On the Propagation Characteristics of the 5 GHz Rooftop-to-Rooftop Meshed Network

Transcription

1 On the Propagation Characteristics of the 5 GHz Rooftop-to-Rooftop Meshed Network Jussi Ojala,Ralf Böhnke, Antti Lappeteläinen, Masahiro Uno Nokia Research Center, Helsinki, Finland antti.lappetelainen@nokia.com,jussi.ojala@nokia.com Sony International (Europe) GmbH, Stuttgart, Germany, boehnke@sony.de,uno@sony.de $%675$&7 The IST project MIND [] aims to ease the creation and provision of broadband services and applications that are fully supported and customised when accessed by users in the future from a wide range of wireless access technologies. As a part of that, techniques for the delivery of broadband services are evaluated. The paper presents pathloss and channel model for the Rooftop-to- Rooftop environment in the 5 GHz band. Based on the obtained models the paper addresses the performance and usability of the H/ physical layer for providing fixed wireless broadband access in the Rooftop-to- Rooftop environment.,,75'8&7, The IST project MIND envisions interesting business scenarios based on rooftop wireless routers providing residential broadband access. The wireless routers would feature full IP stack; thus, they would create a mesh network topology similar of today s wired Internet. On the physical layer the routers utilise OFDM similar to H/. This paper presents the Rooftop-to- Rooftop channel behaviour in the 5 GHz band, which is a crucial factor in the feasibility and performance analysis of the usage of H/ PHYsical layer. The paper is organised as follows. In Chapter II the key business considerations are enlisted together with the respective impacts of the technical realisation of the rooftop wireless access. In Chapter III a comprehensive picture of the measurement set-up and, thus, the limits of the applicability of the pathloss and channel models are given. The main results, the Rooftop pathloss and channel models are derived and presented in Chapter IV. Link layer PER performance of the Rooftop channel model is analysed and compared to BRAN channel models in Chapter V.,,),;(':,5(/(665(6,'(7,$/ %5$'%$'6/87,6 Currently, the residential broadband services based on xdsl access are gaining in popularity. If the physical access were only limited for the wireline alternative the competition might be hindered since the wires are usually owned by one of the carriers. Also, in some environments the realisation of the wireline network may not be cost efficient. Wireless alternatives would address effectively these drawbacks. When offering residential broadband wireless services there is a need to achieve sufficient coverage at low cost without applying unsightly large antennas. Current point-to-multipoint deployments are almost without exception pulling out of the residential market because the poor coverage leads to a lack of customers, unreliable service and expensive installations []. The mesh architecture deployment arouses several relaxations for the radio performance since it shortens the link distances. The mesh architecture results in better coverage due to the possibility to routing around the obstacles and allows scalable on-demand based network deployment. The possible disadvantage is that every node has to broadcast instead of only the node with fixed access in the point-to-multipoint. Thus, in the broadcast high antenna gains due to directional antennae cannot be utilised in the transmission. On the other hand, the needed broadcast distances are not as long as in the point-to-multipoint deployments. IST-MIND also studies mesh topology from the network architecture point of view [3].,,,($685((76<67($' ($685((76 During April-May, a Rooftop radio channel measurement campaign was carried out in Latokaski, suburban area outside Helsinki. Most houses were 5-8m in height. There were trees, taller than houses, in

2 between the houses. The needles and leaves on the trees affected the visibility conditions. The measurement environment as a whole corresponds to mixed propagation scenarios, no clear classification between LOS and NLOS links was possible. The measurements were conducted with a radio channel sounder in the GHz frequency band. The sounder consisted of two separate units, a transmitter (Tx) and a receiver (Rx) controlled by a laptop where the user could view the received signal. The Rx was placed in a fixed location and kept immobile. The Tx was placed at several fixed positions and for each position data was collected. For every Tx/Rx location pair four cases with different antenna heights were considered, Tx/Rx antennas at 5m and/or 8m. Although, the Tx and the Rx were in fixed location wind conditions created movement in antennas and trees i.e. present multipath conditions variants during one measurement. The Tx was a 7dBi dipole antenna and the transmitting power was 7dBm. For the Rx a 4dBi omni antenna was used. The key parameters of the measurement system are presented in Table. 7DEOH5DGLRFKDQQHOVRXQGHUSDUDPHWHUV &HQWUHIUHTXHQF\ 7UDQVPLWWHGSRZHU &RGHOHQJWK 7 &KLSUDWH 6DPSOLQJUDWH 6DPSOHVSHUFKLS 5 7[DQWHQQDKHLJKW 7[5)FDEOH 5[DQWHQQDKHLJKW 5[5)FDEOH HDVXUHPHQWUDWH $ 5.3 GHz 7 dbm 4 MHz MHz HDVXUHPHQWVDQG&DOLEUDWLRQ 5 m and 8 m m 4 db attenuation 5 m and 8 m 5 m db attenuation CIR/λ Prior to the data analysis some initial screening was made to exclude spurious data. This consisted of deletion of CIRs without sufficient dynamics above the noise level. The data analysed here consisted of approximately 6 measurements, each measurement comprising Mb of data corresponding to one Rx/Tx antenna height pair. The data includes about 6 Channel Impulse Responses (CIR) per measurement. One CIR corresponds to the data received by the Rx within a short time interval of length 375ns. The derivations of pathloss and channel models from the CIR are conducted in Chapter IV. Calibration measurements with known cable loss were made in order to neutralise the losses of the measurement equipment to the final results. From the received CIR the calibrated transmitting power is 3 * log ( τ ) + 5, () K τ where K(τ ) is a CIR component with delay τ and the summation is taken over those values of τ for which K (τ ) is higher than peak minus 3 db. This calibrated transmission power was used in the pathloss and channel model calculations reported here.,93$7+/66$'&+$(/'(/ The empirical models are fixed outdoor adaptations of the channel and the indoor propagation models described in [4]. For wideband radio systems both the pathloss and the delay dispersion of the radio channel have to be characterised. The approach used here is to separate small scale and large scale behaviour. The physical reasoning for this is to distinguish the distance and obstacle based attenuation from the multipath fading. Large scale fading includes simple pathloss model with pathloss exponent and STD of pathloss. From the empirical point of view this equals to averaging the received power over one single measurement place with specified antenna heights, to get rid of multipath fading effects. These averaged values are used to obtain the pathloss and the slow fading estimates. The power variations during the averaging period will be used to determine the amplitude behaviour of the channel taps. The small scale behaviour is grossly quantified by a single parameter, the RMS delay spread, and more accurately described by a tapped delay line model. For a fixed number of taps, the excess delays, the normalised amplitudes and amplitude distribution are calculated. $/DUJHVFDOH3DWKORVVPRGHO The large scale pathloss model describes the pathloss as a function of distance between Tx and Rx. The adapted pathloss model is deduced from a semi-empirical approach and it contains the distance and obstacle based attenuation. The form of the pathloss equation is derived from the exponential law of attenuation. Further, the attenuation during the first meter distance is assumed to be as in free space. This results in the equation of the following form: ( D + Q; + ε () 3/ G) In equation (), 3/G is total pathloss >G%@, D is the free space loss at one meter, ; is the distance dependent part, Q is the propagation exponent, and ε is the residual error. More accurately D log (4πG / λ) 46.9 (3) G is one meter, λ is the wavelength.56 [m] and ( G ) ; *log / G (4) where G is the distance in meters. According to the properties of the LSQ estimate the residual error ε approaches a normal distribution with zero mean as number of samples increases. This residual error models obstacle based slow fading.

3 %([WUDFWLRQRIWKH3DWKORVVIURPWKHPHDVXUHGGDWD The large scale pathlosses are obtained by averaging from the instantaneous pathlosses. The instantaneous pathlosses 3/ (G) are obtained from the measured CIRs corresponding to a certain fixed position and fixed Tx/Rx antenna heights at distance G by the following formula τ. (5) 3/ ( G) + 3 log K G % (, ) τ * * + FDEOH _ ORVVHV Here % 5 [db] includes Tx/Rx antenna gains and cable losses and 3 is the calibrated transmission power. However, this still includes the effects of small scale behaviour. It is assumed that variation in multipath fading conditions causes the variation of all the instantaneous pathlosses at one position with fixed antenna height. Hence the calculated large scale pathloss 3/ (G) is the average over all instantaneous pathlosses at one position with fixed antenna height. There are about 6 instantaneous pathlosses to be averaged for one 3/ (G). To achieve the pathloss exponent and standard deviation of pathloss (i.e. slow fading) in model (), a least square fit has been made for 3/ (G). The achieved large scale pathloss model is presented in the following formula. 3/ G) * log ( G) + δ ( (6) This formula is valid for distances 3m < d < 33m. Slow fading δ >G%@is normally distributed with zero mean and standard deviation 7.6 db. Distance based fading and measured pathlosses 3/ (G) are shown in figure. 3 Delays are measured relative to the first detectable signal arriving to the receiver at τ and the average is calculated using the power delay profile as a weight function. The mean RMS delay spread is 49 ns and the cumulative distribution of measured RMS delay is shown in Figure. CDF rms delay spread [ns] )LJXUH&XPXODWLYHGLVWULEXWLRQRIWKH56GHOD\ VSUHDG The tapped delay line model is built for statistical analysis of small scale fading of a wideband channel. Therefore the effects of the distance and the obstaclebased losses have to be eliminated. One obvious way to do that is to normalise CIR components. With these measurements natural normalisation constant is 3/ (G). Thus, at each position with fixed antenna heights the fluctuation from average pathloss value are used to build a tapped delay line model. For a given number of taps the mean amplitude, amplitude distributions and delays are calculated. '7DSSHG'HOD\/LQHRGHO )LJXUH3DWKORVV>G%@YV'LVWDQFH>P@ PHDVXUHPHQWVDQG/64ILW &6PDOOVFDOH&KDQQHOPRGHO The small scale behaviour is first described by RMS delay spread, and then the channel is modelled using tapped delay line model. The RMS delay spread σ τ, i.e. the square root of first central moment of the power delay profile, is calculated for each CIR (τ ) σ τ τ (7) The 4 MHz chip rate gives the 5 ns lower bound for the time-interval that one tap should represent. In the chosen approach each tap represents an equal number of samples. To have an estimate about the maximum number of taps and tap delays the average power delay profile is calculated. All the normalised CIR components that arrive before the peak value and are lower than peak minus 5dB will be discarded and the resulting CIRs are marked delay profile is 3 K # ( Q ) K ( Q τ, # &,5. The average power τ V (8) The time-interval to the last component above 5 db relative to the peak value is considered to be meaningful. Samples in this interval are divided into M groups each containing equal number of elements, say,. The last component is pushed forward so that the division is an integer and together there are, samples. The obtained PDP is shown in Figure 4. The samples in each group yield the parameters of one tap. )

4 that at least taps were required, when the model was built so that all taps represent equal time space. Further, the performance difference in the simulated H/ packet error rates over the measured CIR channel and over the tap model channel were less than.5 db. The estimated parameters are shown in the following table, where delays are rounded to the nearest [ns]. 7DEOH3DUDPHWHUVRI7DSSHG'HOD\/LQHRGHO ZLWKWDSV )LJXUH$YHUDJH3RZHUGHOD\SURILOH The relative tapped delay value in each group/tap is + Q τ 3( Q τ ) τ + (9) 3( Q τ ) where L is the sample index starting group m. For the model the first tap is set to zero so that the actual delay of each tap is τ τ τ. The taps with delay τ have complex amplitude + K ( τ, V) K ( Q τ, V). (), Hence each tap has as many complex amplitude values as there are normalised CIRs. The mean amplitude corresponding to τ is the average of all K ( τ, V) and the tap power is the second power of mean amplitude value. The distribution of amplitude values is assumed to be Ricean/Rayleigh. The Rice factor of each tap is calculated by the following equation [5] [ K ( τ, V) ] ( π [ (, ) ] 4(. + ) ( K τ V ()... + ), +.,., is the first kind modified Bessel function of. exp ( There nth-order. Small Rice factors indicate a distribution close to the Rayleigh distribution. The amplitude distributions were further verified by plotting the calculated CDF of the taps against the theoretically obtained CDF. For the first three taps the theoretical distribution seems to fit well with the measurements. (([WUDFWHGWDSRGHO Earlier subsection showed how the Tapped Delay Line Model was constructed. To decide the number of taps a simpler model is better, however without compromising the accuracy of the model too much. The tap number was decided by comparing the link result given by measurements and different models. The results show Tap No. Delay [ns] P [db] Amplitude Distribution Rice K 7. db 3-6. Rayleigh Rayleigh Rayleigh Rayleigh Rayleigh Rayleigh Rayleigh Rayleigh Rayleigh 9/,./$<(53(5)5$&()7+((: '(/&3$5('7%5$&+$(/ '(/6 $(76,%5$&KDQQHOPRGHOVIRU+ The standardisation committee ETSI BRAN defined several channel models for the H/ link layer evaluation [6, 7]. Those models are based on measurement conducted in various environments (indoor office, large open space, etc.). The models, identified as A, B, C, D and E have RMS delay spreads of 5ns (Channel A), 5ns (Channel C) and up to 5ns (Channel E), typically no LOS component exists. Link Layer results (PER versus CNR) for H/ with the BRAN reference channel models have been widely published [8, 9]. %5RRIWRSWR5RRIWRS&KDQQHOPRGHOFKDUDFWHULVWLF The measured and characterised tap channel model in this paper shows a very small mean RMS delay spread of only 49ns and a strong LOS component (rice factor of 7.dB) which will significantly influence the link layer performance. The fading frequency selectivity is expected to be small. Because of the strong LOS component a performance close to pure a AWGN channel is expected. &6LPXODWLRQFRQGLWLRQ The simulation conditions used for the performance analysis are described in [] and can be summarised as:

5 ƒ H/ PDU train consisting of short transport channels (SCH) and long transport channels (LCH), the packet error rate (PER) of the LCH is observed ƒ Ideal channel estimation and synchronisation ƒ All physical layer parameters conforming the H/ specification (i.e. frame structure, modulation schemes, forward error correction, OFDM processing) ƒ Independent fading between bursts, no fading variation within the burst (Doppler ~ Hz) ƒ Normalised average fading, the power at the output of the tap delay line fading channel averaged over the simulated frames converges to the target CNR (carrier to noise ratio) set for the simulation. '6LPXODWLRQUHVXOWV For the simulations the PER performance with the new Rooftop-to-Rooftop models was analysed and compared to results achieved with a pure AWGN channel and the ETSI BRAN channel model A (with a RMS delay spread of 5ns and no LOS component). )LJXUH3(5LQGLIIHUHQWIDGLQJFKDQQHOV (/LQN/HYHO(QKDQFHPHQW6FKHPHV Adaptive Subcarrier Loading (ASL) has been used in [, ] to significantly enhance the PER performance (3-5dB gain) in pure multipath fading channels with strong frequency selectivity (Channel C). On the other hand, ASL simulations have shown that under the channel model used in this paper insignificant gains are obtained (<db) for the PER of interest (%-%). This is explained by (a) the already close to AWGN performance (strong LOS component) and (b) non frequency selective (flat) fading due to the small RMS delay spread of the fading channel model. 9,&&/86,6$'',6&866, )LJXUH5RRIWRSPRGHODQG$:*SHUIRUPDQFH The results (depicted in Figure 4) for the new channel model show only -3dB degradation compared to ideal AWGN results. The typical performance of H/ is shown in Figure 5, where the channel model A (RMS delay spread of 5ns, no LOS component) and channel model C (RMS delay spread of 5ns, no LOS component) are compared with the performance using the new Rooftop channel model. For the ease of examination only selected PHYsical layer modes (QPSK with coderate ½, 6-QAM with coderate 9/6 and 64QAM with coderate ¾ ) are depicted. The H/ performance in the typical multipath fading channels (without LOS component and increased frequency selectivity) is considerably worse. The performance degradation at % PER varies from 3dB to 7dB for the NLOS multipath fading channels compared to the new channel model. The paper presents results of pathloss and channel models for residential Rooftop-to-Rooftop environment. The models are based on extensive measurements in the 5.3 GHz band. The obtained pathloss exponent is.8. Compared to free space model the environment causes additional attenuation of 6 and db with link distances of m and 3m, respectively. This is well in line with [3]. The measurement has small RMS delay 49 ns. The obtained channel model has LOS component of 7. db. ETSI/BRAN channel models C and D separate the LOS and NLOS cases and that resulted LOS component of db in large open space environment. Smaller LOS component of the model indicates that the line-of-sight path had been somewhat inhibited by trees and hilly terrain. The RMS delay spread of the model is clearly shorter than in the large open space models defined by ETSI/BRAN. However, [6] explains that only few large open measurements were available and they corresponded to average RMS of ns. As a further example, Outdoor to Indoor and Pedestrian A, i.e. microcell model in [4] has average RMS delay spread of 45 ns. The H/ link performance in the Rooftop channel environment shows that guard time of 8 ns is sufficient and the overall link performance is less than 3 db worse than in the AWGN channel. Compared to ETSI/BRAN C model link performance is more than 5 db better in % PER. Using the estimated propagation exponent for the Rooftop

6 environment, the 5 db link gain corresponds to 5% increase in link distance. $&.:/('*((7 This work has been performed in the framework of the IST project IST MIND, which is partly funded by the European Union. The authors would like to acknowledge the contributions of their colleagues from Siemens AG, British Telecommunications PLC, Agora Systems S.A., Ericsson Radio Systems AB, France Télécom S.A., King s College London, Nokia Corporation, NTT DoCoMo Inc, Sony International (Europe) GmbH, T-Systems Nova GmbH, University of Madrid, and Infineon Technologies AG. We would also like to thank Ari Alastalo, Arnaud Brehonnet, Juha Juntunen, Mika Kahola, and Petri Sinisalo from Nokia Research Center for carrying out the radio channel measurements and the preliminary data processing. [5] A. Abdi et al., "On the Estimation of the K Parameter for the Rice Fading Distribution", IEEE communication letters, vol. 5, No 3, March, pp [6] Jonas Medbo (Ericsson Radio Systems), Radio Wave Propagation Characteristics at 5 GHz with Modelling Suggestions, ETSI EP BRAN document 3ERI74A, [7] Jonas Medbo, Peter Schramm (Ericsson Radio Systems), Channel Models for HIPERLAN/, ETSI EP BRAN document 3ERI85B, [8] Angela Doufexi, Simon Armour, Peter Karlsson, Andrew Nix, David Bul, A Comparison of HIPERLAN/ and IEEE 8.a, VTC Fall. [9] J. Khun-Jush, P. Schramm, U. Wachsmann, F. Wegner, Structure and Performance of the HIPERLAN/ Physical Layer, VTC'99 Fall 5()(5(&(6 [] IST Project MIND: Mobile IP based Network Developments (IST--8584) home page [] J. Arrakoski et al. Broadband Radio Access Networks (BRAN); HIPERMAN; Licence exempt FWA for band C (5,75-5,875 GHz), feasibility and initial sharing study ETSI DTR/BRAN- 44 v... [3] Phil Eardley et al. EVOLVING BEYOND UMTS - THE MIND RESEARCH PROJECT Submitted for Conference 3G London May [4] J Kivinen et al., "Empirical Characterization of Wideband Indoor Radio Channel at 5.3 GHz", IEEE transactions on antennas and propagation, vol. 49,, pp [] M. Pauli, B. Wegmann, A. Krämling, A. Kadelka and T. Bing, First Performance Results of BRAIN, IST Summit, Galway, Ireland, Oct. -4,, pp [] R. Grünheid, E. Bolinth, H. Rohling and K. Aretz, Adaptive modulation for the HIPERLAN/ Air Interface, Proc. 5th International OFDM Workshop Hamburg [] E. Bolinth, A. Lappeteläinen, J. Ojala, M. Pauli, A. Krämling, T. Journé, D. Lacroix, R. Böhnke, QoS Enhancements for HIPERLAN/, IST Global Summit, Barcelona, September [3] G. Durgin, T.S Rappaport, H. Xu, "5.85-GHz Radio Path Loss and Penetration Loss Meassurements In and Around Homes and Trees", IEEE communication letters, vol., No 3, March 998, pp [4] Universal Mobile Telecommunications System (UMTS); Selection procedures for the choice of radio transmission technologies of the UMTS, UMTS 3.3 V3.. (997-5) Technical Report.