Analysis of Urban Millimeter Wave Microcellular Networks

Transcription

1 Analysis of Urban Millimeter Wave Microcellular Networks Yuyang Wang, KiranVenugopal, Andreas F. Molisch, and Robert W. Heath Jr. The University of Texas at Austin University of Southern California TheUT authors are funded by U.S. Department of Transportation through D- STOP Tier 1 University Transportation Center and Texas Department of Transportation project CAR-STOP. Dr. Molisch s work is supported by NSF and Samsung.

2 Slides Robert W. Heath Jr. (2016) Manhattan type urban vehicular network Dense skyscrapers:severe signal blockage and attenuation Clustered users at intersections: generate heavy load for BS Most challenging environment for mmwave communication Diverse set of terminals: vehicles, pedestrians, bicyclists Various applications: vehicular safety, infotainment 2

3 Stochastic geometry network models Slides Robert W. Heath Jr. (2016) MmWave w/ blockage [1-2] Manhattan no mmwave [3] V2V no mmwave [4] unblocked blocked unblocked blocked Realistic urban microcell PL model and tractable V2I analysis framework [1] T. Bai and R. W. Heath Jr., Coverage and rate analysis for millimeter wave cellular networks, IEEE Trans. Wireless Comm., [2] M. Kulkarni, S. Singh, and J. G. Andrews. "Coverage and rate trends in dense urban mmwave cellular networks. IEEE GlobeCom, [3] F. Baccelli, and X. Zhang. "A correlated shadowing model for urban wireless networks. IEEE INFOCOM, [4] M. Farooq, H. ElSawy, and M. Alouini. "Modeling inter-vehicle communication in multi-lane highways: A stochastic geometry approach. IEEE VTC fall,

4 Contributions Exploit new urban pathloss model for mmwave microcells Analysis framework for outdoor mmwave using the Manhattan line processes Quantify & compare interference 4

5 System model 5

6 Manhattan distance based pathloss model Slides Robert W. Heath Jr. (2016) Euclidean distance is NOT the right way to characterize pathloss Manhattan distance based PL model LoS segment: 1 st segment of link NLoS PL exponent PL db (d L, D N ) = 10 L log 10 d L + 10 X d2dn N log 10 d + M corner loss LoS PL exponent NLoS segments set: all segments except 1 st # corners A. F. Molisch, A. Karttunen, S. Hur, J. Park and J. Zhang, "Spatially consistent pathloss modeling for millimeter-wave channels in urban environments, EUCAP, Davos, 2016, pp

7 Manhattan poisson line proces (MPLP) 1D-PPPs and stretch out to form MPLP streets BSs are randomly dropped on streets as PPP Vert/hori streets intensity are: λ sv and λ sh Analyze performance of the typical user on horizontal street Street width not considered BS 1D intensity on vert/hori streets are: λ tv and λ th 7

8 Coverage analysis 8

9 Coverage analysis: a new approach (1/3) SINR o = Typical receiver o Transmit power Small-scale fading P t h o u I L + I V + I N + N 0 Three interference Gaussian noise Slides Robert W. Heath Jr. (2016) Associated link path gain u Coverage analysis I L LoS street Step 1: CDF of associated link path gain u I V NLoS vert/cross streets Step 2: coverage probability P c (T, u) conditioned on u I H NLoShori/parallel streets Step 3: coverage probability P c (T) deconditioned on u Assumption: vehicle is associated with the strongest BS (smallest PL) 9

10 Coverage analysis: a new approach (2/3) Slides Robert W. Heath Jr. (2016) 1. CDF of associated link path gain u F (u) =exp 2 th u 1 L exp 2 sv 1 L N 2 tvc 1 L L N u 1 N 2. Conditioned coverage probability P c (T, u) conditioned on u and p c (u, T )=exp( C 1 u 1 )exp( C 2 u 1 L )exp C 1 = TN 0, C 2 =2 th %, C 3 =2 sv 1 L N 2 tvc 1 L % L 3. Deconditioned coverage probability P c (T) P c (T )= Z 1 0 PDF of u N % = Z 1 f U (u)p c (u, T )du 1 C 3 u 1 N, 1 1+T 1 µ L dx 10

11 Coverage analysis: a new approach (3/3) Jensen s inequality on interference Laplace transform Slides Robert W. Heath Jr. (2016) L I V (Tr L ) LoS interference L I L (Tr L ) exp (!E r [2 th r]) exp =exp(!) NLoS vertical exp # sv E r h (2 tv r) L N i # sv tv th L N 1+! L N NLoS horizontal r 2apple L I H (Tu 1 ) 2 sv K 1 2 p 2apple sv sv Modified Bessel function of 1 st order apple very small 1. NLoS-V contributes little to interference 2. LoS interference is still dominant K 1 (µ) µ 1 L I H (Tu 1 ) 1. NLoS-H is negligible 11

12 Numerical results 12

13 Fitting parameters for Euclidean pathloss models Euclidean model 1 PL db (d) = 10 log 10 d + 1 PL exponents and offset to fit Euclidean model 2 Bernoulli RV with parameter of blockage prob. PL db (d) =(1 I(p B (d))) 10 L log 10 d + L 2 + I(p B (d)) 10 N log 10 d + N 2 p B (d) =1 Blockage prob in MPLP 1 exp( 2d( sh + sv )). 2d( sh + sv ) Parameter fitting are divided into LoS and NLoS Parameter fitting: linear regression results Fitting parameters 1 L L 2 N N 2 Values in model in model 1-19dB in model in model 2 0dB in model 2 11 in model 2 5dB 13

14 Comparing pathloss models Coverage Probability Our Model Euc Model 1 Euc Model 2 Ergodic Capacity (bps/hz) Our Model Euc Model 1 Euc Model SINR Threshold (db) SNR at BS (db) Significant differences with Euclidean PL models in ergodic capacity/coverage! 14

15 Association link path gain distribution CDF of Associated Link Channel Gain Simu LOS+V/H-NLOS Simu LOS+V-NLOS Simu LOS Ana LOS+V-NLOS Ana LOS Associated Link Channel Gain Threshold(dB) Analysis and simulation overlap Small gap of CDF w/ and w/o NLoS-V or H LoS association dominates 15

16 Coverage probability Coverage Probability Simu LOS+V-NLOS Simu LOS+V/H-NLOS Ana LOS+V-NLOS New coverage analysis gives exact/concise results Only consider interferers located on the same street SINR Threshold (db) 16

17 Conclusions & future work 17

18 Tractable and realistic MPLP model & Manhattan PL model LoS dominates association & interference More realistic street modeling, e.g., street width Vehicular mobility Multiple vehicles and analysis of system throughput 18

19 Questions? 19