1. PHY Abstraction based on PER Prediction
March 2004
PHY Abstraction based on PER Prediction
John Ketchum, Bjorn Bjerke, Irina Medvedev, Sanjiv Nanda, Rod Walton
Qualcomm Inc.
John Ketchum et al, Qualcomm

2. MAC System Sim Requirements
March 2004
MAC System Sim Requirements
Must have TGn channel model running under MAC
Must provide method for accurately determining post-detection SNRs (after Rx spatial processing) associated with specific channel state at packet arrival instant
Must provide method for predicting PER and generating error process from these SNRs
Details of rate adaptation techniques on which proposal performance is based must be included in simulation
Full disclosure, sufficient for validation, of simulation methodology must be included with simulation results, including
Method of calculating post-detection SNR
Method of calculating/predicting PER from post-detection SNR
John Ketchum et al, Qualcomm

3. March 2004
Basic Approach
Run TGn channel models with PHY abstraction under MAC simulation
Space-frequency channel is explicitly captured in the simulator
Caveat:
With this approach, specifics of PHY abstraction are likely to be dependent on details of coding, modulation, and spatial processing.
At most, TGn should specify general approach, and provide guidelines, suggestions for specifics.
John Ketchum et al, Qualcomm

4. PHY Abstraction Assumptions
March 2004
PHY Abstraction Assumptions
One or more spatially segregated streams resulting from MIMO processing
Each spatial stream consists of sequence of OFDM symbols.
Each spatial stream operates at a specific identifiable coded rate
Per-bin rates also supported
Channel is stationary over individual packet
=>each ofdm bin in each stream can reasonably be treated as an AWGN subchannel
Estimate of SNR in each OFDM bin of each spatial stream is available at the output of Rx spatial processing
John Ketchum et al, Qualcomm

5. PHY Modeling Approach Detailed link level simulation including
March 2004
PHY Modeling Approach
Detailed link level simulation including
TGn channel model with Doppler
Rate control
Spatial processing
RF impairments
Coding/decoding, interleaving/deinterleaving, etc
Used to verify/calibrate PHY abstraction
John Ketchum et al, Qualcomm

6. Month 2002
doc.: IEEE /xxxr0
March 2004
PHY Modeling Approach
Going back to basic principles, average frame error probability in a slowly fading channel is given by where is the frame error probability in an AWGN channel at SNR , and is the pdf of
In other words, if the channel is fading sufficiently slowly, we can treat the channel as if it were an AWGN channel with a fixed SNR over the course of a frame, then average the conditional error probability over the fading statistics.
John Ketchum et al, Qualcomm
John Doe, His Company

7. Month 2002
doc.: IEEE /xxxr0
March 2004
PHY Modeling Approach
Since the TGN channel models are slow fading channels relative to frame durations, we can take this approach in the PHY abstraction:
Treat the channel as an AWGN matrix channel for duration of frame
Post-detection snr determined by spatial processing
Determined by Gaussian noise plus any self-interference
Doppler over duration of frame manifests as channel impairments due to estimation/tracking errors
John Ketchum et al, Qualcomm
John Doe, His Company

8. SNR at Output of Spatial Processing
March 2004
SNR at Output of Spatial Processing
Received signal , where is a transformation on the transmitted sign vector,
Output of received spatial processing: where is transformation on the received signal vector, and is the total noise vector, consisting of both AWGN and residual interference due to incomplete isolation of spatial streams.
John Ketchum et al, Qualcomm

9. SNR at Output of Spatial Processing
March 2004
SNR at Output of Spatial Processing
Then post-detection SNR for stream i is
The transformation is determined by the spatial processing performed at Tx and Rx.
For linear processing, these are simple and well-known.
For non-linear processing (e.g. successive cancellation) exact functions may not be known, and approximations must be used.
John Ketchum et al, Qualcomm

10. Calculating Effective SNR
March 2004
Calculating Effective SNR
Effective SNR for stream i: where a is a constant used to fit the approximation to PHY simulation results.
John Ketchum et al, Qualcomm

11. Calculating PER from SNReff(i)
March 2004
Calculating PER from SNReff(i)
Get AWGN coded bit error probability for effective SNR: , for rate used in stream i, from LUT
Calculate approximate error event probability from Pb(i)
Calculate approximate probability of error in stream i where is the number of coded bits in stream i, , K is the code constraint length, and is the code rate used in stream i.
Then the PER is
John Ketchum et al, Qualcomm

12. March 2004
Results for 4 Cases
Full link sim includes channel estimation procedures at both ends of the link
Frequency and timing offset are set to 0
1 data stream, 18 Mbps:
a = 0
R=3/4, QPSK
2 data streams, 78 Mbps:
a = 0.2 (needs further fine-tuning)
R=5/6, 64-QAM
John Ketchum et al, Qualcomm

13. Results (cont’d) 3 data streams, 108 Mbps: 3 data streams, 156 Mbps:
March 2004
Results (cont’d)
3 data streams, 108 Mbps:
a = (needs further fine-tuning)
R=5/6, 64-QAM
R=3/4, 16-QAM
R=1/2, QPSK
3 data streams, 156 Mbps:
a = 0
R=7/8, 256-QAM
R=3/4, QPSK
John Ketchum et al, Qualcomm

14. March 2004
John Ketchum et al, Qualcomm

15. March 2004
John Ketchum et al, Qualcomm

16. March 2004
John Ketchum et al, Qualcomm

17. March 2004
John Ketchum et al, Qualcomm

18. Summary Verified accuracy of error model
March 2004
Summary
Verified accuracy of error model
Easily integrated into TGn channel model and NS or other system simulator
Incorporates details of TGn channel models into system simulator.
John Ketchum et al, Qualcomm

19. Channel Model Memory Requirements
March 2004
Channel Model Memory Requirements
For each unique link configuration (Ntx,Nrx,Channel model):
1.33 Mbytes persistent
58.3 kbytes local/transient
For each link
65.6 kbytes persistent
1.55 Mbytes local/transient
John Ketchum et al, Qualcomm