Evaluation of HIPERLAN/2 Scalability for Mobile Broadband Systems

Evaluation of HIPERLAN/2 Scalability for Mobile Broadband Systems Ken ichi Ishii 1) A. H. Aghvami 2) 1) Networking Laboratories, NEC 4-1-1, Miyazaki, Miyamae-ku, Kawasaki 216-8, Japan Tel.: +81 ()44 86 226, Fax: +81 ()44 86 8494, e-mail: ishii@bc.jp.nec.com 2) Centre for Telecommunications Research, King's College London Strand, London WC2R 2LS, UK Tel: +44 ()2 7848 2898, Fax: +44 ()2 7848 2664, e-mail: Hamid.Aghvami@kcl.ac.uk ABSTRACT Wireless LAN technologies like HIPERLAN/2 are getting very important to realize mobile broadband systems. However, a shortage of allocated RF channels for the wireless LANs causes interference signals that seriously degrade the system throughput in multi-cell environments. There are two types of approaches to improve the throughput. One is a system level approach which improves CIR level at receiver-sides and the other is a transceiver level approach which improves the interference tolerance of transceivers. In this paper, we perform computer simulations in order to evaluate the throughput of HIPERLAN/2 in the multi-cell environments and clarify the quantitative throughput improvements obtained by the two approaches. Through the simulations, it is shown that HIPERLAN/2 system can realize about 21 Mbps of the throughput in European spectrum allocation. However, in US and Japan, 3 db and 11 db improvements in the interference tolerance are required to achieve the same throughput as in Europe, respectively. 1. Introduction Recently, demand for high-speed Internet access is rapidly increasing and a lot of people enjoy broadband wired Internet access services using ADSL (Asymmetric Digital Subscriber Line) or cable modems at home. On the other hand, the cellular phone is getting very popular and users enjoy its location-free and wire-free services. The cellular phone also enables people to connect their laptop computers to the Internet in location- and wirefree manners. However, current cellular systems like GSM (Global System for Mobile communications) can provide much lower data rates compared with those provided by the wired access systems, over a few Mbps (Mega bit per second). Even in the next generation cellular system, UMTS (Universal Mobile Telecommunications System), the maximum data rate of its initial service is limited up to 384kbps, therefore even UMTS cannot satisfy users expectation of highspeed wireless Internet access. Hence, recently, Mobile Broadband System (MBS)[13][14] is getting popular and important and wireless LAN (Local Area Network) such as ETSI (European Telecommunication Standardization Institute) HIPERLAN (HIgh PErformance Radio Local Area Network) type2 (denoted as HL/2 in the rest of this paper)[1] and IEEE (Institute of Electrical and Electronics Engineers) 82.11[7][8][9] is regarded as a key technology to realize the high-speed wireless access in MBS. Currently, the MBS services are only available at limited small areas covered by one or a few APs (Access Points), e.g. airport lounges or cafés, and portable computer users mainly enjoy the services. In the near future, much smaller terminals like PDAs (Personal Digital Assistance) or cellular phones will have wireless LAN interfaces. Such sophisticated terminals will encourage much more people to enjoy the MBS services. Then MBS will be required to cover much wider service areas such as whole airport terminals or large shopping centers. Because coverage of an AP, i.e. cell, of the wireless LAN is much smaller than that of the cellular systems, the large-scale MBS should be composed of multiple cells. The most serious problem in the multi-cell MBS is degradation of system performances in terms of throughput or delay, which is caused by interference signals from other cells. In order to overcome the performance degradation, there are two approaches, system level and transceiver level approaches. The system level approach improves CIR (Carrier to co-channel Interference signal Ratio) level at receiver-sides. Key factors related to this approach are the number of available channels, location of APs in the service area, frequency selection at each AP and propagation characteristics. On the other hands, the transceiver level approach improves the interference tolerance of transceivers, where error rate characteristics of physical layer and the link adaptation are key factors. To design the large-scale MBS, it is important to appropriately combine the technologies in the two approaches. In this paper, we evaluate the system performances of HL/2 in the multi-cell environments. Through the evaluations, we show quantitative improvements of the performances obtained by the two approaches. Then we examine appropriate combinations of the technologies and clarify requirements for the technologies to realize the large-scale MBS based on HL/2, according to spectrum allocation regulations in Europe, US and Japan. 2. ETSI HIPERLAN type2 (HL/2) 2.1. Protocol stack HL/2 is a part of ETSI BRAN (Broadband Radio Access Network) activities and it includes specifications of PHY (PHYsical) layer, DLC (Data Link Control) layer and CL (Convergence Layer)[1]. MAC (Medium Access Control), EC (Error Control) and RLC (Radio

Link Control) are defined as functions of the DLC layer[4][]. It is specified to use Ethernet CL to transport IP packets in HL/2. In order to evaluate the performances of HL/2 for IP services, we use the Ethernet CL in our evaluations[2][3]. 2.2. The number of available channels In Europe, total frequency bandwidth allocated to wireless LAN in GHz band is 4MHz and the number of available RF channels for HL/2 is 19. In US, the allocated bandwidth is 3MHz and 12 RF channels are available. In Japan, the allocated bandwidth is 1MHz and 4 RF channels are available[1]. 2.3. PHY data rates In HL/2, seven kinds of data rates, 6/9/12/18/27/36/4 Mbps, are defined as its PHY data rate. The lower data rate is the more robust for interference signals. Because the data rate of 4Mbps is optional in HL/2 specification, we do not evaluate the rate in this paper. 2.4. Overheads of HL/2 To evaluate performance of communication systems, it is necessary to consider about overheads in each layers. Each layer adds its own additional information to each transmitting data, and the additional information becomes overhead and degrades the system performance. In this paper, we focus to evaluate HL/2 performance in IP network, therefore we consider the overheads for transmitting IP packets using HL/2. And we do not consider overheads of the higher layers, including IP headers. In the Ethernet CL, we have to consider three kinds of overheads. First overhead is an overhead of Ethernet headers. In the Ethernet CL, IP packet is handled as a payload of an Ethernet frame. As shown in Figure 1, SSCS-PDU (Service Specific Convergence Sublayer Protocol Data Unit) is constructed by 6 bytes of the destination address field, 6 bytes of the source address field and 2 bytes of the type/length field[3]. These fields are added to each payload. Distination Address 6 bytes SSCS-PDU (6-14 bytes) Source Type/ Payload (IP Packet) Address Length 6 bytes 2 bytes 46 - bytes Figure 1: Ethernet frame (SSCS-PDU) The second overhead is the padding (PAD) field and the trailer field for SAR (Segmentation And Reassembly) function. And the third overhead is the SAR header whose length is 12 bits. In DLC layer, DLC header and CRC (Cyclic Redundancy Code) field are added to each SAR-PDU. The DLC header is 12 bits length and the CRC field is 24 bits length. Therefore the overhead of DLC layer is 4. bytes length and the total length of DLC-PDU becomes 4 bytes. In the MAC layer, BCH (Broadcast CHannel), FCH (Frame CHannel), ACH (Access feedback CHannel) and RCH (Random CHannel) do not convey any user data therefore these channels also decrease the system performance. As shown in Figure 2, BCH and ACH have fixed duration of 2 µsec and 12 µsec, respectively. FCH and RCH have variable durations, which depend on the number of MHs in each cell and the number of PDUs carried by each MAC frame. In this paper, we fixed the duration of FCH to 36 µsec and do not take the overheads of RCH into account. 2 ms BCH FCH ACH 2 µsec 12 µsec DL Phase UL Phase RCHs Figure 2: Basic MAC frame structure There are also some overheads in PHY layer[6]. Preambles of PHY layer are added and some gaps to switch the PHY layer between transmitting and receiving modes are required. However these overheads are very small therefore we do not consider them in this paper. When the length of IP packets is fixed to bytes total percentage of these overheads can be calculated as 16.2 %. Therefore, if 36 Mbps is selected as the PHY data rate, the maximum throughput becomes about 3 Mbps even if no error occurs in the PHY layer. 3. Key technologies and environmental factors This section briefly describes technologies and environmental factors in the system and the transceiver level approaches on which we focus to perform evaluations. 3.1. System level approach 3.1.1. Environmental factors There are many kinds of environmental factors that influence the CIR level at the receiver-sides, e.g. cell structure, the number of channels and the path loss component. In this paper, we evaluate the influences caused by effects of the number of channels and the path loss component on the CIR levels as the environmental factors. 3.1.2. System level technology As the system level technologies, frequency selection scheme is a key technology to dominate the system performances and several kinds of frequency selection schemes can be adopted. In this paper, we compare the system performances obtained by Fixed Frequency Selection (FFS), Random Frequency Selection (RFS) and Dynamic Frequency Selection (DFS). In FFS, channels (frequencies) are allocated to each cell as shown in Figure 3. This allocation is almost ideal to minimize the interference signal levels among the cells and realizes the maximum system performances. Therefore we apply FFS as a default frequency selection scheme in our evaluations. In RFS, each AP randomly selects a channel at the initial phase of each simulation. In the specification of HL/2, though protocols and frame formats for DFS are defined, the algorithm of DFS is out of scope. There are some studies which evaluate the effect of DFS in HL/2[11]. In this paper, we assume an

algorithm described below, where each AP selects a channel with the least interference level. Step 1: APs are randomly ordered. Step 2: According to the order, select an AP. Step 3: The AP estimates total power of interference signals from other APs which have already selected a channel and selects a channel with the least interference signal power. (Repeat step 2 and 3 until all APs select their channels.) Interference signals from MHs are not take into account in this algorithm. And we assume that each AP selects a channel only once and never changes the channel during simulations. Figure 3: Fixed frequency selection (4 channels) Table 1: CIR ranges for Link Adaptation CIR range PHY data rate CIR < 3dB 3 db CIR < 8dB 8 db CIR < 1dB 1 db CIR < db db CIR 4. Simulation Environments and Parameters 4.1. Simulation Environments 4.1.1. Cell structure Each AP forms an own cell as its service area. In the simulation, square cells are used and each AP is located at the center of each cell. 144(12 x 12) of APs are placed and distribution of MHs (Mobile Hosts) is uniform. 4.1.2. PER (PDU Error Rate) performance To emulate PDU losses in the PHY layer, we use PER (PDU Error Rate) performances shown in Figure 4[1]. Receivers decide the loss of each receiving PDU according to the PER performance and average CIR of the PDU. Some points of the graphs in Figure 4 are extrapolated from the original data shown in [1]. The maximum PER and the minimum PER are assumed 9 % and %, respectively. 1 3.2. Transceiver level approach In multi PHY data rates supporting systems, a technology called as link adaptation selects appropriate PHY data rates in accordance with receiving signal qualities. The link adaptation is one of key technologies to improve the system performances. In the transceiver level approach, we evaluate the system performance improvement produced by the link adaptation. 3.2.1. Since algorithm of the link adaptation is out of scope of the HL/2 specifications, there have been some studies of the link adaptation algorithm for HL/2[4]. In our simulation, we use the following algorithm. Step 1: Based on the relations of the throughput and the CIR level for each PHY data rate [1][12], we define CIR ranges for each PHY data rate as shown in Table 1. Step 2: After the frequency selection at all APs, each MH calculate its expected CIR level based on estimated power of desired signal from its designated AP and estimated total power of interference signals from other APs which use the same channel with the MH. Step 3: According to the expected CIR level and Table 1, each MH decides its PHY data rate. In the interference signal power estimation of Step 2, we consider that the average interference signal power from MHs in a cell can be represented by the interference signal power from the AP of the cell, because we assume the MHs are uniformly distributed in the cell. PDU Error Rate (PER).1.1 9Mbps.1 1 2 2 3 CIR (db) Figure4:PDUErrorRatevs.CIR 4.2. Simulation parameters Some important simulation parameters are summarized in Table 2 and Table 3. These tables indicate parameters fixed in all simulations and parameters varied in some simulations, respectively. The values denoted by bold characters in Table 3 are used as default value of the parameter.

Table 2: Fixed parameters Parameters Value Number of Access Points 144(12 x 12) Number of Mobile Hosts 144 Mobile Host Location Uniform IP Packet Length bytes Noise Level -9dBm Transmitting power dbm max Antenna Gain dbi Target level of -dbm Transmitting Power Control Traffic Downlink 7% Uplink 3% IP Packet lifetime ms Table 3: Variable parameters Parameters Value Number of channels 4, 9, 16 Cell radius 3m, 1m, m Path loss component (α) 2., 3.,3. Frequency Selection Fixed, Random,Dynamic PHY Data Rate Without link adaptation (DL/UL phase) (BCH/FCH/ACH) With link adaptation 6 to (DL/UL) (BCH/FCH/ACH). Simulation Results.1. Effect of system level approach.1.1. The number of channels Figure shows relations of throughput and the average delay in the different number of channels. The throughput and the delay are measured in IP level. As in Figure, increase of the number of channels can improve the system performances, because the increase enlarges the reuse distance of the same channel and then the interference signal level decreases at receiver-sides. For reference, the performance without PDU error is also shown in the figure. 2.1.2. The path loss component In Figure 6, the path loss component, α, for the desired signal is fixed as 2. and α for the interference signals is varied from 2. to 3.. The increase of α for the interference signal improves the performances as show in Figure 6 because the interference signal level decreases at the receiver-sides. In Figure 7, α for the interference signal is fixed as 2. and α for the desired signal is varied from 2. to 3.. With the increase of α for the desired signal, the desired signal level decreases and the performances are significantly degraded as shown in the figure. These contrastive effects of α result interesting performances when α for the desired signals and α for the interference signals are varied simultaneously. As shown in Figure 8, the increase of α from 2. to 3. improves the performances. However, the performances become much worse in the case of α=3.. In this case, the desired signal levels at MHs located at the edge of cells become very low and the PDU loss rate at the MHs becomes quite high. The worse performances of the MHs dominate total system performances seriously. In the case where the path loss component is high, such as dense building areas or indoor areas with walls or obstacles, we have to reconsider to change some of system design parameters, e.g. the cell radius or the number of channels, or to use lower PHY data rates which are more robust to the interference and the noise. 2 alpha=2. alpha=3. alpha=3. 1 Radius=3 m 1 2 2 Figure 6: Performance with different α for interference signal (α for desired signal=2.) 2 1 Radius=3 m α= 2. 4 channels 9 channels 1 2 2 Figure : Effect of the number of channels 1 Radius=3 m alpha=2. alpha=3. alpha=3. 1 2 2 Figure 7: Performance with different α for desired signal ( α for interference signal=2.)

2 1 alpha=2. alpha=3. alpha=3. Radius=3 m 1 2 2 Figure 8: Performance with different α 2 1 Radius=3 m α= 2. 9Mbps 1 2 2 Figure 1: Performance of different PHY data rates.1.3. Frequency selection schemes As shown in Figure 9, the performances of DFS are very close to those of FFS, while the performances of RFS are very poor compared to the others. From the results, it is obvious that the frequency selection scheme affects the system performance quite seriously. In this paper, we assume almost ideal estimation of the expected CIR level for DFS, therefore we need further study how to estimate the level precisely in real systems. 2 1 Radius=3 m α= 2. Fixed Selection Random Selection Dynamic Selection 1 2 2 Figure 9: Effect of frequency selection scheme.2. Effect of transceiver approach.2.1. PHY data rate As shown in Figure 1, when the PHY data rate is from 6 Mbps to 27 Mbps, the maximum throughput is about 7% of each PHY data rate. However, when the PHY data rate is 36 Mbps, the throughput is about % of the PHY data rate and the throughput is the almost same with the throughput of 27 Mbps, because the PHY data rate of 36 Mbps is more sensitive to the CIR level. The CIR levels at MHs located around the edge of cells are not enough high to use the rate and the performances at the MHs dominate the throughput performances..2.2. As shown in Figure 11, the link adaptation can improve the performances. When the link adaptation is applied, the maximum throughput is higher than the both throughputs of 27 Mbps and 36 Mbps. 2 1 only only Radius=3 m α= 2. 1 2 2 Figure 11: Effect of Link Adaptation In order to investigate the performance improvements in detail, we demonstrate some behaviors of the link adaptation. Figure 12 shows relations between the average PDU loss rate and the distance between MH and AP. From the figure, it is shown that the PDU loss rate with the link adaptation is the almost same with the PDU loss rate of in the area where the distance is shorter than 3 m. This means the PHY data rate of 36 Mbps is selected by the link adaptation at the most of MHs in this area. Correspondingly, the link adaptation lets MHs located at the area where the distance is longer than 3m select the PHY data rate of. As shown in the figure, by applying the link adaptation, the PDU loss rate averaged over all MHs is worse than that in the case where only PHY date rate of is applied. However, by applying the higher data rate to the MHs located in the middle of the cell, the link adaptation can improve the throughput performances as shown in Figure 11.

Average PDU Loss Rate.7.6..4.3.2.1 only only Radius=3 m α= 2. 1 2 2 3 3 4 Distance between MH and AP (m) Figure 12: Average PDU loss rate vs. Distance between MH and AP Figure 13 shows relations between the number of available channels and distributions of the CIR levels estimated in the link adaptation algorithm. As in the figure, when are available, the estimated CIR is higher than 1 db and PHY data rates of 27 Mbps and 36 Mbps can be provided to MHs. On the other hand, the lower PHY data rates are selected in the cases of 9 channels and 4 channels. The transmitting power control mechanism adjusts the maximum level of the desired signals at the receivers to dbm, therefore there are upper limits of CIR levels in the distributions. Probability.3.2.2..1. 9 channels 4 channels Radius=3 m α= 2. 1 2 2 3 CIR level (db) Figure 13: Distribution of estimated CIR level.3. Overall performance evaluations In the previous subsections, we have independently evaluated the improvements of the system performances achieved by the system level and the transceiver level approaches. In this section, we evaluate the system performances in the case where both approaches are adopted. Through the evaluations, we clarify the relation between the number of available channels and the system throughput. Moreover, we reveal required additional interference tolerance to provide higher throughput, when the number of channel is not enough. In the following evaluations, we assume DFS as the frequency selection scheme and the link adaptation is applied..3.1. Relation between the throughput and the number of channels Figure 14 shows the throughput performances with the different number of channels. As shown in this figure, HL/2 can provide about 21 Mbps in Europe where 19 channels are available. If 12 channels and 4 channels are available as in US and Japan, the throughput is only about Mbps and 4 Mbps, respectively. 3 3 2 2 1 Japan US Europe Radius=3 m α= 2. Dynamic Frequency Selection 1 2 2 3 Number of channels Figure 14: Throughput vs. the number of channels.3.2. Additional interference tolerance As shown above, the throughput is much affected by the number of channels. However, the number of channels is restricted by the regulation of spectrum allocation in each region or country. To provide higher throughput without increase of the number of channels, we need additional interference tolerance which improves the PER performances described in Figure 4. Sophisticated PHY technologies such as interference cancellers, adaptive array antennas or powerful FEC decoding algorithms can improve the interference tolerance. It is worth while indicating required improvement of the interference tolerance quantitatively for studies on the PHY technologies. Figure shows relations between the throughputs performance and the additional interference tolerances. If 19 channels are available as in Europe, the throughput reaches to the maximum with 1 db of the additional interference tolerance. The maximum throughput is the almost same with the maximum throughput of 36 Mbps described in section 2.3. This is because that all MHs can select the data rate of with the help of the additional interference tolerance more than 1 db. However if only 12 and 4 channels are available as in US or Japan, 14 db and 2 db of the additional interference tolerances are required to realize the maximum throughput at least. It is also shown, to realize the same throughput in US and Japan as that in a manner without any additional interference tolerance in Europe, 3 db and 11 db of the additional interference tolerances are required, respectively.

3 2 2 1 Radius=3 m α= 2. Dynamic Frequency Selection Europe(19 channels) US(12 channels) Japan(4 channels) 1 2 2 Additional Interference Tolerance (db) Figure : Throughput vs. additional interference tolerance 6. Conclusion System throughput of HL/2 in multi-cell environment was evaluated in this paper. The simulation results show that the number of channels and the path loss components much affect to the throughput and the frequency selection scheme is also important to realize high throughput. By assuming 2. of the path loss component, 3 m of the cell radius and bytes of the IP packet size, and by applying DFS and the link adaptation, it is shown that the HL/2 can achieve the throughput of 21 Mbps in Europe where 19 of RF channels are available. However, in US and Japan where 12 and 4 of RF channels are available, the throughput is limited about Mbps and 4 Mbps, respectively. The number of channels depends on a spectrum allocation regulation in each country or region. Based on the current regulations in US and Japan, to realize the same throughput as in Europe, it is necessary to improve the interference tolerance by at least 3 db and 11 db, respectively. The improvements can be achieved by sophisticated PHY technologies such as interference cancellation or adaptive array antenna technologies. For future work, we have to consider and investigate the effects of more realistic traffic model and advanced link adaptation algorithms. In this paper, we assumed simple random distributed traffic model, but real data traffic is more bursty and it was studied the burstness affects the system performance[][16]. Also we adopted simple link adaptation algorithm, however there are many studies about the advanced link adaptation algorithms which we will have to take into account[17]. REFERENCES [1] ETSI TR 11 683 V1.1.1 (2-2), Broadband System Overview. [2] ETSI TS 11 493-1 V1.1.1 (2-4), Broadband Packet based Convergence Layer; Part 1: Common Part. [3] ETSI TS 11 493-2 V1.1.1 (2-4), Broadband Packet based Convergence Layer; Part 2: Ethernet Service Specific Convergence Sublayer (SSCS) [4] ETSI TS 11 761-1 V1.1.1 (2-4), Broadband Data Link Control (DLC) Layer; Part 1: Basic Data Transport Functions [] ETSI TS 11 761-2 V1.1.1 (2-4), Broadband Data Link Control (DLC) layer; Part 2: Radio Link Control (RLC) sublayer [6] ETSI TS 11 47 V1.1.1 (2-4), Broadband Physical (PHY) layer. [7] IEEE 82-11 1999(E), Part11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications [8] IEEE 82.11a-1999, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications High-speed Physical Layer in the GHz Band [9] IEEE 82.11b-1999, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications Higher-Speed Physical Layer Extension in the 2.4 GHz Band. [1] Khun-Jush J.; Malmgren G.; Schramm P.; Torsner, J., Overview and performance of HIPERLAN type 2 - a standard for broadband wireless communications, Proceedings IEEE VTC 2 spring. [11] Leaves P., Ghaheri-Niri S., Tafazolli R., Christodoulides L., Sammut T., Staht W., Huschke J., Dynamic spectrum allocation in a multi-radio environment: concept and algorithm, 3G Mobile Communication Technologies, Second International Conference on 21. [12] Zihuai Lin; Malmgren G.; Torsner J., System performance analysis of link adaptation in hiperlan type 2, Proceeding of IEEE VTC 2 fall. [13] J. Mikkonen et al., Emerging Wireless Broadband Net-works, IEEE Commun. Mag., vol. 36, no. 2, Feb. 1998, pp. 112 17. [14] M. Dinis, J. Fernandes, Provision of Sufficient Transmission Capacity for Broadband Mobile Multimedia: A Step Toward 4G, IEEE Commun. Mag., vol. 39, no. 8, Aug. 21, pp. 46 4. [] J. Rapp, Hiperlan/2 System Throughput and QoS with Interference Improving Strategies, IEEE 3rd Vehicular Technology Conference, VTC 21 Spring, Rhodes, Greece, May 21. [16] J. Rapp, Increasing Throughput and QoS in a HIPERLAN/2 System with Co-Channel Interference, IEEE International Conference on Networking (ICN), pp. 727-736 (Part I), Colmar, France, July 21. [17] S. Simoens, D. Bartolome, Optimum Performance of Link Adaptation in HIPERLAN/2 Networks, IEEE 3rd Vehicular Technology Conference, VTC 21 Spring, Rhodes, Greece, May 21.