1. The “Black Box” PHY Abstraction Methodology
Month 2004
doc.: IEEE /0172r0
The “Black Box” PHY Abstraction Methodology
Jeff Gilbert, Won-Joon Choi, Qinfang Sun, Ardavan Tehrani, Huanchun Ye Atheros Communications
B.Jechoux, H.Bonneville Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

2. PHY Abstraction problem
Month 2004
doc.: IEEE /0172r0
PHY Abstraction problem
PHY / MAC Interface can drastically impact overall results:
Time varying channel creates time varying PER
Time varying channel could affect systems with feedback
This affects overall delay, jitter and throughput
Challenge
Properly model detailed PHY characteristics
Keep flexibility to adapt to various PHYs
Keep simulation effort reasonable
Jeff Gilbert et. al., Atheros / Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

3. Two Basic Approaches Model PHY as black box using tables (more here)
Month 2004
doc.: IEEE /0172r0
Two Basic Approaches
Model PHY as black box using tables (more here)
Allows use of full-accuracy PHY model
PHY model used “as-is” – no formulas or approximations required
Approximations made at PHY/MAC boundary
Incorporate simplified PHY into MAC sim (Intel)
Use derived, approximate model of PHY
Incorporate directly into MAC/System simulations – interface cleaner
Jeff Gilbert et. al., Atheros / Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

4. The Black Box PHY Method
Month 2004
doc.: IEEE /0172r0
The Black Box PHY Method
Consider PHY and Channel Model combo as a “black box” from MAC perspective
Critical to allow accurate modeling of all proposals’ PHY in an accurate and automated manner
Use of look-up tables giving PHY performance vs. range / environment
Bursty aspects of the PER modeled by PER distribution plus channel coherence time
Rate adaptation modeled in the black box as well to allow rich interaction between PHY and rate adaptation
Jeff Gilbert et. al., Atheros / Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

5. MAC / System Model w/ Rate Adaptation
Conventional Table-Based PHY Simulation
Pre-generates table for MAC simulations
Conventional Table-Based MAC Simulation
Uses PHY simulation data for MAC simulation
Distance & Model Num
Table
Channel Model
Statistics of PERs per data rate and MPDU size
Channel
Black Box
Data rates
Randomly choose pass / fail based on per-rate statistics
PHY Model
Statistics of PERs per data rate and MPDU size
Data rate
Pass/Fail
MAC / System Model w/ Rate Adaptation
Table
Conventional table-based phy simulations have difficulties simulating systems with many rates (ABL, MIMO etc) since PHY sims scale with the number of rates
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

6. Including Rate Adaptation w/ PHY
Month 2004
doc.: IEEE /0172r0
Including Rate Adaptation w/ PHY
Typical table-based systems record PER statistics for each data rate
For MIMO with independent rates on each stream, the number of rate combinations is NumRatesNumTxStreams
For Adaptive Bit Loading, rate set is continuous
This is solved by including rate adaptation w/ PHY
Number of runs does not grow with number of data rates
Richness of PHY / rate adaptation interface is not limited by storing in table
Jeff Gilbert et. al., Atheros / Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

7. Incorporating Channel Variation
Month 2004
doc.: IEEE /0172r0
Incorporating Channel Variation
Two types of channel variation are incorporated:
Micro-variation
Channel variation seen over the time of few packets
Required to exercise and evaluate rate adaptation
Captured by evolving channel between pkts in PHY sims
Macro-variation
Channel variation seen over long time scales
Accounts for run to run variations and outage statistics
Captured by starting w/ several “representative” chans in PHY sims
Results from mix of representative chans used in MAC simulation
Jeff Gilbert et. al., Atheros / Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

8. Proposed PHY / Rate Adaptation Sim Proposed MAC Simulation
Pre-generates table for MAC simulations
Proposed MAC Simulation
Uses PHY simulation data for MAC simulation
Distance & Model Num
Table
Statistics of pairs of “data rates” / PERs per representative channel
Channel Model
Random selection of which representative channel(s) to use
“Representative” channels
Interpolate between representative channels
Black Box
Feedback
Rate Adaptation
PHY Model
Data rates and PERs
Rate Selection
Randomly choose data rate, pass / fail based on statistics
Statistics of pairs of “data rates” / PERs per representative channel
Data rate, Pass/Fail
MAC / System Model
Table
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

9. Month 2004
doc.: IEEE /0172r0
The Table Data
For each location pair, the table contains statistics for several “representative channels”
For each representative channel, store histogram of rates selected and their PERs
The table is used to generate packets in MAC / System simulation by selecting a rate per packet based on statistics and using its PER to determine if the packet will succeed
Jeff Gilbert et. al., Atheros / Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

10. Choosing Representative Channels
Generate a large number of channel realizations for given n channel model,
Compute MIMO capacity for each channel,
Sort channels in ascending order of capacity, ordered channels:
Choose channels
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

11. PHY / MAC Table Data Structure
Channel Quality
Q=0.00
Q=0.25
Q=0.50
Q=0.75
Q=1.00
6 Mbps PASS
9 Mbps PASS
9 Mbps FAIL
12 Mbps FAIL
9 Mbps: PASS
9 Mbps: FAIL
12 Mbps: PASS
12 Mbps: FAIL
18 Mbps: PASS
24 Mbps: FAIL
18 Mbps: FAIL
24 Mbps: PASS
36 Mbps: PASS
Sequences
6M: 40%, PER 0.0
9M: 40%, PER 0.5
12M: 20%, PER 1.0
9M: 60%, PER 0.3
12M: 40%, PER 0.5
12M: 60%, PER 0.3
18M: 20%, PER 0.0
24M: 20%, PER 1.0
12M: 20%, PER 0.0
18M: 40%, PER 0.5
24M: 40%, PER 0.5
24M: 60%, PER 0.3
36M: 20%, PER 0.0
Statistics
Per-quality summary statistics include prob. of occurrence and PER for the data rates used
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

12. Many-Rate PHY Operation
Month 2004
doc.: IEEE /0172r0
Many-Rate PHY Operation
PHY simulation time is independent of the number of data rates
This is why rate adaptation needed to be incorporated into PHY simulation
If many different rates are selected, the statistics on each rate may be coarsely sampled but when aggregated they will be accurate
I.e. the PER accuracy scales with the total number of packets simulated, and not the number of packets per rate as with conventional table methods
Jeff Gilbert et. al., Atheros / Mitsubishi ITE
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

13. PHY Simulation Details
Run detailed PHY simulations using the 5 representative channels as initial conditions, one set of simulation per initial condition and MPDU size.
Each set of PHY simulation shall include:
Time variation due to small scale fading
N=100 packets with specified packet spacing
Rate adaptation/feedback
Output of each set of PHY simulation:
(Ri, di), 1 i  N, where Ri is the data rate of packet i and di is pass or fail.
Condense into: (Rk, rk, Pk), where k is an arbitrary data rate index, pk is the probability of using rate k, and pk is the PER of rate k.
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

14. PHY / Rate Adaptation Simulation
Distance & Model Num
PHY / Rate Adaptation Simulation
Channel Model
Channel characteristics
Choose R=5 channels representing different quality points DT
“Representative” channels
Add Micro-variation
Channel sequence
Feedback
Rate Adaptation
PHY Model
Run N=100 pkts
Rate
Data rate, PER, Prob of occurrence sets per “Representative” channel
MPDU size(s)
Table
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

15. Choosing DT
The PHY / rate adaptation needs to know an inter-packet interval to incorporate the correct amount of inter-packet variation
This must be determined prior to the PHY simulations heuristically from the simulation scenarios
The DT value only affects channel variation scaling
The DT could be a fixed value or distribution
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

16. Table
MAC Simulation
Pick random channel “quality” index between 0.0 and 1.0.
Data rate, PER, Prob of occurrence sets per “Representative” channel
New Quality = f(DT, tc ,old Quality)
Quality
Interpolate between two closest “quality” channels
Data rate, PER, Prob. occur set
Randomly choose data rate / PER pair based probability of occurrence
Data rate, PER
Randomly select fail succeed based on PER
Data rate, Pass/Fail
MAC Model
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

17. Channel Evolution Function
f(DT, tc ,old Quality) is used to capture macro scale variations
Some options include:
Generate random channel qualities and low-pass filter based on channel coherence time (tc )
Markov models to model the channel variation (ST Microelectronics – 11-04/0064)
Evolve full channels based on channel model, determine capacity, map to quality and use this quality index
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

18. Quality Interpolation / Packet Sequence Generation
Channel Quality
Q=0.00
Q=0.25
Q=0.50
Q=0.75
Q=1.00
6M: 40%, PER 0.0
9M: 40%, PER 0.5
12M: 20%, PER 1.0
9M: 60%, PER 0.4
12M: 40%, PER 0.5
12M: 60%, PER 0.4
18M: 20%, PER 0.0
24M: 20%, PER 1.0
12M: 20%, PER 0.0
18M: 40%, PER 0.5
24M: 40%, PER 0.5
24M: 60%, PER 0.3
36M: 20%, PER 0.0
To generate Q=0.4:
40% weight
60% weight
9 Mbps: PASS
12 Mbps: PASS
12 Mbps: FAIL
24 Mbps: FAIL
18 Mbps: PASS ….
9M: 24%, PER 0.40
12M: 52%, PER 0.44
18M: 12%, PER 0.00
24M: 12%, PER 1.00
Sequence of packet for MAC simulation
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

19. Issues
Co-channel or adjacent channel interference would have to be modelled independent of MAC
However usage models do not include much
CSMA/CA still handled correctly in MAC simulation
Rate adaptation approximations
Collision effects incorporated in MAC correctly result in packet losses but do not affect rate adaptation
The DT used in PHY / rate adaptation simulations determined a priori
Channel variation is present but not exact
Jeff Gilbert et. al., Atheros / Mitsubishi ITE

20. Conclusions
The “Black Box PHY” methodology allows arbitrary PHYs to be included in MAC/System simulations with little MAC sim computation
Incorporating rate adaptation into PHY simulations facilitates the use of systems with many rates (MIMO, Adaptive Bit Loading)
Channel variation is incorporated in an approximate manner
Some approximations in PHY / MAC interface
Jeff Gilbert et. al., Atheros / Mitsubishi ITE