1. A New Methodology for the Design of High Speed Wireless Communication Systems Based on Experimental Results Tim Gallagher University of Kansas July 21, 1998

2. 2 Presentation Outline b Introduction b Background b Description of Measurement System and Experimental Procedure b Results b Conclusions

3. 3 Introduction b Broadband wireless: WLL & WLAN b Bit Error Rate vs. Packet Error Rate: WATM b Line-of-sight, Obstructed and All location cases b Carrier frequency: U-NII band b Characterizing the channel

4. 4 Introduction: Characterizing the Channel b Wireless system propagation anomalies Multipath dispersionMultipath dispersion –Multiple versions of the transmitted signal with various phases and amplitudes add at the receiver –Motion implies short-term fading ShadowingShadowing –Channel is not uniform in all directions –Variation in long-term average power

5. 5 Background b Better way to determine percent coverage area than by assuming a Rayleigh channel BER vs. PERBER vs. PER Received power - log-normalReceived power - log-normal Theoretical fading model - Rician distributionTheoretical fading model - Rician distribution K distribution - log-normalK distribution - log-normal Using this information about the channel, can determine the percent coverage areaUsing this information about the channel, can determine the percent coverage area

6. 6 Background: Bit Error Rate (BER) vs. Packet Error Rate (PER) b Non-fading channel equivalent PER b Fading channel - Errors are bursty Can have more bit errors without increasing the PER, so we need lessCan have more bit errors without increasing the PER, so we need less

7. 7 Independent bit decisions in slow flat Rayleigh fading

8. 8 Background: Received Power Distribution b Log-normal shadowing Received signal is the product of many transmission factors -- In dB, it is the sumReceived signal is the product of many transmission factors -- In dB, it is the sum Central Limit Theorem implies GaussianCentral Limit Theorem implies Gaussian Cramer-von Mises goodness of fit testCramer-von Mises goodness of fit test b Received signal over distance Power LawPower Law Two-ray modelTwo-ray model

9. 9 Background: Receive Power Distribution (cont.) b Power law b Two-ray model

10. 10 Background: Fading Model b Construct CDF from collected data b Compare to a theoretical model b Kolmogorov-Smirnoff goodness of fit test b Fading model -- Rician distribution

11. 11 Background: Fading Model (cont.) b Rayleigh Distribution b Rician Distribution:

12. 12 Background: Distribution of K b K is specular-to-random ratio b A 2 and should be log-normally distributed b K(dB) should be normally distributed Cramer-von Mises goodness of fit testCramer-von Mises goodness of fit test

13. 13 Background: Percent Coverage Area b Determine required received signal level link equation (function of K)link equation (function of K) b Determine K and received power correlation b Develop formula for percent coverage area

14. 14 Background: Percent Coverage Area (cont.) b Link equation

15. 15 Background: Percent Coverage Area (cont.) b K and Pr Correlation

16. 16 Background: Percent Coverage Area (cont.) b Development of Percent Coverage Area Formula Bivariate Gaussian pdfBivariate Gaussian pdf

17. 17 Background: Percent Coverage Area (cont.) b Development of Percent Coverage Area Formula Want probability that Pr > Pr req (K) for a specific distanceWant probability that Pr > Pr req (K) for a specific distance Integrate joint pdf over all K and from Pr req (K) to infinityIntegrate joint pdf over all K and from Pr req (K) to infinity

18. 18 Bivariate Gaussian pdf with received signal power threshold

19. 19 Background: Percent Coverage Area (cont.) b Percent Coverage Area Formula

20. 20 Background: Percent Coverage Area (cont.) b Percent Coverage Area Formula

21. 21

22. 22 Measurement System and Experimental Procedure (cont.) b Procedure Determine fading for each location (K)Determine fading for each location (K) Determine average K over distanceDetermine average K over distance –test for normality Determine path loss exponentDetermine path loss exponent –test for normality Close link for average K and path lossClose link for average K and path loss Determine correlation coefficient, K & PrDetermine correlation coefficient, K & Pr Determine percent coverage areaDetermine percent coverage area

23. 23 Results: Fading

24. 24 Results: Distribution of K Distribution of K: LOS locations

25. 25 Distribution of K: OBS locations

26. 26 Distribution of K: All locations

27. 27 Log-normal K

28. 28 Results: Average Receive Power b Log-normal shadowing C-vM statistic = 0.049C-vM statistic = 0.049

29. 29 Results: Average Receive Power b Two-ray model

30. 30 Results: Closing the Link

31. 31 Results: Correlation Between K and Shadowing

32. 32 Results: Percent Coverage Area b BER vs. PER b LOS, OBS, ALL comparison b K and receive power joint distribution Distance trade-offDistance trade-off Power trade-offPower trade-off b ERP family of curves

33. 33 BER vs PER: Packet size = 424 bits

34. 34 LOS, OBS and ALL comparison: BER = 10 -5, ERP = 20 dBW

35. 35 Distance Trade-off: LOS locations, BER = 10 -5, ERP = 20 dBW

36. 36 Power trade-off: LOS locations, BER = 10 -5, Cell radius = 500 m

37. 37 ERP family of curves: All Locations, BER = 10 -5

38. 38 Conclusions b K is log-normal - as predicted b Pr is log-normal - as expected b Two-ray model is better for LOS, but sensitive b PER better than BER -- as expected b K and Pr correlated - as expected b Use of real data better than Rayleigh assumption