1. 1 Data link layer protocol for wireless TCP K.S. Chan EEE Department The University of Hong Kong

2. 2 Outlines Introduction Multi-dimensional zigzag code Data link layer protocol Performance evaluation Automatic adaptation Conclusions

3. 3 Introduction Packet loss in wireless TCP: –congestion: TCP reacts properly –Random loss: reducing congestion window will cause performance degradation New schemes differentiating congestion packet loss and random loss needed Existing schemes: –Wireless aware TCP: two connections TCP modification needed –Wireless unaware TCP: Link layer: airmail, unlimited ARQ Split connection: I-TCP, M-TCP Proxy: snoop, new snoop

4. 4 TCP aware solution Same fraction of ACKs received: packet loss due to congestion Acknowledged fraction significantly different, random loss Control connection user connection

5. 5 Introduction Packet loss in wireless TCP: –congestion: TCP reacts properly –Random loss: reducing congestion window will cause performance degradation New schemes differentiating congestion packet loss and random loss needed Existing schemes: –Wireless aware TCP: two connections TCP modification needed –Wireless unaware TCP: Link layer: airmail, unlimited ARQ Split connection: I-TCP, M-TCP Proxy: snoop, new snoop

6. 6 TCP unaware solution: airmail Not efficient TCP DLC with strong FEC

7. 7 Introduction Packet loss in wireless TCP: –congestion: TCP reacts properly –Random loss: reducing congestion window will cause performance degradation New schemes differentiating congestion packet loss and random loss needed Existing schemes: –Wireless aware TCP: two connections TCP modification needed –Wireless unaware TCP: Link layer: airmail, unlimited ARQ Split connection: I-TCP, M-TCP Proxy: snoop, new snoop

8. 8 TCP unaware solution: I-TCP TCP end-to-end semantics violated Huge buffer at base station TCP TCP or other protocols

9. 9 Introduction Packet loss in wireless TCP: –congestion: TCP reacts properly –Random loss: reducing congestion window will cause performance degradation New schemes differentiating congestion packet loss and random loss needed Existing schemes: –Wireless aware TCP: two connections TCP modification needed –Wireless unaware TCP: Link layer: airmail, unlimited ARQ Split connection: I-TCP, M-TCP Proxy: snoop, new snoop

10. 10 TCP unaware solution: snoop Function for snoop module: –Buffer new packets from sender –Suppress duplicated ACK and retransmit lost packets TCP Snoop module

11. 11 Introduction (cont’d) Our scheme: –Link layer –Hybrid ARQ with limited re-transmission times –No transport layer activities –Adaptive to time-varying channel conditions TCP DLC with limited retransmission of hybrid ARQ

12. 12 Multidimensional Zigzag code d(1,1) d(1,3) d(1,4) d(1,5) d(1,2) p(1) d(3,1) d(3,2) d(3,3) d(3,4) d(3,5) d(I,1) d(I,2) d(I,3)d(I,4) d(I,5) d(2,2) d(2,3) d(2,4) d(2,5)d(2,1) p(I-1) p(2) p(3) p(I)

13. 13 An example of zigzag code 1 1 1 0 0 p(1)=1 0 1 1 1 0 1 0 1 0 1 0 1 0 0 1 p(4)=1 p(2)=1 p(3)=0 p(5)=0 1 1 1 1 1 P=[11010] T I=5  5

14. 14 Multi-dimensional zigzag code D1D1 P1P1 J I D2D2 P2P2 J I DNDN PNPN J I D J I P1P1 P2P2 PNPN (a) (b)

15. 15 DLC Protocol Description User process TCP IP Wireless DLC Wireless MAC High-speed radio IP Wireless DLC Wireless MAC High-speed radio Mobile terminalBase station

16. 16 Hybrid ARQ for TCP connections D J I P1P1 P2P2 P m1 P m2 P m2+1 P m1+1 PNPN G1G1 G2G2 GmGm Level 1 Level 2Level m W-DLC headerTCP packetCRC1CRC2Physical preamble (a) (b) MAC header CRC2Physical preambleMAC header bb n

17. 17 Operation Estimate the allowed retransmission times When a new packet is received, encode it N- dimensional zigzag codeword, and divide the N parity vectors into m groups Transmission counter set to 0 Level 1 transmission: increase the transmission counter by 1, and transmit the information matrix with parity vector group 1 Level i, 1<i<m+1: increase the counter by 1. If the retransmission limit is met, end. Otherwise, transmit the parity vector in group i.

18. 18 An example for ARQ D 80 150 P1P1 P2P2 P7P7 P8P8 P9P9 G1G1 G2G2 W-DLC headerTCP packetCRC1CRC2Physical preamble (a) (b) W-MAC header CRC2Physical preambleW-MAC header bb 12,000G1G1 G2G2

19. 19 Performance Evaluation Additive White Gaussian channel Server-client connection: 32Mbytes, packet size: 12,000bits Goodput measurement 100 m 200 m 1.544Mbps

20. 20 Goodput for different SNRs

21. 21 Goodput for different coding schemes SNR: 5 dB

22. 22 Automatic Channel Adaptation Small number of possible states A state suitable for a wide range of channel conditions Un-match detectable

23. 23 Adaptation 8 states Two level transmission criteria for state change: –If consecutively 100 information packets can be correctly decoded by only (L 1 -1) parity check vectors, increase the state by 1 –Information part of 5 packets out of 100 consecutive packets needed to be retransmitted, decrease the state by 1

24. 24 state 1 2 3 4 5 6 7 8 I k1k1 k2k2 N 100 50 25 20 5 7 4 5 4 5 9 3 4 7 5 7 6 6 2 3 1 4 7 2 2 1 2 1 Table 1: the encoder states for non-real-time services

25. 25 Example for adaptation A FEC connection if 3 packets out of consecutively 60 packets are discarded, state is decreased by 1 if consecutively 40 packets can be decoded with less than N r parity vectors, state is increased by 1 simulation condition: –SNR=5.4dB, coding state: 1 (coding rate: 0.36) –SNR change 0.1dB roughly per 150 packets sent

26. 26 state 1 2 3 4 5 6 7 I NrNr 100 50 25 20 6 9 7 7 7 5 4 Table 2: the encoder states for real-time services

27. 27 Encoder’s adaptation in the time-varying channel.

28. 28 Conclusions Data link layer protocol for TCP over wireless links proposed –Hybrid ARQ with limited retransmission times –Adaptive

29. 29 Thank You