CHANNEL ESTIMATION AND

1
CHANNEL ESTIMATION AND LOW COMPLEXITY DETECTION
FOR HIGH SPEED WLANs A THESIS Submitted
by MUTHURAJA N In partial fulfillment for the
award of the degree of MASTER OF SCIENCE (BY
RESEARCH)
FACULTY OF INFORMATION AND COMMUNICATION
ENGINEERING ANNA UNIVERSITY CHENNAI 600
025 APRIL 2006
2
Outline

• Introduction
• Channel estimation for MIMO-OFDM systems
• MIMO Detection methods
• Summary

3
IEEE 802.11 evolution
4
Evolution of IEEE 802.11

• The effective application rate offered by the
  existing WLANs was still far lesser than the LAN
  (by 2004)
• WLAN standard has evolved from the basic IEEE
  802.11 (which supports to 1Mbps) to 54 Mbps by
  modifying PHY and MAC layer.
• IEEE 802.11a/g was the standard widely used for
  WLANs (by 2004)
• IEEE 802.11n taskgroup (TGn) was formed with the
  goal of increasing the application throughput to
  atleast 100Mbps by making changes in the PHY and
  MAC layer
• Support to Legacy stations

5
IEEE 802.11n Main Features
802.11n WLAN
MAC
Multiple antennas
Efficient OFDM

• Aggregation
• Block Ack
• Advanced Power save

• MIMO
• 2 Antennas at AP and
• 1 antenna at the user

• Reduced guard interval
• Reduced guard band
• Modulation and Coding

6
IEEE 802.11n standard

• The basic technology for increasing the rate is
  the use of multiple antennas for spatial
  multiplexing (SM). However, transmit diversity
  with space time coding, beamforming, and SVD
  based schemes are also proposed as optional
  features.
• The standard proposes the use of upto 4 antennas
• The number of useful subcarriers is increased to
  52
• There is an optional mode with 40MHz BW, wherever
  the regulatory body allows it
• Shortened GI, code rate upto 7/8, advanced FEC
  coding are other optional features.
• Rates supported vary from 6.5Mbps to 500 Mbps.

7
802.11n PHY layer
Multiple antennas
Efficient OFDM
8
MIMO-OFDM systems

• The orthogonal frequency division multiplexing
  (OFDM) transmission scheme is an efficient
  technique to combat ISI and simplify the
  equalization problem
• The use of multiple antennas at transmitter
  and/or at the receiver helps in many ways such as
  diversity gain, spatial multiplexing and
  beamforming
• The MIMO signaling can easily be overlayed on an
  OFDM based system.
• The MIMO signaling treats each subcarrier in OFDM
  as an independent narrowband frequency flat
  channel. It can be viewed as N parallel MIMO
  systems operating with flat fading channel
  coefficients.
• MIMO-OFDM system offers an increase in rate by
  employing SM at the same time as we combat the
  ISI problem in an elegant way

9
Spatial Multiplexing

• General MIMO example

Uncorrelated channels
Received signal at the antennas is the
combination of the spatially multiplexed data
from the different transmit antennas.
Matrix channel is an important parameter in
analysis and design
10
MIMO-OFDM systems

• Challenges in MIMO-OFDM systems
• Channel estimation (CE)
• Synchronization
• MIMO detection
• Channel estimation to estimate the channel
  coefficients corresponding to all transmit
  receive antenna pair and on all subcarrier
  positions.
• MIMO detection also becomes computationally
  intensive as it has be applied on all the
  subcarriers.

11
Contribution of the thesis

• The main focus is towards analysing the various
  PHY layers proposed for 802.11n
• The thesis covers two portion
• Channel estimation for 802.11n
• MIMO detection schemes
• The performance of several preambles used for
  MIMO channel estimation and different schemes are
  analysed
• Low complexity way of implementing the CE schemes
  are also discussed by exploiting the SFCF.
• System performance of various MIMO detection
  schemes is presented.

12
Section2 Channel estimation for 802.11n systems
13
Outline

• Channel estimation in MIMO-OFDM systems
• Different kind of preambles
• TM method
• SM method and its variation
• TO method
• SO method
• Preambles in IEEE 802.11n
• TGn sync SM (twice)
• WWise SO
• EWC TO

14
Outline

• IEEE 802.11n channel model
• Different MIMO preambles
• Preambles in IEEE 802.11n proposal
• Channel estimation schemes
• LS
• LMMSE
• Interpolation based estimation (LCCE)
• TMMSE
• ML method
• Complexity of CE schemes
• Performance for various CE schemes
• Mean square error
• System performance in terms of BER
• MSE results for TGn channels for TGn sync, WWise
  and EWC proposals

15
TGn channel model

• IEEE 802.11n TGn channel model
• MIMO channel model for indoor and typical office
  environment in LOS and NLOS conditions
• Cluster based channel model

16
Cluster based channel model
Modification of Saleh Valenzula model - By
adding of arrival statistics
The complete impulse response with respect to
both time and angle

• The time of arrival of the ith cluster is Ti
• tj,i is the time of arrival of the ijth path.

The clusters and the rays within the cluster
decay in amplitude and time
The decay rate of cluster and the rays are ?, ?,
17
Cluster based channel model
The angle of arrival statistics
Angular domain Impulse response
Ti uniform 0,2p - mean angle of arrival in
ith cluster. ?ij - correspond to the jth ray
angle in ith cluster, modeled as
Laplacian distributed random variable.
Each cluster has the following angular
statistics Mean angle of arrival (AoA) Mean
angle of departure (AoD) Azhimuth angluar spread
(AS) Elevation angular spread
18
Power Angular spectrum
The angle of arrival statistics within a cluster
- Laplacian distribution

s - Angular spread
19
Cluster based channel model
The complex correlation coefficients PAS, AS,
AoA and Individual tap powers
RXX crosscorrelation function between the
real/imag parts RXYcross correlation between the
real part and imaginary part
To calculate the numerical values of
correlation matrices we use a Matlab program
developed and distributed by L. Schumacher
20
TGn channel model - Channel parameters
For each cluster in a channel model, AStx ASrx AoA
AoD are specified
21
Power delay profile
Cluster 2
Cluster 1
150ns rms delay spread
22
Channel generation steps
23
(No Transcript)
24
Spaced frequency correlation
TGn channel models B toF, NLOS conditions
25
MIMO-OFDM system
Y1(k)
Y2(k)
26
MIMO OFDM system
Received signals at kth subcarrier in a simple
2x2 system
Rx. Ant 1
Rx. Ant 2
In matrix representation
ESTIMATE THE CHANNEL COEFFICIENTS AT ALL
SUBCARRIER POSITIONS
27
Time Multiplexed method
S0
S1
SN-1
SN-2
..
Ant 1
S0
S1
SN-1
SN-2
..
Ant 2
MLTF1
MLTF2

• Time Multiplexed (TM) method
• In each MLTF Transmission from one antenna
• Simple channel estimation LS estimate

28
Time Multiplexed method
Training symbol at k the subcarrier from the
two antennas
The received signal at kth subcarrier
29
Time Multiplexed method
The channel estimates at Kth subcarrier is given
by
Mean square error
Average transmit power
Total Energy required
MSE is inversely proportional to SNR
30
Subcarrier Multiplexed method

• Subcarrier multiplexed method
• Odd subcarriers Transmitted from Antenna 1
• Even subcarriers Transmitted from Antenna 2
• Interpolation needs to be done to estimate
  channel on all subcarriers

31
Subcarrier Multiplexed method
Training symbol at the kth subcarrier from the
two antennas
Ant 1
Ant 1
Ant 2
Ant 2
MLTF1
MLTF1
The received signal at kth subcarrier
32
Subcarrier Multiplexed method
In all odd subcarrier positions
In all even subcarrier positions
Even subcarriers of channel coefficients
corresponding to TX.ant 1 are obtained by
Interpolation.
Odd subcarriers of channel coefficients
corresponding to TX.ant 2 are obtained by
Interpolation.
33
Subcarrier Multiplexed method
Average transmit power
Total Energy required
34
Subcarrier Multiplexed method - twice
Training symbol at k the subcarrier from the
two antennas
Ant 1
Ant 1
Ant 2
Ant 2
MLTF1
MLTF2
MLTF1
MLTF2
The received signal at kth subcarrier
35
Time orthogonal method
S0
S1
.
SN-2
SN-1
S0
S1
.
SN-2
SN-1
Ant 1
S0
S1

SN-2
SN-1
-S0
-S1
..
-SN-2
-SN-1
Ant 2
MLTF1
MLTF2
36
Time Orthogonal method
Training symbol at k the subcarrier from the
two antennas
Ant 1
Ant 2
MLTF2
MLTF1
The received signal at kth subcarrier
37
Time Orthogonal method
The channel estimates at Kth subcarrier is given
by
Mean square error
Average transmit power
Total Energy required
MSE is inversely proportional to SNR
38
Subcarrier orthogonal method
S0
S1
.
SN-2
SN-1
S2
Ant 1
S0
-S1

SN-2
-SN-1
S2
Ant 2
MLTF1
39
Subcarrier Orthogonal method
Training symbol at k the subcarrier from the
two antennas
Ant 1
Ant 1
Ant 2
Ant 2
MLTF1
MLTF1
The received signal at kth subcarrier
40
Subcarrier Orthogonal method
The channel estimates can be obtained by
41
Preambles used in IEEE 802.11n proposals

• Preambles used for channel estimation
• TGn sync SM (twice)
• WWise SO method
• EWC TO method

42
TGn sync proposal
43
Packet structure
Ant 1
LSTF
LLTF
HT-SIG
LSIG
HT STF
HTLTF1
DATA
HTLTF2
8µs
8µs
4µs
8µs
2.4µs
7.2µs
7.2µs
Ant 2
LSTF
LLTF
HT-SIG
LSIG
HT STF
HTLTF1
DATA
HTLTF2
8µs
8µs
4µs
8µs
2.4µs
7.2µs
7.2µs
MIMO Channel estimation Is done during this part
CDD
Simplified PPDU format in 2x2 system-TGn sync
proposal for IEEE 802.11n
44
Long preamble structure in TGn sync
Time domain view
Set 1
Set 1
GI
Set 2
Set 2
GI
Ant 1
Set 1
Set 1
SI
Set 2
Set 2
GI
Ant 2
HTLTF1 (7.2µs)
HTLTF2 (7.2µs)
Subcarrier domain view
K0
Set 1
S-26
0
S-24
0
S-2
0
0
0
S2
0
S24
0
S26
..
..
Set 2
S-25
0
S-23
0
S-1
0
0
0
S1
0
S23
0
S25
..
..
45
Least squares channel estimation
Linear relationship between the channel and the
received signal
Solving the linear equations leads to Least
squares (LS) channel estimates
Mean square error (MSE) is directly proportional
to the noise variance
46
WWise proposal
47
WWise preamble Mixed mode
Mixed mode
SS20
LS20
SIG-MM
LS20
DATA
Ant 1
SIG-N
Ant 2
SS20
LS20
SIG-MM
LS20
DATA
SIG-N
8µs
8µs
4µs
4µs
8µs
Cyclic delay of 400ns
Cyclic delay of 1600ns
Cyclic delay of 3100ns
48
WWise preamble Green field mode
Green field mode
SS20
SIG-N
DATA
Ant 1
LS20
Ant 2
SS20
SIG-N
DATA
LS20
8µs
8µs
4µs
Cyclic delay of 400ns
Cyclic delay of 1600ns
49
WWISE method
Training symbol at k the subcarrier from the
two antennas
Ant 1
Ant 1
Ant 2
Ant 2
MLTF1
MLTF1
The received signal at kth subcarrier during
first repetition
50
CE method for WWISE
The channel estimates are obtained by
51
MSE closed form
The spaced frequency correlation is obtained from
F.T of PDP
52
EWC proposal
53
EWC preamble Mixed mode
Mixed mode
L-STF
L-LTF
LSIG
HTSIG
HT STF
HT LTF1
DATA
Ant 1
HT LTF2
Ant 2
L-STF
L-LTF
LSIG
HTSIG
HT STF
HT LTF1
DATA
HT LTF2
8µs
8µs
4µs
4µs
4µs
4µs
8µs
Cyclic delay of 400ns
Cyclic delay of 200ns
54
EWC preamble Green field mode
Green field mode
L-STF
HTSIG
HT LTF2
DATA
Ant 1
HTLTF 1
Ant 2
L-STF
HTSIG
HT LTF2
DATA
HTLTF 1
8µs
4µs
8µs
8µs
Cyclic delay of 200ns
Cyclic delay of 400ns
55
EWC
Time domain view
Set 1
Set 1
GI
-Set 1
GI
Ant 1
Set 2
GI
Set 2
Set 2
GI
Ant 2
HTLTF1 (7.2µs)
HTLTF2 (4µs)
Set 2 is Cyclic shifted by 400ns of Set 1
Subcarrier domain view
Set 1
-Set 1
S1-26
S1-25
S1-24
...
S124
S125
S126
-S1-26
-S1-25
-S1-24
...
-S124
-S125
-S126
Set 2
Set 2
S2-26
S2-25
S2-24
..
S224
S225
S226
S2-26
S2-25
S2-24
..
S224
S225
S226
HTLTF 2
HTLTF 1
56
Mixed mode Least squares
Received signal at Kth subcarrier is given by
The channel estimates at Kth subcarrier is given
by
57
Green field mode Least squares
Received signal at Kth subcarrier is given by
The channel estimates at Kth subcarrier is given
by
58
Enhanced CE schemes

• LMMSE
• Interpolation based estimation (LCCE)
• TMMSE
• ML method

59
LMMSE channel estimation
The spaced frequency correlation in the channel
is used to get better estimate compared to LS
estimate.
Bx1 vector
Autocorrelation, R matrix captures the
frequency domain correlation in the channel
LMMSE estimate
60
LMMSE channel estimation

• LMMSE filter requires the autocorrelation matrix
  and the noise variance
• Imperfect estimation of R and the noise variance
  leads to the irreducible error floor in the MSE
• Computational complexity of the LMMSE scheme is
  very high as it requires B 2 multiplications and
  a matrix inversion
• A block wise LMMSE - Reduce the complexity at
  the expense of performance degradation

61
LMMSE channel estimation
The spaced frequency correlation in the channel
is used to get better estimate compared to LS
estimate.
Bx1 vector
Autocorrelation, R matrix captures the
frequency domain correlation in the channel
LMMSE estimate
62
Blockwise LMMSE
BL Reduced block length B Original Block
length NB Number of BL blokcs in B
The autocorrelation matrix block of length BL
LMMSE estimate for p th block
63
Interpolation based Low complexity channel
estimation (LCCE)

• Interpolation based channel estimation
• Correlation among the adjacent subcarriers are
  used without the need for the autocorrelation
  matrix, R and huge computations.
• Channel estimates are got by weighted average of
  LS estimates and the interpolation estimates

64
Block diagram
HTLTF 1
HTLTF 2
RX. 1 LS est
RX. 2 LS est
RX. 1 LS est
RX. 2 LS est
Int
Int
1-W
W
1-W
W
65
Low complexity channel estimation
The final channel estimates is the weighted
average of direct LS estimate and the
interpolated estimate
Simple linear interpolation filter Low
computational overhead
Linear interpolation
Weights of the linear interpolation are chosen to
be powers of 2 to use shifting instead of
multiplication
Other Interpolations like cubic, spline can be
done Complexity increases
66
Low complexity channel estimation
Error floor is directly proportional to the RMS
delay spread
67
MSE closed form for linear interpolation method
The spaced frequency correlation is obtained from
F.T of PDP
68
Truncated MMSE (TMMSE) - CE
Smoothing LS estimates by weight values
obtained from MMSE solution
The MMSE solution matrix Vp of a truncated R
matrix is obtained as follows
Rp is the correlation matrix of dimension PxP
The middle row of Vp matrix is used as weight
vector
69
TMMSE - CE
Filter the LS CEs using these weight values as
filter coefficients
There is a loss in performance compared to
LMMSE, due to truncation and smoothing with less
number of weights
To reduce complexity the modulus of the complex
weights is considered and quantized to the
nearest power of 2
LCCE method is a special case of TMMSE method
when the weights are real
70
ML method
ML channel estimation with assumption that the
maximum length of the channel impulse response
is not greater than the guard time.
Step 1
Step 2
Step 3
Step 4
Where, F is the Fourier matrix and Fred is the
reduced Fourier matrix whose dimension is L x L
Suitable only for symbol spaced channel
71
Computational complexity
B is the number of subcarrier For TGnSycn B52,
WWISE EWC B 56
72
Results for all the methods

• Performance of various preambles
• Performance - TGn sync preamble
• Performance - EWC preamble
• Performance - WWISE preamble
• Effect of various channel estimation schemes on
  system performance interms of BER PER Section3

73
Simulation model
,
The SNR used here refers to signal to noise
ratio per receive antenna per subcarrier
The performance measure is the MSE of the channel
estimate
are the ideal and estimated CEs on the kth
subcarrier
74
Performance of various preambles
75
TGn sync
76
Performance of LCCE method
77
LMMSE method
78
LMMSE method Mismatch Correlation matrix
79
(No Transcript)
80
TMMSE method, P3
81
TMMSE method, P5
Figure 2.25 MSE performance of TMMSE scheme with
P5 for TGn sync preamble, Channel D NLOS
82
Performance of various CE methods
83
Gain at 0dB
84
Cutoff point of all schemes
85
WWISE
86
Performance of smoothing windowmethod for all
channel models
87
WWISE MSE Performance for different schemes
88
Performance of ML based method
89
Enhanced Wireless Consortium
90
LCCE scheme
91
EWC - Performance for various CE schemes
92
(No Transcript)
93
Effect of CE errors on BER and PER
94
BER performance Uncoded system
95
BER performance TGn sync systemQPSK ½ rate
96
PER performance TGn sync systemQPSK ½ rate
97
Complexity of CE schemes
98
Section2 MIMO Detection methods for 802.11n

• gtgt A simple uncoded system
• gtgt TGn Sync system

99
MIMO Detection schemes

• MIMO detection schemes
• Decorrelator / ZF
• MMSE
• Successive Interference Cancellation (SIC)
• ZF/MMSE VBLAST (Ordered SIC)
• Maximum Likelihood (ML)
• Explain in detail about each of these schemes.

100
MIMO Detection schemes
Maximum Likelihood (ML) Optimum and most complex
detection method
Zero-Forcing (ZF) Pseudo inverse of the channel,
simplest detection method
Minimum mean-squared error (MMSE) Intermediate
complexity and performance
101
MIMO Detection schemes
V-BLAST Ordered successive interference
cancellation (SIC) detector
102
Non-feedback MIMO Receivers (contd..)
103
V-BLAST 2 x 2 Example
104
V-BLAST 2 x 2 Example
105
Mean square error in detection

• The mean square error (MSE) between the
  transmitted data symbols and the output of the
  detection algorithm is a good measure for the
  performance of MIMO detection algorithms
• MSE easy to derive for MIMO detection.
• From simulation, the reduction in MSE leads to
  BER reduction.

106
Low complexity MIMO detection

• We need to employ N independent MIMO detectors in
  a MIMO system with N subcarrier.
• The frequency correlation among the subcarriers
  can be used to reduce the complexity of the
  MIMO-OFDM system
• Instead of independently employing MIMO detector
  in all subcarriers, only the solution for the
  MIMO detector on alternate subcarrier positions
  are found
• The solution for the other subcarriers is found
  by interpolating the solutions obtained for the
  neighboring subcarriers.

107
Low complexity MIMO detection

• Linear interpolation using weights which are
  simple to implement can be used.
• Let k-1 and k1 be the subcarrier positions where
  the direct solution
• let k be the subcarrier position in which the
  solution is obtained by linear interpolation
• Where Vk is the matrix solution for MIMO
  detection
• 50 reduction in the complexity when compared to
  the normal MIMO-OFDM detection methods
• This idea can be used for ZF, MMSE, MMSE-SIC,
  ZF-SIC detection method.
• It cannot be directly applied to the VBLAST based
  detection schemes, since the order in which the
  detection is performed varies for each subcarrier.

108
Complexity comparison

• Number of complex multiplication is considered

109
Complexity comparison
110
A simple MIMO-OFDM system
111
Simulation results and discussion

• Uncoded system
• Simulation results for MIMO detection algorithms
• Effect of CE on the system performance
• Simulation parameters
• Number of Subcarriers, N 64
• Cyclic prefix 16 samples
• BW 20MHz
• QPSK modulation
• TGn channel D NLOS
• Results presented in terms of MSE performance,
  BER and PER.

112
MIMO detection schemes for 2x2 and 4x4
2x2
4x4
113
Low complexity MIMO detection scheme
MSE
BER
114
BER performance with different CE
115
TGn Sync system System model
116
TGn Sync - Simulation results
Info bits
TGn sync Tx
TGn sync Rx.
Channel
Channel Estimation using preambles
Figure Simulation model
Results presented in terms of BER and PER
117
MIMO detection schemes BER PER
118
LC-MIMO detection schemes BER PER
119
Effect of CE schemes on system performance
120
Effect of CE schemes on system performance for
all channel models
Gp is defined as the Loss in performance in terms
of SNR at a cut off point of 10-5 BER
121
Thank You
122
Summary

• Various MIMO detectors are discussed and their
  performance in uncoded system in terms of MSE and
  BER is presented based on simulation.
• The effect of CE on the performance of uncoded
  MIMO system is also presented.
• A low complexity solution for MIMO-OFDM detection
  is proposed and it reduces the computational
  complexity by 50.
• The performance of the TGn sync system is
  presented for various MIMO detection methods in
  terms of BER and PER.
• It is shown by simulations that the LC MIMO
  detectors result in very less performance
  degradation for practical channel conditions.
  Thus, the LC MIMO detectors can be used for IEEE
  802.11n proposals as they reduce the
  computational complexity load at the receiver.
• The effect of various CE schemes on the
  performance of IEEE 802.11n TGn sync proposal is
  presented in terms of BER and PER.
• The results indicate that the LS scheme results
  in about 3 dB loss in performance at 10-5 BER
  point, while the low complexity CE schemes such
  as LCCE, TMMSE have less performance degradation.
  Thus, the TMMSE and LCCE CE schemes can be used
  for IEEE 802.11n proposals leading to fewer
  computations and less performance degradation.