Wireless communications lab, AU-KBC Research Centre TIME SYNCHRONIZATION AND LOW COMPLEXITY DETECTION FOR HIGH SPEED WIRELESS LOCAL AREA NETWORK V. Sathish,

Wireless communications lab, AU-KBC Research Centre

TIME SYNCHRONIZATION AND LOW COMPLEXITY DETECTION FOR HIGH SPEED

WIRELESS LOCAL AREA NETWORK

V. Sathish, 2004438105Supervisor: Dr.S.Srikanth AU-KBC Research centre,

MIT Campus,Chennai, India


Timing synchronization in IEEE 802.11n systems


Presentation Outline

• Abstract

• IEEE 802.11n standard, goals and its challenges

• Review of IEEE 802.11a preamble and its usage

• 802.11n operating modes and frame formats

• Timing synchronization

– Literature survey

– Proposed coarse timing estimation

– Proposed fine timing estimation

• Simulation setup and results discussion

• Conclusion


Abstract• A low complexity timing synchronization method for the systems

leased on the MIMO-OFDM1 based 802.11n standard is proposed

• Two high throughput operating modes in IEEE 802.11n: – Mixed mode where 802.11a/g legacy systems and 802.11n based MIMO-

OFDM systems shall co-exist– Greenfield mode where only 802.11n enabled MIMO-OFDM systems exists

• For timing synchronization purposes,– Mixed mode : short training field (STF) and long training field (LTF) in

preamble– Greenfield mode : Only short training field in preamble

• Essentially, two time sync algorithms are needed for MIMO modes

• Proposed algorithm uses only STF for timing synchronization and achieves same performance as LTF based algorithm

• The STF structure is same on both the modes, so a single time sync algorithm can be implemented for all the high throughput modes.

1MIMO-OFDM Multiple input multiple output – Orthogonal frequency division multiplexing


WLAN standards

• Wi-Fi standards- IEEE 802.11 standard, 1997; 2 Mbps, 2.4GHz,

CSMA/CA

- IEEE 802.11b std, 1999; 11 Mbps, 2.4GHz, CSMA/CA

- IEEE 802.11a std, 1999; 54 Mbps, 5GHz, CSMA/CA

- IEEE 802.11g std, 2003; 11 Mbps & 54 Mbps, 2.4 GHz, CSMA/CA

- IEEE 802.11n draft, 2006; 500 Mbps, 2.4 GHz, CSMA/CA


802.11n standard Goals and its challenges

• Achieve higher data rates (around 500 Mbps)– Use of MIMO-OFDM technology– Supports 20MHz and 40 MHz bandwidth operation

• Interoperable with 802.11a/g legacy systems

• Increased complexity– Multiple radio frequency (RF) and baseband (BB) chains

required– Spatial detection techniques

• Backward compatibility– MIMO-OFDM system should be able to decode the legacy

packets– Legacy system should atleast know about the MIMO-OFDM

transmission to avoid collision– Design of preamble impacts on initial receiver tasks


Review of IEEE802.11a frame

Short training field Long training field

SS SS SS GI LS1 LS2 SIG

Signal Field

0.8 s 1.6 s 3.2 s

Data

Short symbols

1. Start of packet (SOP)

detection

2. Automatic gain control

(AGC)

3. Coarse timing estimation

4. Coarse frequency offset

estimation

Long symbols

5. Fine timing estimation

6. Fine frequency offset

estimation

7. Channel estimation8. Data detection

Receiver tasks


Initial receiver tasks

AGC &Synchro.

Mode

Ch. Estimati

on Mode

Correction &

Tracking mode

Startof

packet

Acquisition modePacket detected

Time & frequencyAcquired

Channel estimatedOffsetupdate

Data detection

End of packet


802.11n frame formats

Non-High Throughput frame format

Short training field Long training field

SIG DATASS SS LSLSCPSS

• Used in the legacy network where only the 802.11a/g enabled devices are present

• Content is identical to the frame defined in the IEEE 802.11a standard

• STF – Short training field

• LTF – Long training field


802.11n frame formats – contd.

DATA

Legacy format Preamble High throughput Preamble

L-STF L-LTF L-SIG HT-SIG HT-STF HT-LTF HT-LTFn

High throughput mixed frame format

• High throughput stations and legacy stations shall co-exists

• MIMO stations should transmit and receive the legacy frames and HT frames

• For compatibility reasons, Initial preamble part is provided with the first three fields of non-HT preamble

• HT-SIG, HT-STF and HT-LTFs are used decoding the MIMO packets

• If the tranmission is intended for MIMO_OFDM system, then based on the number of TX antennas cyclic shift is applied as shown in table1


802.11n frame formats – contd.

H-STF

High throughput Preamble

H-LTF H-SIG HT-LTF HT-LTFn DATA

High throughput Greenfield frame format

• Only HT MIMO-OFDM stations can exist • All the training fields specific to MIMO-OFDM systems • HT-STF is identical to the L-STF field of mixed mode and is used for

timing acquisition, AGC and frequency acquisition • For TH-SIG demodulation, channel estimates are obtained from

first HT_LTF fields• Remaining HT-LTFs are used for estimating the channels across

multiple transmit and receive antennas• Frames in different TX antennas are cyclically shifted based on

table2 before transmission


Cyclic shift for HT frame transmission

ns

Number of

Transmit chain

Cyclic shift for

Tx chain1

( )

Cyclic shift for

Tx chain2

( )

Cyclic shift for

Tx chain3

( )

Cyclic shift for

Tx chain4

( )

1 0

2 0 -200

3 0 -100 -200

4 0 -50 -100 -150

Table1. Cyclic shift for the non-HT portion of the packet

Table2. Cyclic shift for the HT portion of the packet

Number of

Transmit chain

Cyclic shift for

Tx chain1

( )

Cyclic shift for

Tx chain2

( )

Cyclic shift for

Tx chain3

( )

Cyclic shift for

Tx chain4

( )

1 0

2 0 -400

3 0 -400 -200

4 0 -400 -200 -600

ns ns ns

ns ns ns ns


For Backward compatibility

802.11nAccess point

Legacy mode

Green field mode

Mixed mode

Only frames in legacy format

Preambles that are specificto MIMO-OFDM systems

• Preamble should be compatible to legacy stations

• Should work better for MIMO-OFDM systems


Typical 802.11n network

802.11nAccess point

802.11g

802.11g

802.11g

802.11n

802.11n

802.11g

802.11g

802.11n

Active nodeInactive node

Legacy mode

Green field mode

Mixed mode

802.11n


Typical MIMO-OFDM system model

X

( )tN

X k

Spatial Demux

OFDM TX

Spatial Detection

OFDM TX

OFDM RX

OFDM RX

Spatial Mux

1( )X k1( )x n

( )tN

x n

1( )v n

( )rN

v n

X̂

1 ( )tN

h n

1 ( )rN

h n

11( )h n

( )t tN Nh n

( )rN

y n

1( )y n

tN rN

1 1

channelTransmitter Receiver

1(1)Y

1( )Y N

(1)rN

Y

( )rNY N

NtxNr MIMO-OFDM system


Received signal model

1 12

0 0

( ) ( ) ( ) ( )tN P

j nr t rt r

t p

y n x n h n p e v n

( )tx n thtis the transmitted signal from the TX antenna where

( )rth n thtis the impulse response of the channel between the

transmit and receive antennathr

Received signal at the receive antenna thr

( )rv n is the AWGN at the 2v RX antenna with zero mean and variance

thr

is the normalized frequency offset

Pis the channel length and remains static across n

tNThe total power transmitted is normalized across the transmit antennas and is given as

2

1

( ) 1tN

tt

E x n


Timing synchronization• Timing synchronization

– To estimate the sampling time of the OFDM symbol

– The start of OFDM symbol varies based on the strongest path of the fading channel

– Non-optimal sampling causes ISI and ICI

– Done in two steps• Coarse timing offset (CTO) estimation• Fine timing offset (FTO) estimation

• Coarse timing offset estimation– Rough estimate is obtained– After start of packet detection and AGC, timing estimator is

triggered

• Fine timing offset estimation– Optimal starting of OFDM symbol is obtained


Literature survey

• In [4], T. M. Schmidl and D.C. Cox had proposed a maximum likelihood (Ml) synchronization timing estimation method for a SISO-OFDM system.

• An extension of this method for MIMO-OFDM system was proposed in [5] by A. N. Mody and G.L. Stuber, and in [6] by A. Van Zelst and Tim C. W. Schenk.

• The drawback of these methods is that the preambles assumed in the papers are not the same as in the 802.11n standards.

• In [7], Jianhua Liu and Jian Li presented a timing synchronization technique for a preamble that is similar to the one in the 802.11n standard.

• However, the computational complexity of this method is high due to the cross correlation performed on the LTF for fine timing estimation.


Coarse timing offset estimation

• The objective of the CTO estimator is to find the rough starting position of any of the short symbol

• Typically 5-6 blocks of SS is taken for AGC operation

• Coarse timing estimation can be performed only after AGC convergence.

• An easy way is to find the end of the STF by using the autocorrelation property of the received signal.


Proposed Coarse timing offset estimation


Proposed coarse timing estimation technique

• A metric is calculated from the instant k at which the AGC is converged

• This metric is similar to the one in [7] and is given as

( )S n

where

1

0

( )1( )

( )

rNr

r rr

P nS n

N R n

1

*

01 1 1

2 2

0 0 0

( ) ( ) ( )

( ) ( ) ( )

s

s t

N

r r r s

dN N P

t rt r

d t p

P n y n d y n N d

x n d p h p n

1

*

01 1 1

2 2

0 0 0

1( ) ( ) ( )

1( ) ( ) ( )

s

s t

N

r r s r ss dN N P

t rt rs d t p

R n y n N d y n N dN

x n d p h p nN

and

( )r n is the value of the cross correlation between the signal and noise terms

( )r nis the sum of noise energy and value of cross correlation between the signal and noise terms

Step1:


Proposed CTO estimator – contd..

( ) 1 8 sS n n N

( ) 1 8 sS n n N K 0 sK N with

The metric will form the end of the plateau and could be noisy due to AWGN and multipath fading conditions

( )S n

To have a smooth plateau, the current metric is filtered through a weight filter and is given as

' '( ) ( 1) (1 ) ( )S n S n S n

(1 )Where is the weight factor given to previous value and is the weight applied to the current metric

( )S nThe value of metric can take different values based on the index.

( )r n is the sum of the cross correlation of the signal and noise terms, and cross correlation between samples from STF and LTF.

8 sn N( )rP n

( )r nSince the fields STF and LTF are highly uncorrelated, the parameter decreases withthereby reduction in


Plot of metric1

Reference for metric

Metric

The falling end of plateau is noisy and getting a coarse timingestimates will be erroneous

Threshold based detection

Metric forms a Plateau - 2x2 system under the channel D with SNR=10dB


Proposed CTO estimator – contd.

112

0 0

1( ) ( ) ( )

sr NN

r r sr s r d

D n y n d y n N dN N

22 v

( )D n nThe value of metric depends on the instant

8 sn N

8 sn N K For with the metric will be represented as 0K

2( ) 2 ( ) ( )vD n n n

and represents the averaged power of the STF and LTF respectively ( )n ( )n

The total averaged power of the difference signal will increase as n increases. This is because of the contributions from LTF

A smoothing operation is done on the metric by weighted averaging and is given as ' '( ) ( 1) (1 ) ( )D n D n D n

( )D nA new metric which is the average power of a difference signal over a window of samples is defined from the instant

sN

The metric is given as

Step2:


Plot of CTO metrics

Intersection point

M2

Metric plotted for a 2x2 system under the channel D without noise


Proposed CTO estimator – contd.

The metrics and can be used to get a reliable estimate of the CTO ' ( )S n ' ( )D n

Steady increase in metric2 from and steady decrease in the value of metric1 from

8 1sN 8 1sN

The intersection point between these two metrics is estimated as the coarse time

The instant should lie within the range [ , ] 8 1sN 9 sN

At low SNR, both the metrics will be noisy and fluctuating and this would result in wrong estimate

There might more than one intersecting point due to fluctuations

To avoid this a simple condition is proposed

Let be the intersecting point then this instant will be chosen as the CTO estimate when the conditions given below are satisfied

2M

' '2( ) ( )D M D n 2 2 2{ ,..., 1, }n M Q M M ,

' '2( ) ( )D M D n 2 2 2{ ,..., 1, }n M M Q M Q ,

Where is the number of samples used to make sure that the estimate is not a false alarm due to noise

Q


Plot of metrics

Metric 1

Reference for metric 1

Reference for metric 2 Metric 2

2x2 MIMO-OFDM system;Channel model D; SNR=10dB


Proposed Fine timing offset estimation


Proposed fine timing estimator

The objective of the fine timing offset estimator is to find the exact start of the OFDM symbol

In multipath channel conditions this might not be possible because the strongest path could occur at non-zero delays

In the proposed FTO estimator, we find an index in the starting of the 9th SS where the sum of channel impulse response energy is maximum between the receive antenna and transmit antenna

This is achieved by using the correlation property of the STF and the advantages of the cyclic shift

Achieved in two steps


Proposed FTO estimator – contd.

Step1:

A simple cross correlation is performed between the received signal and the transmit signal

The fine timing offset estimation algorithm is triggered from the index 2 8M

The received signal at each receive antenna is correlated with all the transmit signals tN

1*

0

1ˆ( ) ( ) ( )

sN

rt r ts d

g n y n d x dN

Then, the cross correlation output between RX antenna and TX antenna is given asthtthr

Let be the received signal at the RX antenna after coarse frequency offset correction,thrˆ ( )ry n

0 sn N

Since the received signal at each receive antenna contains multiple versions of the transmit signal in cyclically shifted manner, the cross correlation between the received signal and the transmit signal will result in multiple peaks


Each peak corresponds to the total channel energy between transmit and receive antennas

The position corresponding to the first peak of the first receive antenna output sequence is the fine timing estimate

For example

Let us assume the coarse timing estimate and all the channel impulse responses have the strongest path at zero delay

2 8 8sM N

For the 4x4 mixed mode system

The cross correlation output between the first transmit antenna signal and the first receive antenna signal will have 4 peaks placed consecutively from 8 1,..,8 4s sN N

00 ( )g n

Detecting the first peak is quite tricky due to multiple peaks that corresponds to different channel power between transmit and receive antennas

To choose the first peak, we propose a simple technique



Cross correlated output - Example

For a 4x4 system

0 1 2… 13, 14, 15

0 1 2… 13, 14, 15 0 1 2… 13, 14, 15

0 1 2… 13, 14, 15

( )rtg n ( )rtg n

( )rtg n ( )rtg n

Antenna1

Antenna2 Antenna4

Antenna3



1 4

22 0 1

0 0

( ) (( )) ((12 ))s sr N r N

r m

G q g q m g q m

2 4

33 0 1

0 02

( ) (( )) ((14 ))

((12 ))

s s

s

r N r N

r mr N

G q g q m g q m

g q m

3 3

44 0 1

0 02 3

( ) (( )) ((15 ))

((14 )) ((13 ))

s s

s s

r N r N

r mr N r N

G q g q m g q m

g q m g q m

0,..,15q

With reference to the table1a for mixed mode, we propose the metrics , and for different antenna configurations as shown below

22 ( )G q33 ( )G q 44 ( )G q

The cyclic shift 50us, 100us, 150us and 200us applied at the transmit antenna corresponds to numerical shift 15, 14, 13 and 12 that is applied at the correlated output obtained from different transmit signals.

The index corresponding to the maximum of absolute of the metric is determined as the fine timing offset.

Step2:


Complexity analysisIn case of the conventional LTF based FTO estimator, the complex cross correlation should be performed between 64 samples length long symbol and the received signal.

64 64t rN N

In the proposed FTO estimator, the cross correlation is performed between 16 samples length short symbol and the received signal

16 16t rN N


Simulation and performance analysis


Performance of coarse timing estimator

• Probability distribution of CTO estimate is plotted

• Compared to the performance of threshold based technique

• System model– 2x2, 3x3 and 4x4 antenna configuration

– MIMO Channel model• TGn channel models

– SNR = 8dB


Parameters of coarse timing estimator

• For threshold based technique as in [7]– Mixed mode and green field mode

• Threshold c2=0.6 and Q2=15 samples

• For proposed technique– Mixed mode and green field mode

• Threshold =0.45 and Q=8 samples• Smoothing filter weight = 0.5 for both the metrics


Probability of coarse timing offset estimate of conventional and the proposed technique.

Estimation accuracy of the CTO estimator is [0, ]

1sN

Probability of getting zero CTO is high for the algorithm proposed in threshold based technique

Significant probability of the CTO obtained using this algorithm is going beyond the defined estimation accuracy

In the proposed algorithm, estimates are more stable and lie within the estimation range


Comparison of probability of CTO estimates for different antenna configurations

Probability of CTO estimates within the estimation accuracy

Proposed algorithm performs better at the lower SNR values as compared to the CTO estimation algorithm in [7]

As the number of antenna increases, the spatial diversity is leveraged resulting in a better performance for higher antenna configuration


Impact of channel models

Probability of CTO estimates within the estimation accuracy for proposed algorithm in different channel models

The maximum probability is achieved at 10dB SNR for a 2x2 system

Motivation to use only the STF for the fine timing offset estimation


Performance of fine timing estimator

• Probability distribution of fine timing estimate is plotted

• Compared to the performance of simple cross correlation based technique using LTF

• System model– 2x2, 3x3 and 4x4 antenna configurations– MIMO Channel model

• TGn channel models


Comparison of probability of FTO estimates with LTF based FTO estimator

The estimation accuracy is defined with the range [0, 3].

Computationally complex LTF based FTO LTF will have slightly better performance as compared to proposed technique

The probabilities of the FTO estimates within the estimation accuracy is plotted for the 3x3 and 4x4 systems of mixed mode.

Due to better noise averaging


Conclusion

• A low complexity time synchronization algorithm is proposed

• The proposed techniques performs better even at lower SNRs.

• Using only STF, a single coarse and fine timing estimation technique will be used for both the high throughput modes

• Same performance is achieved as LTF based timing synchronization

• Thereby reducing total complexity of the system


References[1]. IEEE P802.11n™/D2.00, “Draft standard for Information Technology-

Telecommunications and information exchange between systems-Local and metropolitan area networks-Specific requirements-“, Feb 2007

[2]. IEEE 802.11a standard, ISO/IEC 8802-11:1999/Amd 1:2000(E), http://standards.ieee.org/getieee802/download/802.11a-1999.pdf

[3]. IEEE 802.11g standard, Further Higher-Speed Physical Layer Extension inthe2.4GHzBand, http://standards.ieee.org/getieee802/download /802. 11g-2003.pdf

[4] T. M. Schmidl and D.C. Cox, “Robust Frequency and Timing Synchronization for OFDM”, IEEE Trans. on Communications, vol. 45, no. 12, pp. 1613-1621, Dec. 1997.

[5]. A. N. Mody and G.L. Stuber, “Synchronization for MIMO-OFDM systems,” in Proc. IEEE Global Commun. Conf., vol. 1, pp.509-513, Nov.2001

[6] A. Van Zelst and Tim C. W. Schenk, “Implementation of MIMO-OFMD based Wireless LAN systems”, IEEE Trans. On Signal Proc. Vol. 52, No.2, pp. 483-494, Feb 2004

[7] Jianhua Liu and Jian Li, “A MIMO system with backward compatibility for OFDM based WLANs”, EURASIP journal on Applied signal processing. Pp. 696-706, May 2004

[8] IEEE P802.11 TGn channel models, May 10 2004,http://www.ece. ariz ona.edu/~yanli/files/11-03-0940-04-000n-tgn-channel-models.doc


Low Complexity MIMO-OFDM System for High Speed WLANs


Presentation Outline

• Introduction

• System model and channel model

• MIMO-OFDM1 detection techniques

• Proposed Group ordered MMSE V-BLAST2 detection

• Simulation results

• Conclusion

1MIMO-OFDM Multiple input multiple output – Orthogonal frequency division multiplexing2MMSE V-BLAST Minimum mean square error – Vertical bell labs layered space time system


Introduction

• MIMO-OFDM is a promising technique to increase data transmission rate in wireless frequency selective fading channels[1,2]

• The key technique behind the MIMO-OFDM system is the spatial detection at the receiver

X

( )tN

X k

Spatial Demux

OFDM TX

Spatial Detection

OFDM TX

OFDM RX

OFDM RX

Spatial Mux

1( )X k1( )x n

( )tN

x n

1( )w n

( )rN

w n

X̂

1 ( )tN

h n

1 ( )rN

h n

11( )h n

( )t tN Nh n

( )rN

y n

1( )y n

tN rN

1 1

channelTransmitter Receiver

1(1)Y

1( )Y N

(1)rN

Y

( )rNY N


802.11n MIMO OFDM baseband transmitter

1

Spa

tial

m

appi

ngStream Parser

FE

C

Enc

oder

Enc

oder

P

ars

er

Scrambler

1

FE

C

Enc

oder

Interleaver

QAM Mapper

Interleaver

QAM Mapper

IFFT&CP

IFFT&CP

ESN

1

ssN

1

tN

802.11n MIMO-OFDM baseband transmitter


802.11n MIMO-OFDM baseband receiver

11

802.11n MIMO OFDM baseband receiver

CP&

FFT

Spa

tial

D

etec

tor

and

dem

appi

ng

(Zer

o fo

rcin

g,

MM

SE

, S

IC, e

tc)

CP&

FFT

rN

QAM De-Mapper

Deinterleav

er

QAM De-Mapper

Deinterleav

er

De

scr

am ble rM U X

D EC O D ER

D EC O D ER

Stream

De-parser

ssN

ESN

1

Decoded bits

RX antenna

s


Signal model and MIMO channel

1

( ) ( ) ( ) ( )tN

m l lm ml

y n x n h n w n

( ) ( ) ( ) ( )k k k k Y H X W

( ) [ ( ) ( ). ( )]r

T1 2 Nk Y k Y k Y kY . . .

( ) [ ( ) ( ). ( )]t

T1 2 Nk X k X k X kX . . .

( ) [ ( ) ( ). ( )]r

T1 2 Nk W k W k W kW . . .

After removing cyclic prefix and FFT operations, the received signal vector corresponding to subcarrier (bar over a variable represents vector)

Received signal:

where

Transmit signal vector

Additive white Gaussian Noise

thk

(1)


11 12 1

21

1

( ) ( ) . ( )

( ) . . .( )

. . . .

( ) . . ( )

r

t t r

N

N N N

H k H k H k

H kk

H k H k

H

Channel matrix at the subcarrier

Signal model and MIMO channelthk

• MIMO detection is done in all the subcarriers in a similar fashion.

• For simplicity, we drop the index ‘k’ and the received signal is given as

• The elements in are independent and identically distributed (iid) zero mean and circularly symmetric complex Gaussian random variables with variance2

v

Y HX W

W


MIMO Detection Techniques

MIMO Detection techniques

Non-linear(ML)

Low BERHigh complexity

Embedded(V-BLAST)Low BER

Moderate complexity

Linear(MMSE, ZF)

High BERLow complexity

Modified

Group ordered MMSE V-BLAST

Low BERLow complexity

Proposed system

MMSE

V-BLASTProposed

BER

SNRPerformance

Complexity

V-BLAST

Proposed

MMSE

Complex computations

SNR



• Zero Forcing • MMSE

2argmin -

est

X AX = Y HX Ais the constellation set

Complexity , M is the order of the constellation )O(M tN

• ML Detection

estX = GYwhere where

1)H H= ( G H H H 2 1)H Ht v= ( N G H H + I H

Complexity 3 )tO( N Complexity 3 )tO( N

Noise enhancement Noise variance computation is an overhead

estX = GY



Successive Interference cancellation (SIC):

With ordering : Order of detection based on SINR, stream with largestSINR is selected in each iteration [4] (V-BLAST with MMSE/ZF solution)

Without ordering : Order of detection is selected randomly

MMSE V-BLAST:

Combined MMSE and iterative SIC

Transmit signal from each antenna is detected at each iteration

Interference due to the detected signal is cancelled form

the received signal

Repeat the iteration until all the signals are transmitted


MMSE V-BLAST

MMSE solutionand ordering

criterion

Sort in ascending and store the index

in Interference cancellation

H 2 -1

vH H + σM I)= (

HG= MHp Re{diag(M)}

H2v

p

q

Y

estX (q)

Values in P represents the SINR for each

stream

First value in q corresponds to the stream with largest

SINR

Detect the stream corresponding to

first index in q

Detect the stream

Corresponding to q(1)

Obtain , ,

2H2G

2p2M

H2v

Repeat the steps until all the streams are

detected

Y


MMSE V-BLAST Algorithm

( ) ( )Hest qX q Quant g Y

2t v( + N σ )H -1 HG = H H I H

( )estX q qY Y h

Complexity 3 2

3 2

1

6 22

tNt t

i

N Ni i i

thqH H

j

q= arg min diag( )M where 2 1( )Ht vNM H H + I

1. Obtain MMSE solution

2. Find the detection order using the criterion below [5]

3. Initial Nulling and detection

4. Interference cancellation

5. RecursionObtain new by replacing the column of with

zeros. Repeat from step 1 until all the streams are detected.


Group Ordered MMSE V-BLASTConcept of proposed detector

1. Group the streams that face similar channel conditions

2. Use same MMSE solution to detect all the streams in that group

3. SIC is applied inside and across the groups

4. Since the MMSE solution is calculated for each group,

there is a reduction in the complexity of detection.

GO MMSE V-BLAST can be implemented in 2 ways

1. Fixed method

2. Adaptive method


Group ordered MMSE- V-BLAST (fixed)

MMSE solutionand ordering

criterion

Sort and store the index

in

SIC

Interference cancellation

Obtain , ,

Group 1

Group 2

q(0)

tNq -1

2

tNq

2

1)tq(N

SIC

2YH 2 -1

vH H + σM I)= (

HG= MHp Re{diag(M)}

H2v

p

q

Y

2H2G

2p2M

H2v

est (q)X

est (q)X



• Find MMSE solution2 1( )H H

t vN G H H I H

Re{diag( )}p M• Calculate , sort it and store the index in p q

where

• Grouping: Group1 – Streams corresponding to Group2 – Streams corresponding to

( ) 0,1,... / 2 1tq v v N ( ) / 2,... 1t tq v v N N

• Apply ordered SIC to detect streams in Group1 using the MMSE solution and store it in G ( ) (0), (1),.. ( / 2 1)est tX v v q q q N

• Cancel the interference due to from ( )estX v[ / 2 1]

[0]

( )tq N

m estm q

X m

2Y Y h

Algorithm :

(2)

Y



• Obtain the MMSE solution for Group2 2 1( )H H

t vN 2 2 2 2G H H I H

where is obtained by replacing the columns of corresponding to index of detected streams with zeros

2H H

• Using and , apply ordered SIC to detect streams in Group2 and store it in

t testX (v) v = q(N /2),...q(N - 1)

2G2Y

• Complexity 3 2

3 2

/ 2,

6 22

t t

t t

i N N

N Ni i i


Group ordered MMSE- V-BLAST (adaptive)

• Obtain and from G p

• Normalize with respect to its minimum valuep

normmin

p

pp• Grouping:

Group streams corresponding to and store the index in (threshold is1.5, 1.25)

norm threspq

Y• Using and , apply ordered SIC to detect streams in the group and store it in

G

estX• Cancel the interference due to from YestX• Obtain , , and corresponding to 2GY 2H normp

2G

Repeat from step 3 until all the streams are detected

Algorithm :


Simulation and Discussion

Uncoded system:Number of transmit antennas = 4Number of receive antennas = 4Modulation = QPSKNumber of subcarriers = 64Cyclic prefix length = 16 samplesMIMO channel – TGn channel model D Max Delay spread of channel D = 390nsSpatial distance between antennas = 0.5For adaptive scheme thres = 1.75 and 2


GO MMSE V-BLAST (fixed)

•In uncoded MIMO-OFDM system the fixed group ordering performs better than MMSE

•The computations required isslightly more than MMSE butless than MMSE V-BLAST


GO MMSE V-BLAST (Adaptive)

•In uncoded MIMO-OFDMsystem the adaptive groupordering almost approaches the performance of original V-BLAST• As thres value decreases, the performance approaches the MMSE V-BLAST

• When thres=1, the performance of proposed scheme is similar to MMSE V-BLAST


Performance under various channel models

• In channel C and B, the system performs poorly due its high condition numbers

• Performance of the system inthe most representative channelmodel D is good.

• SNR at BER=10-4 for fixed scheme and adaptive scheme under all the channel models


Coded GO MMSE VBLAST

• 4x4 MIMO OFDM system from EWC proposal for 802.11n standardization [7]

• Convolutional encoder with coding rate = ½• Interleaving – across the streams and across

subcarriers• QSPK Modulation • Uses 56 subcarriers for useful data with 16 samples as

cyclic prefix length• Channel model D with maximum delay spread of 390

ns• Spatial distance between antennas = 0.5• For adaptive scheme thres = 1.75 and 2


Coded GO MMSE VBLAST (fixed)

•In coded MIMO-OFDM system the fixed group ordering performs better than MMSE and is very close to MMSE V-BLAST

•Coding and interleaving exploits the frequency diversity and provides this performance


Coded GO MMSE VBLAST (Adaptive)

•In coded MIMO-OFDMsystem the adaptive groupordering performs similar to original V-BLAST

•When thres=1, the performance of proposed scheme is similar to the performance MMSE V-BLAST


Performance under various channel models

• In channel C and B, the system performs poorly due its high condition numbers

• Performance of the system inthe most representative channelmodel D is good.

• SNR at BER=10-4 for fixed scheme and adaptive scheme under all the channel models


Complexity Comparison

Proposed fixed scheme requires 3360 extra

computations compared to MMSE

For Adaptive schemes, they are computations are variable

for each subcarrier.

Spatial detection technique No of complex operations

MMSE 22400

MMSE V-BLAST 36400

Fixed GO MMSE V-BLAST

25760

And as threshold decreases, the computations required

also increases.


Conclusion

• A Group ordered MMSE V-BLAST with low complexity has been proposed.

• The complexity required for the proposed system is slightly larger than the MMSE but less when compared to MMSE VBLAST.

• The performance difference between group ordered and MMSE V-BLAST is slightly large in uncoded system whereas in coded system difference merges.

• The proposed technique can be potentially used as a detection technique for high speed WLANs


Acknowledgement

The authors like to acknowledge AAU-CSys and FUNDP-INFO for providing the MATLAB implementation of the IEEE 802.11 HTSG channel model. They would also like to thank Professor Laurent Schumacher for guiding in channel model simulations.


References

[1]. G. J. Foschini and M. J. Gans, “On the limits of wireless communications in a fading environment when using multiple antennas”, Wireless Personal Communications, vol. 6, no. 3, pp. 311-335, 1998.

[2]. http://www.wwise.org/11-05-0149-01-000n-wwise- proposal- high-throughput-extension-to-802-11-standard.doc

[2]. LaurentSchumacher, Klaus I. Pedersen, Preben E. Mongensen, “From antenna spacings to theoretical capacities – guidelines for simulating MIMO systems, Proc. PIMRC 2002, ” pp. 587- 592, vol.2,

[3]. P. W. Wolniansky, G. J. Foschini, G. D. Golden and R. A. Valenzuela, “V-BLAST: an architecture for realizing very high data rates over the rich-scattering wireless channel”, in Proc. ISSSE, pp. 295-300, 1998

[4]. Babak Hassabi, “A efficient square root algorithm for BLAST”, Proc. International Conference on Acoustics, Speech and Signal Processing 2000, pages 737-740.

[5]. IEEE P802.11 TGn channel models, May 10 2004,http://www.ece. ariz ona.edu/~yanli/files/11-03-0940-04-000n-tgn-channel-models.doc

[6]. http://www.enhancedwirelessconsortium.org/home/EWC_PHY_spec_V113.pdf


Publications

• V.Sathish, S.Srikanth, “Low complexity MIMO detection technique for high speed WLANs”, pp. 63-67, Proc. National Conference RF & Baseband systems for wireless applications, TIFAC core, Madurai, India, Dec 11-12, 2005.

• Published a tutorial in www.wirelessnetdesignline.com with title “Tutorial on IEEE 802.11n systems”


Wireless communications lab, AU-KBC Research Centre TIME SYNCHRONIZATION AND LOW COMPLEXITY DETECTION FOR HIGH SPEED WIRELESS LOCAL AREA NETWORK V. Sathish,

Documents