1. Infostation overlays in cellular systems Joan Borràs & Roy Yates

2. Outline zCellular systems zInfostations zCell area vs Coverage area zSystem model zPerformance measures zNumerical example

3. Wireless Data over 2G Cellular zCellular Voice: yAnytime Anywhere ySlow: 10K bps yHigh cost/bit v cents/min voice = 13v cents/MB zData over Cellular: ySlow and Expensive: 20 cents/min voice = $2.60/MB 100X too high

4. Wireless Data over 3G Cellular z2X or 3X increase in BW efficiency  reduced cost/bit zHigher speed: 144K or 384K bps zBut voice is still 10K bps  v cents/min = 13v cents/MB z $/MB still to high for anytime/anywhere data

5. DATA can tolerate delay! zMessaging services: e-mail, voice mail, fax, maps, non-interactive web pages,... zWe don’t need ubiquitous coverage zReduced coverage  Higher data rates zA little wait might be worthwhile!

6. Infostations zNetwork of wireless ports zIrregularly distributed zDiscontinuous coverage zAsymmetric link zHigh data rate transmission zMessaging services

7. Infostation System Download a map Internet access Low bit-rate cellular E-mail voice mail fax

8. Isolated Infostation Optimization Problem: With finite energy, how many bits can be delivered? Solution: Use small coverage

9. Highway Infostations Distance attenuation  time varying channel Xmit power profile = Time Average Capacity

10. Cellular Infostations Infostations at cellular sites

11. Traditional Cellular zModulation and BER  SIR threshold zSIR  Cluster size N (frequency reuse) zFor worst-case location!

12. Cell area  Coverage area zFor Infostations, there is no need to provide ubiquitous coverage zBetter SIR conditions zSmaller cluster size zMore bandwidth available HIGHER DATA RATE!

13. SIR in 2D Infostation system

14. Possible operating points SIR with respect to r/R and N 2D

15. Which system is better? 2M bps 6M bps

16. System model zEvery Infostation is modeled as an M/M/1 queue with reneging 

17. Parameters for 2D zUser density, u zCell radius, R zMobile speed, v, f V (v) zCoverage radius, r zData rate, c zMessages per user per second, u zMessage size, m  = (2  r)(u/2) E[V]  1/  = E[X] = = u t c (m/c)  = 2E[V]/  r R r  

18. 2D Cycle Time Bias: f  (  )=cos(  )/2, 0<|  |<  /2 Prob(-  <  <  )=r/Rt c = (  R 2 )/(2r E[V])

19. Mailbox zBetween Infostation visits, messages accumulate in mailbox zAfter successful visit, ready for next zAfter unsuccessful visit, either ythe residual contents of a mailbox are discarded, or ythe mailbox will not accept new messages until it is emptied

20. 2D Performance zThroughput yc (1-p 0 ) zDelay  t c / (1-P out ) + t Q t c = (  R 2 /2r E[V])1-P out = (  / )(1-p 0 ) t Q calculated for M/M/1 FCFS with reneging

21. 2D Numerical Example u=(0..0.0004), R=500, E[V]=2.5, m=2e6, u =0.01

22. 2D Throughput: 64QAM (r/R) 2 6.25% 17.6% 24.0% 42.2% 72.2%

23. 2D Reneging probability: 64QAM

24. 2D Delay: 64QAM t c / (1-P out ) + t Q

25. 2D Throughput: N=1 (r/R) 2 6.25% 13.7% 31.4% 43.6%

26. 2D Reneging probability: N=1

27. 2D Delay: N=1 t c / (1-P out ) + t Q

28. Conclusions zReduced coverage, allowing higher level modulations and smaller cluster sizes, provides increased throughput. zFor 1D, delay is also better (see WPMC’99) zFor 2D, as user density (offered load) increases, delay performance gets relatively better.