Markov Modeling of the Channel for HEW System

September 2013
doc.: IEEE 11-13/1080r0
Markov Modeling of the Channel for HEW System
Level Simulations
Date: 2013-09-17
Authors:
Name
Affiliations
Address
Joseph LEVY
InterDigital
2 Huntington
+1 631 622
Quadrangle,
4139
Melville, NY 11747
Frank LA SITA
InterDigital
Pengfei XIA
InterDigital
Fengjun XI
InterDigital
Submission
Slide 1
Phone
email
[email protected]
om
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Abstract
This contribution proposes that a Markov Model be used
in system level simulations to provide an accurate and
efficient means of including fast fading of outdoor
channels for HEW system level modelling. Similar
methods have been used by 3GPP for LTE modelling
and 802.16 for 802.16e modelling.
Submission
Slide 2
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Motivation
To provide an accurate and efficient outdoor physical
channel model for HEW system level simulations.
Markov modeling techniques can provide such a system
level simulation model. ([2], [5], [10])
(10^5) can be used for any
The Markov modeling technique
agreed channel model (e.g. WINNER2, ITU, or
customized)
The modeling approach and channel model(s) used for
system level simulations are important considerations
which should be agreed to allow for meaningful
comparison of performance results.
Submission
Slide 3
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Discussion
This contribution provides:
• A description of the proposed Finite State
Markov Chain (FSMC) system level modeling
• Examples of calculated transition probability
matrix (TPM) for some channel models of
(10^5)
interest.
Submission
Slide 4
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Description of the proposed Finite State
Markov Chain (FSMC) system level
modeling
Submission
Slide 5
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Finite State Markov Chain
Model the channel SNR as a finite-state Markov chain (FSMC)
Each state represents a given value (range) of the SNR
The following example has four different SNR states
1.4
1.2
instantaneous SNR
1
0.8
0.6
0.4
0.2
0
Submission
Slide 6
0
1
2
3
4
5
6
Sample index
7
8
9
10
4
x 10
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Use of FSMC in System Level Simulations
Submission
Slide 7
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Transition Probability Matrix (TPM)
Transition probability matrix
pi,j : prob that the jth state is next, given that the ith state is current
0
p 1,1 p 1, 2
p
p 2, 2 p 2,3
2 ,1


 0
...
...

0 p N 1, N
 0
0 
0 
... 

p N ,N 
𝜋𝑖 = 𝜋𝑖1 𝜋𝑖2 ⋯ 𝜋𝑖𝑁 be the prob distribution function when
current state is the ith state
The underlying physical channel is fully characterized by the
corresponding TPM matrix.
Submission
Slide 8
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Multi-Carrier SNR Mapping
In modeling the physical channels, we need to convert multiple SNR
values (one per subcarrier) into one single SNR value
Potential approaches
Throughput averaging (algo 1)
Straightforward, not MCS dependent
| H o |2 effective channel amplitude square
1
N
 log 1    | H |   log 1    | H | 
2
2
i
2
2
o
i
SNR averaging (algo 2)
Used by OPNET for LTE downlink/uplink
Exponential effective SNR mapping (EESM), received bit mutual
information (RBIR), Mean mutual information per bit (MMIB) [10]
MCS dependent (not studied herein)
Submission
Slide 9
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Examples of calculated transition
probability matrix (TPM) for some channel
models of interest.
Submission
Slide 10
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Proposed HEW Use Cases [9]
1
High density of APs and high
density of STAs per AP
2
High density of STAs – Indoor
3
High density of APs (low/medium
density of STAs per AP) – Indoor
4
High density of APs and high
density of STAs per AP – Outdoor
5
High throughput demanding
applications
Submission
a
b
c
d
e
f
a
b
c
d
a
stadium
airport/train stations
exhibition hall
shopping malls
E-Education
Multi-media Mesh backhaul
dense wireless office
public transportation
lecture hall
Manufacturing Floor Automation
dense apartment building
b
Community Wi-Fi
a
Super dense urban Street
b
Pico-cell street deployment
c
a
b
Macro-cell street deployment
surgery/health care (similar to 2e from 11ac)
production in stadium (similar to 1d-1e from 11ac)
c
smart car
Slide 11
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Urban Micro Channels in HEW
• Urban Micro cellular environment fits well in HEW [8]
• “The microcellular test environment focuses on small cells and high
user densities and traffic loads in city centers and dense urban
areas. The key characteristics of this test environment are high
traffic loads, outdoor and outdoor-to-indoor coverage. This scenario
will therefore be interference-limited, using micro cells.” [7]
WINNER 2 model
Metropolitan (C2)
Typical Urban (B1, B4)
Indoor to outdoor (A2)
Rural macro (D1)
Submission
ITU model
Urban macro (UMa)
Urban micro (UMi)
Indoor (InH)
High speed (RMa)
Slide 12
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
TPM Generation Algorithm
Generate multiple (10^5) pairs of channel samples, each pair are
separated by N OFDM symbols (N * 3.2 ms)
Each consists of multiple taps according to the channel model, e.g.
WINNER 2 or ITU
Will serve as current state and next state in statistics collection
For each sample in the channel sample pair,
Convert it to freq domain using DFT
Convert multiple subcarrier SNRs into one effective SNR (either SNR
averaging or throughput averaging)
Quantize the effective SNRs into P (16) equi-prob states
Find the probability of each state transitioning into other states
accordingly
Submission
Slide 13
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Simulation Assumptions
1) Channel Scenarios: WINNER 2 B1, ITU UMi
Directly comparable channel scenarios
2) Single transmit and single receive antenna
3) OFDM symbol separation between current and next state: 10 or 100
4) Large scale signal to noise ratio: 0 or 20dB
5) Mobile velocity: 3 or 30 km/hr
6) Two algorithms considered for determination of effective SNR:
1) Throughput averaging
2) SNR averaging
Submission
Slide 14
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
ITU/WINNER 2 100
difference,
100 symbol,
dB snr,3030km/h,
km/hr,Throughput
algo 1
ITU/WINNER 2 Comparison,
symbols.
20dB20SNR,
Averaging
Diagonal max diff 0.10
Diagonal mean diff 0.05
Off-diagonal max diff 0.09
Off-diagonal mean diff 0.01
ITU Channel
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0
5
10
15
10 11
8 9
7
6
4 5
3
1 2
12
15
13 14
16
WINNER 2 Channel
5
10
15
10 11
8 9
7
6
4 5
3
1 2
14
12 13
The generated TPM are more or less similar for WINNER 2 B1 and
ITU UMi
Submission
Slide 15
Joseph Levy, InterDigital Communications Inc..
15 16
September 2013
doc.: IEEE 11-13/1080r0
ITU/WINNER 2 ITU/WINNER
Comparison,
100 symbols.
20dB
30km/hr,
km/h,algo
SNR
2 difference,
100 symbol,
20 dBSNR,
snr, 30
2 Averaging
Diagonal max diff 0.18
Diagonal mean diff 0.06
Off-diagonal max diff 0.07
Off-diagonal mean diff 0.01
ITU Channel
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0
5
10
15
9 10
7 8
6
5
3 4
2
1
11 12
15
13 14
WINNER 2 Channel
5
16
10
15
9 10
7 8
6
5
3 4
2
1
13
11 12
14 15
The generated TPM are more or less similar for WINNER 2 B1 and
ITU UMi
Submission
Slide 16
Joseph Levy, InterDigital Communications Inc..
16
September 2013
doc.: IEEE 11-13/1080r0
SNR Comparison, 20dB
vs.vs0dB,
100 symbols,
30 km/h,
WINNER
Throughput
20dB
0 dB comparison,
100 symbol,
30 km/hr,
WINNER22 B1
Channel
B1, algo 1 Averaging
Diagonal max diff 0.05
Diagonal mean diff 0.02
Off-diagonal max diff 0.06
Off-diagonal mean diff 0.00
20 dB
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0
5
10
15
10 11
8 9
7
5 6
3 4
2
1
14
12 13
0 dB
5
15 16
10
15
9 10
7 8
6
5
3 4
1 2
13
11 12
14 15
Large scale SNR does not change the TPM noticeably (throughput
averaging)
Submission
Slide 17
Joseph Levy, InterDigital Communications Inc..
16
September 2013
doc.: IEEE 11-13/1080r0
Discussion on Multi-Carrier SNR Mapping
SNR averaging
Simple, independent of large scale SNR
SNR averaging occurs in the linear domain
Throughput averaging
Strictly speaking depending on large scale SNR
This dependence is weak though (see comparison on previous page)
and may be removed for simplicity
The mapping may thus be approximated by SNR averaging in the
dB domain
1
N
 log 1    | H |   log 1    | H | 
2
2
2
o
Throughput averaging
i
1
N
Submission
2
i
 log | H |   log | H | 
2
2
i
2
2
o
i
Slide 18
SNR averaging
in the dB domain
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Simulation Summary
Markov modeling of PHY multipath channels in HEW
TPM more or less similar for ITU UMi and WINNER 2 B1 channel
Faster velocity leads to more state changes in TPM
Larger separation leads to more state changes in TPM
Converting multiple subcarrier SNRs into a single value
SNR averaging in the linear domain
Throughput averaging
may be approximated by SNR averaging in the dB domain
More complex averaging may be used
The general method may be applied to any other indoor or outdoor channels
for HEW system level simulations
ITU channels
WINNER 2 channels
and others
Submission
Slide 19
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Potential Items of Agreement
1. Use of Markov modeling of PHY multipath channels
•
Channel model(s) to be used (e.g. ITU UMi and WINNER 2 B1)
•
Velocities to be considered
•
SNR to be considered
•
Number of Symbols to be averaged
2. Method of converting multiple subcarrier SNR
•
•
Throughput averaging
More complex averaging
3. TPM for each agreed configuration
•
•
Generate multiple (e.g. 10^5) pairs, number of symbols separating pairs, N
Number of equi-probable states, P (e.g. 16)
Submission
Slide 20
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
References
[1] IEEE 802.11-13/0722r1, “HEW SG Evaluation Methodology”, Intel.
[2] IEEE 802.11-04/0184r0, “802.11n TGn proposal for PHY abstraction in MAC
simulators,” ST Microelectronics.
[3] R. Yaniv et. Al., “CINR measurements using the EESM method”, IEEE C802.16e05/141r3.
[4] L. Hentilä, P. Kyösti, M. Käske, M. Narandzic , and M. Alatossava. (2007, December.)
MATLAB implementation of the WINNER Phase II Channel Model ver1.1 [Online].
Available: https://www.ist-winner.org/phase_2_model.html
[5] OPNET Technologies Inc., “LTE PHY Multipath Fading Models – Design Document”.
[6] Software implementation of IMT.EVAL channel model, doc num: IST-4-027756
[7] Report ITU-R M.2135-1 (12/2009) Guidelines for evaluation of radio interface
technologies for IMT Advanced
[8] IEEE 802.11-13/0996r1, . “Outdoor Channel Model Candidates for HEW”, K. Josiam, R.
Taori, and F. Tong,
[9] IEEE 802.11-13/0657, “Usage models for IEEE 802.11 High Efficiency WLAN study
group (HEW SG) – Liaison with WFA”, Laurent Cariou
[10] IEEE 802.16m-08/004r5, “IEEE 802 16m Evaluation Methodology Document (EMD)”
Submission
Slide 21
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Additional TPM plots
Submission
Slide 22
Joseph Levy, InterDigital Communications Inc..
September 2013
doc.: IEEE 11-13/1080r0
Velocity Comparison,
30 30km/hr
km/h vs.vs33km/hr,
km/h,100
100symbol,
symbols,
WINNER
2 B1,
B1algo
Throughput
Velocity comparison
20 dB20dB,
snr, WINNER
2 Channel
1
Averaging
Diagonal max diff 0.60
Diagonal mean diff 0.47
Off-diagonal max diff 0.23
Off-diagonal mean diff 0.03
3 km/hr
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0
5
10
15
10 11
8 9
7
6
4 5
3
1 2
14
12 13
30 km/hr
5
15 16
10
15
9 10
7 8
6
5
3 4
1 2
13
11 12
14 15
Faster velocity leads to more state transitions
Submission
Slide 23
Joseph Levy, InterDigital Communications Inc..
16
September 2013
doc.: IEEE 11-13/1080r0
duration comparison
10 vssymbols,
100 symbols,
20 dB30
snr,km/h,
30 km/hr,
WINNER 22Channel
B1, algo 1 Averaging
Packet DurationSymbol
Comparison,
10 vs. 100
20dB,
WINNER
B1 Throughput
Diagonal max diff 0.59
Diagonal mean diff 0.47
Off-diagonal max diff 0.23
Off-diagonal mean diff 0.03
10 symbol
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0
5
10
15
10 11
8 9
7
5 6
3 4
2
1
14
12 13
100 symbol
5
15 16
10
15
9 10
7 8
6
5
3 4
1 2
13
11 12
14 15
Larger separation leads to more state transitions
Submission
Slide 24
Joseph Levy, InterDigital Communications Inc..
16
September 2013
doc.: IEEE 11-13/1080r0
AlgorithmAlgo
Comparison,100
symbols,
20dB,
30 km/h,
WINNER
difference, 100 symbol,
20 dB snr,
30 km/hr,
WINNER
2 Channel 2B1B1
Diagonal max diff 0.25
Diagonal mean diff 0.17
Off-diagonal max diff 0.09
Off-diagonal mean diff 0.01
Algo 1: Throughput Averaging
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0
5
10
15
Submission
11
9 10
8
6 7
4 5
3
1 2
12
15
13 14
Algo 2: SNR Averaging
5
16
10
15
Slide 25
11
9 10
8
6 7
4 5
3
1 2
14
12 13
Joseph Levy, InterDigital Communications Inc..
15 16
September 2013
doc.: IEEE 11-13/1080r0
Algorithm Comparison,100
symbols,
km/h,
Channel
UMi
Algo difference, 100 symbol,
20 dB20dB,
snr, 3030
km/hr,
ITUITU
Channel
UMi
Diagonal max diff 0.18
Diagonal mean diff 0.09
Off-diagonal max diff 0.08
Off-diagonal mean diff 0.01
Algo 1: Throughput Averaging
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
0
5
10
15
Submission
10 11
8 9
7
6
4 5
3
1 2
14
12 13
Algo 2: SNR Averaging
5
15 16
10
15
Slide 26
9 10
7 8
6
5
3 4
1 2
13
11 12
14 15
Joseph Levy, InterDigital Communications Inc..
16