老師:高永安 學生:蔡育修

34
ICI Mitigation for Pi lot-Aided OFDM Mobile Systems Yasamin Mostofi, Member, IEEE and Donald C. Cox, Fellow, IEEE IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VO L. 4, NO.2, MARCH 2005 老老 老老老 老老 老老老

Upload: keane-garrison

Post on 01-Jan-2016

27 views

Category:

Documents


1 download

DESCRIPTION

ICI Mitigation for Pilot-Aided OFDM Mobile Systems Yasamin Mostofi, Member, IEEE and Donald C. Cox, Fellow, IEEE IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO.2, MARCH 2005. 老師:高永安 學生:蔡育修. Outline. Introduction System model Piece-Wise Linear Approximation Method I - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: 老師:高永安 學生:蔡育修

ICI Mitigation for Pilot-Aided OFDM Mobile SystemsYasamin Mostofi, Member, IEEE and Donald C. Cox, Fellow, IEEEIEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO.2, MARCH 2005

老師:高永安學生:蔡育修

Page 2: 老師:高永安 學生:蔡育修

Outline

Introduction System model Piece-Wise Linear Approximation

Method I

Method II Mathematical Analysis and Simulation Result Noise/Interference Reduction Simulation Results and Conclusion

Page 3: 老師:高永安 學生:蔡育修

Introduction

Transmission in a mobile communication environment is impaired by both delay and Doppler spread.

As delay spread increases, symbol duration should also increase.

reasons---1.near-constant channel in each frequency subband. 2.prevent ISI.

OFDM system become more susceptible to time-variations as symbol length increases. Time-variations introduce ICI. be mitigated to improve the performance.

Page 4: 老師:高永安 學生:蔡育修

We introduce two new methods to mitigate ICI. Both methods use a piece-wise linear model to approximate channel time-variations.

Page 5: 老師:高永安 學生:蔡育修

System model

: the transmitted data point in the th frequency subband

: the received data point in the th frequency subbandi

i

X i

Y i

Assume perfect timing synchronizaton

Page 6: 老師:高永安 學生:蔡育修

21

0

0 1 1j kiNN

i kk

X x e i N

is the cyclic prefix vector with length G(assume the length of

the channel is always less than or equal to G)

0 -1 2

: the time duration of

p

p N G i

x

x i x i G

T

one OFDM symbol after adding the guard

interval.

Page 7: 老師:高永安 學生:蔡育修

ths

s

th

: the channel tap at time instant

is the sampling period

for - 1 and 0 1 represents the channel

tap in the guard and data interval.

ik

ik

h k t i T

T T N G

h G i i N k

0

0 1 3

: a cyclic shift in the base

: a sample of additve white Gaussian noise

N

Gi

i k ii kk

N

i

y h x w i N

N

w

The channel output y

Page 8: 老師:高永安 學生:蔡育修

1

,0 ,1

th

21

0

0 1 4

Define as the FFT of the channel tap with respect to time-

variant:

0 & 0

N

N

i i i i d ii dd

ICI

m

j ukNu N

m mu

Y H X H X W i N

F m

F k h e m G

2

,0

1

1 0 , N-1

j m i dGN

i d mm

k N

H F d e i dN

The FFT of sequence y

Page 9: 老師:高永安 學生:蔡育修

2

,00

1

0

where

1

j miGave N

i mm

Nuave

m mu

H h e

h N h

Furthermore,

Page 10: 老師:高永安 學生:蔡育修

Pilot Extraction

We insert equally spaced pilots, , at subchannels /

for 0 -1 il iL P l i N L

i L

An estimate of Hi,0 can then be acquired at pilot:

,0 ,0

21

,0

0

0 1 5

: ICI at th subcarrier

Through an IFFT of length , the estimate of :

1

i ilili li

li li

i i

avek

j ikLaveL

lik

i

I WYH H i L

P P

I l

L h

h H eL

0 -1 6k L

Page 11: 老師:高永安 學生:蔡育修

In the absence of mobility, L pilots would have been enough to estimate the channel.

However, in the presence of Doppler, due to the ICI term,

using them for data detection results in poor perfor-mance.

This motivates the need to mitigate the resultant ICI.

Page 12: 老師:高永安 學生:蔡育修

Piece-Wise Linear Approximation We approximate channel time-variations with a piece-

wise linear model with a constant slope over the time duration T.

Page 13: 老師:高永安 學生:蔡育修

2thFor the channel tap, E is minimized for 1

2save

k k

Nk h h s

12

Nave

kkh h

For normalized Doppler of up to 20%, linear approxi-

mation is a good estimate of channel time-variations.

We will derive the frequency domain relationship.

Therefore, we approximate

Page 14: 老師:高永安 學生:蔡育修

1

2

th

1 0 -1 72

: the slope of the k channel tap in the OFDM symbol

Ni

k k k s

k

Nh h i T i N

8

where , and are 1 vectors

y x w

y x w N

midh M A������������������������������������������

������������������������������������������

Then,

we will have

Page 15: 老師:高永安 學生:蔡育修

12

k-m

,

0 & - -1 =

0 else

,

0 & - -1 =

0

N

N

N

k m

k m

h k m G N k m N G

k m

k m G N k m N G

midh

A

1

else

M is a diagonal matrix with diagonal elements of M ,

for 12k s

k k

NT k k N

Futhermore,

Page 16: 老師:高永安 學生:蔡育修

1 1 12 2 2

0 1

0 1

9

where

diag ... 0 ... 0

diag ... 0 ... 0

N N N

G

G

Y X X W

FFT h h h

FFT

mid slope

mid

slope

H C H

H

H

��������������������������������������������������������

n,m

2

k

where is the FFT of

1 0

= 1 0.5 0

n m

j k

Ns

BB

N

kB T N e

k

C

An FFT of y:

Page 17: 老師:高永安 學生:蔡育修

To solve for X, both Hmid and Hslope should be estimated.

Matrix C is fixed matrix and Hmid is readily available.

So we show how to estimate Hslope with our two methods.

Page 18: 老師:高永安 學生:蔡育修

Method I: ICI Mitigation Using Cyclic Prefix

2

1

10

where

, 1 1 2

: the transmitted cyclic prefix of the previous OFDM

G

pP

G i

i j G

prep

p

prep

y p w

i j h i G j G

xp

x

x

Q

Q

������������������������������������������

��������������

0 1 G

symbol

... ��������������

The output prefix vector

Page 19: 老師:高永安 學生:蔡育修

2

12

11

where

, for 1 & 1 2

, for 12

Re 1: 1

Re 2 :

mat

G

mat

ppp

N

i j G

s

t

t

p

y p w

i j h i G j G

Ni i T i G i G

v p G

v p G

R D X

R

D

X

��������������������������������������������������������

��������������

��������������

Rev J : reverses the order2,

of elements through of vector J

Re : 2t

i j

i j

v p G G

��������������

Then,

Page 20: 老師:高永安 學生:蔡育修

Equations (9) and (11) provide enough information to solve for X.

We use a simpler iterative approach to solve for X.

Page 21: 老師:高永安 學生:蔡育修

Method II: ICI Mitigation Utilizing Adjacent Symbols This can be done by utilizing either the previous symbol

or both adjacent symbols.

A constant slope is assumed over the time duration of

T+(N/2)*Ts for the former and T for the latter.

Page 22: 老師:高永安 學生:蔡育修

2

2

1 , 12 2

1

0

where

: the slope of the th channel tap in region 2

simiarly, , the slope in region 1

N Nnext

rk k

k

r

k

r

k

h hk G

T

k

Estimate of the slopes in region 2:

Page 23: 老師:高永安 學生:蔡育修

Utilizing two slopes introduces a minor change in (8).

1 1 2 2

1

method

where

: channel slope matrix in the region 1 2

, & 0 1 , 2

0

II

m

r r r r

r th

r

y x w

m m

Ni j i j i

i jelse

midh M A M A

A

MM

������������������������������������������

2

, & 1 , 2

0

r

Ni j i j i N

i jelse

MM

Page 24: 老師:高永安 學生:蔡育修

1 1 2 2method

: diagonal matrix for the slopes of the region

1 2

II

m

r r r rslope slope

r thslope

Y X X W

m

m

midH C H C H

H

��������������������������������������������������������

It can be easily shown the frequency domain relationship

Page 25: 老師:高永安 學生:蔡育修

1

2

r

22 2

s

r

22 2

s

,

1 1.5

1 1 =T

1

4 8

,

1 1.5

1 1 =T

n m

j n m j n mN N

n m

j n m j n mN N

n m

n m

e N e

Nn m

n m

n m

e N e

C

C

1

4 8

Nn m

Page 26: 老師:高永安 學生:蔡育修

Method I and Method II can handle considerably higher

delay and Doppler spread at the price of higher compu-

tation complexity.

Page 27: 老師:高永安 學生:蔡育修

Mathematical Analysis and Simulation Result We define SIRave as the ratio of average signal power

to the average interference power.

Our goal is to calculate SIRave when ICI is mitigated and

compare it to the that of the “no mitigation” case.

Page 28: 老師:高永安 學生:蔡育修

Noise/Interference Reduction

if 0, 0 -1ave ave

k kh Threshold h k L

Estimated channel taps are compared with a Threshold.

Let MAV represent the tap with maximum absolute value.

All the estimated taps with absolute values smaller than

MAV/γ for some γ>=1 will be zeros.

Page 29: 老師:高永安 學生:蔡育修

Simulation Results

System parameters

Page 30: 老師:高永安 學生:蔡育修

The power-delay profile of channel#1 has two main taps

that are separated by 20μs.

The power-delay profile of channel#2 has two main clus-

ters with total delay of 36.5μs.

Page 31: 老師:高永安 學生:蔡育修

Each channel tap is generated as Jakes model.

To see how ICI mitigation methods reduce the error floor.

in the absence of

noise for both channels.

Page 32: 老師:高永安 學生:蔡育修

To see the effect of noise for fd,norm = 6.5%

Page 33: 老師:高永安 學生:蔡育修

To see how ICI mitigation methods reduce the required

received SNR for achieving a Pb = 0.2.

Page 34: 老師:高永安 學生:蔡育修

Conclusion

Both methods used a piece-wise linear approximation to

estimate channel time-variations in each OFDM symbol.

These methods would reduce average Pb or the required

received SNR to a value close to that of the case with no

Doppler.

The power savings become considerable as fd,norm incre-

ases.