IEEE C802.16m-08/1346
Project
IEEE 802.16 Broadband Wireless Access Working Group
<http://ieee802.org/16>
Title
Codebook comparison for OL and CL MIMO
Date Submitted
2008-10-31
Source(s)
Senjie Zhang
Guangjie Li
Hongming Zheng
Shanshan Zheng
Feng Zhou
Yang-Seok Choi
Minnie Ho
senjie.zhang@intel.com
Intel Corp.
Re:
The TGm Call for Contributions and Comments 802.16m-08/040
Abstract
Precoder comparison for Open-Loop Single-User MIMO
Purpose
Discussion and Approval in 16m
Notice
Release
Patent Policy
This document does not represent the agreed views of the IEEE 802.16 Working
Group or any of its subgroups. It represents only the views of the participants
listed in the “Source(s)” field above. It is offered as a basis for discussion. It is not
binding on the contributor(s), who reserve(s) the right to add, amend or withdraw
material contained herein.
The contributor grants a free, irrevocable license to the IEEE to incorporate
material contained in this contribution, and any modifications thereof, in the
creation of an IEEE Standards publication; to copyright in the IEEE’s name any
IEEE Standards publication even though it may include portions of this
contribution; and at the IEEE’s sole discretion to permit others to reproduce in
whole or in part the resulting IEEE Standards publication. The contributor also
acknowledges and accepts that this contribution may be made public by IEEE
802.16.
The contributor is familiar with the IEEE-SA Patent Policy and Procedures:
<http://standards.ieee.org/guides/bylaws/sect6-7.html#6> and
<http://standards.ieee.org/guides/opman/sect6.html#6.3>.
Further information is located at:
<http://standards.ieee.org/board/pat/pat-material.html> and
<http://standards.ieee.org/board/pat>.
Codebook comparison for OL and CL MIMO
Senjie Zhang, Guangjie Li, Hongming Zheng,
Shanshan Zheng, Feng Zhou, Yang-Seok Choi and Minnie Ho
Intel Corporation
Codebook selection for OL SU MIMO
In section 11.8.2.1.1 of SDD (IEEE 802.16m-08/003r5), the precoding matrix P for Open1
IEEE C802.16m-08/1346
loop SU-MIMO is defined using the following equation:
P(k) = W(k),
where the matrix W(k) is an NT × NS matrix, NT is the number of transmit antennas and
NS is the numbers of streams. The matrix W(k) is selected from a predefined unitary
codebook. The detailed unitary codebook is FFS.
Currently two types of codebook are discussed in the SDD in TGm: Grassmannian
codebook (e.g. 802.16e codebook) and DFT codebook. In closed-loop MIMO, the
Grassmannian codebook benefits the low correlation scenario and the DFT codebook
benefits the high correlation scenario.
However, in high correlation scenario, DFT codebook leads to performance loss in OL
SU MIMO (with distributed resource mapping), because it creates sharp spatial beams for
a few subcarriers and loss of power in other subcarriers.
In the simulation, the basic assumption are
1) Tone based DRU are assumed,
2) AoD -60~60, random select for each snapshot.
3) DFT matrix: 4x2 precoder matrix selected from 1,3 or 2,4 vector pair of pure 4x4 DFT
Matrix.
4) Non-ideal CE.
5) 120km, vehical A channel with high correlation and low correlation.
As Fig-1 shown, DFT codebook loses about 1.5dB in high correlation scenario.
10
-1
PER
10
Rank-1 4x2 OL SU MIMO codebook comparion, Noise-Limited
Low-Corr eITU-VehA, 120km/h, Nonideal CE
0
PER
10
10
-2
-5
10
Ran k - 1 4 x2 OL S U M I MO c o d e bo o k c o mp a r i o n , No i se - Li m i ted
H i gh - C o r r e I TU- V eh A , 1 2 0 km / h, N on i d ea l C E
0
-1
16e-RBF, QPSK 1/2
1 6 e- R B F, Q P SK
16e-RBF, 16QAM 1/2
1 6 e- R B F, 1 6 QAM 1 /2
16e-RBF, 64QAM 1/2
1 6 e- R B F, 6 4 QAM 1 /2
DFT-RBF, QPSK 1/2
10
-2
1 /2
D FT - R B F, Q P SK
1 /2
DFT-RBF, 16QAM 1/2
D FT - R B F, 1 6 QAM 1/ 2
DFT-RBF, 64QAM 1/2
D FT - R B F, 6 4 QAM 1/ 2
0
5
SNR in dB
Fig-1
10
15
-5
0
5
S NR i n d B
10
15
802.16e codebook vs. DFT codebook
PAPR comparison for codebook based precoding in DL MIMO
Precoding/beamforming in DL seems to introduce PAPR problem for DL MIMO.
And different codebook will have different PAPR, but how serious the problem has not
been addressed clearly.
In this contribution, the evaluation of PAPR from different codebook are introduced,
and shows that the PAPR from precoding in DL MIMO can be ignored, and PAPR should
not be treated as one of the criteria for codebook selection.
DFT codebook is constant module, which seems has smaller PAPR than 16e based
2
IEEE C802.16m-08/1346
codebook. However the evalution shows the difference is smaller than 0.1dB, which can
be ignored.
The PAPR definition is
In each Tx antenna, in time domain, the peak power within each OFDM symbol
divided by the average power (across enough long time) is calculated.
The CDF is drawn for all the samples (each Tx antenna and each OFDM symbol).
95% and 99% CDF of the PAPR are used for comparison.
In the evaluation, 10MHz system, 48PRU are used, and 4 continuous PRU use one
precoder, the precoder and MCS are from the SLS scheduler (MIMO schemes is MUZF).
Random data is generated for modulation and precoding, and then IFFT the frequency
domain signal to time domain.
In figure 2, the PAPR distribution is shown, and open loop MIMO has the lowest
PAPR, while 16e codebook based MUZF precoding has highest PAPR, however the gap
is very small, only 0.2 dB, which will not has any impact on the PA design and
performance.
In figure 3, the distribution of the power of time domain samples are shown, where
the average power is 1.
According to the result, even with 8dB back-off, there are only 0.2% samples will be
clipped.
(1) CDF of PAPR
PAPR of Different Codebook Comparison
PAPR of Different Codebook Comparison
1
1
0.9
0.99
0.8
0.7
0.98
CDF
CDF
0.6
0.5
0.3
0.2
0.1
0
0.97
Open-loop
MUZF 0.5Lambda 6bits 16e
MUZF 0.5Lambda 4bits DFT
MUZF 0.5Lambda 6bits trans16e
MUZF Uncorrelated 6bits 16e
MUZF Uncorrelated 4bits DFT
MUZF Uncorrelated 6bits Trans16e
0.4
4
5
6
7
8
9
10
PAPR(dB)
11
12
13
Open-loop
MUZF 0.5Lambda 6bits 16e
0.96
MUZF 0.5Lambda 4bits DFT
MUZF 0.5Lambda 6bits trans16e
0.95
MUZF Uncorrelated 6bits 16e
MUZF Uncorrelated 4bits DFT
MUZF Uncorrelated 6bits Trans16e
0.94
9.8
14
10
10.2
10.4
PAPR(dB)
10.6
10.8
11
Figure 2, CDF of PAPR for different codebook based precoding
PAPR
MUOL
Scenario
95% of CDF
9.97
Uncorrelated
MUZF 16e
(6bits)
MUZF DFT
(4bits)
MUZF
Transformed
16e
(6.23bits)
10.22dB
High
Correlated
10.09 dB
10.17 dB
10.20 dB
99% of CDF
10.6
Uncorrelated
10.93 dB
High
Correlated
10.77 dB
10.03 dB
10.89 dB
10.71 dB
10.02 dB
10.93 dB
10.71 dB
3
IEEE C802.16m-08/1346
(2) The power distribution of time domain samples.
The CDF of every sample power
1.005
1
CDF
0.995
0.99
0.985
MUZF DFT
OL
MUZF 16e
0.98
0.975
6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 10.210.510.811
The power of Sample(dB)
Figure 3 The power distribution of time domain samples
The ratio of clipping
sample
Open-loop
MUZF DFT codebook
MUZF 16e codebook
8dB clipping
10dB clipping
0.0018
0.0023
0.0024
4.6E-5
8.8E-5
9.9E-5
Codebook performance evaluation in SLS
In this section, DFT based codebook [1] and 16e codebook are compared in SLS for
CL SU MIMO and MU ZF MIMO. Transformation method [2] is also evaluated.
Both SU and MU result show 16e codebook outperform DFT codebook in
uncorrelated and low correlated channel, while in highly correlated channel, DFT is
better.
16e codebook with transformation method can show the best performance in any
scenarios.
4x2 antenna configuration is used.
The result is shown in figures and listed in the following tables
(1) Tx 4 SUCL
4
IEEE C802.16m-08/1346
Closed-loop SU Rank 1/2 Adaptation in [4Tx,2Rx]
11
6bits 802.16e
4bits DFT
3.23bits Transformed
6.23bits Transformed
10
9
Gross SE(bps/Hz/cell)
8
7
6
5
4
3
2
1
0
Uncorrelated Channel
Low Correlated Channel High Correlated Channel
Different Scenario
Antenna
SUCL
Codebook Gain (%)
Uncorrelated
Channel
(Zero
correlation)
[4Tx 2Rx]
(bits/sub-band/user/frame)
802.16e (6 bits)
DFT (4 bits)
3bits 16e Transform (3.23* bits)
6bits 16e Transform (6.23* bits)
0
-3.42%
-4.76%
2.38%
Lowly
Correlate
d Channel
(4Lambda
3degree)
0
-1.44%
0.14%
3.02%
Highly
Correlated
Channel
(0.5Lambda
3degree)
0
3.45%
4.14%
4.41%
(2) Tx 4 MUZF
MUZF in [4Tx,2Rx]
12
Gross SE(bps/Hz/cell)
10
6bits 802.16e
4bits DFT
3.23bits Transformed
6.23bits Transformed
8
6
4
2
0
Antenna
Uncorrelated Channel
Low Correlated Channel High Correlated Channel
Different Scenario
MUZF
Codebook Gain (%)
Uncorrelat
ed
Channel
(Zero
5
Lowly
Correlated
Channel
(4Lambda
Highly
Correlated
Channel
(0.5Lambda
IEEE C802.16m-08/1346
[4Tx 2Rx]
(bits/sub-band/user/frame)
802.16e (6 bits)
DFT (4 bits)
3bits 16e Transform (3.23 bits)
6bits 16e Transform (6.23 bits)
(3). Rican Channel SLS of MUZF
Antenna
[4Tx 2Rx]
(bits/sub-band/user/frame)
correlation
)
0
-4.56%
-5.03%
2.67%
3degree)
3degree)
0
8.70%
9.42%
14.78%
0
25.49%
39.53%
48.24%
MUZF
Codebook Gain (%)
Lowly Correlated Channel
(4Lambda 3degree)
802.16e (6 bits)
DFT (4 bits)
6bits 16e Transform
(6.23 bits)
0
4.35%
13.04%
The performance of DFT drops in Rican channel. The gain of DFT codebook compared to
16e codebook is less than Rayleigh channel.
Nested structure
The codebook with nested structure means the higher rank codebook cover the
information of lower rank codebook, and lower rank codebook can be obtained from
higher rank codebook.
The property seems can reduce the complexity for rank search, however the
computation complexity of rank search and codebook search is trivial compared with
MIMO detection block, because
1)
The rank/codebook search work on midamble, which is less frequent than data,
and the equation is much simpler than MIMO detection module.
2)
The rank search for rank adaptation is not so frequent as codebook search, for
example every frame MS will check the rank adaptation, however every
subframe, codebook search is necessary.
Nested structure is not efficient for the codebook design compared with flat structure.
For example, with nested structure, rank 2 PMI index contain the rank 1 PMI
information, while MS most time don’t need the information of both PMIs (only one
CQI is reported ), which introduce the waste of information. Meanwhile, the flat
structure will not have such waste.
Nested structure is not necessary for the consideration of codebook.
Conclusion
The performance and complexity is most important for codebook design and
comparison, and the following issue should be considered
1. Codebook design should consider the impact on OL SU MIMO, and DFT
6
IEEE C802.16m-08/1346
codebook will degrade the performance of OL SU MIMO.
2. Constant module property can be ignored when design codebook.
3. Nested structure is not necessary for codebook design
Performance advantage of 16e and DFT is mixed.
The transformation method can improve the codebook performance.
SLS Simulation parameters
Parameter
Assumption
OFDM parameters
10 MHz (1024 subcarriers)
Number of OFDM symbols per
6
subframe
Permutation
Localized
Number of total RU in one subframe
48
Number of RU
4 which is same as in IEEE 802.16e
for PMI and CQI calculation
CQI, PMI feedback period
Every 1 frame (5ms)
Feedback delay
1 frame (5ms)
QPSK 1/2 with repetition 1/2/4/6, QPSK 3/4,
Link Adaptation
16QAM 1/2, 16QAM 3/4, 64QAM 1/2, 64QAM
2/3, 64QAM 3/4, 64QAM 5/6
MIMO receiver
Linear Minimum Mean Squared Error (LMMSE)
Data Channel Estimation
Perfect data channel estimation.
Feedback Channel Measurement
Perfect feedback channel measurement.
Cellular Layout
Hexagonal grid, 19 cell sites, wrap-around,
3 sectors per site
Distance-dependent path loss
L=130.19 + 37.6log10(.R), R in kilometers
Inter site distance
1.5km
Shadowing standard deviation
8 dB
Antenna pattern (horizontal)
(For 3-sector cell sites with fixed
   2

 , Am 
A    min 12

  3dB 

antenna patterns)
3dB = 70 degrees, Am = 20 dB
Users per sector
10
Scheduling Criterion
Proportional Fair
7
IEEE C802.16m-08/1346
Appendix
[1] C802.16m- 08/1187, Samsung, “Evaluation of CL SU and MU-MIMO codebooks.”
[2] C802.16m-08_1345r1 “Transformation method for codebook based precoding”
Proposed SDD text remedy
-------------------------------------Begin of proposed remedy------------------------------------In page 80 line 5~6, add the following sentence as shown:
[The detailed unitary codebook, and the parameter u and v are FFS. The unitary
codebook could be the same or different from the one used for CL SU MIMO.]
-------------------------------------End of proposed remedy -------------------------------------
8