coverage area, and the received power is estimated to about 80 dbm. As a result of multipath the attenuation factor, n > 2. The multipath and obstacles (some leaves on the birch) result therefore in an additional attenuation of about 20 db compared with free space. Figure 6-7 The photo is taken from Teleskolen radio tower at Grimstad towards the building near the central area of the town with the WB-S Outdoor test with 14.5 dbi beam antennas and LOS The configuration here is the same as above. The wireless bridge slave terminal (WB-S) is now located in a small boat 5 km from the wireless master bridge (WB-M). In this case it s absolutely LOS. One problem is the waves on the sea, which influence on the pointing accuracy. As long as the pointing accuracy is within an acceptable margin, however, the connection is excellent. In the SMC antenna/bridge specifications max range Point-to-Point is 14 km. The assumption is then neither obstacles nor multipath. In our case with transmission over sea, multipath can t be neglected. Consequently the attenuation factor is normally n > 2 and the max range less than 14 km. OBAN-WP4-SIN-049g-D Page 120 of 177 OBAN Consortium

127 Figure 6-8 The radio path between Teleskolen radio tower at Grimstad and a boat 5 kilometres from the tower Figure 6-9 The photo is taken from Teleskolen radio tower at Grimstad towards the boat with the WB-S OBAN-WP4-SIN-049g-D Page 121 of 177 OBAN Consortium

128 Outdoor test with 14.5 dbi beam antennas with optical LOS and pine trees in the path area The WLAN connection is established between two buildings at distance 210 meters as shown in Figure The terrain between the buildings is fairly flat, and with a football ground in the middle. In the neighbourhood of each building there are some trees, mostly pine, as shown in Figure 6-11, Figure 6-12 and Figure Both sides of the link are equipped with 14.5 dbi beam antennas. It s impossible to establish connection with antennas in a height where the radio beam hit the crown of the trees. To establish a connection it was necessary to lower the antenna position to a level such that the radio beam was shot between the trunks. To maintain a stable connection an EIRP = 27.5 dbm was necessary! In this case it seems to be limited number of obstacles between the access point and terminal point. The distance is only 210 meters. Still it s impossible to establish a stable connection with a legal EIRP and antennas with 14.5 dbi gain. Figure 6-10 The map shows the buildings with the access point and terminal point OBAN-WP4-SIN-049g-D Page 122 of 177 OBAN Consortium

129 Figure 6-11 The camera is placed near the terminal antenna. The picture is taken towards the access point Figure 6-12 This picture is taken from the football ground in a position near the optical line between the antennas. The picture is taken towards the terminal antenna OBAN-WP4-SIN-049g-D Page 123 of 177 OBAN Consortium

130 Figure 6-13 This picture is taken from the football ground in a position near the optical line between the antennas. The picture is taken towards the access point antenna Conclusions The examples above show how the WLAN coverage areas depend on the surroundings. The surroundings often changes all the time; people, trees, etc. are in movement. It seems to be hard to calculate the WLAN coverage and interference based on a few simple parameters. May be it s better to got knowledge about indoor coverage by field trials in different typical buildings. For outdoor coverage different build-up areas with different vegetation and topology is actual. A standardized way for collecting knowledge about propagation in different model areas seems to be important. For example test/measurements could be done by means of standardised test WLAN equipment with well defined output power, sensitivity and antenna gain. The measure results could be presented in maps/drawings with supplementary descriptions and photos. OBAN-WP4-SIN-049g-D Page 124 of 177 OBAN Consortium

131 6.3. TUB measurements Introduction This Section describes the so far conducted measurements which were done by TUB within WP4 Activity 2. Basically, single and multi-user scenarios have been setup for WLAN/WiFi b usage in the presence and absence of OBAN and non-oban interference in close frequency bands (basically microwave ovens and Bluetooth devices). Measurements took place on the IP level in terms of effective throughput, response times and transaction rates. Additionally, signal-level measurements took place, that is to say, SNR measurements related to the signal receive power and noise level. Most of the traffic was UDP-based Hardware constrains All measurements were conducted using the following equipment: Mobile Nodes IBM T40 Laptops with built-in WLAN (centrino) adapters o No external antennas were attached to the client adapters Cisco Aironet a/b/g client adapters o No external antennas were attached to the client adapters Cisco b/g Aironet AP 1200 access points o For the access points the standard antennas which are included with the delivered hardware will be used Constraints and Scenarios This Chapter describes the IP performance measurements and different scenarios how the IP performance tests were conducted. The main objective of setting up and running these scenarios is to rebuild a more or less realistic environment which covers as many real-life use cases as possible. On the radio side we have conducted measurements with regard to SNR and signal quality for the described scenarios. On the IP performance side throughput for each individual end-device in addition to response times and transaction rates have been recorded. The following scenarios describe measurements with up to several mobile nodes Scenarios for IP performance measurements Assuming the Visitor is walking along the footbath and he is going to establish a connection to check his mails or to get an idea where he actually is and to find a map in the internet to find the way back to the hotel. During his walk he will probably pass different RGWs (Residential Gateways) in different environments. Sometimes he will get excellent link quality associated with a very good IP performance, sometimes may not. However, the reason is because of the changing conditions between the Visitor and the RGW. The case to get a real LOS (line of sight) connection is occurring very seldom. Therefore, we defined three different basic scenarios which intents to cover situations you will frequently meet in real life. The scenarios containing: Measurements with line of sight connection Measurements with non line of sight connection with walls and if possible also trees in between Measurements with non line of sight connection with walls and if possible also trees and human beings in between These basic scenarios do not include all the possible variations of situations in a real environment. So in addition we were defining four different test cases to cover a broader set of eventualities. OBAN-WP4-SIN-049g-D Page 125 of 177 OBAN Consortium

132 The four different test cases for the scenarios: Measurements with 1 to 6 users in parallel Interference Measurements RGW stand alone and together with Bluetooth (only makes sense if b or g is used) with several different RGW with low distance in between Measurements with different distances In conclusion, the three scenarios in conjunction with the four test cases resulting in twelve basic test cases including the specific conditions of each test case The basic scenarios Scenario 1 Measurements with line of sight connection The first scenario is very basic and includes a quasi LOS connection. There is no wall between the RGW considered and the Visitor. For the OBAN environment this should be the ideal use case. There is nothing else in between the user s terminal and the RGW. How we take care about different wall thicknesses and building materials may affecting link quality and IP performance is described below. Figure 6-14 Scenario1 LOS between RG and Visitor Scenario 2 Measurements with non line of sight connection with walls and trees in between The second scenario is similar to the first scenario mentioned above. It is different due to walls and if possible also plants and trees between the user s terminal and the next RGW. It is expected, that the performance will decrease due to the barriers within the wireless link, mainly due to reflections and interferences. We choose this scenario to simulate situations in urban environments. OBAN-WP4-SIN-049g-D Page 126 of 177 OBAN Consortium

133 Figure 6-15 Scenario 2 NLOS between RG and Visitor with walls and if possible also trees in between Scenario 3 Measurements with non line of sight connection with walls, and if possible also trees and human beings in between The third scenario is describing the same environment as described before, apart from the fact that there are human beings in addition to the walls, plants and trees in between the wireless link. We choose this scenario to simulate situations in leisure parks or in the near of public parks and streets with many green surrounding the places. Figure 6-16 Scenario 2 NLOS between RG and Visitor with walls, trees and human beings in between OBAN-WP4-SIN-049g-D Page 127 of 177 OBAN Consortium

134 Detailed Scenario Descriptions and Measurement Results Single User Scenarios The measurements in those sections are of two types: IP-level with mostly UDP traffic and signal-level (SNR). Data traffic is mostly downstream, that is to say, the traffic flows from the 1 st end point which is connected to the WLAN AP, and the termination points of the segments (2 nd end point) are the client devices. The access scheme used is CSMA/CA, due to the fact that mainly b APs were used, regardless of the client end points, the RTS/CTS wasn t needed. Concerning data rates, the scheme used was to flood the whole available bandwidth, that is maximize the data rate, and clients were competing rather than cooperating to maximize their used rates. PURE WLAN b and g - Point-to-point measurements of traffic for single user connected to a WLAN AP and performing handovers, mobility between 2 APs with Case A: same signal strength (configurable) o Case 1: stationary Signal to Noise Ratio (SNR) values throughput fluctuation delay variation o Case 2: mobile at pedestrian speed SNR values throughput fluctuation delay variation Case B: different signal strength (configurable on hardware) o Case 1: stationary SNR values throughput fluctuation delay variation o Case 2: mobile at pedestrian speed SNR values throughput fluctuation delay variation OBAN-WP4-SIN-049g-D Page 128 of 177 OBAN Consortium

135 Figure 6-17 SNR of an b AP with an average value of dbm in absence of Bluetooth SNR of a WLAN AP (802.11b), same as the one used for most single and multi-user measurements in this set of experimental trials. It achieves and average value of dbm which shall be compared to the value of the same parameter in the case when BT is on (induced interference for a sufficient period from a single source). WLAN Interference with Bluetooth (stationary source) - Measurements due for throughput drop on the IP level with UDP traffic upon turning on Bluetooth; additionally, signal level measurements have been conducted to demonstrate the SNR drop when Bluetooth is turned on. Please note that SNR is measured in dbm and that the x-axis of all graphs corresponds to measured time units, that is instances when measurements took place rather than seconds; tests were conducted e.g. over a duration of 7 minutes with Bluetooth interference either constantly present with a high data rate e.g. for a file transfer and in other cases device search or short signalling took place to emulate temporary interference. Case A: same signal strength (configurable) o Case 1: stationary SNR values throughput fluctuation delay variation o Case 2: mobile at pedestrian speed SNR values throughput fluctuation delay variation Case B: different signal strength (configurable on hardware) o case 1: stationary SNR values throughput fluctuation delay variation o case 2: mobile at pedestrian speed SNR values throughput fluctuation delay variation OBAN-WP4-SIN-049g-D Page 129 of 177 OBAN Consortium

136 Figure 6-18 SNR of an b AP with an average value of dbm in presence of Bluetooth The figure above shows the SNR of the WLAN AP (802.11b) in the presence of Bluetooth interference. A significantly higher value is recorded ( dbm) compared to the case of a pure WLAN signal where a value of dbm was recorded. A consequent measurement in the figure shows even a slightly higher value of dbm. It is difficult to explain that the SNR (which should in fact be the signal to noise plus interference ratio) increases when interference is added. The result is probably linked to the way the equipment measures the SNR. The co-existence of Bluetooth and WLAN and its consequences have to be analyzed from the effective throughput point of view rather than only signal strength due to the following reasons: WLANs use 20 MHz DSSS (Direct Sequence Spread Spectrum) or OFDM whereas Bluetooth uses FHSS (Frequency Hopping Spread Spectrum) with signal bandwidth 1 MHz. The degree of interference is not very severe but is significant because it pushes the effective throughput to a pretty low level compared the initial one; e.g. from 12.5 Mbps down to 4-5 Mbps for a single user case and single source of interference for the user. The relationship between the physical layer and the IP-throughput where packet collisions are the additional factor in addition to backoff times, hardware properties shall be analyzed in greater detail to draw some scientific conclusions OBAN-WP4-SIN-049g-D Page 130 of 177 OBAN Consortium

137 Figure 6-19 SNR in presence of Bluetooth Figure 6-20 Throughput fluctuation with interference OBAN-WP4-SIN-049g-D Page 131 of 177 OBAN Consortium

138 Figure 6-21 Delay variation fluctuation for a mobile user using and exposed to interference Figure 6-22 Throughput fluctuations multiple source interference OBAN-WP4-SIN-049g-D Page 132 of 177 OBAN Consortium

139 Figure 6-23 Delay variations effects WLAN/BT WLAN Interference with a Microwave Oven (stationary source) Point to point measurements of traffic for single user connected to a WLAN AP and performing handovers, mobility between 2 APs with Case A: same signal strength (configurable) o Case 1: stationary SNR values throughput fluctuation delay variation o Case 2: mobile at pedestrian speed SNR values throughput fluctuation delay variation Case B: different signal strength (configurable on hardware) o Case 1: stationary SNR values throughput fluctuation delay variation o Case 2: mobile at pedestrian speed SNR values throughput fluctuation delay variation OBAN-WP4-SIN-049g-D Page 133 of 177 OBAN Consortium

140 Note: All above scenarios were indoor measurements; walls: material brick, with plastic paint coating and thickness of 12 cm; these measurements were based on a single user scenario The diagrams to follow depict the impact of turning on a microwave oven in the vicinity of a WLAN AP of the type b, with clients attached to the WLAN AP and having the available bandwidth flooded with UDP (downlink): Microwave-WLAN test case 1: The topology is depicted below followed by the recording of the observed drop in throughput upon turning on the microwave oven at the time instants: 1 min and turning it off at the instant 3 minutes (2 minutes overall duration of active microwave radiation); The traffic is UDP for downlink with traffic flowing from the AP to the client device, flooding the whole bandwidth portion available. Figure 6-24 Topology of test case 1 and 2 Figure 6-25 Throughput for one user under influence of microwave oven interference. The interference is turned on after 1 minute and turned off after 3 minutes. OBAN-WP4-SIN-049g-D Page 134 of 177 OBAN Consortium

141 Microwave-WLAN test case 2: The same topology is used for the test case 2, but the microwave oven is turned on 2 times (at 2 different intervals), from 1 minute till 2 minutes, and from 3 minutes to 4 minutes; this is reflected very well in the figure to reflect the almost 25 % temporary loss in bandwidth. Figure 6-26 Throughput for one user under influence of microwave oven interference. The interference is turned on after 1 minute, off after 2 minutes, on again after 3 minutes and finally off after 4 minutes. Microwave-WLAN test case 3: The topology with the spacing is depicted in the figure below. Figure 6-27 Topology for test cases 3 and 4. OBAN-WP4-SIN-049g-D Page 135 of 177 OBAN Consortium

142 Figure 6-28 Throughput for one user under influence of microwave oven interference. The interference is turned on after 1 minute and off after 3 minutes. Microwave-WLAN test case 4: Figure 6-29 Throughput for one user under influence of microwave oven interference. The interference is turned on after 1 minute, off after 2 minutes, on again after 3 minutes and finally off after 4 minutes. OBAN-WP4-SIN-049g-D Page 136 of 177 OBAN Consortium

143 Microwave-WLAN test case 5: The topology is depicted in the figure below. Figure 6-30 Topology for test cases 5 and 6. Figure 6-31 Throughput for one user under influence of microwave oven interference. The interference is turned on after 1 minute and off after 3 minutes. OBAN-WP4-SIN-049g-D Page 137 of 177 OBAN Consortium

144 Microwave-WLAN test case 6: Figure 6-32 Throughput for one user under influence of microwave oven interference. The interference is turned on after 1 minute, off after 2 minutes, on again after 3 minutes and finally off after 4 minutes. Microwave-WLAN test case 7: Topology: Figure 6-33 Topology for test case 7. OBAN-WP4-SIN-049g-D Page 138 of 177 OBAN Consortium

145 Figure 6-34 Throughput for one user under influence of microwave oven interference. The interference is turned on after 1 minute, off after 2 minutes, on again after 3 minutes and finally off after 4 minutes. Some specifics to the settings: The microwave oven used was operated with a dissipation power of 750 Watts and at a frequency of 2.45 GHz. We observed that what matters most is the distance from the WLAN AP to the microwave oven. It was also noted that the percent of bandwidth lost during microwave operation was always in the range of 25-30%. Beyond a distance of 7 or 8 m from the AP, the influence of a microwave oven becomes much weaker on the IP throughput Multi-user Scenarios Similar setup to the above with the additional functional descriptions: Remark: The topology is approximately circular with distances used in meters to designate how far the end devices are from the WLAN AP. This is depicted in the diagram below. The presence of barriers in between has been explained for each distance throughout the text; [in short, within 3-4 meters distance, no barriers or lass door, above 4 m, 2 glass doors or a stone wall, to emulate real life situations] OBAN-WP4-SIN-049g-D Page 139 of 177 OBAN Consortium

146 Figure 6-35 Setup for multi-user scenario Scenario 1: 2 residential users connected to WLAN, no Bluetooth When one user is 2 meters away from the access point and the other is 7 meters away; this case is well suited for the scenario when one user is at home and the other is roaming outside; the test has been conducted once with a glass door as a barrier and the second time with a wooden door as a barrier to match real-life scenarios. Another practical case where such a scenario is encountered is when 2 home users are in different rooms, one within the room where the access point is located and the other in a neighbouring room. Figure 6-36 Throughput OBAN-WP4-SIN-049g-D Page 140 of 177 OBAN Consortium

147 Figure 6-37 Response time Scenario 2: 3 residential users connected to WLAN, no Bluetooth Tests have been conducted with 3 end devices, namely laptops with Intel internal wireless LAN adapters and Cisco Aironet WLAN cards. We did measurements for the following distances in meters of end points to the access point: [each table will show the system behaviour via a throughput plot for the various clients and a delay variation] Figure 6-38 Throughput (1,1,1) OBAN-WP4-SIN-049g-D Page 141 of 177 OBAN Consortium

148 Figure 6-39 Response time (1,1,1) Figure 6-40 Throughput (1,1,3) OBAN-WP4-SIN-049g-D Page 142 of 177 OBAN Consortium

149 Figure 6-41 Response time (1,1,3) Figure 6-42 Throughput (1,4,6) OBAN-WP4-SIN-049g-D Page 143 of 177 OBAN Consortium

150 Figure 6-43 Response time (1,4,6) Figure 6-44 Throughput (4,4,4) OBAN-WP4-SIN-049g-D Page 144 of 177 OBAN Consortium

151 Figure 6-45 Response time (4,4,4) Figure 6-46 Throughput (4,6,6) OBAN-WP4-SIN-049g-D Page 145 of 177 OBAN Consortium

152 Figure 6-47 Response time (4,6,6) Figure 6-48 Throughput (6,6,6) OBAN-WP4-SIN-049g-D Page 146 of 177 OBAN Consortium

153 Figure 6-49 Response time (6,6,6) Scenario 3: Same as Scenario 2, but with Bluetooth interference Description: The so-far-agreed-upon method for authenticating roaming/mobile/visiting users in OBAN is EAP-SIM initiated on a mobile phone which is either the OBAN terminal itself in some rare cases or an additional device which interfaces the OBAN terminal such as a PDA or laptop via Bluetooth. This has some direct implications on the environment where the OBAN service usage takes place due to the very close used frequency ranges by BT and WLAN. We conducted some tests using the following techniques (to simulate Bluetooth usage): Using a Motorola A830 UMTS mobile phone to search for devices and generate active BT-based search for a duration of about (12 on the average seconds); 2 cases have been tested: The A830 is stationary during the test period The A830 is moving at a slow pace to simulate a walking user authenticating and working most probably on his PDA; in this case the most important piece of information is the fact that the distance between the WLAN AP and the BT device is slightly growing or lessening (in our case growing) Using a constant BT interference generator on an Acer laptop trying to detect all devices in its vicinity (constant traffic with a low bandwidth) Using 2 BT supporting laptops (of the used 3 overall end-devices) to transfer a large file to generate heavy payload traffic and see its impact on the WLAN throughput with the various test-case topologies described earlier. The same cases (topologies as in section 4.2.2/scenario 2) will now be plotted with an average 12-sec Bluetooth interference (numbers are end point distances in meters of the respective terminals from the access point; distances below 4 meters are indoor; 5 m is outdoors with a glass or wooden door, 6, 7 m is definitely outside with a stone wall or 2 glass/wooden barriers in between, based on the environment where we conducted our tests: OBAN-WP4-SIN-049g-D Page 147 of 177 OBAN Consortium

154 Figure 6-50: Throughput (1,1,1) Figure 6-51: Response time (1,1,1) OBAN-WP4-SIN-049g-D Page 148 of 177 OBAN Consortium

155 Figure 6-52 Throughput (1,1,3) Figure 6-53 Response time (1,1,3) OBAN-WP4-SIN-049g-D Page 149 of 177 OBAN Consortium

156 Figure 6-54 Throughput (1,4,6) Figure 6-55 Response time (1,4,6) OBAN-WP4-SIN-049g-D Page 150 of 177 OBAN Consortium

157 Figure 6-56 Throughput (4,4,4) Figure 6-57 Response time (4,4,4) OBAN-WP4-SIN-049g-D Page 151 of 177 OBAN Consortium

158 Figure 6-58 Throughput (4,6,6) Figure 6-59 Response time (4,6,6) OBAN-WP4-SIN-049g-D Page 152 of 177 OBAN Consortium

159 Figure 6-60 Throughput (6,6,6) Figure 6-61 Response time (6,6,6) Scenario 4: Constant Bluetooth signals (throughout all the simulation, not limited in time) This case corresponds to INDOOR Bluetooth usage with cases such as file sharing, device detection, etc. Therefore, different tests where Bluetooth has been activated for different intervals of time and with different loads (volume of BT traffic) have been conducted. The results are depicted below to show various impacts of Bluetooth indoors with multiple users. Some tests involve up to 4 end-devices. This is very much inline with a scenario when a home user has some personally known guests indoors as visiting users using the public VLAN of the AP but physically as good as collocated with the home user. Results are shown below: OBAN-WP4-SIN-049g-D Page 153 of 177 OBAN Consortium

160 Case A: 4 end-devices, BT started towards the end of the 7 min test interval, which allows starving clients to achieve more throughput on the expense of the 2 dominant ones as seen in the figure; BT blunts all signals and then allows a more fair distribution, with a little overall aggregate throughput. Figure 6-62 Throughput 4 end devices and BT Figure 6-63 Response time (4 end devices and BT) OBAN-WP4-SIN-049g-D Page 154 of 177 OBAN Consortium

161 Case B: A very good test case, which is showing fair behaviour and distribution of capacity among the 4 clients especially after BT is on. Figure 6-64 Throughput (4 end devices with BT) Figure 6-65 Response time (4 end devices with BT) Case C: Same as in B OBAN-WP4-SIN-049g-D Page 155 of 177 OBAN Consortium

162 Figure 6-66 Throughput (3 end devices with BT) Figure 6-67 Response time (3 end devices with BT) Case D: variation in effective throughput for clients is reduced after turning on Bluetooth OBAN-WP4-SIN-049g-D Page 156 of 177 OBAN Consortium

163 Figure 6-68 Throughput (3 end devices with BT) Figure 6-69 Response time (3 end devices with BT) OBAN-WP4-SIN-049g-D Page 157 of 177 OBAN Consortium

164 Some special case analyses on particular issues concerning performance and coverage WLAN clients coordination concerning back-off times and channel allocation and sharing to minimize packet collisions Due to the CSMA nature of the channel, the backoff time is the deciding factor for sending stations; in other words, upon the presence of multiple WLAN supporting clients, backoff times and other parameters have to be coordinated in such a way to minimize collisions. This takes a certain amount of time as shown in the figure below: Figure 6-70 Effective aggregated bandwidth in a 3-user scenario Effect of strong Bluetooth Interference on a single client, leading sometimes to deep fades and very high delays This is depicted in the figure below showing a scaled combined graph with the throughput and the delay variation; a very unstable signal level and throughput level severely impacts applications running on the end device. Such a case is quite realistic as the figure shows; deep fades with almost complete bandwidth loss or extremely high delays can be observed: OBAN-WP4-SIN-049g-D Page 158 of 177 OBAN Consortium

165 Figure 6-71 WLAN signal fluctuation with strong BT interference Conclusions of the TUB measurements The conclusions of the TUB measurements can be summarised as follows: For the measurements done with interference from microwave ovens, we concluded that what matters most is the distance from the WLAN AP to the microwave oven. It was also noted that the percent of bandwidth lost during microwave operation was always in the range of 25-30%. We used mainly UDP traffic schemes; the reason is that in practice it has been verified that UDP and TCP are identical when the wireless channel is relatively good, whereas TCP has a somewhat higher throughput data rate under congested channel conditions; which shall not be the case for OBAN, normally. The hardware supplier plays a significant role in terms of performance, because some algorithms such as reducing the amount of transmitted data as a reaction to interference is totally hardware dependent. The number of users that an access point can support is quite limited if a certain level of performance is to be preserved. In our case, more than 4 competing users for one access point already showed degraded performance in terms of per-user effective throughput. Measurements of performance serve quite well as a support for network planning in terms of capacity analysis, and estimation of the potential of an OBAN-like network in supporting users and delivering an acceptable level of quality. For the purpose of getting a clear view on what to expect from the OBAN-WP4-SIN-049g-D Page 159 of 177 OBAN Consortium