PER Prediction for 802.11n MAC Simulation March 2004 PER Prediction for 802.11n MAC Simulation John S. Sadowsky ( john.sadowsky@intel.com ) Intel John S. Sadowsky, Intel
Overview Review of methodology PHY Model Fit Example Summary March 2004 Overview Review of methodology PHY Model Fit Example Summary References 11-03/0863 (Sadowsky & Li) 11-04/0174 (Ketchum, Bjerke, Nanda, Walton & Sadowsky) John S. Sadowsky, Intel
Freq. Selective Fading & Interference March 2004 Freq. Selective Fading & Interference John S. Sadowsky, Intel
PER Prediction from PSymb March 2004 PER Prediction from PSymb Psymb = prob. of a Viterbi decoder error within the duration of a single OFDM symbol Psymb is independent of packet length Allows scaling to arbitrary packet lentghs Basic Assumption: symbol errors are ~ independent OFDM symbols > several constraint lengths good approx. See 11-03-0863 for validation John S. Sadowsky, Intel
March 2004 One OFDM Symbol John S. Sadowsky, Intel
March 2004 Psymb Calculation John S. Sadowsky, Intel
Soft Bit SNRs as delivered to Viterbi decoder March 2004 Soft Bit SNRs as delivered to Viterbi decoder OFDM Symbol OFDM Symbol OFDM Symbol The OFDM symbol window is the natural block size for PER prediction because the soft bit SNRs, as presented to Viterbi decoder, are periodic with this with period = to this block size. John S. Sadowsky, Intel
Post Detection SNRs Channel Linear Equalizer March 2004 John S. Sadowsky, Intel
Post Detection SNRs Example: Ideal Zero Forcing (unbiased) March 2004 Post Detection SNRs Example: Ideal Zero Forcing (unbiased) Example: Zero Forcing with Channel Estimation Error where channel estimation error added as a random matrix of variance determined by the estimators processing gain John S. Sadowsky, Intel
Parametric Model Fit = mean capacity March 2004 Parametric Model Fit Capacity statistics calculated from subcarrier-spatial stream capacities = mean capacity CV = capacity coefficient of variation (std. deviation / mean) John S. Sadowsky, Intel
Example Model Fit MIMO Receiver = MMSE Two MIMO Spatial Streams March 2004 Example Model Fit MIMO Receiver = MMSE random channel estimation errors (PG = 3 dB) Two MIMO Spatial Streams 2x2 configuration no diversity 2x3 configuration Rx diversity 64 QAM, Rate ¾ 576 coded bits, 432 data bits Channel Models: B, D & F (NLOS) John S. Sadowsky, Intel
PHY Simulations For k = 0, …, N generate a channel realization March 2004 PHY Simulations For k = 0, …, N generate a channel realization calculate and CV for current channel simulate with fixed channel - stop after 500 packet errors store , CV and estimates Packet size = 1000 bytes 19 symbols per packet John S. Sadowsky, Intel
March 2004 ~400 data points John S. Sadowsky, Intel
Same Data – organized by CV (instead of B-D-F and 2x2 v 2x3) March 2004 Same Data – organized by CV (instead of B-D-F and 2x2 v 2x3) John S. Sadowsky, Intel
Parametric Model Summary March 2004 Parametric Model Summary One Fit works for ALL Channel Models This is a worst case example! (weak coding and no diversity w/ 2x2) Quality of Fit RSS for = 0.0236 +40% or -30% standard error on Fit parameters RSS = Residual Sum of Squares John S. Sadowsky, Intel
March 2004 John S. Sadowsky, Intel
Summary Methodology Advantages TGn channels generated in MAC simulator March 2004 Summary Methodology TGn channels generated in MAC simulator PHY abstraction at FEC decoder Receiver captured in MSE calculations MSE calculation subcarrier SNR subcarrier capacity Subcarrier-spatial stream capacity statistics Receiver captured in Predict symbol error prob. PER Advantages Simple and accurate PER prediction NO lookup tables! Common fit across all channel models! All MAC functions implemented in MAC simulator eg. rate adaptation is NOT fold into an ensemble average LUT John S. Sadowsky, Intel