Performance of DS-CDMA Protocols in Wireless LANS M.Parikh, P.Sharma, R.Garg, K. Chandra, C. Thompson Center for Advanced Computation and Telecommunications University of Massachusetts Lowell, Lowell, MA
OUTLINE Description of IEEE802.11b/DSSS Measurements of TCP Packet Transmission using b Products Computational Model for Multipath Estimation Statistical Analysis of Location Dependent effects on TCP Performance Conclusions
IEEE b Direct Sequence Spread Spectrum IEEE b specifies MAC and PHY layer specifications for DSSS PHY layer – FH, DS, IR Synchronization (128 bits) SFD (16) Signal (8) Service (8) Length (16) CRC (16) Payload (variable) PLCP Preamble PLCP header IEEE PHY frame using DSSS PHY PLCP -provides carrier sense and CCA PMD - BPSK/QPSK modulation - Wireless encoding
Transmission Specifics CSMA/CA DCF : Distributed Control Function : Contention based channel access Interval durations: DIFS= 50 µs SIFS= 10 µs Slot Time= 20 µs Defer Access Slot time Select slot and decrement backoff Backoff windowNext Frame Contention Window DIFS PIFS SIFS Busy Medium DIFS Immediate access if medium is free >= DIFS
Transmisson Specifics Packets retransmitted from PHY layer if receiver ACK not received within SIFS duration : Assumes collision Packets retransmitted from higher layers (TCP) when –channel access fails –MAC/PHY layer buffers overflow Protocol parameters fixed Our Objective: Examine performance in spatially varying multipath channels
TCP Performance Measurements Generate TCP flows between server on wired network & mobile client –Server Access Point (AP) -> Mobile Terminal (MT) –AP : Apple Airport TM –MT: Apple Ibook with Airport TM card WIN/XP & Linux m/cs with Lucent Orinoco Gold Card Capture TCP headers at Server, AP & MT using tcpdump utility Jointly Analyze TCP logs to determine retransmission and collision statistics
Channel Impulse Response Computation Channel impulse response computed using method of images - Walls of enclosure replaced by point sources of varying strength and location - Image source amplitude chosen by continuity of transverse electric and magnetic fields across reflecting boundary - Incorporates influence of varying dielectric constant for six walls - Rectangular room geometry; Empty Room; - Model details described in paper Generates a sampled approximation of bandpass impulse response GHz center frequency, 40 MHz bandwidth Multipath impulse responses computed for locations A, B and C
CHANNEL IMPULSE RESPONSE
Signal & Noise Statistics Local and Remote Signal and Noise power distribution averaged across 6 measurement ensembles - noise range fairly stationary –-98 : -81 dBm : Local –-102 : -93 dBm: Remote Distinct bi-modal noise distribution at source suggests : multipath noise may be distinct from environmental noise - signal range : dBm: Local dBm: Remote
TCP Statistics Average TCP throughput stationary across all locations : 4.2 – 5 Mbps TCP retransmission & packet collision statistics highly location dependent -Retransmission detected from server log -Collision assumed when single occurrence of sequence number in server log, but >1 in Mobile TCP log: Assume ACK from MT is lost/delayed
Position dependent Retransmission & Collision Statistics LOC TCP Retransmissions µ iat (Rx) secs PHY Collisions µ iat (Col) secs ANone- - B C Mean interarrival time for the mobile terminal ~ 2ms Retransmissions from the PHY layer for location B occur at a faster rate than location C Mean interarrival time for retransmissions for loc B is deterministic
TCP time-sequence plots
CONCLUSIONS Performance degradation in b products can be attributed to: - Retransmissions for the PHY layer due to a lost ack - TCP retransmissions - Protocol overhead - Multipath effect Effective throughput decreases to 50 % due to the combined effects of the degradation attributes Future work: Improve the protocol performance due to the multipath effect