doc.: IEEE /257 Submission Slide 1 May 2001 Coffey et al, Texas Instruments Multipath comparison of IEEE802.11g High Rate Proposals Sean Coffey, Anuj Batra, Srikanth Gummadi, Chris Heegard, Matthew Shoemake Texas Instruments 141 Stony Circle, Suite 130 Santa Rosa California (707) ,

doc.: IEEE /257 Submission Slide 2 May 2001 Coffey et al, Texas Instruments Contents  PBCC-22’s qualifications for the IEEE g High Rate Standard  Receiver structures  Multipath performance comparisons  Discussion, how pure 11a OFDM is different  Conclusions

doc.: IEEE /257 Submission Slide 3 May 2001 Coffey et al, Texas Instruments IEEE b High Rate Task Group  Goal of Task Group: A high rate, > 20 Mbps, extension of the existing b standard –must be backwards compatible with 11b

doc.: IEEE /257 Submission Slide 4 May 2001 Coffey et al, Texas Instruments PBCC-22 features  Excellent performance in full range of multipath conditions - much better than CCK-OFDM at comparable rates –This is the central technical point in dispute  Range advantage of PBCC 22 Mbps over CCK- OFDM 24 Mbps in multipath conditions is 30-40%.  These claims documented later; standard IEEE models were used.

doc.: IEEE /257 Submission Slide 5 May 2001 Coffey et al, Texas Instruments Multipath comparison of the proposals First, for each proposal, assume same ground rules: –floating point implementation –full channel knowledge –standard IEEE multipath model –off-the-shelf algorithms –assume each uses receiver structure presented by proposers

doc.: IEEE /257 Submission Slide 6 May 2001 Coffey et al, Texas Instruments PBCC-22 Receiver:  treat multipath and code as forming a composite state machine, or “super code”  decode the “super trellis” using any standard reduced state algorithm Simulation results here assume whitened matched filter plus M-algorithm; standard material, very well understood: –whitened matched filter - Forney, –M-algorithm - Anderson, 1969.

doc.: IEEE /257 Submission Slide 7 May 2001 Coffey et al, Texas Instruments M-algorithm decoder background:  M-algorithm operates like regular trellis decoder, but retains only best “M” paths at each depth –No restrictions on choice of M –straightforward way of trading performance versus complexity –natural receiver upgrade path –Main results use M = 64 we also present M = 8, M = 16, M = 32, M = 128.

doc.: IEEE /257 Submission Slide 8 May 2001 Coffey et al, Texas Instruments M-algorithm decoder:  Assume the “state” consists of input data bits at last 8 time units –Compare last 4 time units to represent pure code state –Choice of 8 is arbitrary, other values possible  Assume each “state” remembers the full impact of the past on the future –Curve shown in Doc. 01/140 assumes instead that multipath is regenerated from last 8 time unit inputs

doc.: IEEE /257 Submission Slide 9 May 2001 Coffey et al, Texas Instruments Baseline comparisons, IEEE multipath model, 100 ns 5 dB Ideal channel knowledge, floating point implementations

doc.: IEEE /257 Submission Slide 10 May 2001 Coffey et al, Texas Instruments Baseline comparisons, 100 ns, contd.

doc.: IEEE /257 Submission Slide 11 May 2001 Coffey et al, Texas Instruments Implications of 5 dB advantage: 5 dB translates to a factor of 3.1  For similar throughput and range, PBCC requires 3 times less received power than CCK-OFDM 24 Mbps – translates to greater battery life  For similar throughput and received power, PBCC has 40% more range than CCK-OFDM 24 Mbps –Assuming the “power of 3.3” model for path loss – this is the standard model used in (Doc /138r0)

doc.: IEEE /257 Submission Slide 12 May 2001 Coffey et al, Texas Instruments 100ns: 40% PBCC range advantage PBCC 22 Mbps CCK-OFDM 24 Mbps Double the coverage

doc.: IEEE /257 Submission Slide 13 May 2001 Coffey et al, Texas Instruments Relative throughputs:

doc.: IEEE /257 Submission Slide 14 May 2001 Coffey et al, Texas Instruments Actual PBCC receiver algorithms, 100 ns 2.9 dB Ideal channel knowledge, floating point implementation for CCK-OFDM

doc.: IEEE /257 Submission Slide 15 May 2001 Coffey et al, Texas Instruments Baseline comparisons, contd: 250 ns 4.5 dB Ideal channel knowledge, floating point implementations

doc.: IEEE /257 Submission Slide 16 May 2001 Coffey et al, Texas Instruments Baseline comparisons, 250 ns, contd.

doc.: IEEE /257 Submission Slide 17 May 2001 Coffey et al, Texas Instruments What is wrong with CCK-OFDM that is right with pure 11a OFDM?

doc.: IEEE /257 Submission Slide 18 May 2001 Coffey et al, Texas Instruments The CCK-OFDM Dilemma Any receiver requires overhead for channel estimation, tracking, etc: –802.11a is “pay as you go”  ultra-short preamble, 16 musecs –802.11b and PBCC-22 are “pay all up front”  11b “short” preamble, 96 musecs –CCK-OFDM is “pay up front and again as you go”  11b “short” preamble, plus (non-standard) OFDM preamble, 110 musecs “Double the pain”

doc.: IEEE /257 Submission Slide 19 May 2001 Coffey et al, Texas Instruments Overhead and Data Payloads CCK-OFDM PBCC 11a No acks, 500 byte packets

doc.: IEEE /257 Submission Slide 20 May 2001 Coffey et al, Texas Instruments Relative throughputs, contd.

doc.: IEEE /257 Submission Slide 21 May 2001 Coffey et al, Texas Instruments Packet size & system performance Compare performances – results are critically dependent on packet size –Short packets (e.g., MPEG-4 packets of 188 bytes) strongly favor “pay-as-you-go” approach  a/Hiperlan 2 –Long packets increasingly favor “pay-all-up-front” approach”  PBCC-22 aimed at this application –Short or long, it won’t work well if it’s CCK-OFDM

doc.: IEEE /257 Submission Slide 22 May 2001 Coffey et al, Texas Instruments Conclusions  PBCC-22 has a natural superiority over CCK-OFDM in multipath. Established IEEE multipath model used.  CCK-OFDM tries to juggle two incompatible things – makes an underperforming system out of a merger of two otherwise good components  A better way of doing things – PBCC a!