Spatial Modulation

2228 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO.
4, JULY 2008
Spatial Modulation
Raed Y. Mesleh, Member, IEEE, Harald Haas, Member, IEEE, Sinan Sinanović,
Chang Wook Ahn, Member, IEEE, and Sangboh Yun, Member, IEEE
Abstract—Spatial modulation (SM) is a recently developed I. I NTRODUCTION

transmission technique that uses multiple antennas. The basic idea
is to map a block of information bits to two information carrying
units: 1) a symbol that was chosen from a constellation diagram
and 2) a unique transmit antenna number that was chosen from
T HE NEED for high data rate and high spectral effi-
ciency are the key elements that drive research in future
wireless communication systems [53]. Adaptive coding and
a set of transmit antennas. The use of the transmit antenna num- modulation, iterative (turbo) decoding algorithms, space–time
ber as an information-bearing unit increases the overall spectral
efficiency by the base-two logarithm of the number of transmit coding (STC), multiple antennas and multiple-input–multiple-
antennas. At the receiver, a maximum receive ratio combining output (MIMO) systems, multicarrier modulation, and ultra
algorithm is used to retrieve the transmitted block of information wideband radio are examples of enabling technologies for next-
bits. Here, we apply SM to orthogonal frequency division multi- generation wireless communication. Among the set of existing
plexing (OFDM) transmission. We develop an analytical approach technologies, MIMO orthogonal frequency division multiplex-
for symbol error ratio (SER) analysis of the SM algorithm in
independent identically distributed (i.i.d.) Rayleigh channels. The ing (MIMO–OFDM) with adaptive coding and modulation is
analytical and simulation results closely match. The performance a promising candidate for future wireless systems. A MIMO
and the receiver complexity of the SM–OFDM technique are system boosts spectral efficiency by using multiple antennas
compared to those of the vertical Bell Labs layered space–time to simultaneously transmit data to the receiver [1]–[4]. OFDM
(V-BLAST–OFDM) and Alamouti–OFDM algorithms. V-BLAST converts a frequency-selective channel into a parallel collection
uses minimum mean square error (MMSE) detection with ordered
successive interference cancellation. The combined effect of spatial
of frequency flat-fading subchannels, in which the available
correlation, mutual antenna coupling, and Rician fading on both bandwidth is very efficiently used [5]. The OFDM technique
coded and uncoded systems are presented. It is shown that, for the has been adopted in several wireless standards such as digital
same spectral efficiency, SM results in a reduction of around 90% audio and video broadcasting, the IEEE 802.11a standard [6],
in receiver complexity as compared to V-BLAST and nearly the the IEEE 802.16a metropolitan area network standard, and the
same receiver complexity as Alamouti. In addition, we show that
SM achieves better performance in all studied channel conditions,
local area network standard [7].
as compared with other techniques. It is also shown to efficiently There are three main categories of MIMO techniques. The
work for any configuration of transmit and receive antennas, even first category improves power efficiency by maximizing spatial
for the case of fewer receive antennas than transmit antennas. diversity, e.g., using delay diversity [8], [9]. In such systems
Index Terms—Interchannel interference (ICI), multiple-input– (e.g., STC), the capacity improvement results from diversity
multiple-output (MIMO), orthogonal frequency division multi- gain, which reduces the bit error probability for the same spec-
plexing (OFDM), receiver complexity, space–time coding (STC) tral efficiency. However, the maximum spectral efficiency of
coded modulation, spatial modulation (SM), vertical Bell Labs full-diversity STC systems is one symbol per symbol duration
layered space–time (V-BLAST).
for any number of transmit antennas [9]. These systems can
be designed to achieve full diversity gain with very low re-
ceiver complexity. In addition, STCs are well known to combat
channel imperfections that exist in real-time implementations
of MIMO systems [10], [11]. The second category of MIMO
Manuscript received June 12, 2006; revised February 20, 2007, June 29, techniques exploits knowledge of the channel at the transmitter.
2007, September 4, 2007, and September 18, 2007. This work was supported It decomposes the channel matrix by using singular value de-
by the Samsung Advanced Institute of Technology, Suwon, Korea. The review
of this paper was coordinated by Prof. E. Bonek.
composition and uses the resulting unitary matrices as prefilters
R. Y. Mesleh is with the School of Electrical Engineering and Computer Sci- and postfilters at the transmitter and receiver, respectively, to
ence, Jacobs University, D-28759 Bremen, Germany (e-mail: r.mesleh@jacobs- achieve capacity gain [12], [13]. In this paper, channel in-
university.de).
H. Haas is with the Institute for Digital Communications, University of
formation is assumed to be known only at the receiver, with
Edinburgh, EH9 3JL Edinburgh, U.K., and also with Jacobs University, no channel information at the transmitter. Therefore, these
D-28759 Bremen, Germany (e-mail: h.haas@ed.ac.uk). techniques are not implemented here, but possible scenarios
S. Sinanović is with the Institute for Digital Communications, University of
Edinburgh, EH9 3JL Edinburgh, U.K. (e-mail: s.sinanovic@ed.ac.uk). will briefly be discussed in future works.
C. W. Ahn was with Gwangju Institute of Science and Technology, The third type of MIMO technique uses a layered space–time
Gwangju 500-712, Korea. He is now with the Department of Computer En- approach to transmit multiple independent data streams over
gineering, Sungkyunkwan University, Suwon 440-746, Korea (e-mail: cwan@
evolution.re.kr). the antennas to increase capacity. A well-known technique
S. Yun is with the Telecommunication R&D Center, Samsung Electronics is the Bell Labs layered space–time (BLAST) architecture
Company, Ltd., Suwon 442-600, Korea (e-mail: sbyun@samsung.com). [3]. The BLAST scheme demultiplexes a user data stream
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. into a number of substreams that are equal to the number of
Digital Object Identifier 10.1109/TVT.2008.912136 transmission antennas. Two types of BLAST realizations have
0018-9545/$25.00 © 2008 IEEE
Authorized licensed use limited to: GLASGOW CALEDONIAN UNIVERSITY. Downloaded on June 02,2020 at 13:24:56 UTC from IEEE Xplore. Restrictions apply.
MESLEH et al.: SPATIAL MODULATION 2229
widely been discussed: 1) diagonal BLAST (D-BLAST) [3] and by SM to boost the spectral efficiency. That is why SM is
2) vertical BLAST (V-BLAST) [14]. different from other MIMO techniques such as space–time bit-
The D-BLAST architecture is considered the reference in interleaved coded modulation [24], in which the antenna pattern
performance for MIMO systems, since it can reach capacities is recognized as a spatial constellation but is not used as a
near the Shannon limit [15]. The D-BLAST system has a source of information.
diagonal layered STC architecture with sequential nulling and In SM, a block of any number of information bits is mapped
interference cancellation decoding. However, it suffers from into a constellation point in the signal domain and a constella-
boundary wastage at the start and at the end of each packet, and tion point in the spatial domain. At each time instant, only one
its complexity is very high to be practical [16]. The V-BLAST transmit antenna of the set will be active. The other antennas
architecture is a simplified version of D-BLAST, which tries will transmit zero power. Therefore, ICI at the receiver and
to overcome its limitations. However, in doing so, the trans- the need to synchronize the transmit antennas are completely
mit diversity is lost. It has been demonstrated that, with the avoided. At the receiver, maximum receive ratio combining
V-BLAST algorithm, spectral efficiencies of 20–40 b/s/Hz can (MRRC) is used to estimate the transmit antenna number, after
be achieved in an indoor rich scattering propagation environ- which the transmitted symbol is estimated. These two estimates
ment, assuming a practical signal-to-noise ratio (SNR) range are used by the spatial demodulator to retrieve the block of
and bit error performance, respectively [14]. information bits.
Several problems are encountered in the development of mul- In this paper, a closed-form analytical symbol error ratio
tiple antenna transmission schemes [17]–[20]. These problems (SER) of SM in independent identically distributed (i.i.d.)
arise from several sources, among which are the following. Rayleigh channels is derived. Analytical and simulation results
closely match over a wide range of SNR values. The assump-
1) BLAST transmission systems suffer from high inter-
tion of i.i.d. channel conditions is idealized, and we chose it to
channel interference (ICI) at the receiver due to simulta-
simplify the analytical calculations; it is not a practical model
neous transmissions on the same frequency from multiple
for MIMO–OFDM systems. However, further investigation in
antennas.
this paper includes nonidealistic channel conditions, in which
2) The high ICI requires a complex receiver algorithm,
the BER is simulated for SM–OFDM, V-BLAST–OFDM [25],
which increases the overall system complexity.
and Alamouti–OFDM [26]. We also compare receiver com-
3) System performance is traded off with receiver complex-
plexity for these systems. The combined effect of nonidealistic
ity. In addition, although BLAST systems achieve a rela-
channel conditions, including spatial correlation (SC), mutual
tively good performance in ideal channel conditions, their
antenna coupling (MC), and Rician fading, on both coded and
performance severely degrades under nonideal channel
uncoded systems is discussed.
conditions [17], [18].
The rest of this paper is organized as follows. Section II
4) With full-diversity STCs, these limitations are overcome.
presents the SM idea and its application to OFDM transmis-
In addition, due to their orthogonal design, they can
sion. Analytical calculation of SER for SM is, then, shown in
easily be decoded at the receiver side. STCs are also
Section III, as well as a comparison between the analytical and
robust in the presence of channel imperfections [10],
the simulation results. The channel model and the modeling of
[11]. However, the maximum spectral efficiency of full-
Rician fading, SC, and MC channel imperfections are discussed
diversity STC systems is one symbol per symbol duration
in Section IV. Simulation results and the receiver complexity
for any number of transmit antennas. In other words, for
comparison follow in Sections V and VI, respectively. Finally,
full-diversity STCs to achieve spectral efficiency that is
Section VII concludes this paper.
similar to that of BLAST techniques, they need to use
higher modulation orders.
5) For efficient operation of BLAST techniques, the number II. SM–OFDM S YSTEM M ODEL
of transmit antennas must be less than or equal to the
The following notations are used throughout this paper.
number of receive antennas [16]. In STCs, the situation is
Bold and lowercase letters denote vectors, whereas bold and
different. Generally, STCs can be designed for different
capital letters denote matrices. The notations (·)+ , (·)H , and
numbers of transmit and receive antennas and can effi-
(·)T denote the pseudoinverse, Hermitian, and transpose of a
ciently work, even if the number of receive antennas is
vector or matrix, respectively, and (·)−1 denotes the inverse of
less than the number of transmit antennas. However, an
a matrix.
orthogonal design of full-rate-code STC is only known
The SM–OFDM system model is shown in Fig. 1.
for the case of two transmit antennas, and there is no
Q(k) is an m̃ × n binary matrix that will be transmitted
known solution for a higher number of transmit antennas.
in one OFDM symbol, where m̃ is the total number of bits
Therefore, the design of STCs for more than two transmit
per symbol per subchannel, and n is the total number of
antennas must sacrifice a portion of the data rate to
OFDM subchannels. The SM maps this matrix into another
achieve full orthogonality and, hence, full diversity.
matrix X(k) of size Nt × n, where Nt is the total number of
One approach for dealing with these issues is to use spatial transmit antennas, by using the SM mapping table shown in
modulation (SM) [21]–[23]. In this case, only one transmit Fig. 1. This table maps each column in Q(k) into a binary
antenna is active at any instant. The active transmit antenna phase-shift keying (BPSK) constellation point and a single
number is an added source of information that is exploited transmit antenna number from a set of four antennas. BPSK
2230 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 4, JULY 2008
Fig. 1. SM–OFDM system model.
and four transmit antennas are considered as an example here. TABLE I

SM MAPPING TABLE: 3 b/SYMBOL/SUBCHANNEL
In general, any number of transmit antennas and any digital
modulation scheme can be used. The constellation diagram and
the number of transmit antennas determine the total number
of bits to be transmitted on each subchannel at each instant.
The combination of BPSK and four transmit antennas in this
illustration results in a total of three information bits to be trans-
mitted on each subchannel. Instead, four quadrature-amplitude
modulation (QAM) and two transmit antennas can be used to
transmit the same number of information bits, as shown in
Table I. The number of bits that can be transmitted on each
OFDM subchannel for a system that uses a QAM constellation
diagram of size M (m = log2 (M )) and Nt transmit antennas
is [22]
m̃ = log2 (Nt ) + m. (1)
This shows that the constellation diagram and the number of user system. For instance, two competing approaches have been
transmit antennas can be traded off for any number of trans- proposed for the MIMO-oriented version of the IEEE 802.11n
mitted information bits. In addition, SM increases the spectral standard: 1) one with a 2 × 2 MIMO matrix and 2) another
efficiency by the base-two logarithm of the total number of with a 4 × 4 matrix. The current 802.11n draft provides for up
transmit antennas. This can be viewed as a disadvantage for to four transmit antennas, even though compliant hardware is
a large number of transmit antennas as compared to, for exam- not required to support that many antennas [27].
ple, V-BLAST. Note that, in V-BLAST, the spectral efficiency With SM mapping, the matrix X(k) has one nonzero element
linearly increases with the number of transmit antennas. For in each column at the position of the mapped transmit antenna
example, consider a MIMO system with eight transmit and number. All other elements in that column are set to zero. For
receive antennas. If V-BLAST is used with 16 QAM, a spectral instance, in Fig. 1, an input bit sequence of [0 1 1]T [highlighted
efficiency of 32 b/s/Hz can be achieved. However, if SM is column vector in Q(k)] is mapped to the BPSK symbol +1 and
used with the same configuration and modulation order, the the second transmit antenna by using the SM mapping table.
spectral efficiency is only 7 b/s/Hz. In order for SM to achieve This means that only the second antenna transmits this symbol
the spectral efficiency of V-BLAST with 16 QAM, it requires on the first OFDM subchannel, whereas all other antennas
228 transmit antennas, which is not feasible. This means that transmit zero power. As a result, the first column vector in X(k)
SM cannot compete with V-BLAST when a large number of is [0 + 1 0 0]T . The second bit sequence is [1 1 1]T and is
antennas and high modulation orders are involved. However, it mapped to [0 0 0 + 1]T , and so on. The resulting symbols in
is generally accepted that a large number of transmit antennas each row vector xκ (k) are the data that will be transmitted on all
is impractical with current technology, particularly when con- subchannels and from antenna κ. Then, each row vector xκ (k)
sidering the cost that comes from adding antennas for an end- is modulated using an OFDM modulator.
The resulting output vectors at the OFDM modulator are the following derived equation is the upper bound on the true
simultaneously transmitted from the Nt transmit antennas over performance in such channel conditions. The bits are correctly
the MIMO channel H(τ, t). At the receiver, the rows of the recovered only if both estimates are correct. To compute the
received matrix Y(t) = H(τ, t) ⊗ S(t) + R(t), where S(t) is overall probability of error Pe , let Pa denote the probability
a matrix that contains all OFDM symbols that are transmitted that the estimate of the antenna number is incorrect, and let
from all transmit antennas, R(t) is the additive white Gaussian Pd be the probability that the transmitted symbol estimate is
noise (AWGN) matrix, and ⊗ denotes time convolution, are incorrect. Then, the retrieved SM bits are correct if and only if
demodulated using Nr OFDM demodulators. In the following, the estimates of the antenna number and the transmitted symbol
discrete time representation is considered. The output from the are both correct. The probability of that is
OFDM demodulators is a matrix Y(k) of size Nr × n, each
column of which corresponds to the received data in the n Pc = (1 − Pa )(1 − Pd ). (5)
OFDM subchannels from the Nr receive antennas.
In the following, MRRC is used to detect the transmit The probability that the demapped bits are incorrect is, then,
antenna number and the transmitted symbol in the frequency 1 − Pc and can be written as
domain for each OFDM subchannel. The following notations
Pe = Pa + P d − P d Pa . (6)
are for a single OFDM subchannel, and the generalization to
multiple subchannels is straightforward by simply adding an If there is only one transmit antenna, then Pa is zero, and
additional subscript. However, for reasons of clarity, this is left the overall probability of error is reduced to the case of MRRC
out. The MRRC algorithm multiplies the Hermitian conjugate with multiple receive antennas, which is an upper bound for SM
of the frequency response channel matrix for each subchannel, performance. However, the use of multiple transmit antennas in
which is assumed to be known at the receiver, with the received SM results in an increase in the overall probability of error. This
column vector at this particular subchannel, i.e., increase in Pe is evident by noting that 1 ≥ (Pa and Pd ) ≥ 0.
g(k) = HH (k)y(k) (2) Hence, SM increases the SER by Pa − Pd Pa ≥ 0 as compared
to MRRC.
where H is the Nr × Nt discrete time-invariant frequency re- In Sections III-A–C, the SER of each estimation process is
sponse channel matrix, and y is the corresponding demodulated considered separately.
OFDM vector of length Nr . In an ideal scenario, where there is
perfect time and frequency synchronization and no noise, g(k) A. Analytical SER of the Transmitted Symbol Estimation
is the same as x(k), since by definition, x(k) contains only one Process
element that is different from zero. Therefore, in the presence
of AWGN, the estimated transmit antenna number ˜ at time As discussed in Section III, the estimation of the transmitted
instant k is the index or position of that element in g(k), whose symbol for any M -QAM is a 1 × Nr MRRC detection, since
absolute value is maximum, i.e., only the corresponding element in the resulting vector is con-
sidered for the estimation process. The average SER of a square
˜ = arg max (|gi (k)|) i = 1, . . . , Nt . (3) M -QAM over generalized fading channels is [28]
∀i
π

Nr
Assuming that the estimate of the transmit antenna number is 4 1 2 −gQAM
correct, the transmitted symbol at this instant can be estimated Pd = 1− √ Mil ; γ̄l dφ
π M 0 l=1
sin2 (φ)
as follows:
2 π

Nr
4 1 4 −gQAM
x̃(k) = Q gi=˜(k) (4) − 1− √ Mil ; γ̄l dφ (7)
π M 0 l=1
sin2 (φ)
where gi=˜ is the element number ˜ in the column vector g(k),
and Q(·) is the constellation quantization (slicing) function. where
These two estimates are, then, used by the SM demodulator 3
to retrieve the transmitted information bits on this particular gQAM = ,
2(M − 1)
subchannel by an inverse mapping process that uses the same
−1
mapping table that was used at the transmitter. −gQAM gQAM γ̄l
Mil ; γ̄l = 1 +
sin2 (φ) sin2 (φ)
III. A NALYTICAL SER C ALCULATION OF SM is the moment-generating function for Rayleigh flat-fading
Computation of the analytical performance of SM is not channels, and γ̄l is the average SNR at each receive antenna.
straightforward. There are two estimation processes that are
involved: 1) The transmit antenna number is estimated, and
B. Analytical Error Calculation of the Transmit Antenna
2) the transmitted symbol is estimated. The two processes are
Number Estimation Process
assumed to be independent in the calculation. However, this
is not generally correct. For instance, if the channel paths are In the following, the computation of Pa is considered. For
correlated, the two estimation processes will be dependent, and simplicity, only four transmit antennas are considered for the
derivation, and the result is later generalized to any number of

transmit antennas.
The antenna number estimate is the position of the maximum
absolute value of all elements in the vector that results from
MRRC. This corresponds to finding the element with the great-
est magnitude in the spatial domain among a set of Nt elements.
At a specific time instant and for a 4 × 4 transmission
scenario, let H = [h1 , h2 , h3 , h4 ] be the channel matrix, and let
hi = [h1i , h2i , h3i , h4i ]T be the corresponding channel vector
from each transmit antenna to all receive antennas. To illustrate,
assume that a sequence of data bits is mapped to symbol s2 from
a square QAM constellation and to the second transmit antenna.
Then, the received vector y = h2 s2 + n appears at the input of
the receive antennas. Applying MRRC to the received vector y,
as in (2), results in the following vector:
 H 
hH
1 h2 s2 + h1 n
Fig. 2. Order statistics pdfs of four random variables. The main figure shows
 hH h s + hH 2 n
g =  2H 2 2 . (8) the pdfs at SNR = 15 dB for 64 QAM and for µ1 = 1, whereas the upper main-
h3 h2 s2 + hH
3 n corner figure shows the same pdfs at the same SNR but with µ3 = 5. The small
hH H
4 h2 s2 + h4 n
figure inside the corner figure shows the crossing point between the pdfs by
plotting the y-axis on log scale. The bold circles in the main figure are the three
intersection points of the pdfs. It is clearly shown that the error contribution
Under the assumption of unity channel gains and for i.i.d from the second maximum is the largest, then from the third maximum,
channels, we have and so on.
Gaussian distribution with mean µi and variance σn2 . Then, the

1, if i = k
E hH
i hk = δi,k , δi,k = (9) probability density functions (pdfs) for v and x for each µi are
0, otherwise.
1 −
(v−µi ) 2
Therefore, if the noise is assumed to be AWGN with zero fV v|µi , σn2 = √ exp 2σn2
mean and σn2 variance, then three elements in the vector g have σn 2π

zero mean and σn2 variance. The other element, i.e., the second 1 −
(x−µi )2
−
(x+µi )2
element in (8), has mean s2 and variance σn2 . The square QAM fX x|µi , σn = √
2
exp 2σn 2
+ exp 2
2σn
.
σn 2π
signal can be decomposed into two independent but identical
amplitude modulated signals: 1) in phase I and 2) quadrature (12)
Q. In what follows, only the real positive part of the QAM
constellation is considered for the calculation. Assume that µi The second step in estimating the antenna number is finding
is the absolute value of the real part of the transmitted symbol the position of the element in g with a maximum absolute
s2 . Then, µ is a vector of length c = 2(m/2)−1 , which contains value. This is done by computing the pdfs of the sorted Nt
the positive real-part elements of the constellation diagram. Let random variables, where each has a pdf as given in (12) but
P (µi ) denote the probability that the antenna number estimate with different means. This problem can be treated with order
is incorrect when transmitting µi . Then, the average overall statistics [30].
probability of error for the antenna number estimate, when Let X(1) , . . . , X(Nt ) denote the order statistics of random
considering the real part Par , is given by samples from a continuous population with a cumulative dis-
tribution function FX (x|µi , σn2 ) and a pdf fX (x|µi , σn2 ), where
X(Nt ) > X(Nt −1) > · · · > X(1) . Then, the pdf of X(j) is
1
c
Par = P (µi ). (10)
c i=1 n!
fX(j) x|µi , σn2 = fX x|µi , σn2
(j − 1)!(Nt − j)!
The imaginary part is identical to the real part and can be j−1
calculated in a similar way. The probability that the detection × FX x|µi , σn2
is correct for both real and imaginary parts is the product of Nt −j
× 1 − FX x|µi , σn2 . (13)
two probabilities, namely, (1 − Par )(1 − Par ). As a result, the
overall probability of error, when considering both real and Considering the current case of four transmit antennas, Fig. 2
imaginary parts, is given by [29] shows the order statistics pdfs of the four random variables,
which result from taking the maximum of the absolute value
Pa = 1 − (1 − Par )2 = 2Par − Par
2
. (11) of each element in the vector, which results from MRRC at the
receiver. If the order statistics pdfs are assumed to be statisti-
The detection of the transmit antenna number is given in (3). cally independent, the probability that the antenna number esti-
Let x = |v|, where v is a random variable that follows a mation is incorrect can be found by numerically integrating the
intersection areas of fX(4) (x|µi , σn2 ) with all other distributions

for each mean value. However, the assumption of statistical in-
dependence is an approximate, since the pdfs are derived based
on the conditional probability that X(Nt ) > X(Nt −1) > · · · >
X(1) . Nevertheless, it will be shown in Section III-C that
both analytical and simulation results demonstrate a very close
match, which indicates that the previous approximations are
valid.
Let x3 , x2 , and x1 indicate the intersection points (bold
circles in Fig. 2) between fX(4) (x|µi , σn2 ) and fX(3) (x|0, σn2 ),
fX(2) (x|0, σn2 ), and fX(1) (x|0, σn2 ), respectively. Fig. 2 shows
that x3 > x2 > x1 > 0, which indicates that the error contri-
bution from the second largest sample is always the highest,
followed by that of the third largest sample, and so on. The
probability of error for each µi is, then, given by averaging the
multiple hypothesis errors as follows:
x Fig. 3. SMs of 16- and 64-QAM 4 × 4 analytical and simulation SER.
3 x2
1
P (µi ) = fX(4) x|µi , σn dx + fX(4) x|µi , σn2 dx
2
3
0 0

x1

+ fX(4) x|µi , σn2 dx . (14)
0
For any number of transmit antennas, (14) can be written as

 
Nt−1 xi
1 
P (µi ) = fX(Nt ) x|µi , σn2 dx . (15)
Nt − 1 i=1
0
Knowing P (µi ) for ∀i, Par is calculated as in (10). Par is, then,
used to compute Pa as in (11). Both Pa and Pd are used to
calculate the overall probability of error as in (6).
Fig. 4. Order statistics pdfs of four random variables. The main figure shows
the pdfs at SNR = 30 dB for 64 QAM and for µ2 = 3, whereas the other figure
C. Analytical and Simulation Results (figure inside the box) shows the same pdfs at the same SNR but with µ3 = 5.
The figures clearly show that numerical integration is not possible at this high
We next consider the simulation and analytical SER of SNR, since the crossing points between the pdfs no longer exist.
SM over i.i.d. Rayleigh flat-fading channels for different SM
system configurations. The results for a 16-QAM 4 × 4 SM iterative detection and decoding receiver. Turbo-BLAST works
(resulting in m̃ = 6 bits) and a 64-QAM 4 × 4 SM (resulting with any configuration of transmit and receive antennas and has
in m̃ = 8 bits) are depicted in Fig. 3. Additional simulation a better performance than traditional BLAST techniques. This,
and analytical results for a 16-QAM 4 × 3 SM (resulting in however, comes at the expense of an immense increase in the
m̃ = 6 bits) are depicted in Fig. 5. overall system complexity. An alternative solution is to use SM,
The simulation and analytical results, as shown in Fig. 3, as shown in Fig. 5. The results in Fig. 5 show the special case
are in close agreement. At relatively high SNR values, it is where there are more transmit antennas than receive antennas.
not possible to compute the numerical integration, since the The analytical calculation is still valid, and the analytical and
crossing points no longer exist, as in, for example, Fig. 4. For simulation results are almost the same.
a higher modulation order, the crossing points disappear at a
higher SNR.
IV. C HANNEL M ODELS
An efficient and practical MIMO architecture must handle
any configuration of transmit and receive antennas, including The multipath frequency selective and time-variant channel
the case of fewer receive antennas than transmit antennas. This model, as well as the Rician fading, SC, and MC channel
is, indeed, required in most cellular systems, since the base models, are presented in this section. These models are only
station can usually accommodate more transmit antennas than relevant to the simulations and the critical assessment of SM
mobile transceivers. The BLAST techniques efficiently work as presented in Section V, but they are clearly not a necessary
for Nt ≤ Nr . These techniques result in a poor performance requirement for SM to work.
if Nt = Nr and have an error floor if Nt > Nr [31]. Turbo- The channel matrix H(τ, t) is a block matrix and can be
BLAST is based on a random layered space–time code and an viewed as a collection of Nr × Nt vectors of length p, where
A. Rician Fading Channel

The standard statistical model for a multipath fading channel
with a line-of-sight (LOS) component follows a Rician distri-
bution or embodies Rician fading. In Rician fading, the channel
impulse response between each transmit antenna κ and receive
antenna ν is modeled as the sum of a fixed component and a
random multipath channel component [35], which is given by

K 1
ĥν,κ (τ, t) = hLOS (τ, t) + hν,κ (τ, t) (19)
1 + K ν,κ 1+K
ν,κ = [1 0 . . . 0] is a 1 × p vector, with all components,

where hLOS
except for the first one, being set to zero for all ν and κ. K/(1 +
K) is the mean power of the LOS component, and 1/(1 + K)
Fig. 5. SM of 16 QAM 4 × 3 analytical and simulation SER. is the mean power of the random component. K is defined as
the Rician factor.
p is the number of channel paths for each channel link between
each transmit antenna and receive antenna. Each vector gives
B. SC Model (Kronecker Model)
the multipath channel gains between each transmit antenna and
receive antenna as follows: The channel correlation depends on both the environment
 h (τ, t)  and the spacing of the antenna elements. A terminal, which is
1,1 h1,2 (τ, t) ··· h1,Nt (τ, t) surrounded by a large number of local scatterers, can achieve
 h2,1 (τ, t) h2,2 (τ, t) ··· h2,Nt (τ, t)  relatively low correlation values, even if the antennas are sep-
H(τ, t) = 
 .. .. .. .. .

. . . . arated by only half the wavelength. In outdoor base stations,
hNr ,1 (τ, t) hNr ,2 (τ, t) ··· hNr ,Nt (τ, t) the antennas are significantly higher than the scatterers, and a
(16) sufficiently low correlation is likely to require ten wavelengths
between neighboring antenna elements. In indoor base stations,
Here, hν,κ (τ, t) is a channel vector of size p × 1 between the required antenna separation is likely to be between these
receive antenna ν and transmit antenna κ, which contains all two extremes [36].
the multipath channel gains and can be written as To incorporate the SC into the channel model, the correlation
T among channels at multiple elements needs to be calculated.
hν,κ (τ, t) = h1ν,κ (τ1 , t) h2ν,κ (τ2 , t), . . . , hpν,κ (τp , t) . (17) The correlated channel matrix is, then, modeled using the
Kronecker model for its straightforward mathematical descrip-
In this paper, the multipath channels between different links tion [37]. It is recognized that this model does not capture
are taken to be statistically independent and are modeled by the all possible correlation scenarios to their full extent [38], but
Monte Carlo method [32], [33]. An indoor multipath channel is since the interest is a comparative study of different spatial
considered. Each channel path gain is given by multiplexing techniques, its use is deemed justified. Thus
1
Nh
hϕ
ν,κ (τ, t) = √ ρ[ϕ] ej(2πfϕ,q t+θϕ,q ) δ(τ − τϕ ) (18) Hcorr (τ, t) = R1/2
rx H(τ, t)Rtx .
1/2
(20)
Nh q=1
where fϕ,q = fd sin(2πuϕ,q ), θϕ,q , and Nh are the discrete The correlation matrices can be generated using the spatial
Doppler frequencies, the Doppler phases, and the number of channel model [39] or can analytically be computed based on
harmonic functions, respectively. The propagation delay that the power azimuth spectrum (PAS) distribution and array geom-
is related to the ϕth channel path is τϕ . The quantities uϕ,q etry [37]. In this paper, the latter approach is used, assuming
are independent random variables, with each having a uniform uniform linear arrays with isotropic antenna elements at the
distribution in the range (0, 1] for all ϕ = 1, 2, · · · , p and are transmitter and receiver. In addition, a clustered channel model
independently generated for each link. The maximum Doppler is assumed, as shown in Fig. 6, in which groups of scatterers
frequency of the frequency selective multipath channel is given are modeled as clusters that are located around the transmit and
by fd . Finally, the coefficients of the discrete multipath profile receive antennas. The calculation of the correlation matrices at
are modeled by ρ[ϕ]1 [34]. the transmitter and the receiver follows the procedure discussed
in [37], which is derived based on the PAS distribution and the
array geometry. The PAS is modeled as a truncated Laplacian
1 The channel profile that was used in this paper is ρ = distribution over (−π, π], since it best fits the measurement
[1, 0.8487, 0.7663, 0.788, 0.66578, 0.5643, 0.5174, 0.0543, 0.04652]. results in urban and rural areas [40].
Fig. 6. Geometry of the cluster channel model—SC between transmit/receive signals and MC due to signal reradiations.
C. MC Model TABLE II
SIMULATION CONFIGURATIONS. TRANSMISSIONS OF 6 AND
A radio signal that impinges upon an antenna element in- 8 b/s/Hz, USING SM, V-BLAST, AND ALAMOUTI,
ARE CONSIDERED FOR SIMULATION
duces a current in that element, which, in turn, radiates a field
that generates a surface current on the surrounding antenna
elements. This effect is known as MC. Such a coupling in-
fluences the radiation pattern and the antenna correlation. The
parameters that affect MC are element separation, frequency,
and array geometry [20].
The modified MIMO channel, in the presence of MC at delay
time τ , is given by [20]
detection that was used for flat Rayleigh flat-fading channels
Zr H(τ, t)Zt
Hcoup (τ, t) = = Crxr H(τ, t)Ctxr (21) and can be applied on each subchannel [25]. The Alamouti
cr ct
detection for OFDM follows the same principle as in [8] and
where Crxr = Zr /cr and Ctxr = Zt /ct are receiver and trans- [26]. For the Alamouti simulation, a quasistatic channel is
mitter coupling matrices, and Zr and Zt are the overall assumed, which remains constant for an Alamouti codeword
impedance matrices that are shown at the receiver and the period and is equal to the Alamouti codeword length.
transmitter, respectively. ct and cr are normalization factors that In the following, the bit error rate (BER) performance
guarantee that the input and output voltages are the same for of coded and uncoded SM–OFDM, Alamouti–OFDM, and
zero MC. The impedance matrix for an N -element array with V-BLAST–OFDM systems are compared under ideal channel
dipole antenna length dl and dipole radius dr is calculated as in conditions and the combination effects of Rician fading, SC,
[20] and [41]. and MC. In the simulation, all compared systems are selected
such that they achieve the same spectral efficiency, as shown in
Table II. Two spectral efficiencies are of interest: 1) 6 bits/s/Hz,
V. S IMULATION R ESULTS
and 2) 8 bits/s/Hz.
In the simulation, a carrier frequency of 2 GHz with a
20-MHz system bandwidth and 256 OFDM subchannels is
A. Ideal Channel (No Channel Imperfection)
assumed. A time-variant multipath channel with a 0.45-µs max-
imum propagation delay, a 5-Hz Doppler frequency, a 0.5-µs Figs. 7 and 8 show the simulation results under ideal channel
guard interval, and 20 OFDM symbols per frame is considered. conditions for 6- and 8-b/s/Hz transmissions, respectively.
The multipath channels of different links are statistically inde- For both fixed spectral efficiencies, all schemes show approx-
pendent. The total signal power is the same for all transmis- imately similar performance at a low SNR (SNR < 10 dB).
sions. The noise is additive Gaussian, which is spatially and For SM transmission, the 4 × 4 system with a 6-b/s/Hz-
temporally white. Perfect time and frequency synchronization transmission and the 8 × 4 system with an 8-b/s/Hz trans-
is assumed. mission start to show significantly better performance than
The V-BLAST system uses minimum mean square error V-BLAST at SNR > 10 dB, whereas the other SM systems
(MMSE) detection with ordered successive interference can- (i.e., the 2 × 4 system with a 6-b/s/Hz transmission and the
cellation decoding and assumes knowledge of the SNR at 4 × 4 system with an 8-b/s/Hz transmission) in both figures
the receiver. The substream with the strongest SNR is first show better performance gains than V-BLAST at SNR > 20 dB.
detected, followed by demodulation and subtraction from the This can be explained due to the use of a lower modula-
initial signal. The detected substream is nulled, and the process tion order in the first set of systems as compared to the
is iteratively repeated for all other substreams. In addition, the other set. This also explains the behavior of the Alamouti
V-BLAST detection for OFDM is the same as the V-BLAST scheme, where Alamouti shows poor performance as compared
Fig. 7. BER versus SNR for the case of a 6-b/s/Hz transmission (ideal Fig. 9. In SM systems, signal modulation and SM can be traded off to achieve
channel). better performance. The figure shows that antenna error dominates for large
SNR values, and hence, the 4 × 4 SM system performs worse than the 2 × 4
SM system.
which is the upper bound, as discussed in Section III. However,

at a higher SNR, the performance of SM systems starts to
be poorer than that of MRRC. The 4 × 4 SM system starts
to deviate from MRRC at an SNR of about 16 dB, and at
a BER of 10−5 , it loses around 5 dB SNR, as compared to
MRRC. However, the 2 × 4 SM system demonstrates better
behavior and achieves almost the same performance as MRRC
up to an SNR of 23 dB. However, the performance slightly
degrades at SNR > 23 dB. This behavior can be explained
by the existence of two estimation processes in SM: 1) the
antenna number and 2) the transmitted symbol. For a low SNR,
the error is dominated by the estimation of the transmitted
symbol. For a high SNR, however, the error is dominated by
the estimation of the antenna number. Therefore, the likelihood
of erroneous antenna detection increases with an increasing
Fig. 8. BER versus SNR for the case of an 8-b/s/Hz transmission (ideal
channel).
number of transmit antennas. Hence, the 2 × 4 SM system
performs better than the 4 × 4 SM system at a high SNR. This
also explains the crossing points of the BER curves of the SM
to SM and V-BLAST. The main reason behind this degradation systems in both Figs. 7 and 8.
in Alamouti’s performance is the use of a higher constellation This is a significant finding, which opens a new area of
size to achieve the same spectral efficiency as in SM and adaptive modulation, in which the number of transmit antennas
V-BLAST. The performance of SM in both Figs. 7 and 8 and the size of the constellation diagram can be traded off
demonstrates a major enhancement over V-BLAST and the to achieve better performance. For instance, if the channel is
Alamouti scheme. For example, at a BER of 10−3 and for a bad (resulting in a low SNR at the receiver) and the antennas
6-b/s/Hz transmission, SM outperforms V-BLAST by around are almost uncorrelated, it is better to use a smaller signal
3 dB in the best case and also outperforms Alamouti by around constellation size and more transmit antennas. On the other
10 dB. At the same BER and for an 8-b/s/Hz transmission, hand, if the channel is good, then fewer antennas and a larger
8 × 4 and 4 × 4 SM systems outperform V-BLAST by signal constellation size achieve better performance.
around 4 dB and 1 dB, respectively. The Alamouti scheme
demonstrates poor performance for the 8-b/s/Hz transmission,
B. Uncoded System With All Imperfections
and the BER was still greater than 10−3 , even at SNR = 30 dB.
The particular behavior of SM systems is further analyzed, and The effect of all channel imperfections on the performance
the results are presented in Fig. 9. of SM–OFDM, Alamouti–OFDM, and V-BLAST–OFDM is
In Fig. 9, the BER performance of SM and MRRC, under studied in the following. The Rice factor K in (19) is set to 2 in
similar modulation orders and the same number of receiving the presence of LOS. In an indoor environment, the measured
antennas, is depicted. At a relatively low SNR, the performance values of K are close to the selected value in this paper [42].
of SM systems approaches that of the corresponding MRRC, SC is present in all systems and at both the transmitter and
and depends on the constellation size that was used, as well

as the value of the Rician factor K [10], [11], [43]. The first
important observation is that, for the 6-b/s/Hz transmission
(Fig. 10), Alamouti performs slightly worse than SM at a low
SNR but has an almost-similar performance at a high SNR. In
addition, due to full diversity gains, Alamouti is robust to the
presence of SC and MC imperfections [44], [45]. Furthermore,
SM performs 1 dB better than Alamouti for SNR < 25 dB.
The gain of SM over Alamouti for the 8-b/s/Hz transmission
(Fig. 11) is around 5 dB at a BER of 10−2 .
For a 6-b/s/Hz transmission and a BER of 10−2 , the com-
bined effect degrades the 4 × 4 SM performance by about 5 dB
and the 2 × 4 SM performance by 3 dB. The effect on
V-BLAST’s performance is more severe, and a loss in SNR
of approximately 8 dB at a BER of 10−2 is noticed. For an
8-b/s/Hz transmission at the same BER, the 4 × 4 and the
Fig. 10. BER versus SNR for the case of a 6-b/s/Hz transmission (combined 8 × 4 SM systems suffer from 2- and 5-dB losses in SNR, re-
channel imperfections).
spectively. Again, the effect on V-BLAST is more pronounced,
and a loss in SNR of approximately 10 dB is noticed. The high
loss in V-BLAST’s performance in the presence of all channel
imperfections explains why the Alamouti scheme has a better
performance than V-BLAST in such channel conditions. At a
BER of 10−2 , SM outperforms V-BLAST by about 8 dB in both
6- and 8-b/s/Hz transmissions.
In Figs. 10 and 11, both SM schemes show approximately the
same BER versus SNR behavior. This is in contrast to previous
results, in which SM systems with a low number of transmit
antennas performed better than SM systems with a high number
of transmit antennas. An explanation is given by noting that
the degradation of SM systems, as discussed in the previous
paragraph, in the presence of all channel imperfections, in-
creases with an increasing number of transmit antennas. This
means that the presence of all channel impairments affects the
detection of the transmit antenna number to a greater extent.
Therefore, the BER improvement for SM systems with a low
Fig. 11. BER versus SNR for the case of an 8-b/s/Hz transmission (combined number of transmit antennas at a high SNR is considerably
channel imperfections).
less marked than for an ideal channel. For this reason, the SM
systems perform almost the same at a high SNR.
the receiver arrays. Antenna spacing and the angular values
for the simulation of SC are taken from the Third-Generation
C. Coded System With All Channel Imperfections
Partnership Project (3GPP) spatial channel model [39]. The
transmit array with an element spacing of 0.5λ, mean angle of For coded systems, we consider a convolutional channel
arrival of 20◦ , and an angular spread (AS) of 35◦ is assumed. encoder, followed by a block interleaver. Each transmitted
The receive array has an element spacing of 10λ, a mean OFDM frame is independently encoded by the channel encoder
angle of departure of 20◦ , and an AS of 5◦ . The MC effect is and is, then, interleaved by the block interleaver. A nonrecursive
negligible for an antenna spacing that is beyond λ. Therefore, rate of 1/2 convolutional encoder, with an overall constraint
MC is only present at the transmitter in all systems. To simulate length of 3, is used. The data that were received after OFDM
the MC matrix, a uniform linear array with identical dipole demodulation are block deinterleaved and are, then, decoded
antenna elements with an isotropic radiation pattern is assumed. using a hard Viterbi decoder.
A dipole antenna length of 0.5λ and a radius of 3.33 × 10−3 λ The BER performance of coded SM, Alamouti, and
are considered. V-BLAST with all channel imperfections is depicted in Figs. 12
The effect of all channel imperfections on the performance and 13 for 6- and 8-b/s/Hz transmissions, respectively.
of SM, V-BLAST, and Alamouti is depicted in Figs. 10 and 11 First, a coding gain of approximately 5 dB is evident in both
for 6- and 8-b/s/Hz transmissions, respectively. figures, as compared to the previous uncoded results. Second,
Combined imperfections degrade the performance of SM and the SM results show the best performance, and they outperform
V-BLAST, whereas it enhances Alamouti’s performance. The both Alamouti and V-BLAST. For the case of the 8-b/s/Hz
enhancement in Alamouti’s performance, as compared to the transmission, as shown in Fig. 13, SM outperforms Alamouti
ideal case, comes from the presence of the LOS component by 5 dB and outperforms V-BLAST by 7 dB at a BER of 10−3 ,
in [48] reduces the receiver complexity by 30% for a 4 × 4

system and by 20% for a 2 × 4 system. It will be shown in
the following that SM yields a greater reduction in receiver
complexity than the variants of V-BLAST. In particular, SM
has around 90% and 80% reduction in receiver complexity,
as compared to MMSE V-BLAST and to V-BLAST-based
QR decomposition, respectively, without sacrificing spectral
efficiency. Furthermore, as shown in Section V, SM results in
better BER performance for the studied scenarios which do
incorporate realistic assumptions regarding the channel. The
Alamouti and STCs techniques, on the other hand, are well
known to achieve very low receiver complexity. It will be
shown that SM achieves complexity that is comparable to those
techniques, while superior BER performance is reported in
Section V.
Only multiplication and addition of complex numbers are
Fig. 12. BER versus SNR for the case of a 6-b/s/Hz transmission (coded considered as operations. The MMSE criterion requires two
combined channel imperfections). matrix multiplications: 1) one inversion and 2) one addition.
It is assumed that the matrix transposes are not explicitly
computed. The first matrix multiplication requires Nt2 Nr op-
erations, the matrix addition requires Nt2 operations, and the
matrix inverse needs 4Nt3 (using Gaussian elimination) op-
erations [50], [51]. The second matrix multiplication takes
Nt3 operations. Therefore, a total of (5Nt3 + Nr Nt2 + Nt2 )
complex operations are needed. For V-BLAST, these steps
are repeated for i = 1, . . . , Nt . This means that the inverse is
computed for a deflated matrix, with decreasing dimensions
of (Nr × (Nt − ξ)), ξ = 0, . . . , Nt − 1. As a result, the total
number of complex operations is given by

Nt
(5i3 + Nr i2 + i2 ).
i=1
For V-BLAST with QR decomposition [52], the computa-

tional time is dominated by finding the QR factorization.
Fig. 13. BER versus SNR for the case of an 8-b/s/Hz transmission (coded The QR decomposition can be done by directly applying the
combined channel imperfections). Householder unitary transformations, which costs 2Nt2 (Nr −
(Nt /3)) complex operations [51]. Filtering the received vector
respectively. For the 6-b/s/Hz transmission, SM and Alamouti through QH requires Nr2 operations. For backward and forward
have almost the same performance at a high SNR, whereas SM sweeps, twice this number of complex operations is needed.
outperforms Alamouti by about 1 or 2 dB for SNR < 23 dB. Thus, a total of 2(2Nt2 (Nr − (Nt /3)) + Nr2 ) complex opera-
Finally, SM outperforms V-BLAST by about 7 dB in the best tions are needed.
case at a BER of 10−4 , as shown in Fig. 13. SM applies MRRC at the receiver. MRRC needs Nt Nr
complex multiplications and Nt (Nr − 1) complex additions.
VI. R ECEIVER C OMPLEXITY Thus, a total of
In this section, we compare the complexity (the number (2Nt Nr − Nt )

of complex operations) of the V-BLAST, Alamouti, and SM
algorithms that were presented in this paper. There are many complex operations are required.
variants of the original V-BLAST algorithm, some of which, in The Alamouti detection process requires Nr2 complex multi-
fact, result in reduced receiver complexity. These algorithms are plications and 2(2Nr − 1) complex additions for two transmit-
based on QR decomposition [46], [47], Gram–Schmitt Orthog- ted symbols. Therefore, a total of 0.5(Nr2 + 4Nr − 2) complex
onalization, which substitutes the computation of pseudoin- operations are required at the receiver for the detection of one
verse in finding the weight vectors [48], and the use of a OFDM subchannel.
recursive matrix update, as proposed in [49]. However, the The required number of complex operations per OFDM
reduction in receiver complexity is only significant for a large subchannel for transmitting six information bits at a time
number of transmit and receive antennas, whereas it is in- for SM–OFDM, Alamouti–OFDM, and V-BLAST–OFDM are
ferior for small numbers. For instance, the proposed scheme shown in Table III and Fig. 14.
TABLE III gram, APL Submarine Technology Department, John Hopkins

RECEIVER COMPLEXITY COMPARISON FOR A 6-b/s/Hz TRANSMISSION
University, for his very useful comments.
R EFERENCES
[1] E. Telatar, “Capacity of multi-antenna Gaussian channels,” Eur. Trans.
Telecommun., vol. 10, no. 6, pp. 558–595, Nov./Dec. 1999.
[2] G. J. Foschini and M. J. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” in Wirel. Pers.
Commun., vol. 1.6. Norwell, MA: Kluwer, 1998, pp. 311–335.
[3] G. J. Foschini, “Layered space-time architecture for wireless communi-
cation in a fading environment when using multi-element antennas,” Bell
Labs Tech. J., vol. 1, no. 2, pp. 41–59, Sep. 1996.
[4] Y. J. Zhang and K. Letaief, “Adaptive resource allocation for multiaccess
MIMO/OFDM systems with matched filtering,” IEEE Trans. Commun.,
vol. 53, no. 11, pp. 1810–1816, Nov. 2005.
[5] Y. Li, J. Winters, and N. Sollenberger, “MIMO-OFDM for wireless com-
munications: Signal detection with enhanced channel estimation,” IEEE
Trans. Commun., vol. 50, no. 9, pp. 1471–1477, Sep. 2002.
[6] IEEE 802.11a, IEEE Part II: Wireless LAN Medium Access Control
(MAC) and Physical Layer (PHY) Speciﬁcations: High-Speed Physical
Fig. 14. Receiver complexity comparison for the 6-b/s/Hz transmission that
Layer in the 5-GHz Band, 1999. Retrieved Oct. 10, 2006 from [Online].
uses MMSE V-BLAST, V-BLAST-based QR decomposition, SM, and the
Available: http://standards.ieee.org/getieee802/802.11.html
Alamouti algorithm.
[7] IEEE 802.16a, Local and Metropolitan Area Networks—Part 16: Air
VII. C ONCLUSION Interface for Fixed Broadband Wireless Access Systems, 2001. Retrieved
Oct. 10, 2006 from [Online]. Available: http://standards.ieee.org/catalog/
A practical multiple antenna transmission approach, called olis/lanman.html
[8] S. Alamouti, “A simple transmit diversity technique for wireless commu-
SM, has been applied to OFDM and has been presented in this nications,” IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp. 1451–1458,
paper. The closed-form analytical performance of SM in i.i.d. Oct. 1998.
Rayleigh flat-fading channels has been derived, and analytical [9] V. Tarokh, N. Seshadri, and A. Calderbank, “Space-time codes for
high data rate wireless communication: Performance criterion and code
and simulation results were shown to closely agree. The SM construction,” IEEE Trans. Inf. Theory, vol. 44, no. 2, pp. 744–765,
scheme was shown to be more robust to the presence of Rician Mar. 1998.
fading, SC, and MC, as compared to V-BLAST. In addition, [10] H. El Gamal, “On the robustness of space-time coding,” IEEE Trans.
Signal Process., vol. 50, no. 10, pp. 2417–2428, Oct. 2002.
when comparing the performance of SM and Alamouti, SM [11] M. Godavarti, T. Marzetta, and S. Shamai, “Capacity of a mobile multiple-
was shown to have a better or, in the worst case, a similar antenna wireless link with isotropically random Rician fading,” IEEE
performance for all the simulated cases. The performance of Trans. Inf. Theory, vol. 49, no. 12, pp. 3330–3334, Dec. 2003.
[12] P. Viswanath, D. Tse, and V. Anantharam, “Asymptotically optimal wa-
SM degrades in the presence of a LOS component, whereas the ter filling in vector multiple-access channels,” IEEE Trans. Inf. Theory,
performance of Alamouti improves in such channel conditions. vol. 47, no. 1, pp. 241–267, Jan. 2001.
Another limitation of SM is the increase in spectral efficiency [13] G. Raleigh and J. Cioffi, “Spatio-temporal coding for wireless communi-
cation,” IEEE Trans. Commun., vol. 46, no. 3, pp. 357–366, Mar. 1998.
by the base-two logarithm of the total number of transmit anten- [14] P. Wolniansky, G. Foschini, G. Golden, and R. Valenzuela, “V-BLAST:
nas, as compared to a linear increase for the V-BLAST system. An architecture for realizing very high data rates over the rich-
The obtained gain in the SNR for the same spectral efficiency scattering wireless channel,” in Proc. URSI ISSSE, Sep. 29–Oct. 2, 1998,
pp. 295–300.
of SM over V-BLAST at a BER of 10−4 varies from 3 dB in [15] S. Haykin and M. Moher, Modern Wireless Communications.
ideal channel conditions to 8 dB in the presence of all channel Englewood Cliffs, NJ: Prentice-Hall, 2003.
imperfections, along with channel coding. In addition, around [16] H. Jafarkhani, Space-Time Coding: Theory and Practice. Cambridge,
U.K.: Cambridge Univ. Press, 2005.
90% of reduction in receiver complexity is achieved. As com- [17] A. Goldsmith, S. Jafar, N. Jindal, and S. Vishwanath, “Capacity limits of
pared to Alamouti, different performance gains in SNR can be MIMO channels,” IEEE J. Sel. Areas Commun., vol. 21, no. 5, pp. 684–
noticed. In the presence of all channel imperfections, along with 702, Jun. 2003.
[18] M. Damen, A. Abdi, and M. Kaveh, “On the effect of correlated fading
channel coding, SM gains 1 and 5 dB in SNR over Alamouti for on several space-time coding and detection schemes,” in Proc. IEEE 54th
6- and 8-b/s/Hz transmissions, respectively. Moreover, SM and Veh. Technol. Conf., Oct. 7–11, 2001, vol. 1, pp. 13–16.
Alamouti have comparable receiver complexity. An additional [19] M. Chiani, M. Win, and A. Zanella, “On the capacity of spatially corre-
lated MIMO Rayleigh-fading channels,” IEEE Trans. Inf. Theory, vol. 49,
significant advantage of SM is the freedom to work with any no. 10, pp. 2363–2371, Oct. 2003.
system configuration, even for the case where there are more [20] T. Svantesson and A. Ranheim, “Mutual coupling effects on the capacity
transmit antennas than receive antennas. of multielement antenna systems,” in Proc. IEEE ICASSP, May 7–11,
2001, vol. 4, pp. 2485–2488.
Future work will concentrate on the investigation of adaptive [21] S. Ganesan, R. Mesleh, H. Haas, C. W. Ahn, and S. Yun, “On the per-
algorithms to trade off “SM” against signal modulation based formance of spatial modulation OFDM,” in Proc. 40th Asilomar Conf.
on the actual signal-to-interference-plus-noise ratio, as well as Signals, Syst. Comput., Oct. 29–Nov. 1, 2006, pp. 1825–1829.
[22] R. Mesleh, H. Haas, C. W. Ahn, and S. Yun, “Spatial modulation—A
new antenna number detection algorithms. new low-complexity spectral efficiency enhancing technique,” in Proc.
CHINACOM, Oct. 25–27, 2006, pp. 1–5.
[23] R. Mesleh, H. Haas, C. W. Ahn, and S. Yun, “Spatial modulation–
ACKNOWLEDGMENT OFDM,” in Proc. 11th InOWo, Aug. 30–31, 2006, pp. 288–292.
[24] A. Tonello, “Space-time bit-interleaved coded modulation with an it-
The authors would like to thank Prof. McDonough, who erative decoding strategy,” in Proc. IEEE 52nd Veh. Technol. Conf.,
was the former vice chair of the Electrical Engineering Pro- Sep. 24–28, 2000, vol. 1, pp. 473–478.
[25] R. Piechocki, P. Fletcher, A. Nix, C. Canagarajah, and J. McGeehan, [52] M. Damen, K. Abed-Meraim, and S. Burykh, “Iterative QR detection for
“Performance evaluation of BLAST-OFDM enhanced hiperlan/2 using BLAST,” Wirel. Pers. Commun., vol. 19, no. 3, pp. 179–191, Dec. 2001.
simulated and measured channel data,” Electron. Lett., vol. 37, no. 18, [53] H. Haas and S. McLaughlin, eds., Next Generation Mobile Access
pp. 1137–1139, Aug. 30, 2001. Technologies: Implementing TDD. Cambridge, U.K.: Cambridge Univ.
[26] K. Lee and D. Williams, “A space-frequency transmitter diversity tech- Press, Jan. 2008.
nique for OFDM systems,” in Proc. IEEE GLOBECOM, Nov. 27–
Dec. 1, 2000, vol. 3, pp. 1473–1477.
[27] BROADCOM Corp., 802.11n: Next-Generation Wireless LAN Technol-
ogy, Apr. 2006. White paper, retrieved Aug. 4, 2006. [Online]. Available:
http://www.broadcom.com/docs/WLAN/802-11n-WP100-R.pdf
[28] M.-S. Alouini and A. Goldsmith, “A unified approach for calculating error Raed Y. Mesleh (S’00–M’08) received the B.Sc.
rates of linearly modulated signals over generalized fading channels,” degree in communication engineering from Yarmouk
IEEE Trans. Commun., vol. 47, no. 9, pp. 1324–1334, Sep. 1999. University, Irbid, Jordan, in 2000, the M.Sc. degree
[29] J. G. Proakis, Digital Communications, 4th ed. New York: McGraw- in communication technology from Ulm University,
Hill, 2000. Ulm, Germany, in 2004, and the Ph.D. degree in elec-
[30] G. Casella and R. L. Berger, Statistical Inference, 2nd ed. ser. Duxbury trical engineering from Jacobs University, Bremen,
Advanced. Pacific Grove, CA: Duxbury, 2002. Germany, in 2007.
[31] M. Sellathurai and S. Haykin, “Turbo-BLAST for wireless communica- From 2000 to 2002, he was with Orange—a
tions: Theory and experiments,” IEEE Trans. Signal Process., vol. 50, mobile operator company in Amman, Jordan—as
no. 10, pp. 2538–2546, Oct. 2002. a Systems Engineer. From 2002 to 2004, he was
[32] M. Pätzold, Mobile Fading Channels. Chichester, U.K.: Wiley, 2002. partly with LogicaCMG as a protocol and testing
[33] P. Höher, “A statistical discrete-time model for the WSSUS multi- engineer. Since September 2007, he has been with the School of Electrical
path channel,” IEEE Trans. Veh. Technol., vol. 41, no. 4, pp. 461–468, Engineering and Computer Science, Jacobs University, where he is currently
Nov. 1992. a Postdoctoral Fellow. His research interests include wireless communications
[34] J. Medbo and P. Schramm, Channel Models for HIPERLAN 2, 1998. and communication signal processing.
ETSI/BRAN Document No. 3ERIO85B. Dr. Mesleh received the Best Student Awards from Yarmouk University and
[35] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Ulm University.
Wireless Communications. Cambridge, U.K.: Cambridge Univ. Press,
2003.
[36] A. Hottinen, O. Tirkkonen, and R. Wichman, Multi-Antenna Tansceiver
Techniques for 3G and Beyond. Hoboken, NJ: Wiley, 2003.
[37] A. Forenza, D. Love, and R. Heath, Jr., “A low complexity algo- Harald Haas (S’98–A’00–M’03) received the Ph.D.
rithm to simulate the spatial covariance matrix for clustered MIMO
degree from the University of Edinburgh, Edinburgh,
channel models,” in Proc. IEEE VTC—Fall, May 17–19, 2004, vol. 2,
U.K., in 2001.
pp. 889–893.
From 2001 to 2002, he was a Project Manager
[38] H. Özcelik, M. Herdin, W. Weichselberger, J. Wallace, and E. Bonek, “De-
with Siemens AG (Information and Communica-
ficiencies of ‘Kronecker’ MIMO radio channel model,” Electron. Lett.,
tion Mobile Networks) for an international research
vol. 39, no. 16, pp. 1209–1210, Aug. 2003.
project on new radio access technologies for beyond
[39] 3GPP, “Spatial Channel Model for Multiple-Input Multiple-Output third-generation (3G) wireless systems. The project
(MIMO) simulations (Release 6),” 3GPP TR 25.996 V 6.1.0(2003-09),
involved several Chinese and German universities.
2003. Retrieved Sep. 1, 2006 [Online]. Available: www.3gpp.org/specs/
In September 2002, he joined the International Uni-
[40] K. Pedersen, P. Mogensen, and B. Fleury, “A stochastic model of the
versity Bremen (now Jacobs University), Bremen,
temporal and azimuthal dispersion seen at the base station in outdoor Germany, where he is currently an Associate Professor of electrical engineer-
propagation environments,” IEEE Trans. Veh. Technol., vol. 49, no. 2,
ing. In June 2007, he joined the Institute for Digital Communications, Univer-
pp. 437–447, Mar. 2000.
sity of Edinburgh. He is a coauthor of the Handbook of Information Security
[41] C. A. Balanis, Antenna Theory Analysis and Design. New York:
(Wiley) and Next Generation Mobile Access Technologies: Implementing TDD
Wiley, 1997. (Cambridge University Press). His work on optical wireless communication
[42] M. Carroll and T. Wysocki, “Fading characteristics for indoor wireless
was selected for publication in “100 Produkte der Zukunft” by T. W. Hänsch.
channels at 5-GHz unlicensed bands,” in Proc. IEEE Joint 1st Workshop
He is the holder of several patents in wireless communications. His research
SympoTIC, Oct. 26–28, 2003, pp. 102–105.
interests include wireless systems engineering and digital signal processing,
[43] S. Parker, M. Sandell, and M. Lee, “The performance of space-time codes in particular, MAC protocols, multiuser access, link adaptation, scheduling,
in office environments,” in Proc. 57th IEEE Semiannual VTC, Apr. 22–25,
dynamic resource allocation for TDD-based systems, multiple antenna systems,
2003, vol. 1, pp. 741–745.
and optical wireless communication.
[44] Y. Kim, H. Kang, and K. Kim, “Performance of space-time block codes
Dr. Haas received the Best Paper Award at the 10th International Symposium
with QAM under the spatially correlated channels,” in Proc. IEEE 59th
on Personal, Indoor, and Mobile Radio Communications (PIMRC 1999),
VTC—Spring, May 17–19, 2004, vol. 2, pp. 670–673.
Osaka/Japan.
[45] A. Abouda, H. El-Sallabi, and S. Haggman, “Effect of mutual coupling on
BER performance of Alamouti scheme,” in Proc. IEEE Antennas Propag.
Soc. Int. Symp., 2006, pp. 4797–4800.
[46] D. Wubben, R. Bohnke, V. Kuhn, and K.-D. Kammeyer, “MMSE exten-
sion of V-BLAST based on sorted QR decomposition,” in Proc. IEEE
VTC—Fall, Oct. 6–9, 2003, vol. 1, pp. 508–512. Sinan Sinanović received the B.S.E.E. degree
[47] K. Kusume, M. Joham, and W. Utschick, “MMSE block (summa cum laude) from Lamar University,
decision-feedback equalizer for spatial multiplexing with reduced Beaumont, TX, and the M.S. and Ph.D. degrees in
complexity,” in Proc. IEEE GLOBECOM, Nov. 29–Dec. 3, 2004, vol. 4, electrical and computer engineering in 2006 from
pp. 2540–2544. Rice University, Houston, TX.
[48] W. K. Wai, C.-Y. Tsui, and R. Cheng, “A low-complexity architecture of In 2006, he was with Jacobs University, Bremen,
the V-BLAST system,” in Proc. IEEE WCNC, Sep. 23–28, 2000, vol. 1, Germany, as a Postdoctoral Fellow. In 2007, he
pp. 310–314. joined the University of Edinburgh, Edinburgh, U.K.,
[49] J. Benesty, Y. Huang, and J. Chen, “A fast recursive algorithm for opti- where he is currently a Research Fellow with the
mum sequential signal detection in a BLAST system,” IEEE Trans. Signal Institute for Digital Communications. He was with
Process., vol. 51, no. 7, pp. 1722–1730, Jul. 2003. Texas Instruments, Dallas, TX, working on ASIC for
[50] N. J. Higham and R. S. Schreiber, “Fast polar decomposition of an ar- the central office modem. While working for Halliburton Energy Services,
bitrary matrix,” SIAM J. Sci. Stat. Comput., vol. 11, no. 4, pp. 648–655, Houston, he developed an acoustic telemetry receiver. He is the holder of a
Jul. 1990. patent for a directional acoustic telemetry receiver (U.S. patent 7 158 446).
[51] G. H. Golub and C. F. V. Loan, Matrix Computations, 3rd ed. Baltimore, Dr. Sinanović is a member of the Tau Beta Pi and Eta Kappa Nu Electrical
MD: Johns Hopkins Univ. Press, 1996. Engineering Honor Societies.
Chang Wook Ahn (S’02–M’05) received the B.S. Sangboh Yun (M’97–S’99–M’02) received the B.S.,
and M.S. degrees in electrical engineering from M.S., and Ph.D. degrees in electrical engineering
Korea University, Seoul, Korea, in 1998 and 2000, from Korea University, Seoul, Korea in 1994, 1998,
respectively, and the Ph.D. degree from Gwangju In- and 2006, respectively.
stitute of Science and Technology (GIST), Kwangju, From 1994 to 2000, he was with Daewoo Telecom,
Korea, in 2005. Inc., as a Research Engineer. From 2000 to 2001, he
In 2003, he was a Visiting Scholar with the was the Chief Technical Officer with NeoSolution,
Illinois Genetic Algorithms Laboratory (IlliGAL), Inc., where he is also a Cofounder. In 2001, he
University of Illinois, Urbana-Champaign. From joined Samsung Advanced Institute of Technology,
2005 to 2007, he was a Research Staff Member with Kiheung, Korea, as a member of Technical Staff.
the 4G Communication Research Group, Samsung In 2006, he joined the Telecommunication R&D
Advanced Institute of Technology. From 2007 to 2008, he was a Research Center, Samsung Electronics Company, Ltd., Suwon. His research interests
Professor with GIST. Since 2008, he has been with the Department of Computer include IMT-advanced wireless communication systems, multihop relay, and
Engineering, Sungkyunkwan University, Suwon, Korea, as a Faculty Member. radio resource management.
His research interests include intelligent wireless networks, machine learning,
and evolutionary algorithms.

Spatial Modulation

Uploaded by

Spatial Modulation

Uploaded by

2228 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO.

Abstract—Spatial modulation (SM) is a recently developed I. I NTRODUCTION

0018-9545/$25.00 © 2008 IEEE

Fig. 1. SM–OFDM system model.

and four transmit antennas are considered as an example here. TABLE I

m̃ = log2 (Nt ) + m. (1)

derivation, and the result is later generalized to any number of

Gaussian distribution with mean µi and variance σn2 . Then, the

intersection areas of fX(4) (x|µi , σn2 ) with all other distributions

For any number of transmit antennas, (14) can be written as

A. Rician Fading Channel

ν,κ = [1 0 . . . 0] is a 1 × p vector, with all components,

which is the upper bound, as discussed in Section III. However,

and depends on the constellation size that was used, as well

in [48] reduces the receiver complexity by 30% for a 4 × 4

For V-BLAST with QR decomposition [52], the computa-

In this section, we compare the complexity (the number (2Nt Nr − Nt )

TABLE III gram, APL Submarine Technology Department, John Hopkins

You might also like

Spatial Modulation

Uploaded by

Spatial Modulation

Uploaded by

2228 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO.

Abstract—Spatial modulation (SM) is a recently developed I. I NTRODUCTION

0018-9545/$25.00 © 2008 IEEE

Fig. 1. SM–OFDM system model.

and four transmit antennas are considered as an example here. TABLE I

m̃ = log2 (Nt ) + m. (1)

derivation, and the result is later generalized to any number of

 Gaussian distribution with mean µi and variance σn2 . Then, the

intersection areas of fX(4) (x|µi , σn2 ) with all other distributions

For any number of transmit antennas, (14) can be written as

A. Rician Fading Channel

ν,κ = [1 0 . . . 0] is a 1 × p vector, with all components,

which is the upper bound, as discussed in Section III. However,

and depends on the constellation size that was used, as well

in [48] reduces the receiver complexity by 30% for a 4 × 4

For V-BLAST with QR decomposition [52], the computa-

In this section, we compare the complexity (the number (2Nt Nr − Nt )

TABLE III gram, APL Submarine Technology Department, John Hopkins

You might also like

Gaussian distribution with mean µi and variance σn2 . Then, the