Spatial Modulation - A New Low Complexity
Spectral Efficiency Enhancing Technique
R. Mesleh, and H. Haas Chang Wook Ahn, and Sangboh Yun
School of Engineering and Science C & N Lab., Samsung AIT
International University Bremen 11-4, Nongseo-ri, Youngln-si
28759 Bremen, Germany Kyeonggi-Do, KOREA
{[Link] & [Link]}@[Link] [Link]
Abstract- The multiplexing gain of multiple antenna transmis- efficiencies of 20 -40 bps/Hz can be achieved in an indoor
sion strongly depends on transmit and receive antenna spacing, rich scattering propagation environment assuming a practical
transmit antenna synchronization, and the algorithm used to SNR range, and bit-error performance respectively [7]. The
eliminate interchannel interference (ICI) at the receiver. In this high spectral efficiencies stem from parallel signal transmis-
paper, a new transmission approach, called spatial modulation,
that entirely avoids ICI and requires no synchronization between sion resulting in a multiplexing gain. Other multiple antenna
the transmitting antennas while maintaining high spectral effi- techniques employ, for example, space-time coding (STC)
ciency is presented. A block of information bits is mapped into a techniques in order to increase the SNR at the receiver by
constellation point in the signal and the spatial domain, i.e. into exploitation of spatial diversity [8]. In this case the system
the location of a particular antenna. The receiver estimates the improvement results from a diversity gain.
transmitted signal and the transmit antenna number and uses An alternative transmission approach that entirely avoids
the two information to de-map the block of information bits. ICI at the receiver input was proposed in [9], [10] for
For this purpose, a novel transmit antenna number detection
algorithm called iterative-maximum ratio combining (i-MRC) is BPSK and QPSK transmission respectively. The idea is to
presented. Spatial modulation is used to transmit different compress a block of Nt symbols into a single symbol prior
number of information bits and i-MRC is used to estimate both to transmission, where Nt indicates the number of transmit
the transmitted signal and the transmit antenna number. The antennas. Information is retained by an algorithm that maps
results are compared to ideal V-BLAST (Vertical-Bell Lab Lay- this single symbol to one, and only one, of the Nt antennas.
ered Space-Time) and to MRC. Spatial modulation outperforms The task of the receiver is twofold: First, to estimate the
MRC. The (bit-error-ratio) BER performance and the achieved single symbol, and second, to detect the respective antenna
spectral efficiency is comparable to V-BLAST. However, spatial
modulation results in a vast reduction in receiver complexity. number from which the symbol is transmitted. By combining
both information and by carrying out the inverse encoding
I. INTRODUCTION operation, the receiver is able to retrieve the entire block of
Nt symbols. Thereby, multiplexing gain is achieved, but ICI
The need to improve the spectral efficiency and reliability in the MIMO transmission is completely avoided. However,
of radio communication is driven by the ever increasing the scheme in [10] suffers from a loss in spectral efficiency
requirement for higher data rates and improved QoS (Quality as compared to V-BLAST due to the use of parity symbols.
of Service) across wireless links. Higher data rate and bet- Moreover, a general solution for different modulation schemes
ter spectral efficiency are of paramount importance in next is not provided in [10].
generation cellular communication. MIMO (multiple input The proposed concept solves this problem in a novel
multiple output) technology is one solution to attain this by fashion. Traditionally, modulation techniques such as BPSK
transmitting multiple data streams from multiple antennas [1]. (binary phase shift keying), QPSK (quadrature phase shift
However, the capacity gain results from MIMO transmission keying), 8PSK, 16QAM (quadrature amplitude modulation),
depends strongly on transmit and receive antenna spacing [2], etc. map a fixed number of information bits into one symbol.
[3], transmit antenna synchronization [4], [5], and the used Each symbol represents a constellation point in the complex,
algorithm to reduce the interchannel interference (ICI) at the two dimensional signal plane. In the following this is referred
receiver input. Copious ICI reduction algorithms are reported to as signal modulation. In this paper, a novel approach is
in literature. The proposed Bell Labs Layered Space-Time proposed in which this two dimensional plane is extended to
Architecture (BLAST) [6] is one of the most promising MIMO a third dimension - the spatial dimension. In the following this
detection algorithms. In its most basic form known as vertical is referred to as spatial modulation. It is demonstrated that this
(V)-BLAST [7], multiple transmitted data streams are sepa- new approach will result in a very flexible mechanism which is
rated and detected successively using a combination of array able to achieve high spectral efficiency and very low receiver
processing (nulling) and interference cancellation techniques. complexity, and that is ideally suited for high speed uplink
It was demonstrated that with the V-BLAST algorithm spectral transmission.
1-4244-0463-0/06/$20.00 ©2006 IEEE
The idea of using the transmit antenna number as an modulation, m = log2 (M) is the number of bits/symbol. The
additional source of information is utilized in spatial modula- spatial modulation system model is shown in Fig. 1. q(k)
tion. The number of information bits that can be transmitted is a vector of n bits to be transmitted. The binary vector is
using spatial modulation depends on the used constellation mapped into another vector x(k) of size Nt such that only one
diagram and the given number of transmit antennas. For element in the resulting vector is different from zero. Symbol
instance, six information bits can be mapped into 32QAM number f in the resulting vector x(k) is xe, where f is the
and two transmit antennas. Alternatively, if the channel and mapped transmit antenna number f C [1: Nt]. The symbol
interference environment do not allow the use of 32QAM, the x. is transmitted from the antenna number f over the MIMO
same spectral efficiency can be achieved with 16QAM and channel, H(k). H(k) can be written as a set of vectors where
four transmit antennas, etc.. each vector corresponds to the channel path gains between
In view of the fact that information is not only included in transmit antenna v and the receive antennas as follows:
the transmitted symbol but also in the actual physical location
of the antenna, estimation of the transmit antenna number H = [h1 h2 ... hN, ] (1)
is of key importance. In this paper, an innovative transmit
antenna number detection algorithm, called iterative-maximum where:
ratio combining (i-MRC) is presented. The i-MRC algorithm is hv =
[hi, 1, h2,z, ...
hN,, Ij (2)
based on the fact that only one antenna transmit at a time. The
antenna number may change at the subsequent transmission The received vector is then given by y(k) = h(= )xf +w(k);
instants, but at any given time only a single transmit antenna where w(k) is the additive white Gaussian noise vector.
is transmitting. The channel vectors between each transmit The number of transmitted information bits, n, can be
antenna and the number of receive antennas are considered adjusted in two different and independent ways - either by
separately at the receiver. The receiver iteratively computes changing the signal modulation and/or changing the spatial
the MRC results between the channel paths from each transmit modulation. For example, three bits per symbol could be send
antenna to the corresponding receive antennas. Assuming full from four transmit antennas using BPSK modulation as shown
knowledge of the channel at the receiver, the receiver chooses in Fig. 2. Alternatively, using two transmit antennas instead of
the transmit antenna number which gives highest correlation. four, three bits could be send if the modulation technique is
In addition to eliminating ICI at the receiver, spatial modu- changed to 4QAM as shown in Fig. 2. Similarly, in order to
lation produces no correlation between the transmit antennas transmit four bits, BPSK modulation and eight antennas can
and it requires no synchronization between them. In general, be used or 4QAM modulation and four transmit antennas. The
1/2 wavelength spacing can achieve uncorrelated fading but same spectral efficiency can be achieve if two antennas and
this is affected by factors such as the proximity to the 8QAM are used.
human body and other objects. On the other hand, lack of In general, the number of bits that can be transmitted using
synchronization is shown to have a major effect on system spatial modulation is given as follows:
performance [4], [5]. Furthermore, in spatial modulation, the log2(Nt) + m
symbol duration is unchanged while the transmitted symbol
n= (3)
carries a higher number of information bits (note that the III. TRANSMIT ANTENNA NUMBER ESTIMATION
distance between the signal constellation points is not altered)
due to the novel extension of modulation to the spatial domain. With spatial modulation data is encoded in an information
As a consequence, an improvement in spectrum efficiency is symbol and an antenna number. Hence, estimation of the
obtained. transmit antenna number is an important task. In the following,
The rest of the paper is organized as follows: In Sec- a novel antenna number detection algorithm is presented.
tion II, the spatial modulation system model is presented. In The received vector y(k) is iteratively multiplied by the
Section III, the i-MRC antenna number detection algorithm channel path gains, which are assumed to be known at the
is presented. Simulation results and a receiver complexity receiver, in order to estimate both: the transmitted symbol and
comparison are shown in Section IV. Finally, conclusions are the transmit antenna number:
provided in Section V.
II. SYSTEM MODEL hjHy, Forj =:Nt (4)
Throughout the paper, the following notations and assump- g = [91 92 9Ntl] (5)
tions are used. Bold and small letters 'a' denote vectors.
Bold and capital letters 'A' denote matrices. The notations,
£ = argmax*g
i
(6)
( *, (.)+, (.)H* and (.)T denote conjugate, pseudoinverse, 'zf = Q(g(j=i)) (7)
Hermitian, and transpose, of a matrix or a vector respectively,
and () -1 denotes the inverse of a matrix. hV, T is the channel where Q(.) is the constellation quantization (slicing) function.
gain between transmit antenna T and receive antenna v. Assuming correct estimates for £ and x., the receiver can
Nr is the number of receive antennas. Finally, for MQAM straightforwardly de-map the original information bits.
Modulation
q(k)= [O I ..] =[O 1 ...I
q(k) x(k)= [0 x, ... 0] n bits
Nt symbols
Ihl hl2
Hk=h2l h22
-hNrl hNr 2
Fig. 1. Spatial Modulation System Model
o o aa
antt,
nt', /
% o-1 f ant:±
4t 0
0 O 0 ant'1 Symbol.4
I
0 1/arn' W f(
antl~ 1 6A
0
a nt it
0_
a /
t4t1 2
ant
2an
an>t
l /M...... 0
an antatn 0
[Link]
[Link] 1odulation:
Mouainnbt 2bistansissonsinBPKadfurranmitantnna,o,4AM
rnmsinuigBS n ortasi ant nnas or1QM nndworanm
w tasiatna
A. Example: 3 bits spatial modulation transmission and detec- y(k) [0.4149 + 0.6606i, 0.4486 + 0.6415i, -0.7304
tion using 2x4 antenna configuration and 4QAM transmission 0.1944i, 0.4348-O.1361i]'.
The mapping table for 3 bits transmission using 4QAM The i-MRC detection algorithm is considered in the fol-
signal modulation and 2x4 MIMO antenna configuration is lowing. The channel is assumed to be known at the receiver.
shown in Fig. 2. applying the i-MRC detection algorithm to the received vector
Assume the following sequence of bits to be transmitted, y(k) results in g = [-1.0000- [Link], -0.3271 -0.2978i].
q(k) = [0, 1, 1]. Mapping this to a 4QAM symbol and two Then, the antenna number can be estimated by computing
transmit antennas results in x(k) [1i, 0]T where i eqn. (6) and results in £ = 1. Similarly, the transmitted
is the imaginary number, i 1. The vector x(k) is symbol can be estimated by computing eqn. (7) and results
transmitted over the 2x4 MIMO channel H(k). Note that only in xe =-1 -i. Using the respective look-up table in Fig. 2,
antenna number one will be transmitting the symbol x. and the transmitted 3-bit sequence can be de-mapped into q(k)
antenna number two will be transmitting zero energy. For [0,1,1], which exactly corresponds to the transmitted bits.
example, assume the following channel matrix and noise free The estimation of the transmit antenna number is based
transmission: on the cross correlation between different channel paths.
Therefore, like in other spatial multiplexing techniques, the
-0.5377 - 0.1229i -0.6175 + 0.1516i performance of the algorithm is dependent on the channel
- 0.0964i -0.3271 - 0.0006i
H(k)= -0.54500.4624 - -0.2680i 0.2058 + 0.3171i
correlation. However, with the proposed new technique, the
correlation only depends on the characteristics of the channel
0.2854 - -0.1493i -0.5190 + 0.2767i and not on the spacing between the transmit antennas since
Then, the received vector at the receiver input equals only a single antenna transmit at any given time. This is
deemed an important feature - especially for uplink transmis- 16QAM and the 4x4 8QAM spatial modulation results also
sion. match very closely. The spatial modulation technique outper-
forms lx4 32QAM MRC by about 4 dB.
IV. RESULTS
In the simulation a flat Rayleigh fading channel is assumed Spatial Modulation, 5 bits transmission
100 F.
with additive white Gaussian noise (AWGN). The receiver is
assumed to have full channel knowledge. The receive antennas
are assumed to be separated wide enough to avoid correlation. 1o-10
The total transmit power, P, is the same for all transmission
and the SNR at each receiver input is given by .2, where (2 10-2
is the noise power. For V-BLAST transmission, the transmit 0
antennas are assumed to be synchronized and separated wide .75 10-3
enough to avoid correlation. Finally, the SNR is assumed to
m
be known at the receiver when using the MMSE (minimum 10-4
mean square estimation) algorithm for V-BLAST detection.
MRRC 32QAM 1x4
A. Three bits transmission 10-5 spatial modulation 8QAM 5bits 4x4
The BER (bit-error-ratio) of 4x4 BPSK and 2x4 4QAM . spatial modulation 16QAM 5bits 2x4
V-BLAST MMSE 8QAM 2x4
spatial modulation is plotted in Fig. 3. The 3x4 BPSK MMSE 1o-6
0 5 10 15 20 25
V-BLAST and lx4 8QAM MRC results are plotted in the SNR
same figure. 4x4 BPSK spatial modulation performs nearly
the same as the lx8 8QAM MRC and 2x4 4QAM spatial Fig. 4. Five bits transmission using spatial modulation and MRC
modulation performs almost the same as 3x4 BPSK MMSE
V-BLAST. 4x4 BPSK spatial modulation shows a performance C. Six bits transmission
degradation of 2 dB as compared to 2x4 4QAM spatial
modulation and 3x4 BPSK MMSE V-BLAST. Fig. 5 shows the BER performance of spatial modula-
tion 4x4 16QAM transmission and 2x4 32QAM transmission
Spatial Modulation, 3 bits transmission
compared to 2x4 8QAM and 3x4 4QAM MMSE V-BLAST
lo' t::::::::::::: .............:: i: ::
: :: : ....: algorithm and to lx4 64QAM MRC transmission. 4x4 16-
QAM transmission and 2x4 8QAM V-BLAST transmission
1o-1 perform nearly the same. 3x4 4QAM MMSE V-BLAST trans-
mission shows a degradation in performance at high SNR
due to the presence of error propagation and the high ICI
1 o-2
0
at the receiver. The new spatial modulation 4x4 1 6QAM
transmission outperforms lx4 64QAM MRC transmission by
10 approximately 7 dB.
-3
10 Spatial Modulation, 6 bits transmission and V-BLAST Comparison
100 ;...........
u6-5 MRRC 8QAM 1x4
spatial modulation 3bits 4-QAM 2x4 10 1
. = spatial modulation 3bits BPSK 4x4
V-BLAST MMSE BPSK 3x4
io-6 10o2
0 5 10 15 20 25
SNR 0
*t -3
10
Fig. 3. Three bits transmission using spatial modulation, V-BLAST, and
MRC
10-5
MRRC 64QAM 1x4
B. Five bits transmission spatial modulation 16QAM 6bits 4x4 X
I. X
10-6
Five bits transmission BER using 4x4 8QAM and 2x4 spatial modulation 32QAM 6bits 2x4
V-BLAST MMSE 8QAM 2x4 ..
16QAM spatial modulation and lx4 32QAM MRC are plotted V-BLAST MMSE 4QAM 3x4 :::::
in Fig. 4. It is difficult to directly compare this constellation to 10
0 5 10 15 20 25 30
V-BLAST as 5-bit transmission is not possible. Therefore, the SNR
V-BLAST transmission with six bits is chosen, i.e., 2x4 8QAM
MMSE. The results are plotted in Fig. 4. Spatial modulation Fig. 5. Six bits transmission using spatial modulation, V-BLAST, and MRC
and V-BLAST perform virtually the same. In addition, 2x4
D. System Complexity Estimation a special transmit antenna number detection algorithm tailored
The system complexity of the discussed transmission tech- for the new spatial modulation technique has been proposed.
niques in this paper is considered in the following. While the new spatial modulation technique shows almost the
Consider only multiplication and addition of a complex same BER performance for the same spectral efficiency as
numbers as an operation. ideal V-BLAST, it results in a significant reduction in receiver
The MMSE criterion requires two matrix multiplications, complexity. In addition, only one RF (radio frequency) chain is
one inversion and one addition. It is assumed that the matrix required at the transmitter because at any given time only one
transposes are not computed explicitly. The first matrix multi- antenna transmits. This means that the cost of the transmitter
plication requires Nt2Nr, the matrix addition requires Nt2 and can be significantly reduced. Spatial modulation is expected
the matrix inverse needs 4Nt3 (using Gaussian Elimination) to outperform V-BLAST under more realistic assumptions,
operations [11], [12]. The second matrix multiplication takes i.e. taking into consideration transmit antenna correlation and
Nt3 operations. Therefore, a total of (5Nt3 + NrNt2 + Nt2) frequency and antenna synchronization issues. This is subject
complex operations are needed. For V-BLAST these steps are to future research.
repeated for i = 1 ... Nt [13]. This means that the MMSE ACKNOWLEDGEMENT
is computed for deflated H but with decreasing dimensions The authors would like to thank Samsung, and in particular
of (Nr x (Nt- L)), L = 0 ... .Nt- 1. As a result, the total the Samsung Advanced Institute of Technology (SAIT) in
number of complex operations is given by: Korea for supporting this work.
Nt
REFERENCES
( + Nri2 + i2). (8)
i=l [1] E. Telatar, "Capacity of Multi-Antenna Gaussian Channels," European
Transaction Telecommunication, vol. 10, no. 6, pp. 558-595, Novem-
i-MRC requires NtNr complex multiplications and ber/December 1999.
Nt(Nr -1) complex additions. So, a total of: [2] G. J. Foschini, M. J. Gans, and J. M. Kahn, "Fading Correlation and
its Effect on the Capacity of Multielement Antenna Systems," IEEE
[2NtNr -Nt] (9) Transactions on Communications, vol. 48, pp. 502-513, March 2000.
[3] S. Loyka and G. Tsoulos, "Estimating MIMO System Performance
complex operations are required. Using the Correlation Matrix Approach," IEEE Communications Letters,
In the following, let ( be the number of complex operations. vol. 6, no. 1, pp. 19-21, January 2002.
The receiver complexity comparisons between the spatial [4] M. Chiani, M. Z. Win, and A. Zanella, "On the Capacity of Spatially
Correlated MIMO Rayleigh-Fading Channels," IEEE Transactions on
modulation and V-BLAST are shown in Table I for three bits Information Theory, vol. 49, no. 10, pp. 2363-2371, October 2003.
transmission and in Table II for six bits transmission. [5] S. Catreux, P. F. Driessen, and L. J. Greenstein, "Simulation Results
for an Interference-Limited Multiple-Input Multiple-Output Cellular
TABLE I System," IEEE Communications Letters, vol. 4, no. 11, pp. 334-336,
THREE BITS TRANSMISSION REVIVER COMPLEXITY COMPARISON November 2000.
[6] G. J. Foschini, "Layered SpaceTtime Architecture for Wireless Commu-
V-BLAST Spatial Modulation nication in a Fading Environment when Using Multi-Element Antennas,"
Bell Labs Technical Journal, vol. 1, no. 2, pp. 41-59, September 1996.
MMSE BPSK 4QAM [7] P. W. Wolniansky, G. J. Foschini, G. D. Golden, and R. A. Valenzuela,
( ll 560 28 14 "V-BLAST: An Architecture for Realizing Very High Data Rates Over
the Rich-Scattering Wireless Channel," Proceedings of the IEEE Inter-
national Symposium on Signals, Systems, and Electronics (ISSSE), no. 2,
TABLE II pp. 295-300, September/October 1998.
SIX BITS TRANSMISSION REVIVER COMPLEXITY COMPARISON [8] V. Tarokh, N. Seshadri, and A. R. Calderbank, "Space-Time Codes for
High Data Rate Wireless Communication: Performance Criterion and
8QAM V-BLAST 4QAM V-BLAST Spatial Modulation Code Construction," IEEE Transactions on Information Theory, vol. 44,
no. 2, pp. 744-765, March 1998.
MMSE MMSE 16QAM 32QAM
[9] H. Haas, E. Costa, and E. Schulz, "Increasing Spectral Efficiency by
110 560 28 14 Data Multiplexing Using Antenna Arrays," Proceedings of the Interna-
tional Symposium on Personal, Indoor and Mobile Radio Communica-
tions (PIMRC 2002), vol. 2, pp. 610-613, September 15 - September
V. CONCLUSION 18 2002.
[10] R. Mesleh, H. Haas, Y Lee, and S. Yun, "Interchannel Interference
A novel spectral efficient multiple antenna transmission Avoidance in MIMO Transmission by Exploiting Spatial Information,"
scheme that utilizes the spatial information in an innovative Proceedings of the International Symposium on Personal, Indoor and
fashion has been presented. A new modulation scheme, called Mobile Radio Communications PIMRC 2005, September 11 - September
14 2005.
spatial modulation, maps multiple information bits into a [11] N. J. Higham and R. S. Schreiber, "Fast Polar Decomposition of an Ar-
single information symbol and into the physical location of bitrary Matrix," SIAM Journal of Scientific and Statistical Computings,
the single transmitting antenna. Spatial modulation avoids ICI vol. 11, no. 4, pp. 648-655, July 1990.
[12] G. H. Golub and C. F. V. Loan, "Matrix Computations," Johns Hopkins
at the receiver input, produces no correlation between the University Press, Baltimore, MD, 3rd edition, 1996.
transmit antennas and requires no synchronization between [13] W. K. Wai and R. S. C. Chi-Y. Tsui, "A Low Complexity Architecture of
them. Moreover, spatial modulation does not suffer from the the V-BLAST System," IEEE Wireless Communications and Networking
error propagation problem that exists in V-BLAST. In addition, Conference, (WCNC), vol. 1, pp. 310-314, September 2000.