This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LWC.2017.2753800, IEEE
Wireless Communications Letters
1

A Space-Time Transmission Scheme for Large

MIMO Systems
Arti M.K., Member, IEEE
Abstract— In this letter, we propose a space-time transmission time slots for large MIMO systems), iv) STBCs are designed
scheme (STTS) for large multiple-input multiple-output (MIMO) under the constraint of full spatial diversity which is very
communication system. These systems are primarily used to large (almost infinity) for very large dimensions and practically
transmit a huge amount of data. The existing space-time block
codes (STBCs) are not suitable for these systems due to the not needed for large MIMO systems. OSTBCs compromise
fact that these STBCs require very large channel coherence data rate for high spatial dimensions; whereas, non-orthogonal
time (more than the number of transmit antennas), render poor STBCs sacrifice decouple decoding. However, the proposed
data rate (less than one), and aim at full spatial diversity. In scheme compromises diversity order to obtain good data rate
practical large MIMO system with one thousand transmit and and decouple decoding; which is prime requirement of large
one thousand receive antennas, the diversity order of 106 can
be achieved; however, this much high value of diversity order MIMO systems. The proposed scheme may be useful for base
is not required in practical systems. Therefore, a STTS with station (BS) to BS communication.
good data rate, parsimoniously utilized temporal dimensions, and Because of the aforementioned facts, existing STBCs are
symbol-wise decoding is essentially required for practical large mostly not suitable for large MIMO systems. For large MIMO
MIMO systems. In this paper, we emphasize over design of a systems, a space-time transmission scheme (STTS) is required
new STTS by relaxing the constraint of achieving full spatial
diversity. In the proposed STTS, the data rate of the large with high data rate and low decoding complexity. The diversity
MIMO system is increased at the expense of the reduction in order of large MIMO system can be compromised according
spatial diversity order. An approximate expression of symbol to the practical requirements (i.e., required quality-of-service).
error rate of the proposed system is obtained–based on the Therefore, in this letter, we propose a STTS design for
moment generating function of the instantaneous received signal- large MIMO systems. The decoding of transmitted symbols
to-noise ratio. Further, it is shown that the diversity order of the
proposed STTS based system is one less than the number of is performed by using the concept of left null space. It is
receive antennas. shown that symbol-wise decoding is possible by employing
the proposed STTS in large MIMO system. The performance
Index Terms— Space-time transmission scheme, Large MIMO
systems, Performance analysis, M -ary phase shift keying (M - of the proposed STTS based large MIMO system is discussed
PSK), Rayleigh fading, symbol error rate. in terms of symbol error rate (SER) for M -PSK constellation
I. I NTRODUCTION and diversity gain.
A growing demand of data rate requires implementation II. S YSTEM M ODEL AND P ROPOSED STTS
of large multiple-input multiple-output (MIMO) systems, in We consider a large MIMO communication system with Nt
practice. In order to meet the high data rate requirements and antennas at the transmitter and Nr antennas at the receiver,
to conserve energy, a higher spectral efficiency can be obtained where Nt and Nr take very large values. It is assumed that
by using these systems [1]–[5]. Another attractive feature of the perfect channel state information (CSI) is available at the
large MIMO is that a large number of antennas put together receiver. Moreover, exact values of the noise variances are
to transmit a very big amount of data to receiver on the same assumed to be available at the receiver.
time-frequency resources. Space-time block code (STBC) is a
A. STTS for Eight Transmit Antennas
well known strategy for MIMO systems to transmit different
symbols from different antennas in certain time-slots [6]. Let us first take an example of STTS for eight transmit
An omnidirectional STBC design in large MIMO systems to antennas for simplicity. In next subsection, a general STTS
provide omnidirectional signaling with spatial diversity is pre- for Nt transmit antennas is presented. In the proposed scheme,
sented in [7]. An orthogonal STBC (OSTBC) design for large four time slots are needed to transmit eight symbols from eight
MIMO systems by using diminutive OSTBCs is discussed in antennas as follows:
 
[8]. An OSTBC for 32 transmit antennas with real symbols is x1 x1 −x1 x1
presented, in [9], with a code rate of 0.25. STBC has following  x2 x2 −x2 x2 
 
problems in context of large MIMO systems: i) symbol-wise  x3 −x3 −x3 x3 
 
decoding based STBCs do not provide high data rate (rate in  x4 −x4 −x4 x4 
X=  , (1)
general less than one), ii) if high rate STBCs are used, then  x5 −x5 x5 x5 
decoding complexity is very high which is unaffordable for  x6 −x6 x6 x6 
 
very large dimensions, iii) STBCs use temporal dimensions  x7 −x7 x7 −x7 
liberally (temporal dimensions must be more than transmit x8 −x8 x8 −x8
antennas), which requires a significantly large channel coher- where xi s, i ∈ {1, · · · , 8}, are complex symbols from
ence time (i.e., channel must remain constant for thousands M −PSK constellations. In (1), rows of the matrix X show
Arti M.K. is with the Department of Electronics Engineering, YMCA transmit antennas and columns stand for time slots. The
University of Sciences and Technology, Faridabad, Haryana-121006, India complex signal received at the receiver is given by
(e-mail: arti [email protected]).
Y = HX + N , (2)

2162-2337 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LWC.2017.2753800, IEEE
Wireless Communications Letters
2

where H ∈ C Nr ×Nt (C is set of complex numbers) is In this case, B2 is a 7×8 matrix. By performing the whitening
the MIMO channel in between transmitter and receiver. The of noise terms present in (7), we have
entries of H, i.e., channel coefficients are assumed to be  
−1/2 y1 + y2 −1/2
Gaussian distributed with zero mean and unit variance. Further r1 = K1 B2 = K1 B2 h1 x1 + w1 , (8)
it is also assumed that all channel coefficients are independent 2
and identically distributed (i.i.d.); N is additive white Gaussian where w1  is AWGN vector with zero

mean
 and unit variance,
n1 +n2 n 1 +n2
H
noise (AWGN) matrix containing elements with σ 2 variance


K1 = E B2 2 B2 2 , and E(·) denotes
and zero mean. After some algebra, (1) can be rewritten as expectation operator. It can be shown through simulations that
    the value of B2 BH
2 = I, where I denotes an identity matrix (in
y1 y2 y3 y4 = h1 h2 h3 h4 h5 h6 h7 h8
  the considered example, B2 BH 2 is the identity matrix of size
x1 x1 −x1 x1 2
7 × 7); hence, K1 = σ2 I. From (8), the detector of x1 can
x2 x2 −x2 x2 
  be obtained by maximizing the conditional probability density
x3 −x3 −x3 x3 
  function of r1 conditioned on B2 and h1 , as
x4 −x4 −x4 x4   
× x5 −x5 x5
 + n1 n2 n3 n4 , (3) 2
 x5 

x̂1 = arg min r1 − K1
−1/2
B2 h1 x̃1 . (9)
x6 −x6 x6 x6  x̃1
 
x7 −x7 x7 −x7 Similarly, by following the aforementioned steps, the detector
x8 −x8 x8 −x8 of xi , i ∈ {2, · · · , 8} will be
where

2
 arg minx̃ ri/2 − K−1/2 Bi−1 hi x̃i , i = 2Z,

yk ∈ C 8×1 , k ∈ {1, · · · , 4}, hi = [hi1 hi2 · · · hi8 ]T ∈ i i
x̂i= 2 (10)
C 8×1 ((·)T shows the transpose),  and nk ∈ C
8×1
are
 complex arg minx̃i r(i+1)/2 − K−1/2
 Bi+1 hi x̃i ,i = 2Z + 1,
i
valued vectors. Further, Y = y1 y2 y3  y4 ∈ C 8×4 ,
H = h1 h2 h3 h4  h5 h6 h7 h8 ∈ C 8×8 , and where r2 = (−y2 − y3 ) /2, r3 = (y3 + y4 ) /2, r4 =
N = n1 n2 n3 n4 ∈ C 8×4 . From (3), we have 2
(y1 − y4 ) /2, Ki = σ2 I, Bi is the left null of hi , and Z stands
8
X for integer.
y1 = hi xi + n1 , (4a)
B. STTS for Nt Transmit Antennas
i=1
In the proposed STTS, for transmitting N = Nt symbols
2 8
X X from Nt transmit antenna; N/2 time slots are needed. The
y2 = hi xi − hi x i + n2 , (4b)
STTS for Nt transmit antennas can be written as
i=1 i=3
→ Time slots
4 8  
X X x1 x1 −x1 x1 · · · · ±x1
y3 = − hi x i + hi xi + n3 , (4c) x2 x2 −x2 x2 · · · · ±x2 
i=1 i=5  
x3 −x3 −x3 x3 · · · · ±x3 
 
6
X 8
X x4 −x4 −x4 x4 · · · · ±x4 
 
y4 = hi xi − hi x i + n4 . (4d) x5 −x5 x5 x5 · · · · ±x5 
N
 
i=1 i=7 t x6 −x6 x6 x6 · · · · ±x6 
X=   ,
From (4a) and (4b), we get   x7 −x7 x7 −x7 −x7 · · · ±x7 
y x8 −x8 x8 −x8 −x8 · · · ±x8 

y1 + y2 n1 + n2 
· · · · · · · · · 

= h1 x1 + h2 x2 + . (5)
2 2
 
· · · · · · · · · 
 
In order to detect x1 from (5), the contribution of x2 should · · · · · · · · ∓xN −1
be nullified. It can be achieved by pre-multiplying (5) with the xN −xN · · · · · · ∓xN
left null of h2 . Similarly, x2 can be detected by nullifying the (11)
contribution of x1 . It can be observed through simulations that where X ∈ C N ×N/2 , the rows denote transmit antenna, and
the left null of a column vector with more than two elements columns denote time interval. The procedure of transmission
is a matrix. The matrix to be pre-multiplied with (5), i.e., in the proposed STTS is as follows:
B2 = (null(hH 2 ))
H
((·)H shows the hermitian), is chosen such • In first time slot, one symbol from each antenna with its
that h2 falls in the null space of B2 , i.e., B2 h2 = 0; we get original sign is transmitted.
    • In second time slot, the sign of first and second symbols
y1 + y2 n1 + n2
B2 = B2 h1 x1 + B2 h2 x2 + B2 . (6) from first and second antennas are detained (same as in
2 2 previous time slot) and rest of the symbols are transmitted
Because of the property of left null, B2 h2 = 0 in (6). with opposite signs.
Therefore, (6) results into • In next time slot, the sign of third and forth symbol is
    detained and all other symbol’s sign are flipped and so
y1 + y2 n1 + n2 on.
B2 = B2 h1 x1 + B2 . (7)
2 2 • In N/2 time slot, if N is a multiple of four then ‘+’ sign
is taken, otherwise, ‘-’ sign is taken.

2162-2337 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LWC.2017.2753800, IEEE
Wireless Communications Letters
3

C. Detector Design Therefore, from (14) and (16), we get

NX
r −1
The received signal at the receiver is given by 2
    γxi = h̃i1 γ¯i . (17)
y1 · · · yN/2 = h1 h2 · · · · hNt i=1
 
x1 x1 · · · ±x1 Since all channel coefficients are Gaussian distributed with2
x2 PNr −1
 x2 · · · ±x2  zero mean and unit variance, the distribution of i=1 h̃i1
· · · ·    will be central Chi-square with 2(Nr − 1) degrees of freedom.
×  + n1 · · · nN/2 ,(12)
· · · · 
 PNr −1 2

· The PDF of i=1 h̃i1 can be written
as
· · · 
uNr −2 exp − u2
xN −xN · ∓xN fu (u) = Nr −1 . (18)
2 Γ(Nr − 1)
where yk , nk , and hi = [hi1 hi2 · · · hiNr ]T are Nr × 1
From the definition ofZMGF, we can write
vectors; k ∈ {1, · · · , N/2} and i ∈ {1, · · · , Nt }. By following ∞
∆
the steps as given in aforementioned example, the detector of Mγi (s)= exp (−suγ¯i )fu (u)du. (19)
0
symbol xi at receiver can be derived
√ as 2
2 From (18) and (19), we get
x̂i = arg min ri − Bi±1 hi x̃i , (13)
σ exp − u2 − suγ¯i
Z ∞ Nr −2

x̃i u
Mγi (s) = du. (20)
where ‘+’ sign is used for odd values of ‘i’ and ‘-’ sign 0 2Nr −1 Γ(Nr − 1)
−1/2
is
used for even values of ‘i’; ri = Ki Bi±1 yk ± yk+1 /2. After solving the integral in (20) [10], the expression of MGF
1
All symbols except
xN and xN −1 can
be detected either is given by
Mγi (s) = N −1
. (21)
by using yk + yk+1 /2 or − yk + yk+1 /2, 1 ≤ k < N/2, (1 + 2sγ¯i ) r
 of ri in(13). For the detection of xN and xN −1 ,
in place C. Expression of SER
ri = y1 − yN/2 /2 for N = 4Z + , where Z + denotes the The SER of the proposed STTS based system with M -
  PSK constellation can be calculated with the help of following
positive integer; and ri = y1 + yN/2 /2 for N 6= 4Z + .
relation [11]:
1 βM g
Z 
III. P ERFORMANCE E VALUATION M P SK
PM P SK = M γi dθ, (22)
π 0 sin2 θ
In this section, the performance of the proposed STTS
is evaluated by using moment generating function (MGF) where βM = π(M − 1)/M and gM P SK = sin2 (π/M ). An
approach. The analytical performance of the proposed STTS approximation of (22) can be written as [12]:
3
based system is discussed in terms of the error probability and X
diversity gain. PM P SK ≈ ηp Mγi (µp ) , (23)
p=1
A. Expression of Instantaneous Received SNR
The instantaneous received SNR corresponding to xi can where η1 = θM /(2π) − 1/6, η2 = 1/4, η3 = θM /(2π) −
be obtained at the receiver from (13) as 1/4, µ1 = gM P SK , µ2 = 4gM P SK /3, and µ3 =
2 2 gM P SK /sin2 (θM ). An approximated SER of the proposed
2 kBi±1 hi k E |xi | 2
γxi = = kBi±1 hi k γ¯i , (14) system can be obtained in closed-form by using (23) and (21)
σ2 as 3
X ηp
where γ̄i = 2E(|xi | )/σ 2 .
2 PM P SK ≈ Nr −1
. (24)
D. Diversity Order p=1 (1 + µp γ¯i )
B. Expression of MGF
Diversity order is a very important parameter to analyze
For simplifying the problem, we perform singular value
the behavior of a communication system at very high value of
decomposition (SVD) of Bi±1 = UAVH , where U and V
SNR. By substituting very high values of γ̄i in (24), we get
are unitary matrices and A is diagonal matrix. Here, λi s
(i ∈ 1, 2, · · · , (Nr − 1)) are non zero diagonal values of 3
X ηp δ
matrix A, i.e., singular values of Bi±1 . Therefore, λ2i will be PM P SK ≈ Nr −1
= , (25)
(µp γ¯i ) γ¯i dv
an eigenvalue of Bi±1 BH i±1 . By using SVD of Bi±1 in (14),
p=1

we obtain where δ is a constant and dv is diversity order of the system.

2 2 2
kBi±1 hi k = UAVH hi = UAh̃i , (15) It can be noticed from (25) that the diversity order of the
proposed scheme is Nr − 1.
where h̃i = VH hi . Here, h̃i (= VH hi ) has the same distribu- Remark 2: Since the proposed STTS is used for large
tion as hi because V is unitary and hi is circularly symmetric MIMO communication systems, Nr is very large (of the order
complex Gaussian. Since UH U = I, from (15) we have of hundreds). Hence, the diversity order of the system is still
NXr −1
2 2 2 reasonably high, in this case. Note that N symbols can be
UAh̃i = Ah̃i = λ2i h̃i1 . (16) transmitted in N/2 time slots; therefore, the rate of the pro-
i=1
posed STTS is two and symbol-wise decoding of the symbols
Remark 1: It can be shown through simulations that all is also possible. Therefore, the proposed scheme is suitable for
eigenvalues of left null matrix are equal to one. heavy data transmission with sufficient performance in large
MIMO systems.

2162-2337 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LWC.2017.2753800, IEEE
Wireless Communications Letters
4

0 0
10 10

dv=9
−2 N =N =4,Sim.,STBC [6],256−PSK −20
10 t r 10
N =N =4,Sim.,Ext.Sch.[3],BPSK
t r
Nt=Nr=4, Simulation, 8−PSK
−4 Nt=Nr=4, Analysis, 8−PSK −40 dv=29
10 10 d =19
v
Nt=Nr=4, Simulation, QPSK

BER
SER

Nt=Nr=4, Analysis, QPSK

Nt=Nr=4,Sim.,Ext.Sch. [2],BPSK −60 d =29
−6 10 10×10 MIMO System v
10
Nt=Nr=4, Simulation, BPSK 20×20 MIMO System
Nt=Nr=4, Analysis, BPSK 30×30 MIMO System
Nt=Nr=10, Simulation, QPSK 50×50 MIMO System
−80
−8 10 50×80 MIMO System
10 Nt=Nr=10, Analysis, QPSK d =49
v
Ideal diversity (d )
v
N =N =10, Simulation, BPSK
t r
N =N =10, Analysis, BPSK −100
t r 10
−10
10 0 5 10 15 20 25 30 35 40 45 50
−15 −10 −5 0 5 10 15 20 SNR [dB]
SNR [dB]
Fig. 1. SER versus SNR performance of the proposed scheme with Nt = Fig. 2. Diversity order of proposed STBC based large MIMO system for
Nr = 4, 10, existing scheme [2], existing scheme [3], STBC [6], and with Nt = 10, 20, 50, Nr = 10, 20, 30, 50, 80, and with BPSK constellation.
BPSK, QPSK, 8-PSK, 256-PSK constellations.
SER of the considered scheme matches with that of the ideal
IV. N UMERICAL R ESULTS diversity curves, at high SNR, as shown in Fig. 2. Further, it
The analytical and simulated SER versus SNR plots of is verified that the diversity order of the proposed system is
proposed STTS based large MIMO system with Nt = 4, 10, Nr − 1, as indicated in Subsection IV-D.
Nr = 4, 10, BPSK, QPSK, and 8-PSK constellations are
V. C ONCLUSIONS
shown in Fig. 1. Simulation results are generated by using
106 channel realizations. A close matching of the theoretical We have proposed a space-time transmission scheme for
and simulated SER values is evident from the figure. Further, large MIMO systems. The data rate of the proposed scheme
it can be observed from the figure that by increasing the is two and diversity order is equal to number of receive
values of Nt and Nr , the performance of the considered antennas minus one. It has been shown that symbol-wise
system significantly improves. For example, the considered decoding is attainable; therefore, this scheme is very useful
system with Nt = Nr = 10 and QPSK constellation performs for large MIMO systems. It has been shown by the analysis
approximately 5 dB better than that with Nt = Nr = 4, at and simulation that the performance of the scheme can be
SER=10−2 , as seen in Fig. 1. Moreover, the performance of improved by increasing the number of receive antennas.
existing schemes [2] and [3] with Nt = Nr = 4 and BPSK R EFERENCES
constellation is compared with the performance of proposed [1] A. Chockalingam and B. S. Rajan, Large MIMO Systems. Cambridge
STTS with Nt = Nr = 4 employing QPSK constellation; in Univ. Press, Feb. 2014.
order to have the same data rate for all schemes. It can be [2] T. L. Marzetta, “Noncooperative cellular wireless with unlimited num-
bers of base station antennas,” IEEE Trans. Wireless Commun., vol. 9,
observed from the figure that the proposed scheme performs no. 11, pp. 3590–3600, Nov. 2010.
significantly better than [3] and slightly poorer than [2]; but it [3] H. Yang and T. L. Marzetta, “Performance of conjugate and zeroforc-
is worth noticing that in the scheme of [2] transmitted symbol ing beamforming in large-scale antenna systems,” IEEE J. Sel. Areas
Commun., vol. 31, no. 2, pp. 172–179, 2013.
vector is jointly detected so its computational complexity is [4] F. Rusek, D. Persson, B. Lau, E. Larsson, T. Marzetta, O. Edfors, and
very high (of the order of N 2 +2N ). However, in the proposed F. Tufvesson, “Scaling up MIMO: Opportunities and challenges with
STTS, the decouple decoding of the symbols is possible. Exact very large arrays,” IEEE Signal Process. Mag., vol. 30, pp. 40–60, 2013.
[5] Y. Lim, C. Chae, and G. Caire, “Performance analysis of massive MIMO
computational complexity of the proposed scheme is Nt Nr /2. for cell-boundary users,” IEEE Trans. Wireless Commun., vol. 14, no. 12,
On the other hand the computational complexity of [3] is N 3 pp. 6827–6842, Dec. 2015.
due to involvement of zero-forcing algorithm, which is much [6] G. Ganesan and P. Stoica, “Space-time block codes: A maximum SNR
approach,” IEEE Trans. Inform. Theory, vol. 47, no. 4, pp. 1650–1656,
higher than the computational complexity of the proposed 2001.
scheme (of the order of N Nr ). Further, the performance of [7] X. Meng, X.-G. Xia, and X. Gao, “Omnidirectional STBC design in
the proposed STTS with Nt = Nr = 4, 8-PSK constellation is massive MIMO systems,” In Proc. of Global Communications Confer-
ence (GLOBECOM), 2015, 6-10 Dec. 2015, San Diego, CA, USA, pp
also compared to OSTBC based system [6] with Nt = Nr = 4 1–6.
and 256-PSK constellation (to have the same data rate for [8] Arti M.K., “OSTBC transmission in large MIMO systems,” IEEE
both schemes) is also shown in Fig. 1. The proposed scheme Commun. Lett., vol. 20, no. 11, pp. 2308–2311, Nov. 2016.
[9] M. H. Halmi, “Orthogonal space-time block codes for large MIMO
performs significantly better than OSTBC based system [6], systems,” in proc. International Conference on Communications, Man-
as seen from the figure. agement and Telecommunications (ComManTel), 2015.
In Fig. 2, the analytical SER of the proposed STTS based [10] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
Products, 6th ed. San Diego, CA, USA: Academic Press, 2000.
large MIMO system is plotted for large value of the SNR. [11] M. K. Simon and M.-S. Alouini, Digital Communication over Fading
The analytical values are obtained for BPSK constellation, Channels. New Jersey, USA: John Wiley & Sons, Inc., 2005.
Nt = 10, 20, 50, Nr = 10, 20, 30, 50, 80, by using (24). Ideal [12] M. McKay, A. Zanella, I. Collings, and M. Chiani, “Error probability and
SINR analysis of optimum combining in Rician fading,” IEEE Trans.
diversity plots are also shown by using (25), where δ is a Commun., vol. 57, no. 3, pp. 676–687, Mar. 2009.
positive-valued constant. The rate of decay of the analytical

2162-2337 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.