Proceedings of the 2nd International Symposium on Computer, Communication, Control and Automation (ISCCCA-13)

An SC-FDMA Channel Estimation Algorithm Research Based on Pilot Signals

Xue Jianbin, Li Songbai

School of Computer and Communication
Lanzhou University of Technology
Lanzhou, China
xue_jabn@hotmail.com, rackby@163.com

AbstractSC-FDMA as LTE uplink main technology has been effect, which is reflected by the limited length Fourier series
widely studied,at the same time, channel estimation around approximation of Gibbs phenomenon, and shows that
such technology is becoming hot,In order to research and MMSE channel estimation can withstand minimum
increase channel estimation accuracy of LTE uplink system,an sideband recession;R-LS (Regularization-LS) algorithm and
algorithm based on pilot is proposed in this paper,this M-MMSE(Mismatch-MMSE) algorithm are respectively
algorithms main idea is,by setting a certain bit error proposed in literatures[6-7],the feasibility of theoretical
rate,choosing some proper data after channel estimation done analysis is verified shown in its simulation results.The
under R-LS and M-MMSE, combining the chosen data and algorithm proposed in this paper is based on block type pilot
former block pilot and increasing pilot number to do channel
and combines R-LS and M-MMSE,the simulation results
estimation again, system bit error rate and mean square error
results can be got by simulation;at last,the theoretical analysis
show that higher channel estimation precision can be
and simulation results show that this algorithm really can obtained through the limited time iterative estimation, which
make the channel estimation accuracy improved and bit error means that this algorithm is feasible.
rate decreased.
II. SYSTEM MODULE
Keywords-channel estimation; block pilot; bit error rate; SC- SC-FDMA based on OFDMA has a DFT module and an
FDMA;LTE IDFT module respectively in its sender and receiving
end.Signals after being modulated and coded are firstly
I. INTRODUCTION transformed into multipath parallel signals through serial-
parallel transformation,then go into subcarrier mapping after
OFDMA(Orthogonal Frequency Division Multiple N points DFT;multipath signals become a serial signal after
Access) is not suitable for LTE(Long Term Evolution) parallel-serial transformation and M points IFFT,then CP
uplink,because it asks for high power amplifier and (Cyclic Prefix) is added on the signal,finally,such serial
transmitting power from transmitting terminal,such signal is transmitted into wireless channel after DAC(Digital
requirements can hardly be met for a mobile equipment to Analogue Conversion) and RF,receiving end has a nearly
whose direction is smaller,smarter and long battery life.SC- opposite process.System block diagram is shown as Figure.1
FDMA avoids OFDMAs defects without changing OFDM With normal CP,the basic frame structure of SC-FDMA
transmission structure,whats more,it combines traditional in time domain is 10ms long and equally divided into 20
technologies of FDMA and single carrier transmission,and time slots,each slot is 0.5ms long and has 7 SC-FDMA
can let multiuser share communication resources.thats why symbols,in frequency domain,each slot has 12 sub-carriers
SC-FDMA has been chosen as LTE uplink main technology and each subcarrier bandwidth is 15kHz,pilot signal location
and widely used in wireless communication. is shown as Figure 2,block type pilot locates in every slots
With the development of SC-FDMA,channel estimation forth symbol.At anytime t,the received signals under
around such technology is becoming a research focus.LTE multipath can be presented as:
study group has no firm rule on channel estimation methods,
but has a rule that its uplink has massive pilot pattern.At y (t ) = x(t ) * h(t ) + w(t ) . (1)
present,commonly used pilot channel estimation algorithms
include LS(Least Square),MMSE (Minimum Mean Square here,* is convolution operation,h(t) is channel shock
Error),and channel estimation based on transformation response,w(t) is addictive noise, suppose the modulated
domain,etc.Literature[1] proposed a symbol decision signal is x=[x1,x2,, xN]T,after N points DFT:X=FNx,here,FN
algorithm for fast time-varying channel estimation,but its is DFT matrix whose element is:
accuracy is not good enough;literature[2] presented a
1 j 2 nm . (2)
combination channel estimation algorithm that has good ( F ) n ,m = exp( )
performance when SNR is between 7~15dB,but it decreases N N
when SNR is other value;literatures[3-4] had a research on here,n,m [0,N-1],N is the length of SC-FDMA symbol,the
LS, MMSE and decision feedback algorithm,the simulation received signal is
results show that MMSE has a better performance;
TNMSE(Truncated Normalized Mean Squared Error)
algorithm proposed in literature[5] overcomes sideband

Published by Atlantis Press, Paris, France.

the authors, 2013
0179
 Proceedings of the 2nd International Symposium on Computer, Communication, Control and Automation (ISCCCA-13)

y = xh + w . (3) 0
C = L ZC ( M L ZC )
I L ZC . (11)
CP turns linear convolution into circular convolution, and IM 0 ( L + M ) M
ZC

FFT changes circular convolution into frequency domains

I is unit matrix,then keep going as Figure 1,SC-FDMA
product,represented as
signals are produced and transmitted into wireless channel.
Y = XH + W . (4) III. THE ALGORITHMS BASIC IDEA
here,Y is received signal,X is transmitted signal, H is After a range of treatments,SC-FDMA symbols arrive at
frequency domain channel response,W is Gaussian white the receiving end,some symbols that have error bits and are
noise.The structure of SC-FDMA pilot is block type pilot under a certain threshold will be extracted and put into
that can be produced by ZC[8-9] serial,a Lzc long ZC serial qb channel estimator with pilot for quadratic estimate,if
is shown as: necessary,more estimates can be done till achieving desired
results. block digram is:
s 2 2 + sd
exp( j 2 b ), Lzc = 2a
Lzc
qb (s) = . (5)
exp( j 2b s ( s + 1) 2 + sd
), Lzc = 2a + 1
Lzc
here,a is natural number,b {1,2,,LZC-1} is ZC serial radical Fig 3. Algorithm block diagram
exponent,s=0,1, ,LZC-1,d is any integer,for simplicity, In Figure.3,channel response H1 comes out from channel
suppose d=0[10],then signal with pilot is: estimator1,we can get the first channel estimated and

equalized result X 1 by putting H1 and SC-FDMA symbols
X = [ X 1 , X 2 , , X N ]T . (6) in equalizer1if channel estimation accuracy or BER meets
the requirements, keep moving as Figure 1,if not,set a
after subcarrier mapping: certain BER,extract some reliable signals from the first
estimation result and put it in channel estimator2 with
X = AX . (7) pilot,then, get the secondary channel estimation and

here,A is zero padding matrix: equalization results X 2 , more iterative computations can be
made if necessary,the threshold BER comes from the
1 0 0 0 relationship of BER and SNR[11]shown as formula(12)-(16) :
0vN
1
0 BER = [1 ]. (12)
1 0 0 2 2 2
. (8)
A= 0vN here,=SNR/(1+SNR),Suppose Xn is transmitted signal,and

Yn is received signal,Hn is channel response in frequency
0 0 0 1 domain,Wn is Gaussian white noise,the signal received on
the nth subcarrier must be:
0( M ( v +1) N )N M N
Yn = X n H n + W n . (13)
here,v {0,1,,|M/N|} and M>N,There are two subcarrier
mapping ways of SC-FDMA: localized FDMA and after channel estimation,the signal and SNR are:
distributed FDMA,their main difference is zero padding
among sub-carriers, when inserting v 0 among each DFT X n = Yn H n = X n + W n H n . (14)
sampling output signals,its localized FDMA if v=0, and
each users bandwidth is 1/U of system bandwidth;its 2
distributed FDMA if v>0,each users bandwidth is expanded SNR = 2
n ( N2 n H n ) = H ( n ) SNR x . (15)
X n
as long as system bandwidth;this paper adopts the former.
Then,after M points IFFT,we get Obviously,SNR is function of channel grain H(n),if
2
SNRthreshold is determined, and H (n) SNRx SNRthreshold ,
X = F MH X . (9)
what means
H
here, F is IFFT matrix,then,multiplying a stack factor C . (16)
M H (n) H threshold = SNR threshold / SNR x
that means inserting CP to eliminate ISI(Inter Symbol
Interference), then signal with CP is: then,the threshold Hthreshold is determined,so the chosen
~ symbol is proper as long as its grain is higher than this
X = C x. (10) threshold.

Published by Atlantis Press, Paris, France.

the authors, 2013
0180
 Proceedings of the 2nd International Symposium on Computer, Communication, Control and Automation (ISCCCA-13)

IV. PERFORMANCE SIMULATION AND ANALYSIS

MSE M MMSE = E [|| H M MMSE H || 2 ] . (24)
A. Theoretical Analysis simplified as:
~
For LS algorithm,suppose the pilot symbol is g, H is
g MSE M MMSE = R HH R HH (1 +
( R HH ) 1 ) 1 . (25)
g 1Gaussian white noise vector,and its variance is
SNR

2 ,then,the channel estimation H g is

~
here,RHH=E{HHH}is autocorrelation matrix of channel
H g response,then,the average SNR is:
~ ~ 2
H g = H g + H g = Fgh + H g . (17) SNR = E X ( n ) 2
g
. (26)

here,g ( 0, , G ) is useful pilot,h is L 1 channel shock and:

response,L is channel maximum delay length,and LLcp,Fg is 2 2
. (27)
= E X (g ) E1 X (g)
G L matrix,FL is N L matrix which has the lines where 12 12

pilot locates and the former columns of N N DFT here, is a constant that is dependent on modulation,=1
matrix, H G is G 1 Gaussian white noise vector,then the
~
when the modulation is QPSK.
whole channel shock response is:
B. Simulation Results

H LS = FL (F G
H
FG ) 1
F G
H
H G . (18) Simulation in this paper is based on LTE uplink SC-
FDMA,for simplicity,U=2,carrier bandwidth is 5MHz,FFT
as the reference literature [6] shows that parameters chosen sampling points M=512,Lcp=M/8,Doppler frequency shift is
in LTE system makes LS algorithm can not be directly 50Hz, subcarrier number is 300,subcarrier mapping is
applied to SC-FDMA,But classical and robust R-LS localized FDMA,modulation is QPSK,bit error rate
algorithm can.Here,expression ( F GH F G ) 1 is an ill- threshold is 0.005,twice iterative computations,1000 times
conditioned matrix because there are some unmodulated of Monte Carlo simulation on each SNR node,channel is
sub-carriers in SC-FDMA system,this problem can be typical city macro cell model,simulation results are shown as
resolved by adding a normalized matrix IL to avoid Figure 4 and Figure 5.
inversing matrix,here is a constant,usually,it is 0~1,the Figure 4 shows the MSE performance curves of two
system has best overall performance when =1[12],so the kinds of estimation methods,under the condition that channel
estimation performance can be improved in given SNR diameter number is known or estimation is accurate and
range, channel response in frequency domain is: exists error,whether M-MMSE or R-LS, performance of
second estimation result is better than that of first
H

R LS = F L ( F GH F G + I L ) 1 F GH H

G . (19) estimation,and M-MMSE performance has about 2~3 dB
gain than R-LS algorithm,each estimation method has no
its MSE(Mean Square Error)is: error basement,estimation performance gains of both have
obvious rising trend when SNR growing,but that trend slows

MSE = E [|| H R LS H || 2 ] . (20) down when SNR is higher.
R LS
Figure 5 shows system BER curves of two kinds of
simplified as: estimation methods,when SNR is low,estimation accuracy
can not be improved because the signals chosen from
MSE R LS = 2
~ F L ( F GH F G + I L ) 1 F GH . (21) estimator1 are not enough,but estimation precision rises
H g
when the selected symbol number and SNR increasing,
For M-MMSE algorithm,its MMSE estimation result obviously,result from estimator2 is better than that from
is estimator1, and estimator2 of R-LS is 5dB better than
estimator1,such two algorithms nearly have error basement

H MMSE = F L ( F GH F G + 2
C 1
) 1 F GH H

G
. (22) when SNR is close to 18dB,grains of such two algorithms
h
have almost linear change and have no obvious rise when
~
H G

SNR is higher,
here,Ch=E[hhH],MMSE is statistical estimation,which needs
second order statistics(power delay and noise variance).To V. CONCLUSION
avoid second order statistical estimation and real-time
inversing matrix that is for MMSE direct application, Each user in SC-FDMA system occupies only part band
suppose channel power delay spectrum is evenly distributed, and sub-channel has time-frequency related properties,an
then,Ch has identity matrix structure, expression (27)can be SC-FDMA channel estimation based on pilot is proposed,
improved as: from the above theoretical analysis and simulation results,
such channel estimation algorithm is feasible in LTE uplink
H

M MMSE = F L [ F GH F G + ( 2
G2 ) I L ] 1 F GH H

G
. (23) system,because it not only improves channel estimation
precision,but also reduces system BER.The only defect is
~
H G

that such algorithm becomes complex when making iterative

and,its MSE is:

Published by Atlantis Press, Paris, France.

the authors, 2013
0181
 Proceedings of the 2nd International Symposium on Computer, Communication, Control and Automation (ISCCCA-13)

computations,the actual hardware design feasibility and [6] A.Ancora,C.Bona and Slock D.T.M.,Down-Sampled Impulse
economy efficiency may not be obvious,so,how to improve Response Least-Squares Channel Estimation for LTE OFDMAin
Proc.IEEE International Conference on Acoustics, Speech and Signal
estimation precision and reduce the algorithm complexity at Processing (ICASSP2007) (Hawaii,USA),2007(4).
the same time is still a problem that is worth studying. [7] P.Hoher,S.Kaiser and P.Robertson,Pilot--Symbol-aided Channel
Estimation Methods in Time and Frequencyin Proc.Communication
ACKNOWLEDGMENT Theory Mini-Conf.(CTMC)within IEEE Global Telecommun. Conf.
This research is financially supported by the National (Globecom1997) (Phoenix, USA), 1997 (7).
Natural Science Fund Project of China(No.61062002) and [8] J.D.C.Chu,Polyphase Codes with Good Periodic Correlation
Properties.IEEE Trans.on Information Theory, Vol.18, p.531-
the Master Student Tutor Fund Education Department of 532,1972(7)
Gansu Province (No.1014ZTC109). [9] R.Frank,S.Zadoff and R.Heimiller,Phase Shift pulse Codes With
Good Periodic Correlation Properties. IEEE Trans.on Information
REFERENCES Theroy, Vol.8,p.381-382,1962(10).
[10] Stefania Sesia,Issam Toufik,Matthew Baker. LTE-The UMTS Long
[1] Bao Guan,Xiaojun Wang,Xiaowen Zhang.etc.A new decision Term Evolution From Theory to Practice[M].Wiley Publishing,Inc.
feedback detection algorithm of LTE uplink.The Airforce Radars 2011.p.122-163.
Journal,2010.24(4). [11] J G.PROAKIS,Digital communication. New York:5thed. [S.l.]: Mc
[2] Chaowu Zhan.An OFDM Channel estimation Algorithm Based on Graw Hill,2009.
Pilot Signals.TV technology.2010,34 (11). [12] Jan-Japp van de Beek,Ove Edfors,Magnus Sandell, et. On Channel
[3] RANA M M.Channel estimation techniques and LTE.terminal Estimation in OFDM.IEEE.1995.
implementation challenges //Proc. 13th Inter national Conference on [13] 3GPP TS 36.211,3rd generation partnership project; technical
Computer and Information Technology.Dhaka:International specification group radic access network; evolved universal
Conference Press, 2010,p.545-549. terrestrial radio access(E-UTRA); physical channel and modulation
[4] HUANG G,NIX A,ARMOUR S.Decision feedback equalization in (release 9) [S]. 2010.
SC-FDMA//Proc.IEEE19th International Symposium on Penonal,
Indoor ad Mobile Radio Communication. [S.l.]:IEEE Press,2008:1-5.
[5] J.W.Gibbs,Fourier Series.Nature,Uol.59,p.200,1898.

Fig 2. SC-FDMA pilot signal location(include normal CP)

Fig 1. SC-FDMA transmit-receive system(include pilot)

Fig 4.MSE performance curves of two kinds of estimation methods Fig 5.System BER curves of two kinds of estimation methods

Published by Atlantis Press, Paris, France.

the authors, 2013
0182