Performance Comparison of Some Codes in Code

Division Multiplexing based MIMO Channel

Sounder ArchitectureMinjae Kim, Sunghyun Kim, and Hyuckjae Lee

School of EngineeringInformation and Communications University119 Munjiro Yuseong-gu Daejeon, Korea

Email: {mj kim, boofunky, hjlee}licu.ac.kr

Abstract- Code division multiplexing(CDM) based MIMOchannel sounder architecture is excellent at real-time measure-ment. In this paper, Loosely Synchronous(LS), Kasami andChaotic sequences are suggested as candidates for the code thatwill be used in CDM architecture. Among those sequences, then,the best suitable sequence for channel sounding is examinedthrough simulation. And by proposing the combined scheme ofCDM and TDM, when the number of transmit antennas aremore than 4, the decrease of dynamic range due to the highcross-correlation between sequences can be mitigated .

Keywords-MIMO channel sounder, Loosely Synchronous se-quence, Kasami sequence, Chaotic sequence, SCME

I. INTRODUCTION

It is anticipated that MIMO wireless access technologywill play an important role in increasing spectrum efficiencyand link reliability in the next generation wireless commu-nication systems.[1] Because MIMO wireless channel hasspatially directional property as well as temporal propertyunlike SISO channel, the performance of MIMO system islargely susceptible to MIMO channel condition. Therefore anaccurate knowledge about MIMO channel is very importantin developing MIMO communication systems and MIMOchannel field sounding is required for accurate understandingof the channel environment. Channel measurement data isalso used for MIMO channel modeling and evaluating MIMOsystems.

Conventional MIMO channel sounder is based on timedivision multiplexing(TDM) architecture that sounding signalsfrom each transmit antenna are transmitted utilizing syn-chronized switch sequentially, not concurrently and uses PNsequence as a sounding signal as shown in Figure 1.[2,3]TDM architecture has a merit of cost-effective in hardwareimplementation, but this scheme is not suitable for real-timechannel measurement and has some drawbacks such as therequirement of precise synchronization between transmitterand receiver and accuracy reduction during switching time aswell. As a result, reformation about architecture of MIMOchannel sounder is needed for overcoming these limitations ofconventional TDM based MIMO channel sounder.

Hyunbeom Lee and Heung-Ryeol YouInfra Laboratory

Next Generation Mobile Research Department, KT17 Woomyeon-dong, Seocho-gu, Seoul 135-792, Korea

Email: {hlee, hryou}@kt.co.kr

In this paper, as a solution of those limitations, code divisionmultiplexing(CDM) based MIMO channel sounding architec-ture is considered.[4] CDM architecture that sounding signalsfrom all transmit antennas are transmitted simultaneouslyenables to measure the MIMO channel in real time. In thisCDM architecture, the level of interference among differentcodes has a dominant effect on the performance of channelmeasurement because this scheme multiplexes transmit signalsby using codes. Therefore we need a code which has very lowautocorrelation and cross-correlation values between differentcode sets for this CDM architecture.We propose Loosely Synchronous(LS), Kasami and Chaotic

sequence which are known for having a very low-correlationproperty among sequence sets to be used as a pilot sequencefor CDM architecture. After introducing the generation princi-ple and properties of those sequences, the effects that each can-didate sequence has on channel measurements are comparedthrough simulation. In simulation, 3GPP/WINNER SCME isused and RMS(root mean square) delay spread error andangular spread error from transmitting two or four codes si-multaneously using each candidate sequence are computed andrepresented as cumulative distribution function(CDF) graph.And in the situation that the number of transmit antennas isrequired as more than 4, by combining CDM with TDM wecan mitigate a major drawback of CDM architecture whichis the decrease of dynamic range due to the increased cross-correlation between codes.

This paper is organized as follows. It starts with MIMOchannel sounder architecture in section II. In section III, threecandidate sounding sequences are introduced. In section IV,

PN ~ RF module Slidingsequence correlation

MT MR PN

t Rubidium reference synchronization f sequence

ChannelData

Processing

Fig. 1. TDM based MIMO channel sounder architecture

Feb. 17-20, 2008 ICACT 2008ISBN 978-89-5519-136-3 -1343-

Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on November 26, 2009 at 20:27 from IEEE Xplore. Restrictions apply.

Code I RF module I

Code 2 RF module 2 2

Code MT RF module MT

The same sequence:............................................................................................

Lcorrelatin

2 Sliding Channelco rrelation Data

Processing

Codegenerator

Fig. 2. CDM based MIMO channel sounder architecture

Co Cl Co C1-'So Si So -Si1f] 1\MCo CiSro ae pa ctc odse Cl Sa gn-So Sdpar cd

[Co -C1 SO ola C1 Co Cc SO-Sp SoSnSCo Sol [Co -C1 -Co - SO -Si -So -Sij v

IC, SI I SCI Co Cl -C SI So SI -Sol e

SCI Co Si So [Ci Co -C1 Co Si So -Si So

Ic, -CO SI -SoI SCi Co Cl Co Si -So Si So

*Ci, Si t-,1 [CI -o -C1 -C Si -So -Si -Soj

[i ]Mate pair [ii Same-parents codesets iii ]Same-grandparents codesets

(A) Golay complement pair generation

the simulation environments and results are presented andhybrid architecture combining CDM with TDM is proposed.Finally the conclusion is discussed in section V.

II. MIMO CHANNEL SOUNDER ARCHITECTUREIn MIMO channel, the information about spatial characteris-

tics of the channel is considered importantly and is representedas main factors in MIMO channel impulse response as follows.

All zeros

' O.5 T.2

1o

2! °

(a) circular autocorrela

E IFW zone

z 12( 00 -500 0Code shift

(c) circular cross-correlation=0 0.5

8

*0.

° -100 40005Oz Code shift

H(a,T, fR, SOT)L

E1 ai6(T-Ti)6(9OR - OR,i)6(9OT - OT,i)i=l

(1)where ai, Ti, OR,i and OT,i is the amplitude, delay, directionof arrival and direction of departure in i-th multipath channeland L is the number of multipath. Because these MIMOchannel parameters are obtained from channel sounding, moreprecise channel parameters can be anticipated through theimproved sounding technique. As shown at previous section,because TDM based MIMO channel sounder architecture haslimitations to make it more advanced for the next wirelesssystem, CDM sounding architecture is represented as analternative method and shown in Figure 2. In this sounder,removal of switch in transmitter enables to measure the spatial-temporal channel in real-time and it doesn't need to synchro-nize between transmitter and receiver. But considering that thecorrelation between transmitting signals influence on accuracyof this CDM based sounder absolutely, the codes applicablefor CDM architecture have to be selected above all. In thefollowing section, Loosely Synchronous sequence, Kasamisequence and Chaotic sequence are chosen as candidate codesfor CDM architecture and their generation and correlationproperties are examined.

III. PROPOSED SOUNDING SEQUENCESA. Loosely Synchronous Sequence

Loosely Synchronous(LS) sequence can be applied forCDM based channel sounder by its perfect zero-correlationproperty in a specific situation. Composed of ternary, LSsequence is generated by inserting all zero sequences betweenGolay complementary pair C and S which are binary sequenceas shown in Figure 3-(A),(B). Its correlation property isprovided by following equation.

L-1

Rjk(T) = E Cj,iCk,(i+T) mod Li=l

{ L,0,0,

T=O,j= kTO , j#Ak

0 < |T| < TIFW(2)

X (D All zeros S

(B) Loosely Synchronous sequence

ation (b) circular cross-correlation.0 0.5

~0

o

2 K°

500 1000 o _1 oo -500 0z Code shift

[i il pair ) (d) circular cross-correlation.5ol

.m

E42

500 10~00 o -1000 -500 0Code shift

(C) Correlation propoerties of LS sequence

Fig. 3. Loosely Synchronous sequence

where Cm,n is n-th symbol in m-th one in LS sequence sets, Tis the delay shift, and L is the length of the sequence. From (2),LS sequence has zero-correlation within specific zone and thiszone is called interference free window(IFW) zone(= 2TIFW).If any delayed sequence itself or other sequences among LSsequence sets are within this zone, we can perfectly knowour desired sequence without any interference from othersequences. Inside IFW zone, the system is analyzed as severalsingle-input single-output (SISO) components. However, thewidth of IFW zone depends on the number of sequencestransmitted concurrently. In case of using a mate pair of LSsequences, the length of TIFW is L-1 where 2L is the lengthof a complementary pair and the length of zeros is L-1 to getoptimum IFW zone (the total sequence length is 4L-2)[5]. Fordifferent pairs with the same parents codes, the achievable IFWzone is the half of the case using the mate pair. With the samegrandparents codes, the quarter of the case using the matepair is obtained through the different pairs with the differentparents sets as IFW zone. In other words, if we use twice ofthe sequences, we get half of IFW zone as shown in Figure3-(C). In this reason, LS sequence is limited to measure thechannel which has many transmit antennas with longer delays.In later section, we will suggest another scheme to overcomethis drawback.

B. Kasami SequenceKasami sequence is another candidate code known for its

low-correlation property. The correlation between differentKasami sequences or its delayed sequences have only threevalues, {-1,- s(n), s(n) -2} which are very low valuesrelative to the sequence length when the length of the sequence

Feb. 17-20, 2008 ICACT 2008

500 1000

[Ilid pair)

500 1000

ISBN 978-89-5519-136-3 -1 344-

Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on November 26, 2009 at 20:27 from IEEE Xplore. Restrictions apply.

is 2'_ 1, where s(n) = 2n/2 + 1 and even number nis the number of stages.[6] Even though Kasami sequencehas very low correlation values, it also has a dynamic rangelimitation in case of applying for channel sounder because theinterferences are increased in proportion to the increase of thenumber of transmitting sequences. And this sequence has itsown drawback that the selection of the length is limited largelyfrom the length equation 2n_ 1 where n is even number.

C. Chaotic SequenceChaotic sequence is a deterministic sequence which is

generated sequentially by using a mapping function Xn+1 =

f (Xn) and an initial value xo but whose distribution lookslike white noise.[7] Mainly used maps are Logistic map, Cubicmap, Skew tent map, Henon map and the best well-known mapis Logistic map which is expressed as Xk+1 = Xk( -Xk).By choosing 3.5699456... < , < 4, aperiodic and randomsequence is sequentially generated. The distribution of Chaoticsequences is like white noise because of randomness of se-quence generation and this property makes almost uncorrelatedbetween different sequences. This sequence has some meritsthat knowing only two information that are a mapping functionand an initial value, the same sequence can be regenerated andthe selection of the sequence length is not restricted.

IV. SIMULATION AND PROPOSED SCHEME

A. Simulation EnvironmentsIn this section, the effects that each candidate sequence

has on channel measurements are analyzed and comparedthrough simulation. With 3GPP/WINNER spatial channelmodel(SCME)[8,9] as MIMO channel model, root meansquare(RMS) delay spreads and angular spreads are calculatedfrom the received signals when each sequence is transmitted.The number of transmitting sequence is 2 and 4, and througheach simulation in urban macro(delay spreads: 650 ns) andsuburban macro (delay spreads: 170 ns) scenario, we alsowant to know the effects of maximum delays. These scenariosare typical channel environment for mobile communications.

The specific simulation environments are as follows. Cen-ter frequency is 5.3 GHz, the bandwidth of the channel is1OOMHz(chip duration is lOns), mobility is 60 km/h as SCMEinput parameters. Receive antennas are 8-element uniformlinear array, the chip length of the codes are 4094(LS) and4095(Kasami, Chaotic), and EC/No is fixed to 40 dB forfocusing on interference among sequences. Delay spread isobtained from power delay profile with sliding correlationand angular spread is calculated utilizing Unitary ESPRIT[10]algorithm. The calculation of RMS delay spread Trms andRMS angular spread 0rms are as follows.

|(Tmean)2 Pg (T)dT f0Coo TPg (T)dTTrms TTmean 0f '. Pg(T)dTf ' Pg(T)dT

f (0- Omean)2Pg (0) dOorms V Pg(O)d mean

(3)f7_ OPg(0)dOfyT Pg(O)do

(4)

0.9

0.8

0.7

0.6LLO 0.5C-)

0.4

0.3

0.2

0.1

0

Fig. 4.

1e

0.9

0.6

Co05C-)

0.4

0.3

0.2

- a - LS 2-Tx

A ---i LS 4-Tx- 4 - Kasami 2-Tx

Kasami 4-Tx- e -Chaotic 2-Tx--= Chaotic 4-Tx

~~i50 100 150 200 250 300 350 400

RMS Delay Spread Error [ns]

CDF of RMS delay spread error in Urban macro

LS2-TxLS4-Tx

4 - Kasami 2-TxKasami 4-Tx

e - Chaotic 2-TxChaotic 4-Tx

50 100 150 200 250 300 350 400RMS Delay Spread Error [ns]

Fig. 5. CDF of RMS delay spread error in Suburban macro

RMS delay or angular spread error is calculated from theabsolute difference between RMS delay or angular spread ofmeasured channel and SCME, and represented as the form ofCDF. From the CDF graph we can compare RMS delay andangular spread error performance among the codes.

B. Simulation ResultsFigure 4,5 are the CDF of RMS delay spread error when

transmitting candidate sequences in urban macro and suburbanmacro environments. From the results, we can know thatirrespective of the number of transmit antennas and scenarios,channel measurement using LS sequence is more accuratethan that of Kasami and Chaotic sequence because correlationproperties within IFW zone of LS sequence are perfect. Andthe case of Kasami sequence has smaller RMS delay spreaderror than that of Chaotic sequence by and large. From Figure6,7, results of comparison with regard to RMS angular spreaderror represent the same results as the case of RMS delayspread error. And from both simulation results, as the numberof Tx antennas changes from 2 to 4, we can make certain thatthe performance of channel measurement is reduced. However,when the number of Tx antennas are more than about 4 , itis estimated that the use of LS sequence will be impracticaldue to the narrow IFW zone and the amount of interferencesin Kasami and Chaotic sequence also will have serious effectson accuracy of the measurement.

Feb. 17-20, 2008 ICACT 2008ISBN 978-89-5519-136-3 -1345-

Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on November 26, 2009 at 20:27 from IEEE Xplore. Restrictions apply.

0.9

0.8 C

0.7

0.6LLqCD 0.5O 05C-)

0.4

0.3

02

0.1

0

- LS 2-Tx- LS 4-Tx- Kasami 2-Tx- Kasami 4-Tx- Chaotic 2-Tx- Chaotic 4-Tx

0.5 1 1.5 2RMS Angular Spread Error [degree]

2.5 3

Fig. 6. CDF of RMS angular spread error in Urban macro

o00

C)0.5 l

02

001

0 0.5

- LS 2-Tx-LS 4-Tx- Kasami 2-Tx- Kasami 4-Tx- Chaotic 2-Tx- Chaotic 4-Tx

1 1.5 2RMS Angular Spread Error [degree]

2.5 3

Fig. 7. CDF of RMS angular spread error in Suburban macro

some candidate sequences applicable for CDM architectureand compared performance of channel measurement. As aresult of the comparison, LS sequence can be more preferredthan Kasami and Chaotic sequence in 2 to 4 transmit antennas.And by proposing hybrid architecture of CDM and TDMwhen the number of transmit antennas is more than 4, we canmitigate interference problem between different sequences andhardware cost problems.

REFERENCES

[1] J. Kivinen, T. 0. Korhonen, P. AiKio, R. Gruber, P. Vainikainen, S.G.Haggman, Wideband radio channel measurement system at 2GHz, IEEETrans. on instrum. and measur., vol. 48, Issue 1, pp.39-44, Feb. 1999.

[2] MEDAV RUSK MIMO Channel Sounder Manual.[3] Elektrobit PROPsound MIMO Channel Sounder Manual.[4] K. SAKAGUCHI, J. TAKADA, K. ARAKI, "A Novel

Architecture for MIMO Spatio-Temporal Channel Sounder,"IEICE Transactions on Electron, vol.E85-C, No.3, pp.436-431, March2002.

[5] S. Stanczak, H. Boche, M. Haardt, Are LAS-codes a miracle?, IEEEGLOBECOMO0, vol. 1, pp.589-593, Nov. 2001.

[6] A. M. D. Turkmani, U. S. Goni, Performance evaluation of maximal-length, gold and Kasami codes as spreading sequences in CDMA systems,ICUPC.1993, vol.2, pp.970-974, Oct. 1993.

[7] G. Heidari-Bateni and C. D. McGillem, Chaotic sequences for spreadspectrum: An alternative to PN-sequences, Proc. IEEE ICWC-92, 437-440,

[8] "Spatial channel model for multiple input multiple output (MIMO)simulations", 3GPP TR25.996 V6.1.0, Sep, 2003.

[9] D. S. Baum, J. Salo, G. Del Galdo, M. Milojevic, P. Kyosti, and J.Hansen, "An in-terim channel model for beyond-3G systems," in Proc.IEEE VTC'05, Stockholm, Sweden, May 2005.

[10] Martin Haardt and Josef A. Nossek, "Unitary ESPRIT: How to ObtainIncreased Estimation Accuracy with a Reduced Computational Burden,"IEEE Trans. on signal processing, vol. 43, No. 5, pp.1232-1242, May1995.

C. Proposed architectureFigure 8 represents hybrid architecture between CDM and

TDM. When the number of Tx antennas in CDM architecturemore than 4, we propose to use this hybrid architecture.Although the main drawback of TDM such as synchronizationstill remains, this hybrid scheme enables real-time measure-ment by reducing measurement time in proportion to thenumber of concurrent transmit codes and hardware cost whichwas a drawback in CDM can be adjusted by selecting thenumber of transmit codes as 2 to 4.

V. CONCLUSIONSBy using CDM based MIMO channel sounder, we can over-

come some major drawbacks of the conventional TDM-basedMIMO channel sounder scheme. In this paper we examined

Transmit signals sequentiallyin each RF module

I-iYM ~ ~~~~~~correlation-MTIk X 2 Sliding

* correlation ChannelData

Processing

Co ek RFmodulek Slidingcorrelation

I-MTI/kCode

generator

Fig. 8. Hybrid architecture of CDM and TDM

Feb. 17-20, 2008 ICACT 2008ISBN 978-89-5519-136-3 -1 346-

Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on November 26, 2009 at 20:27 from IEEE Xplore. Restrictions apply.