18 views

Uploaded by khajarasool_sk

Attribution Non-Commercial (BY-NC)

- LTE Benefits - LTE Product Design
- Recent Advance in Multi-Carrier Underwater Acoustic Communications
- Sajid Hussain MSc Thesis Report
- MIMO Wireless Communication Systems
- 1.8 MIMO Systems
- ROC ANALYSIS OF SPECTRUM SENSING TECHNIQUES IN COGNITIVE RADIO FOR OFDM SYSTEM
- 256-Channel ADSL - MATLAB & Simulink Example
- ADS-SV-N
- A primer on 3GPP narrowband Internet of Things.pdf
- OFDMA
- Underwater Acoustic Communications and Networking.pdf
- Chapter 6
- adaptive_ofdm
- Chapter 1
- LTE OFDM Principle 32
- 10.1.1.27.6413
- BER and SER Based Performance Analysis of BPSK and QPSK
- 234[1]
- IJETR031933
- Simulation of DVB Ch for Antena Diversity Rx

You are on page 1of 15

www.elsevier.com/locate/sigpro

systems with null sub-carriers

Hyoung-Goo Jeona, Erchin Serpedinb,

a

Department of Information Communication Engineering, Dongeui University, Busan, Republic of Korea

b

Department of Electrical and Computer Engineering, Texas A&M University College Station, TX 77843-3128, USA

Received 10 April 2007; received in revised form 7 August 2007; accepted 19 October 2007

Available online 26 November 2007

Abstract

This paper proposes an efﬁcient scheme to track the time variant channel induced by multi-path Rayleigh fading in

mobile wireless multiple input multiple output–orthogonal frequency division multiplexing (MIMO–OFDM) systems with

null sub-carriers. In the proposed method, a blind channel response predictor is designed to cope with the time variant

channel. The proposed channel tracking scheme consists of a frequency domain estimation approach that is coupled with a

minimum mean square error (MMSE) time domain estimation method, and does not require any matrix inverse

calculation during each OFDM symbol. The main attributes of the proposed scheme are its reduced computational

complexity and good tracking performance of channel variations. The simulation results show that the proposed method

exhibits superior performance than the conventional channel tracking method [Y.G. Li, N. Seshadri, S. Ariyavisitakul,

Channel estimation for OFDM systems with transmitter diversity in mobile wireless channels, IEEE J. Sel. Areas

Commun. 17 (March 1999) 461–471] in time varying channel environments. At a Doppler frequency of 100 Hz and bit

error rates (BER) of 104 , signal-to-noise power ratio (E b =N 0 ) gains of about 2.5 dB are achieved relative to the

conventional channel tracking method [Y.G. Li, N. Seshadri, S. Ariyavisitakul, Channel estimation for OFDM systems

with transmitter diversity in mobile wireless channels, IEEE J. Sel. Areas Commun. 17 (March 1999) 461–471]. At a

Doppler frequency of 200 Hz, the performance difference between the proposed method and conventional one becomes

much larger.

r 2007 Elsevier B.V. All rights reserved.

from each transmit antenna [1]. Recently, orthogo-

The multiple input multiple output (MIMO) nal frequency division multiplexing (OFDM) has

technique represents an efﬁcient method to increase been effectively used for transmitting high speed

data transmission rate without increasing band- data in multi-path fading channel environments. In

OFDM, the high speed data stream is processed in

Corresponding author. Tel.: +1 979 458 2287; parallel and transmitted by N (in general, a power

of 2) orthogonal sub-carriers. The high spectral

fax: +1 979 862 4630.

E-mail addresses: hgjeon@deu.ac.kr (H.-G. Jeon), efﬁciency of OFDM and its robustness to multi-

serpedin@ece.tamu.edu (E. Serpedin). path fading channel environments are the main

0165-1684/$ - see front matter r 2007 Elsevier B.V. All rights reserved.

doi:10.1016/j.sigpro.2007.10.017

ARTICLE IN PRESS

H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016 1003

reasons for its widespread usage in high bit-rate In Section 4, the mean square error (MSE) and

transmissions such as digital audio broadcasting computational complexity of the proposed channel

(DAB), digital video broadcasting (DVB) and tracking scheme are assessed. The performance of

wireless local area networks (WLAN) [2]. The the proposed method is corroborated by computer

combined transmission method of MIMO–OFDM simulations in Section 5. Finally, Section 6 con-

has attracted a lot of attention as a new data cludes the paper.

transmission method in high speed data rate

systems. In MIMO–OFDM receivers, the estimated

2. Channel and MIMO–OFDM system description

channel frequency response is used to separate the

mixed signals received from multiple antennas. An

The channel impulse response of the mobile

important aspect is the fact that the performance of

wireless channel [3,6] can be modeled by

MIMO–OFDM receivers highly depend on the

accuracy of the channel estimator. X

L1

Thus far, numerous studies for channel estima- hðt; tÞ ¼ ak ðtÞdðt tk Þ, (1)

k¼0

tion in MIMO–OFDM systems have been reported

(see e.g., [3–10]). Among the most notable results, Li where ak ðtÞ denotes the complex gain of the kth

proposed a MMSE channel estimation method [3] path, tk represents the delay of the kth path, L is the

that exhibits good accuracy. However, this method number of the multi-paths in the channel and dðtÞ

is computationally very complex due to the inverse stands for the impulse function. The frequency

matrix calculation. In [6], by exploiting the correla- response at time t is given by

tion of the subcarrier responses, Minn et al. Z 1 X

L1

proposed a low complexity channel estimation Hðt; f Þ9 hðt; tÞ ej2pf t dt ¼ ak ðtÞ ej2pf tk .

method which reduced the inverse matrix size by 1 k¼0

half. However, Minn et al. method may cause ð2Þ

channel estimation errors in large delay spread Considering the motion of the mobile station, the

environments. Li also proposed a simpliﬁed channel path gains ak ðtÞs are modeled to be independent

estimation method which required no matrix inver- wide-sense stationary, narrow band complex Gaus-

sion [4]. However, as mentioned in [6], if null sub- sian processes and to have different average powers

carriers are used, Qii ½n of [4] would not be the s2k . With tolerable leakage, the channel frequency

identity matrix, and there may be some performance response can be expressed as [3]

degradation according to the number of null sub-

carriers. Since real OFDM systems have null sub- X

L 0 1

carriers in the guard band, a low complexity channel H½l; k9HðlðT f þ T g Þ; kDf Þ ¼ h½l; nW kn

N , (3)

n¼0

estimation method considering null sub-carriers is

still needed. where h½l; n9hðlðT f þT g Þ; nts Þ, W N 9 expðj2p= NÞ,

As a possible solution to these problems, we are L0 stands for the channel length and depends on the

proposing a novel simpliﬁed channel tracking time dispersion of the wireless channel, N is the

method that relies on a blind channel predictor. number of tones and the fast Fourier transforma-

The proposed method does not require prior tion/inverse fast Fourier transformation (FFT/

channel information or matrix inversion calculation IFFT) size, T f and Df denote the OFDM symbol

at all. In addition, the proposed method can period and sub-carrier spacing of the OFDM

effectively track the nonlinear time varying channel system, respectively, T g represents the guard time

by using a piecewise linear model. To reduce its ðT g 9T f =4Þ and ts is the sample interval given by

computational complexity while maintaining a good ts ¼ 1=ðNDf Þ.

tracking accuracy, the proposed channel tracking In a MIMO–OFDM system, the output signal at

scheme is built by coupling a frequency domain each Rx (receive) antenna is a mixed signal

estimation approach with an MMSE time domain consisting of the data streams coming from all Tx

channel estimation approach. (transmit) antennas. If the channel response

The remainder of this paper is organized as does not change during one OFDM symbol

follows. In Section 2, the MIMO–OFDM system and the cyclic preﬁx is longer than the channel

and channel model are brieﬂy described. Section 3 response length, the receive signal at the jth Rx

introduces the proposed channel tracking method. antenna can be expressed in the frequency domain

ARTICLE IN PRESS

1004 H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016

time synchronization is assumed to be perfect, and

X

Nt

Rj ½l; k ¼ H ij ½l; kX i ½l; k þ W j ½l; k, the maximum likelihood (ML) detection method is

i¼1 used. Tx antennas transmit a long preamble

j ¼ 1; . . . ; N r ; 0pkpN 1, ð4Þ consisting of two training symbols before data

transmission mode, as WiBro and WLAN systems

where H ij ½l; k is the channel frequency response do [2,12]. It is assumed that in the data transmission

corresponding to the kth sub-carrier and the lth mode, N d OFDM symbols are transmitted con-

OFDM symbol transmitted between the ith Tx secutively in each Tx antenna. For unbiased

antenna ði ¼ 1; . . . ; N t Þ and the jth Rx antenna. performance comparison of channel tracking algo-

Also, let N, N r and N t denote the number of sub- rithms, no channel coding is used.

carriers, the number of Rx antennas, and the

number of Tx antennas, respectively. X i ½l; k de-

notes the data transmitted from the ith Tx antenna 3. Proposed channel estimation method

on the kth sub-carrier at the lth OFDM symbol.

W j ½l; k represents the additive white Gaussian noise Since the wireless channel is time-variant, it is

(AWGN) at the jth receiver antenna, with zero necessary to track the channel response continu-

mean and variance s2n , and is assumed to be ously. In addition, since the received signal at each

uncorrelated for different j’s, k’s, or l’s. Under the Rx antenna in MIMO–OFDM systems is a multi-

assumption that the channel stays constant within ple-input single-output (MISO) signal, a time

one OFDM symbol duration but the channel domain channel estimation cannot be directly

changes from symbol to symbol, we will develop a applied on the received signal. In this paper, we

channel tracking scheme with improved perfor- propose a low complexity adaptive channel estima-

mance relative to the conventional scheme [3]. The tion method based on a blind channel prediction

computer simulations, which assume realistic Ray- scheme that is suitable for time variant channel

leigh fading conditions (that are not limited to the environments. The conceptual block diagram of

block fading assumption) [11], corroborate the the proposed channel tracking scheme is shown in

superior performance of the proposed channel Fig. 1. Before channel estimation, the frequency

tracking scheme. The indices n and k denote time domain MISO signal received at the jth Rx antenna

and frequency-domain indices, respectively. The (Rj ½l; k) is converted into the desired single-input

symbols a, ~ a^ and a denote the temporally estimated single-output (SISO) signal (See Section 3.2 for the

value, the estimated value and the predicted value of deﬁnition of desired SISO signal) by canceling the

Delay

device

H Time Hˆij [l,k]

FFT to domain domain Post ML

IFFT FFT

SISO

conversion

Channel Channel ˆij [l,n]

h detector

Est. Estimation

MISO SISO

X˜i [l,k]

detector predictor

ARTICLE IN PRESS

H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016 1005

other interfering signals coming from the other Tx rate. For example, a channel with a short coherence

antennas. When pre-demodulating and converting time will have a small M, and vice versa. In the

the MISO signal Rj ½l; k into the desired SISO piecewise linear model, the condition Dij ½l; k ’

signal, the predicted channel response H ij ½l; k is Dij ½l 1; k is assumed. Therefore, from (5) we infer

used to cope with the time variant channel instead that H ij ½l; k ’ H ij ½l 1; k þ Dij ½l 1; k. If we

of using the previously estimated channel response know H^ ij ½l 1; k and D^ ij ½l 1; k at the time

H^ ij ½l 1; k. Once the SISO signal is obtained, instant corresponding to the lth symbol, then

temporal channel estimation in frequency domain the predicted channel response H ij ½l; k can be

is carried out to remove the inverse matrix calcula- obtained as

tion required in the following time domain channel

estimation. In the time domain channel estimation H ij ½l; k9H^ ij ½l 1; k þ D^ ij ½l 1; k. (6)

block, the channel impulse response is obtained by Referring to Fig. 2 of the piecewise linear model,

minimizing a MSE cost function, considering the D^ ij ½l 1; k can be expressed by using the previously

presence of null sub-carriers. estimated channel responses, since 2Dij ½l 1; k ’

H ij ½l 1; k H ij ½l 3; k and Dij ½l 1; k ’

3.1. A blind channel response predictor design H ij ½l 2; k H ij ½l 3; k. In the case where the

channel response varies linearly during the M

In this paper, we design a blind channel predictor consecutive OFDM symbols, therefore, (6) can be

by exploiting a piecewise linear model for the time expressed by the linear combination of the ðM 1Þ

varying channel. Fig. 2 shows an example of the previously estimated channel responses as follows:

time varying channel frequency response at the kth

sub-carrier. The channel frequency response of each X

M1

H ij ½l; k9 om H^ ij ½l m; k

sub-carrier, H ij ½l; k, varies nonlinearly with time.

m¼1

However, the adjacent channel frequency responses

X

M1

H ij ½l; k and H ij ½l 1; k present a certain correla- ¼ om ðH ij ½l m; k þ Zij ½l m; kÞ, ð7Þ

tion with each other and this relation is expressed as m¼1

H ij ½l; k ¼ H ij ½l 1; k þ Dij ½l; k, (5) where om is a weight value and Zij ½l; k9H^ ij ½l; k

where Dij ½l; k denotes the difference between the H ij ½l; k and denotes the random channel esti-

channel frequency responses corresponding to lth mation error with zero mean and variance s2e (see

and ðl 1Þth symbols at the kth sub-carrier. If a Section 4.2). Let us deﬁne DH ij ½l; k9H ij ½l; k

piecewise linear model [13] is used during the short H ij ½l; k as the channel response prediction random

time of some OFDM symbol periods as shown in error. The weight values can be found by minimiz-

Fig. 2, the nonlinear time varying channel frequency ing the following MSE cost function:

response H ij ½l; k can be treated as a linear model.

x½l; k9EfjDH ij ½l; kj2 g

Using the piecewise linear model, let us assume 8 2 9

that H ij ½l; k varies linearly during the time of the < X

M1 =

consecutive M OFDM symbols. The variable M can ¼ E H ij ½l; k om H^ ij ½l m; k . ð8Þ

: m¼1

;

be set according to the channel response changing

During the M OFDM symbols, H ij ½l; k ’

H ij ½l m; k þ m Dij ½l; k, 1pmoM. Therefore,

Channel

frequency (

Hij [l − 1,k] Hij [l ,k]

response

Hij[l − 2,k]

x½l; k ’ E jH ij ½l M þ 1; k þ ðM 1ÞDij ½l; k

X

M1

om ðH ij ½l M þ 1; k

m¼1

Time(l )

þ ðM 1 mÞDij ½l; k

)

Piecewise linear model

þZij ½l m; kÞj2

Fig. 2. Piecewise linear model for nonlinear time variant channel.

ARTICLE IN PRESS

1006 H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016

( !

X

M1 the proposed predictor requires no additional

¼ 1 om H ij ½l M þ 1; k training process and convergence time.

m¼1

! )2

X

M1

þ M 1 om ðM 1 mÞ Dij ½l; k 3.2. Conversion of MISO Rx signal into SISO

m¼1 desired signal

!

X

M1

þ s2e o2m . ð9Þ As mentioned in the previous section, it is

m¼1 efﬁcient to estimate channel response after convert-

The weight values om can be determined by solving ing the MISO Rx signal into the SISO desired signal

qx½l; k=qom ¼ 0. However, since H½l M þ 1; k, by canceling the interference signal coming from the

D½l; k and s2e are not given due to the blind other Tx antennas. The SISO desired signal between

estimation property, we cannotP solve the equation the ith Tx antenna and the jth Rx antenna is deﬁned

directly. From (9), when M1 by S ij ½l; k9X i ½l; kH ij ½l; k ¼ X i ½l; kðH ij ½l 1; kþ

PM1 ð1 i¼1 om Þ ¼ 0 and

ðM 1 m¼1 om ðM 1 mÞÞ ¼ 0, the cost Dij ½l; kÞ. The temporally estimated SISO signal

function x½l; k can be minimized, and then S~ ij ½l 1; k is given by

P

x½l; k ’ s2e M1 2

m¼1 om . The surface of the cost func- X

Nt

tion is convex with respect to om . Therefore, om can S~ ij ½l; k ¼ Rj ½l; k X~ m ½l; kH^ mj ½l 1; k

m¼1;mai

be numerically determined by ﬁnding the values

minimizing the cost function of Cðo1 ; o2 ; . . . ; oM1 Þ ¼ X i ½l; kH ij ½l; k þ N ij ½l; k, ð11Þ

P PN t

9 M1 m¼1 om

2

while satisfying the conditions 0

where N ij ½l; k9W j ½l; k þ m¼1;mai X m ½l; kDmj ½l; k

PM1 PM1

m¼1 o m ¼ 1 and m¼1 om ðM 1 mÞ ¼ M 1. and D0ij ½l; k is deﬁned by D0ij ½l; k9H ij ½l; k H^ ij

In this paper, the weight values om are obtained ½l 1; k, and X~ i ½l; k denotes the demodulated data

for M ¼ 4, 5 and 6, and the results are given as by using the previous channel response H^ ij ½l 1; k.

follows: For convenience, X i ½l; k ¼ X~ i ½l; k is assumed.

M ¼ 4; o1 ¼ 43; o2 ¼ 13; o3 ¼ 23, Since D0ij ½l; k is proportional to the channel varia-

tion, it can be seen from (11) that the faster the

x½l; k ’ 2:3s2e , channel variations are and the more Tx antennas

M ¼ 5; o1 ¼ 1; o2 ¼ 0:5; o3 ¼ 0, are used, the higher the channel estimation errors

o4 ¼ 0:5; x½l; k ’ 1:5s2e , are. Moreover, if the previous channel response

H^ mj ½l 1; k is used in fast time varying channel

M ¼ 6; o1 ¼ 0:8; o2 ¼ 0:5; o3 ¼ 0:25, environments, X i ½l; k ¼ X~ i ½l; k cannot be assumed

o4 ¼ 0:2; o5 ¼ 0:35; x½l; k ’ 1:11s2e . ð10Þ anymore. In that case, the noise term N ij ½l; k may

increase signiﬁcantly. In this paper, in order to

It is assumed that the initial channel response

reduce the noise term in (11), the predicted channel

values H^ ij ½1; k and H^ ij ½2; k can be obtained by

response H ij ½l; k is used not only to cancel the

using the long preamble signal consisting of two

interference signal in the MISO Rx signal but also

consecutive training OFDM symbols, as is the case

to pre-demodulate X i ½l; k as shown in the block

with WiBro and WLAN systems. After processing

diagram of Fig. 1. Using H ij ½l; k and X i ½l; k, (11) is

the long preamble signal, we have two initial

revised as

channel response values H^ ij ½1; k and H^ ij ½2; k,

but no access yet to H^ ij ½3; k; . . . ; H^ ij ½1 M; k. X

Nt

For this reason, the initial condition is given as S~ ij ½l; k ¼ Rj ½l; k X m ½l; kH mj ½l; k

H^ ij ½2; k ¼ H^ ij ½3; k ¼ ¼ H^ ij ½1 M; k. In the m¼1;mai

conventional channel estimation method [2], the ¼ X i ½l; kH ij ½l; k þ N 0ij ½l; k, ð12Þ

long preamble signal was averaged to reduce noise P

variance. It should be noted that in this paper, where N 0ij ½l; k9W j ½l; k þ N t

m¼1;mai X m ½l; kDH mj

however, the long preamble signal is used for ½l; k. For convenience, X i ½l; k ¼ X i ½l; k is assumed.

channel response prediction in time varying channel If there is no prediction error such that

environments. Since the proposed channel predictor DH ij ½l; k ¼ 0, then S~ ij ½l; k ¼ X i ½l; kH ij ½l; k þ W j

uses piecewise linear model, it can adaptively track a ½l; k, regardless of the channel response variation

nonlinearly varying channel response. In addition, and the number of Tx antennas.

ARTICLE IN PRESS

H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016 1007

3.3. Joint frequency and time domain MMSE guard band. Note that estimating the channel

channel estimator response in the frequency domain removes the need

of calculating an inverse matrix in the next step of

The channel estimation methods [3,6] required time domain channel estimation employing an

inverse matrix calculation in every OFDM symbol. MMSE technique. The time domain channel esti-

As mentioned in [6], Qii ½n of [4] is not the identity mation is performed during the next step, by

matrix in the presence of null sub-carriers and non- considering all the null sub-carriers used in the

constant modulation. For this reason, Li’s simpli- guard band. The time domain channel estimate

ﬁed method [4], which requires no matrix inversion, h^ij ½l; n can be found by minimizing the following

cannot be used in general MIMO–OFDM systems MSE cost function:

with null sub-carriers and employing non-constant

modulation. Therefore, our goal is to derive a novel Cðfh^ij ½l; ng; i ¼ 1; . . . ; N t Þ

low complexity channel estimation method which 2

X

N 1 X L0

can be used in general MIMO–OFDM systems ^

9 Y ij ½l; k nk

hij ½l; nW N Z½k . ð16Þ

while maintaining good performance. In this k¼0

n¼0

section, we propose a frequency domain estimation

approach coupled with an MMSE method that h^ij ½l; n is the estimated channel impulse response

requires no inverse matrix calculation in every and can be determined by solving

symbol. In the proposed method, frequency domain ( )

channel estimation is carried out ﬁrst such that the qCðfh^ij ½l; ngÞ 1 qCðfh^ij ½l; ngÞ qCðfh^ij ½l; ngÞ

subsequent MMSE estimator does not require a 9 j

qh^ij ½l; n0 2 qRðh^ij ½l; n0 Þ qIðh^ij ½l; n0 Þ

matrix inversion during any OFDM symbol. The

proposed method can be used in any MI- ¼ 0, ð17Þ

MO–OFDM system, regardless of null sub-carriers

and non-constant modulation. where RðÞ and IðÞ denote the real and imaginary

Referring to (12), considering the null sub- parts of a complex number, respectively, and

carriers and using the least squares (LS) method in n0 ¼ 0; 1; . . . ; L0 . Direct solving (17) results in

frequency domain, the temporally estimated chan- !

nel frequency response is given by X

N1 X

L0

Y ij ½l; k h^ij ½l; nW kn

N Z½k Z½kW kn

N

0

¼ 0.

Y ij ½l; k ¼ ðS~ ij ½l; k=X i ½l; kÞZ½k k¼0 n¼0

given by

X

N 1

X

Nt q½n9 Z½kW kn

N , (19)

V ij ½l; k ¼ X m ½l; kDH ij ½l; k k¼0

m¼1;mai

!,

þW j ½l; k X i ½l; k. ð14Þ X

N 1

yij ½l; n9 Y ij ½l; kZ½kW kn

N . (20)

k¼0

Z½k is a time invariant function to denote the

pattern of null sub-carriers used in the guard band Then, (18) can be expressed as

and is given by

8 X

L 0 1

>

> 1 if 1pkoN=2 g or h^ij ½l; nq½n0 n ¼ yij ½l; n0 (21)

>

>

< N=2 þ gokpN 1; n¼0

Z½k ¼ (15)

>

> 0 if k ¼ 0 or

>

> for i ¼ 1; . . . ; N t and n0 ¼ 0; 1; . . . ; L0 . Eq. (21) can

: N=2 gpkpN=2 þ g; be expressed in matrix form as

where index k ¼ 0 denotes the DC component and g

stands for the number of null sub-carriers in the Qh^ ij ½l ¼ yij ½l, (22)

ARTICLE IN PRESS

1008 H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016

where be expressed as

0 1

q½0 q½1 q½1 L0 h^ ij ½l ¼ Q1 yij ½l ¼ hij ½l þ Q1 vij ½l. (31)

B C

B q½1 q½0 q½2 L0 C The MSE of the channel impulse response estimate

B C

Q9B .. .. .. C, can be given by

B . . . C

@ A

q½L0 1 q½L0 2 q½0 MSE½l9Efkh^ ij ½l hij ½lk2 g

(23) ¼ EfðQ1 vij ½lÞH Q1 vij ½lg

¼ TracefQ1 Efvij ½lvij ½lH gðQ1 ÞH g. ð32Þ

h^ ij ½l9ðh^ij ½l; 0; h^ij ½l; 1; . . . ; h^ij ½l; L0 1ÞT , (24) H

A generic entry of Efvij ½lvij ½l g is given by

( !

yij ½l ¼ ðyij ½l; 0; yij ½l; 1; . . . ; yij ½l; L0 1ÞT . (25)

X

N 1

n1 k1

Efvij ½l; n1 vij ½l; n2 g ¼ E V ij ½l; k1 Z½k1 W N

Hence, the channel impulse response can be k1 ¼0

estimated by !)

X

N1

2 k2

V ij ½l; k2 Z½k2 W þn

h^ ij ½l ¼ Q1 yij ½l. (26) k1 ¼0

N

N 1

determined by the null sub-carrier pattern, the ¼ EfV ij ½l; k1 V ij ½l; k2 g

k1 ;k2 ¼0

inverse matrix Q1 can be pre-calculated and stored

1 k 1 þn2 k 2

in the memory before beginning the channel Z½k1 Z½k2 W n

N , ð33Þ

tracking. Therefore, no explicit calculation of Q1

where n1 ; n2 ¼ 1; 2; . . . ; L0 . If we substitute (14) into

is required for every OFDM symbol. However, [3]

(33), Eq. (33) can be rewritten as

and [6] require inverse matrix calculation for every

OFDM symbol. Efvij ½l; n1 vij ½l; n2 g

(PN

X

N 1 t

m¼1;mai jX m ½l; kj

2

s2p

k¼0 jX i ½l; kj

)

4.1. MSE calculation s2n ðn1 n2 Þk

þ Z½kW N , ð34Þ

jX i ½l; kj2

In this section, we will derive the MSE of the P

proposed channel estimation scheme. From (20), where s2p 9EðjDH ij ½l; kj2 Þ ¼ s2e M1 2

m¼1 om and DH ij

one infers that ½l; k and W j ½l; k are assumed to be independent of

! each other. Hence, if a constant modulus modula-

X

N X

1 L 0 1

tion is used,

mk

yij ½l; n ¼ hij ½l; mW N þV ij ½l; k Z½kW nk

N

k¼0 m¼0 s2n

Efvij ½lvij ½lH g ¼ ðN t 1Þs2p þ Q: ð35Þ

X

L 0 1

jX i ½l; kj2

¼ hij ½l; mq½n m þ vij ½n; 0pnpL0 ,

m¼0 If no null sub-carriers are used, since Q ¼ NI,

ð27Þ L0 2 1

PN1 MSE½l ¼ ðN t 1Þsp þ . (36)

where vij ½n ¼ nk N SNR

k¼0 V ij ½l; kZ½kW N . Eq. (27) can

be expressed in matrix form as Given a SNR, the MSE½l is dependent on the

yij ½l ¼ Qhij ½l þ vij ½l, (28) prediction error, the channel response length and

the number of Tx antennas.

where

4.2. Mean and variance of random variable Zij ½l; k

hij ½l ¼ ðhij ½l; 0; hij ½l; 1; . . . ; hij ½l; L0 1ÞT , (29)

For convenient comprehension, let us assume

vij ½l ¼ ðvj ½l; 0; vj ½l; 1; . . . ; vij ½l; L0 1ÞT . (30)

that BPSK is used for X i ½l; k, then E½X i ½l; k ¼ 0.

From (28), the channel impulse response estimate X m ½l; k, DH ij ½l; k, W j ½l; k and X i ½l; k are indepen-

corresponding to Tx antenna i and Rx antenna j can dent of each other. It is clear that E½X m ½l; k ¼ 0,

ARTICLE IN PRESS

H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016 1009

E½X i ½l; k ¼ 0 and E½W j ½l; k ¼ 0. Therefore, it is modulus modulations, Li’s simpliﬁed method will

also clear that E½V ij ½l; k ¼ 0, where V ij ½l; k is given not be discussed in this comparison. In the case of

by (14). From (31), the channel estimation error in two Tx and Rx antennas and N sub-carriers, the

time domain can be expressed P as E½h^ ij ½l hij ½l ¼ channel estimation complexity for each method is

Q1 E½vij ½l. Since E½vij ½n ¼ N1 k¼0 E½V ij ½l; kZ½k given in Table 1 (refer also to [6]). In Table 1, FFTN

1

W nk N ¼ 0, Q E½v ij ½l ¼ 0. It means that the mean denotes the number of multiplications required for

of the channel estimation error in time domain is the FFT operation with size N. invðL0 L0 Þ stands

zero. Since the channel frequency response estimate for the number of multiplications required for L0

H^ ij ½l; k is obtained by performing Fourier trans- L0 matrix inversion. N u denotes the number of the

formation for the channel impulse response estimate sub-carriers used. When N u ¼ N, no null sub-

h^ij ½l; n, the mean of the channel estimation error in carrier is used. The number of FFT operations for

frequency domain is also zero. That is, E½Zij ½l; k ¼ Li method [3] can be easily obtained by referring to

E½H^ ij ½l; k H ij ½l; k ¼ 0. Fig. 3 in [3]. Considering Eqs. (15) and (16) in [3],

The MSE of the channel frequency response in the number of multiplications required for calculat-

the kth sub-carrier is deﬁned by E½ðH^ ij ½l; k ing hij ½l can be derived straightforwardly. The

H ij ½l; kÞ2 ¼ s2e . The total MSE of the channel calculation amount for Minn method [6] can be

impulse response in time domain is equal to obtained by considering the similarity with Li

MSE½l given by (36). When the channel impulse method. Note that when N u ¼ N and constant

response h^ij ½l; n is Fourier transformed to obtain modulation is used, Qii ½n is a identity matrix. In

the channel frequency response H^ ij ½l; k, the total that case, Li’s method needs calculating only

MSE in time domain is equal to the total MSE in invðL0 L0 Þ instead of calculating invð2L0 2L0 Þ.

frequency domain. Therefore, if there are N sub- From Table 1, we can see that the proposed method

carriers in the OFDM system, then s2e ¼ MSE½l=N. has the lowest complexity among these methods,

regardless of the non-constant modulation and the

4.3. Computational complexity comparison presence of null sub-carriers.

the schemes proposed in [3,6] and the method

proposed herein are compared brieﬂy. The complex- Computer simulations are carried out to evaluate

ity comparison will be focused on the channel the performance of the proposed method. Two Tx

estimation based on the decision-directed estimation and two Rx antennas are used for the MI-

method, as [6] did. Since Q1 can be pre-calculated, MO–OFDM system. There are a total of 128 sub-

Q1 calculation is not included in the complexity carriers so that the FFT/IFFT size is 128. The DC

comparison. Since Li’s simpliﬁed method [4] may component sub-carrier is not used, and 10 and 9

cause estimation error in MIMO–OFDM systems sub-carriers on each end of the spectral band,

with null sub-carrier and employing non-constant respectively, are used as guard band. The rest of 108

Table 1

Evaluation of computational complexity for each method

Ref. [6] 4:5N þ 2ðL0 Þ2 þ 3FFTN=2 þ invðL0 L0 Þ

Proposed 3N þ ð2L0 Þ2 þ 5FFTN

Constant modulus with N u aN Ref. [3] 3N þ ð2L0 Þ2 þ 5FFTN þ invð2L0 2L0 Þ

Ref. [6] 4:5N þ 2ðL0 Þ2 þ 3FFTN=2 þ invðL0 L0 Þ

Proposed 3N þ ð2L0 Þ2 þ 5FFTN

Non-constant modulus Ref. [3] 5N þ ð2L0 Þ2 þ 5FFTN þ invð2L0 2L0 Þ

Ref. [6] 6N þ 2ðL0 Þ2 þ 4FFTN=2 þ 2 invðL0 L0 Þ

Proposed 3N þ ð2L0 Þ2 þ 5FFTN

ARTICLE IN PRESS

1010 H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016

2 1

Channel response Channel response

Eb/No = 15 dB Estimated channel response Estimated channel response

1.5

fd = 200 Hz 0.5

Eb/No = 15 dB

fd = 200 Hz

Image {H11(l,10)}

1

Real {H11(l,10)}

0

0.5

−0.5

0

−1

−0.5

−1 −1.5

0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100

OFDM symbol index OFDM symbol index

Fig. 3. An example of channel frequency response H ij ½l; k and tracking at f d ¼ 200 Hz, k ¼ 10 and E b =N 0 ¼ 15 dB.

sub-carriers are used to transmit data. The OFDM Fig. 3 shows an example of the proposed channel

symbol rate is 10 Ksps, and the symbol period is tracking and the nonlinear time variant channel

100 ms, including the guard time of 20 ms. The frequency response H 11 ½l; k simulated at the

channel length L0 is assumed to be 18. Modulation given multi-path channel parameters, l ¼ ð1; 2; . . . ;

in sub-carriers is QPSK. The carrier frequency is 100Þ, k ¼ 10, and maximum Doppler frequency

2.4 GHz. The multi-path Rayleigh fading channel f d ¼ 200 Hz. In Fig. 3, solid line is the channel

assumes two rays with equal gain, and each ray has response tracked by the proposed method at

six multi-path delay taps. Each signal path is E b =N 0 ¼ 15 dB. Fig. 3 shows that the nonlinear

assumed to undergo an independent Rayleigh channel response is well tracked by the proposed

fading. The rms delay spread is 1:82 ms. We used method. The performance simulation results are

the Rayleigh fading channel simulator (Jake’s shown in Figs. 4–11. Fig. 4 shows the MSE of the

sinusoid sum method) openly published in proposed method at the conditions of M ¼ 4, 5 and 6,

Ref. [11]. The Doppler frequencies 40, 100 and and at the ﬁxed SNR of E b =N 0 ¼ 25 dB. The range of

200 Hz are used to represent different mobile the normalized Doppler frequency (f d T s ) is given

environments. After completion of channel estima- from 0 to 0:03. In the case of f d T s o0:02, the

tion by using the training signal, the system state is in performance of the proposed scheme (M ¼ 4) is the

data transmission mode. In data transmission mode, worst among all the investigated channel tracking

channel tracking for 20 consecutive OFDM symbols methods. The reason is that in decision directed mode,

is carried out continuously, using a decision directed the performance of the proposed channel response

method in which the demodulated data is used as the estimator is very highly affected by the prediction

reference data. The performance of the system is error caused by the ﬁrst channel prediction just after

measured by the estimator’s MSE and bit error rates two training OFDM symbols. The prediction error

(BER), each averaged over 100,000 OFDM blocks. causes the demodulation error which results in MSE

For unbiased comparison, no channel coding is used performance degradation and propagates into the

in this simulation. In order to track the time varying next channel estimation. As mentioned before, when

channel response, Kalman ﬁlter method may be used two training OFDM symbols are received, only two

for MIMO–OFDM systems as Komninakis did [14]. channel responses H ij ½1; k and H ij ½2; k are

However, the Komninakis method should calculate obtained, and H ij ½3; k; . . . ; H ij ½1 M; k are as-

the inverse matrix in every OFDM symbol to obtain sumed to be equal to H ij ½2; k. Therefore, from (10),

Kalman gain matrix and thereby the calculation the MSEs for the ﬁrst channel prediction are

amount increases signiﬁcantly. For this reason, we proportional to x½l; k / 1:8s2e when M ¼ 4, x½l; k /

compare the proposed method with Kalman ﬁlter s2e when M ¼ 5 and x½l; k / 0:68s2e when M ¼ 6,

estimator, using scalar Kalman ﬁlter in each sub- respectively. There is a larger difference in MSE

carrier [15]. between M ¼ 4 and 5 than between M ¼ 5 and 6.

ARTICLE IN PRESS

H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016 1011

10−1

Proposed(M = 4)

Proposed(M = 5)

Proposed(M = 6)

Li original

10−2

MSE

10−3

10−4

0 0.005 0.01 0.015 0.02 0.025 0.03

fd*Ts

101

Proposed CE

No predict

FDE Eb/No = 25 dB

100 Li original

10−1

MSE

10−2

10−3

10−4

0 0.005 0.01 0.015 0.02 0.025 0.03

fd*Ts

The noise effect in channel prediction can be increased sensitivity to channel time variations.

reduced by increasing M, as shown in Fig. 4. Note that MSEðM ¼ 6Þ is larger than MSEðM ¼

However, increasing M beyond a certain limit may 5Þ at f d T s 40:015. We can see that when M ¼ 5, the

result in performance degradation due to the proposed method shows the best performance in

ARTICLE IN PRESS

1012 H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016

102

Proposed CE

No predict

101 FDE

FED + Km filter

Li original

100

MSE

10−1

10−2

10−3

10−4

0 5 10 15 20 25 30

Eb/No [dB]

100

Proposed CE

No predict

FDE

10−1 FED + Km filter

Li original

Perfect CE

10−2

10−3

BER

10−4

10−5

10−6

10−7

0 5 10 15 20 25 30

Eb/No [dB]

Fig. 4. Hereafter, M is set to 5 in all the simulations 500 ms. Fig. 5 shows the MSE for each method as a

for performance evaluation. When M ¼ 5 and the function of f d T s at the conditions: E b =N 0 ¼ 25 and

OFDM symbol period is 100 ms, the time duration M ¼ 5. As we can see from Fig. 5, the MSE of the

for which a piecewise linear model is assumed is proposed method increases more slowly than the

ARTICLE IN PRESS

H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016 1013

103

Proposed CE

No predict

102 FDE

FED + Km filter

Li original

101

100

MSE

10−1

10−2

10−3

10−4

0 5 10 15 20 25 30

Eb/No [dB]

100

10−1

10−2

BER

10−3

10−4

Proposed CE

No predict

10−5 FDE

FED + Km filter

Li original

Perfect CE

10−6

0 5 10 15 20 25 30

Eb/No[dB]

other methods as f d T s increases. On the other hand, Figs. 6, 8 and 10 show MSE performances at

MSE of Li’s original method increases rapidly when Doppler frequencies of 40, 100 and 200 Hz, respec-

f d T s 40:015. tively. Figs. 7, 9 and 11 show BER performances at

ARTICLE IN PRESS

1014 H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016

102

Proposed CE

No predict

FDE

101 FED + Km filter

Li original

100

MSE

10−1

10−2

10−3

10−4

0 5 10 15 20 25 30

Eb/No [dB]

100

10−1

10−2

BER

10−3

10−4

Proposed CE

No predict

FDE

10−5 FED + Km filter

Li original

Perfect CE

10−6

0 5 10 15 20 25 30

Eb/No [dB]

Doppler frequencies of 40, 100 and 200 Hz, respec- these ﬁgures, ‘FDE’ denotes the frequency domain

tively. In these ﬁgures, BER performance curves for channel tracking method of Ref. [10]. In order to

perfect channel estimation are given to show the compare the effect of channel prediction, BER and

performance in the ideal channel estimation case. In MSE curves for the ‘no predict’ tracking method are

ARTICLE IN PRESS

H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016 1015

also drawn in these ﬁgures. ‘No predict’ means the 5 dB), the channel estimation error may diverge when

proposed method without the channel prediction tracking the channel response. On the other hand,

function. If ‘no prediction’ is used, the previously since the proposed method uses the known inverse

estimated channel value H ij ½l 1; k is used as matrix Q1 which is a constant time-invariant

the current channel value H ij ½l; k. In this case, the matrix, the effect of the erroneous demodulated data

channel estimation error is given by H ij ½l; k is much less signiﬁcant than that of Li’s original

H^ ij ½l 1; k ¼ Dij ½l; k þ Zij ½l; k. When the channel method. The simulation results show that as expected

prediction is used, the channel estimation error is the proposed method does not diverge.

given by DH ij ½l; k ¼ H ij ½l; k H ij ½l; k. The MSE

of ‘no prediction’ is given by E½ðH ij ½l; k 6. Conclusions

H^ ij ½l 1; kÞ2 ’ Var½Dij ½l; k þ s2e . The MSE of

‘prediction’ is E½jDH ij ½l; kj2 ’ 1:5s2e as we can see This paper proposed a novel simpliﬁed channel

from (10). Therefore, the performance of ‘predic- tracking method to reduce the computational

tion’ is better than that of ‘no prediction’ as long as complexity and improve the tracking performance

Var½Dij ½l; k40:5s2e is satisﬁed. Var½Dij ½l; k 0 at in time varying channel environments. In the

a low Doppler frequency ands2e is inversely propor- proposed method, a blind channel response pre-

tional to the E b =N 0 such that s2e 0 at a high dictor is designed to cope with the time variant

E b =N 0 . Therefore, the lower the Doppler frequency channel. The proposed channel tracking scheme

is in the wireless channel, the higher E b =N 0 is consists of a frequency domain estimation approach

required to satisfy the condition of Var½Dij ½l; k that is coupled with an MMSE time domain

40:5s2e . That is the reason why the performance of estimation method, and does not require any matrix

‘no prediction’ is better than that of ’prediction’ at inverse calculation during each OFDM symbol. By

f d ¼ 40 Hz. Note that in Fig. 6, the performance converting the MISO signal into a SISO signal and

gap in E b =N 0 between ‘no prediction’ and ‘predic- performing temporal channel estimation in the

tion’ is getting narrow with increase of E b =N 0 . If we frequency domain before beginning time domain

can know the information about E b =N 0 and the channel estimation, no matrix inversion is required

Doppler frequency, either ‘no prediction’ or ‘pre- anymore. The simulation results show that the

diction’ can be selected to improve the performance proposed method exhibits superior performance

of channel estimator, based on such information. than Li’s original method in time varying channel

At a Doppler frequency of 40 Hz, there is a little environments. At a Doppler frequency of 100 Hz

difference in the MSE and BER performance and BER of 104 , signal-to-noise power ratio

between the proposed method and Li’s original (E b =N 0 ) gains of about 2.5 dB are achieved relative

method. At f d ¼ 100 Hz and BER of 104 , the to Li’s original method. At a Doppler frequency of

performance improvement provided by the pro- 200 Hz, the performance difference between the

posed method is about 2.5 dB in E b =N 0 compared proposed method and conventional one becomes

with that of Li’s original method. However, at a much larger.

Doppler frequency of 200 Hz, the BER performance

difference becomes much larger when compared

Acknowledgment

with those of other methods. As we can see from

Figs. 9 and 11, due to the inter sub-carrier

This work was partially supported by research

interference (ICI) [16], BER performance at a

project 07-03 funded by Electronics and Telecom-

Doppler frequency of 200 Hz is worse than that of

munications Research Institute (ETRI) in Korea.

100 Hz.

From these ﬁgures, we can observe that the

channel estimation error of Li’s original method is References

very large at a given low E b =N 0 . At a given low

E b =N 0 , BER is high and the demodulated data [1] G.J. Foschini, Layered space–time architecture for wireless

X^ ½l; k is erroneous. Since Li’s original method communication in a fading environment using multiple

antennas, Bell Labs Tech. J. 1 (2) (Autumn 1996) 41–59.

calculates the inverse of a matrix made of erroneous

[2] J. Terry, J. Heiskala, OFDM Wireless LANs: A Theoretical

demodulated data X^ ½l; k, the channel estimation and Practical Guide, Sams Publishing, 2002

error is ampliﬁed by noise and the erroneous inverse [3] Y.G. Li, N. Seshadri, S. Ariyavisitakul, Channel estimation

matrix. In the worst case of low E b =N 0 (less than for OFDM systems with transmitter diversity in mobile

ARTICLE IN PRESS

1016 H.-G. Jeon, E. Serpedin / Signal Processing 88 (2008) 1002–1016

wireless channels, IEEE J. Sel. Areas Commun. 17 (March [10] Y. Chen, D. Jayalath, A. Thushara, Low complexity

1999) 461–471. decision directed channel tracking for MIMO WLAN

[4] Y.G. Li, Simpliﬁed channel estimation for OFDM systems system, in: ISISPCS, Hong Kong. December 2005,

with transmit antenna, IEEE Trans. Commun. 1 (January pp. 629–632.

2002) 67–75. [11] H. Harada, R. Prasad, Simulation and Software Radio for

[5] Y. Gong, K.B. Letaief, Low complexity channel estimation Mobile Communications, Artech House, 2002.

for space–time coded widehand OFDM systems, IEEE [12] Telecommunication Technology Association (TTA), Speci-

Trans. Wireless Commun. 2 (September 2003) 876–882. ﬁcation for 2.3 GHz Band Portable Internet Service—

[6] H. Minn, D.I. Kim, V.K. Bhargava, A reduced complexity Physical Layer, June 2004.

channel estimation for OFDM systems with Transmit [13] S. Haykin, Adaptive ﬁlter Theory, third ed., Prentice-Hall,

diversity in mobile wireless channels, IEEE Trans. Commun. Englewood Cliffs, NJ, 1996.

50 (5) (May 2002) 799–807. [14] C. Komninakis, C. Fragouli, A. Sayed, R. Wesel, Multi-

[7] H. Minn, N. Al-Dhahir, Optimal training signals for MIMO input multi-output fading channel tracking and equalization

OFDM channel estimation, Globecom 2004, November using Kalman estimation, IEEE Trans. Signal Process. 50 (5)

2004. (May 2002) 1065–1076.

[8] D. Schafhuber, G. Matz, F. Hlawatsch, Kalman tracking of [15] Z. Yuanjin, A novel channel estimation and tracking method

time-varying channels in wireless MIMO–OFDM systems, for wireless OFDM systems based on pilots and Kalman

in: Asilomar Conference on Signals, Systems and Computers ﬁlter, IEEE Trans. Consum. Electron. 49 (2) (May 2003)

2003, vol. 2 (November 2003) pp. 1261–1265. 275–283.

[9] K.A.G. Teo, S. Ohno, T. Hinamoto, Kalman channel [16] X. Cai, G.B. Giannakis, Bounding performance and

estimation based on oversampled polynomial model for suppressing intercarrier interference in wireless mobile

OFDM over doubly-selective channels, in: IEEE SPAWC OFDM, IEEE Trans. Commun. 51 (12) (Dec 2003)

2005, June 2005, pp. 116–120. 2047–2056.

- LTE Benefits - LTE Product DesignUploaded bychrismorley
- Recent Advance in Multi-Carrier Underwater Acoustic CommunicationsUploaded byijcsis
- Sajid Hussain MSc Thesis ReportUploaded byNikolay Gurin
- MIMO Wireless Communication SystemsUploaded bysujithkumar446
- 1.8 MIMO SystemsUploaded byAngel
- ROC ANALYSIS OF SPECTRUM SENSING TECHNIQUES IN COGNITIVE RADIO FOR OFDM SYSTEMUploaded byseventhsensegroup
- 256-Channel ADSL - MATLAB & Simulink ExampleUploaded byeric
- ADS-SV-NUploaded byprabhu
- A primer on 3GPP narrowband Internet of Things.pdfUploaded byrayed althubiti
- OFDMAUploaded byYasir Azmat
- Underwater Acoustic Communications and Networking.pdfUploaded bysurya
- Chapter 6Uploaded byMnl Kalyani
- adaptive_ofdmUploaded bySaroj Pradhan
- Chapter 1Uploaded byȘtefania Guran
- LTE OFDM Principle 32Uploaded byTharindu Wijegoonasekara
- 10.1.1.27.6413Uploaded byNageswara Rao Challa
- BER and SER Based Performance Analysis of BPSK and QPSKUploaded byASHISH K ADIGA
- 234[1]Uploaded byVijay Kumar Raju Vipparti
- IJETR031933Uploaded byerpublication
- Simulation of DVB Ch for Antena Diversity RxUploaded byStarLink1
- Chen-Energy-Efficient Resource Allocation_IEEE TVT_2014(1)Uploaded bypainkiller000
- ePMP Capacity Planner Guide R2.4.3.pdfUploaded bygustavito92012420945
- OFDMUploaded bytruvuong92
- Sca(Sincronizacion de Paquetes)Uploaded byFrancisco Tello
- Comparative_Study_of_Digital_Modulation (1).pdfUploaded byBindia Sahu
- 1.8 MIMO SystemsUploaded byCathyFer Valles Yanez
- ADSLTech1Uploaded byGaurav Bansal
- BBIC 6 Channel Estimation PptUploaded byYahia Selim
- Research TopicUploaded byReshma Chaturvedi
- A coordinated multi-point-based quality of service provision resource allocation scheme with inter-cell interference mitigation.pdfUploaded byShovon Pal

- Wireless Networked Security System Based on ZigbeeUploaded bykhajarasool_sk
- Wireless Device Identification in MIMO ChannelsUploaded bykhajarasool_sk
- Window Design for SISO and MIMO OFDMWindow Design for SISO and MIMO OFDM SystemsUploaded bykhajarasool_sk
- Video Ads Dissemination Through WiFi-Cellular Hybrid NetworksUploaded bykhajarasool_sk
- Vertical Handover Criteria and Algorithm in IEEE 802.11 and 802.16 Hybrid NetworksUploaded bykhajarasool_sk
- Utility-Based Adaptive Radio Resource Allocation InUploaded bykhajarasool_sk
- User Set Estimation for Adaptive Resource Allocation in a Multiuser OFDM SystemUploaded bykhajarasool_sk
- Uneven Pilot-Assisted Channel Estimation Method for Mobile OFDM SystemsUploaded bykhajarasool_sk
- Two-Layer Spreading CDMA an Improved MethodUploaded bykhajarasool_sk
- Turbo MIMO Equalization and Decoding in FastUploaded bykhajarasool_sk
- Turbo Coded OFDM for Reducing PAPR and Error RatesUploaded bykhajarasool_sk
- Training Symbol Based Coarse TimingUploaded bykhajarasool_sk
- Throughput Evaluation of the the Nonorthogonal CSKCDMA ALOHA With InterferenceUploaded bykhajarasool_sk
- Three-Dimensional Modeling and Simulation OfUploaded bykhajarasool_sk
- Theoretical Studies and Efficient Algorithm OfUploaded bykhajarasool_sk
- The Continuous-Time Peak-To-Average Power Ratio of OFDM Signals Using Complex Modulation SchemesUploaded bykhajarasool_sk
- Synchronization, Channel Estimation, AndUploaded bykhajarasool_sk
- Study of a Novel Zoom Spectrum Analysis Approach for Wireless Communication AnalyzerUploaded bykhajarasool_sk
- Stochastic Differential Equations ForUploaded bykhajarasool_sk
- Space Time Frequency Code ImplementationUploaded bykhajarasool_sk
- Space Shift Keying Modulation for MIMO ChannelsUploaded bykhajarasool_sk
- Smooth Beamforming for OFDMUploaded bykhajarasool_sk
- Single-Symbol ML Decoding for Orthogonal and Quasi-Orthogonal STBC in Clipped MIMO-OfDM Systems Using a Clipping Noise Model With Gaussian ApproximationUploaded bykhajarasool_sk
- Simple MMSE Interference Suppression for Real and Rate Half Complex Orthogonal Space Time Block CodesUploaded bykhajarasool_sk
- Sequence Hopping Cell Search Scheme for OFDM Cellular SystemsUploaded bykhajarasool_sk
- Selective Mapping With Symbol Re-Mapping for OFDM or TDM Using MMSE FDEUploaded bykhajarasool_sk
- Selective Level MappingUploaded bykhajarasool_sk
- IMP Selected Mapping Without Side Information for OFDM Selective ChannelsUploaded bykhajarasool_sk
- Robust Video Broadcasting Over 802.11 in Time Correlated Fading ChannelsUploaded bykhajarasool_sk

- 85fr_ref-coUploaded byRodolfo Tobias
- Characterization of Brighteners in Detergents by High-performance Liquid ChromatographyUploaded byAnastasiya Alexandrova
- BridgesUploaded bykisaki66
- nifty thrifty fiftyUploaded byapi-259179075
- Strongco IdlersUploaded byNicolas Fournier
- Oscillators Simplified - Delton T HornUploaded byrobertlivingstone
- bekcUploaded bypaancute8982
- selauUploaded byapi-295392817
- Tltp St. IvesUploaded byGilberto Ramos
- Hangzhou Fuchuan Electric Equipment Co.,Ltd_PPTUploaded byRahmat Nur Ilham
- Btm101 Syllabus 9-11Uploaded byOmesh KoosvanderMerwe
- board meeting reflectionUploaded byapi-240535342
- Topic7 Edu3108 PpgUploaded byGeena78
- Joint Press Release - IGAD-EU-ADA Sign New Grant Agreement on Peace and Security in the Horn - 23 March 2018Uploaded byAnonymous PLtI5ua
- Carrier Screw ChillerUploaded byShash Sha
- Piece by Piece.pdfUploaded bymirabella johnson
- BS-200EUploaded byASr Asr
- butyl_cellosolveUploaded byHatim Morbi
- 1756-En2T (EN2F) ControlLogix EtherNet-IP Bridge Module Release NotesUploaded byJoshua Patterson
- Nitrogen the Silent KillerUploaded byKyaw Kyaw Aung
- Sustainable Cotton Production of Cotton-BAYERUploaded byguru9anand
- 12.2 Ebusiness Suite - Collecting Fusion Middleware Log FilesUploaded byMohammed Abdul Muqeet
- Cooperatives Europe brochureUploaded byRay Collins
- D&D 3.5 - Deluxe Eberron Player Character SheetsUploaded byZachary Alan Ginsberg
- Power System Dynamic ModelingUploaded bySyed Muhammad Munavvar Hussain
- Cardiogenic Shock.pdfUploaded byAdreiTheTripleA
- 2012 Analysis Epri Txline Polymer Insulator, Fibreglass FailureUploaded byRylai Crestfall
- Advancing Hypoxic Training in Team Sports (1)Uploaded byKaren Ortiz
- 1956 ConstitutionUploaded byZamurrad Awan
- Fluid Velocities -RecommendationUploaded bygrabetty