doc.: ieee 802.11-13/1390r0 submission nov. 2013 yakun sun, et. al.slide 1 phy abstraction for hew...

doc.: IEEE 802.11-13/1390r0

Submission

Nov. 2013

Yakun Sun, et. Al.Slide 1

PHY Abstraction for HEW System Level SimulationDate: 2013-11-11

Authors:

Name Affiliations Address Phone email

Yakun Sun Marvell Semiconductor

5488 Marvell Ln, Santa Clara, CA 95054

1-408-222-3847 [email protected]

Yan Zhang Marvell Semiconductor

Hongyuan Zhang Marvell Semiconductor

Hui-Ling Lou Marvell Semiconductor

Mingguang Xu Marvell Semiconductor

doc.: IEEE 802.11-13/1390r0

Submission

Introduction

• System simulation has been adopted as a powerful tool in investigating network performance.– Critical to evaluate HEW, whose target includes improving system and

edge-of-network throughput .– Simulate multiple BSSs simultaneously on the intra- and inter-BSS

interactions.

• Physical layer abstraction is used to simplify the complicated simulation of a large number of APs and STAs.– Relieve system simulation from transmitting and decoding real PHY

packets, and align simulator behaviors from different companies.– Predict if a packet can be successively received from instantaneous

channel conditions.

Nov. 2013


doc.: IEEE 802.11-13/1390r0

Submission

How Does PHY Abstraction Work?• System simulator transmitter “sends” a virtual encoded packet over frequency-selective

channels.– No encoding or signal generation actually happens. – No packet travels through channels but channel realizations are generated.

• System simulator receiver “receives” the virtual packet by calculating the post-processing SINR values per subcarrier. – Equalizer/MIMO impact on performance kicks in.

• PHY abstraction predicts instantaneous PER based on the SINR values (given the current channel realization). – Namely, a function with a vector of SINR values as input and a PER as output. – This function depends on the coding scheme (BCC, or LDPC) one table per coding scheme.

• System simulator takes the predicted PER to decide if this virtual packet has passed through.– Flip a coin based on PER.

• This approach has been widely used in IEEE 802.16m [1] and 3GPP [2].

Nov. 2013


System Level

· Generate frequency selective channel H(f)

· Determine the received SINR of each sub-carrier

Link adaptation, Scheduling, ARQ, etc.

Link Level

BLERAWGN

(PHY AbstractionMapping)

Mapping Function e.g. MIESM, EESM

of each subchannel

BLER

Throughput, packet error rate, etc.

SINR

doc.: IEEE 802.11-13/1390r0

Submission

Challenge on PHY Abstraction• PHY abstraction function maps a vector to a scalar

– f: RNR; where N is the number of SINRs over frequency/time.

• This is a very challenging task:– It is impossible to pre-store the mapping table due to N-to-1 mapping, as well as arbitrary types of

fading channels.– It is, however, fairly easy to store a set of SNR vs. PER tables for AWGN channels (i.e., 1-to-1

mapping).

• The solution is to find an AWGN channel at an equivalent SNR level having PER performance the same as the fading channel.– In other words, map (compress) a vector of SINR values to a single SNR scalar effective SNR

mapping (ESM).• The key factors of ESM are

– (1) simple, (2) accurate, (3) channel independent (the ESM method, and the parameters do not change across different channel types).

– For example, linear/dB average SINR is NOT a good ESM method.

Nov. 2013


doc.: IEEE 802.11-13/1390r0

Submission

ESM for PHY Abstraction• Effective SINR Mapping has been adopted in system

level simulation for IEEE 802.16m[1] and 3GPP LTE [2,3].

• Effective SINR is an average mapped equalizer-output SINR over all subcarriers.– Hedge factors alpha and beta can be used to calibrate and

compensate any residual errors.

• OFDM transmission is modeled as an AWGN channel with one effective SINR.

Nov. 2013


1

1

1 Nn

effn

SINRSINRN

doc.: IEEE 802.11-13/1390r0

Submission

SINR Mapping Functions

• A list of well-known SINR mapping functions

Nov. 2013


PHY Abstract SINR Mapping

EESM [1, 2, 3] Exponential mapping

MIESM (RBIR) [1, 2, 4]

Mutual information per symbol

MMIB [1, 2, 5] Mutual information per bit

expx x

1 1

1i

M K

b k ki k

x I x a J c xM

222 2

1 1

1log log expM M

U k mm k

x M E U x s s UM

doc.: IEEE 802.11-13/1390r0

Submission

MIESM for BICM

• Suppose a SISO channel,

• RBIR in the previous table is mutual information for such a SISO channel, achieved by coded modulation.

• BICM is widely used for advanced wireless systems including WiFi.– CM based mutual information (RBIR) is overestimated for BICM.

Nov. 2013


2 , 2

222 2

1 1

|; | log log

|

1log log exp

z SS Y

M M

U k mm k

p y zx I S Y SINR x M E

p y s

M E U x s s UM

1~ 0, ;r hs n y s u u CN s SSINR

doc.: IEEE 802.11-13/1390r0

Submission

MIESM for BICM (2)

• Considering BICM, MIESM can be given as [6]– Referred as “RBIR-BICM”

• Mutual information for each bit is given as

• Mutual information for this channel use is given by

Nov. 2013


, 2

|; 1 log

|ib

z Si b Y

z S

p y zI b Y E

p y z

2 2

2

log log 11

2 2 21 1 0

exp1; log log

expib

ik b

M

M M kk

i Ui i b s S

ks S

x s s Ux I b Y M E

M x s s U

doc.: IEEE 802.11-13/1390r0

Submission

Difference of RBIR Mapping

• RBIR and RBIR-BICM are close but with some gap.– At most 1dB apart for 64QAM.

Nov. 2013


-10 -5 0 5 10 15 20 250

1

2

3

4

5

6

SNR (dB)

Mut

ual I

nfor

mat

ion

(bps

)

QPSK

16QAM

64QAM

RBIRRBIR-BICM

doc.: IEEE 802.11-13/1390r0

Submission

Performance of PHY Abstraction• 11ac, 1x1, 8000 bit per packet, MCS0-MCS7, BCC

– EESM is not considered here without well known parameters for BCC.– Channel D-NLOS, AWGN

• Effective SNR vs. PER curves for D-NLOS are referenced to SNR vs. PER curves for AWGN channels.– The closer, the better!

• All three methods (MMIB, RBIR, RBIR-BICM) provides good PER results referenced to AWGN.– RBIR-BICM and MMIB (both bit-level MI) are closer than RBIR

(symbol level MI) to AWGN performance except MCS0.– All three methods perform the same for MCS0 (BPSK).

Nov. 2013


doc.: IEEE 802.11-13/1390r0

Submission

Performance of PHY AbstractionNov. 2013


0 5 10 15 2010

-3

10-2

10-1

100

Effective SNR (dB)

PE

R

AWGN

DNLOS, RBIR-BICM

DNLOS, RBIR

DNLOS, MMIB

• The gap between effective SNR to SNR is no more than 0.6dB across MCSs.

doc.: IEEE 802.11-13/1390r0

Submission

RBIR-BICM Fine TuneNov. 2013


• After applying some hedge factors per MCS (basically dB shift), RBIR-BICM can provides almost exact PER results as AWGN.

0 5 10 15 2010

-3

10-2

10-1

100

Effective SNR (dB)

PE

R

AWGN

DNLOS, RBIR-BICM

DNLOS, RBIR-BICM, retune

doc.: IEEE 802.11-13/1390r0

Submission

Comments on RBIR-BICM

• RBIR-BICM matches AWGN performance better than RBIR.

• RBIR is easier to extend to high modulation than MMIB for the availability of theoretical expressions.– Although still requires numerical evaluation (or via Monte Carlo),

it does not require any curve fitting/parameter (a_k, c_k) optimization as for MMIB.

Nov. 2013


doc.: IEEE 802.11-13/1390r0

Submission

Summary

• Both MMIB and RBIR can effectively predict OFDM performance.

• RBIR-BICM and MMIB perform better than RBIR referenced to AWGN results.

• RBIR-BICM is easier to extend to high modulations than MMIB.

• RBIR-BICM with some dB shift can almost exactly match AWGN performance.

• Suggest to take RBIR-BICM as the PHY abstraction technique for HEW system simulations.

Nov. 2013


doc.: IEEE 802.11-13/1390r0

Submission

References

[1] IEEE 802.16m-08/004r5, Jan. 2009[2] R1-050680, “Text Proposal: Simulation Assumptions and Evaluation for

EUTRA”, 3GPP TSG RAN WG1 #41bis, June, 2005[3] R1-061626, “LTE Downlink System Performance Evaluation Results”, 3GPP

TSG RAN1 #45, May, 2006[4] 11-13-1131-00-0hew-phyabstraction-for-hew-system-level-simulation[5] 11-13-1059-00-0hew-phy-abstraction-for-hew-evaluation-methodology[6] “Bit-Interleaved Coded Modulation”, Giuseppe Caire, Giorgio Taricco, and

Ezio Biglieri, IEEE Trans. Of Info. Theory, 1998.

Sept 2013


doc.: ieee 802.11-13/1390r0 submission nov. 2013 yakun sun, et. al.slide 1 phy abstraction for hew...

Documents