1860 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO.

2, FEBRUARY 2017

Correspondence
Near-Optimal Signal Detector Based on Structured a large number of low-cost antennas for higher spectrum efficiency
Compressive Sensing for Massive SM-MIMO but only one or several power-hungry transmit radio frequency (RF)
chains to save power, whereas the user can compactly employ the
Zhen Gao, Linglong Dai, Chenhao Qi, Chau Yuen, and multiple receive diversity antennas with low correlation [2]. Since
Zhaocheng Wang the power consumption and hardware cost are largely dependent
on the number of simultaneously active transmit RF chains (particu-
larly the power amplifier), massive SM-MIMO outperforms the tra-
ditional MIMO schemes in higher spectrum efficiency, reduced power
Abstract—Massive spatial-modulation multiple-input multiple-output consumption, lower hardware cost, etc. In practice, SM can be adopted
(SM-MIMO) with high spectrum efficiency and energy efficiency has recent- in conventional massive MIMO systems as an energy-efficient trans-
ly been proposed for future green communications. However, in massive
SM-MIMO, the optimal maximum-likelihood detector has the high com-
mission mode. Meanwhile, massive SM-MIMO can be also considered
plexity, whereas state-of-the-art low-complexity detectors for small-scale as an independent scheme to reduce both power consumption and
SM-MIMO suffer from an obvious performance loss. In this paper, by hardware cost.
exploiting the structured sparsity of multiple SM signals, we propose a For massive SM-MIMO, due to the small number of receive an-
low-complexity signal detector based on structured compressive sensing tennas at the user and massive antennas at the BS, the signal detec-
(SCS) to improve the signal detection performance. Specifically, we first
propose the grouped transmission scheme at the transmitter, where mul- tion is a challenging large-scale underdetermined problem. When the
tiple SM signals in several continuous time slots are grouped to carry the number of transmit antennas becomes large, the optimal maximum
common spatial constellation symbol to introduce the desired structured likelihood (ML) signal detector suffers from the prohibitively high
sparsity. Accordingly, a structured subspace pursuit (SSP) algorithm is complexity [3]. Low-complexity signal vector (SV)-based detector has
proposed at the receiver to jointly detect multiple SM signals by leveraging
the structured sparsity. In addition, we also propose the SM signal inter-
been proposed for SM-MIMO [3], but it is confined to SM-MIMO
leaving to permute SM signals in the same transmission group, whereby with a single transmit RF chain. In [4]–[6], the SM is generalized,
the channel diversity can be exploited to further improve signal detection where more than one active antennas are used to transmit indepen-
performance. Theoretical analysis quantifies the gain from SM signal dent signal constellation symbols for spatial multiplexing. Linear
interleaving, and simulation results verify the near-optimal performance minimum mean square error (LMMSE)-based signal detector [1] and
of the proposed scheme.
sphere decoding (SD)-based detector [7] can be used for SM-MIMO
Index Terms—Massive multiple-input multiple-output (MIMO), signal systems with multiple transmit RF chains. However, they are only suit-
detection, signal interleaving, spatial modulation (SM), structured able for well or overdetermined SM-MIMO with Nr ≥ Nt and suffer
compressive sensing (SCS).
from a significant performance loss in underdetermined SM-MIMO
systems with Nr < Nt , where Nt and Nr are the numbers of transmit
I. I NTRODUCTION and receive antennas, respectively. Due to a limited number of RF
chains, SM signals have the inherent sparsity, which can be considered
S PATIAL-MODULATION multiple-input multiple-output (SM-
MIMO) exploits the pattern of one or several simultaneously
active antennas out of all available transmit antennas to transmit
by exploiting the compressive sensing (CS) theory [8] for improved
signal detection performance. By far, CS has been widely used in wire-
extra information [1], [2]. Compared with small-scale SM-MIMO, less communications [9]–[12], and the CS-based signal detectors have
which only introduces the limited gain in spectrum efficiency, massive been proposed for underdetermined small-scale SM-MIMO [11], [12].
SM-MIMO has recently proposed by integrating SM-MIMO with However, their bit-error-rate (BER) performance still has a significant
massive MIMO working at 3–6 GHz to achieve higher spectrum effi- gap compared with that of the optimal ML detector, particularly in
ciency [1]. In massive SM-MIMO systems, the base station (BS) uses massive SM-MIMO with large Nt , Nr , and Nr  Nt .
This paper proposes a near-optimal structured compressive sensing
(SCS)-based signal detector with low complexity for massive SM-
MIMO. Specifically, we first propose the grouped transmission scheme
Manuscript received April 8, 2015; revised October 17, 2015 and January 26, at the BS, where multiple successive SM signals are grouped to
2016; accepted March 31, 2016. Date of publication April 21, 2016; date of carry the common spatial constellation symbol to introduce structured
current version February 10, 2017. This work was supported in part by the
National Natural Science Foundation of China under Grant 61271266 and Grant sparsity. Accordingly, we propose a structured subspace pursuit (SSP)
61302097, by the International Science and Technology Cooperation Program algorithm at the user to detect multiple SM signals, whereby their
of China under Grant 2015DFG12760, by the Singapore A∗STAR Project structured sparsity is leveraged for improved signal detection perfor-
under Grant 142 02 00043, and by the Beijing Natural Science Foundation mance. Moreover, the SM signal interleaving is proposed to permute
under Grant 4142027. The review of this paper was coordinated by Dr. Y. Ma.
Z. Gao, L. Dai, and Z. Wang are with the Tsinghua National Laboratory for SM signals in the same transmission group, so that the channel diver-
Information Science and Technology, Department of Electronic Engineering, sity can be exploited. Theoretical analysis and simulation results verify
Tsinghua University, Beijing 100084, China (e-mail: [email protected]. that the proposed SCS-based signal detector outperforms existing CS-
edu.cn; [email protected]; [email protected]). based signal detector.
C. Qi is with the School of Information Science and Engineering, Southeast
University, Nanjing 210096, China (e-mail: [email protected]). Notation: Boldface lowercase and uppercase symbols represent
C. Yuen is with Singapore University of Technology and Design, Singapore column vectors and matrices, respectively. · denotes the integer floor
138682 (e-mail: [email protected]). operator. The transpose, conjugate transpose, and Moore–Penrose
Color versions of one or more of the figures in this paper are available online matrix inversion operations are denoted by (·)T , (·)∗ , and (·)† , respec-
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TVT.2016.2557625 tively. The lp -norm operation is given by  · p , and | · | denotes the

0018-9545 © 2016 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.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 2, FEBRUARY 2017 1861

active antennas selected from Nt transmit antennas is regarded as

Nt
log 
the spatial constellation symbol. Hence, there are |A| = 2 2 (Na )
kinds of patterns of active antennas, i.e., supp{x} ∈ A. Meanwhile,
the signal constellation symbol of the ith active antenna, which is
denoted by x(i) for 1 ≤ i ≤ Na , is mapped into the M -ary signal
constellation set B. Therefore, the signal detection in SM-MIMO can
Nt
log 
be formulated as the M Na 2 2 (Na ) -hypothesis detection problem.
Clearly, the optimal signal detector to this problem is the ML signal
detector, which can be expressed as [1]
x̂ML = arg min y − Hx2 . (1)
supp(x)∈A,x(i) ∈B,1≤i≤Na

However, the computational complexity of the optimal ML signal

Nt
log 
detector is O(M Na 2 2 (Na ) ), which can be unrealistic when Nt ,
Na , and/or M become large.
To reduce the complexity, the SV-based signal detector has
been proposed [3], but it only considers the case of Na = 1. The
Fig. 1. Spatial constellation symbol and the signal constellation symbol in LMMSE-based signal detector with the complexity of O(2Nr Nt2 +
SM-MIMO systems, where Nt = 4, Na = 1, and QPSK are considered an
example. Nt3 ) [1] and the SD-based signal detector with the complexity of
O(max{Nt3 , Nr Nt2 , Nr2 Nt }) [7] have been proposed for well or
cardinality of a set. E{·}, var{·}, Re{·}, and Im{·} are operators to overdetermined SM-MIMO with Nr ≥ Nt . However, for underdeter-
take the expectation, variance, the real part, and the imaginary part of a mined SM-MIMO systems with Nr < Nt , these detectors suffer from
random variable. Tr{·} is the trace operation for a matrix. If a set has n a significant performance loss [12]. Since only Na transmit antennas
elements,the are active in each time slot for power saving and low hardware cost,
 number of k-combinations is denoted by the binomial co- there are only Na < Nt nonzero entries in x; thus, the SM signal
efficient nk . The index set of nonzero entries of the vector x is called
the support set of x, which is denoted by supp{x}, xi denotes the ith has the inherent sparsity. By exploiting such sparsity, the CS-based
entry of the vector x, and Hi denotes the ith column vector of the ma- signal detectors have been proposed for SM [10]–[12]. In [10], a
trix H. xΓ denotes the entries of x defined in the set Γ, whereas HΓ de- spatial modulation matching pursuit (SMMP) algorithm is proposed
notes a submatrix of H with indexes of columns defined by the set Γ. to detect multiuser SM signals in the uplink massive SM-MIMO
systems. In [11] and [12], the CS-based signal detectors are proposed
II. S YSTEM M ODEL for underdetermined single-user SM-MIMO systems with Nr < Nt
in the downlink. The normalized compressive sensing (NCS) detector
In SM-MIMO systems, the transmitter has Nt transmit antennas (with the complexity of O(2Nr Na2 + Na3 )) in [11] first normalizes the
but Na < Nt transmit RF chains, and the receiver has Nr receive MIMO channels and then uses orthogonal-matching-pursuit algorithm
antennas. Each SM signal consists of two symbols (see Fig.  1): the to detect signals. In [12], a basis pursuit denoising (BPDN) algorithm
spatial constellation symbol obtained by mapping log2 N Nt
a
 bits to (with the complexity of O(Nt3 )) from the classical basis pursuit
a pattern of Na active antennas out of Nt transmit antennas and Na in- algorithm is developed to detect SM signals. However, both NCS and
dependent signal constellation symbols coming from the M -ary signal BPDN detectors are based on the framework of CS theory, and such
constellation set (e.g., quadrature amplitude modulation). Hence,  each CS-based signal detectors still suffer from a significant performance
SM signal carries the information of Na log2 M + log2 N Nt
a
 bits. gap compared with the optimal ML detector when Nt /Nr becomes
At the receiver, the received signal y ∈ CNr ×1 can be ex- large, particularly in massive SM-MIMO systems with Nr  Nt [12].
pressed as y = Hx + w, where x ∈ CNt ×1 is the SM signal
transmitted by the transmitter; w ∈ CNr ×1 is the additive white III. P ROPOSED S TRUCTURED C OMPRESSIVE S ENSING -BASED
Gaussian noise (AWGN) vector with independent and identically S IGNAL D ETECTOR
distributed (i.i.d.) entries following the circular symmetric complex
Gaussian distribution CN (0, σw 2
); H = Rr H̃Rt ∈ CNr ×Nt is
1/2 1/2 In this section, an SCS-based signal detector is proposed for down-
the correlated flat Rayleigh-fading MIMO channel, with entries of H̃ link single-user massive SM-MIMO as shown in Fig. 2.
being subjected to the i.i.d. distribution CN (0, 1); and Rr and Rt are
the receiver and transmitter correlation matrices, respectively [13]. The A. Grouped Transmission and Interleaving at the Transmitter
correlation matrix R is given by rij = r |i−j| , where rij is the ith row We assume that signal constellation symbols in the proposed scheme
and the jth column element of R, and r is the correlation coefficient are mutually independent. Moreover, for the proposed grouped trans-
of neighboring antennas. mission scheme, every G consecutive SM signals are considered as a
It should be pointed out that H should be known by the receiver group, and SM signals in the same transmission group share the same
and can be acquired by channel estimation [13]. To achieve both high spatial constellation symbol, i.e.,
spectrum efficiency and energy efficiency, massive SM-MIMO, which      
supp x(1) = supp x(2) = · · · = supp x(G) (2)
employs massive low-cost antennas but few power-hungry transmit RF
chains at the BS to serve the user with comparatively small number of where x(1) , x(2) , . . . , x(G) are SM signals in G consecutive time
receive antennas, has recently been proposed [1]. However, its signal slots. Due to the conveyed common spatial constellation symbol,
detection is a challenging large-scale underdetermined problem since x(1) , x(2) , . . . , x(G) in the same transmission group share the same
Nt , Nr can be large and Nr  Nt , e.g., Nt = 64 and Nr = 16 are support set and thus have the structured sparsity. It is clear that
considered [1].  Nt  to introduce such structured sparsity, the effective information bits
For x, the spatial constellation symbol of log2 N a
 bits is carried by spatial constellation symbols will be reduced. However,
mapped into the spatial constellation set A, where the pattern of Na as will be demonstrated in our simulations, such structured sparsity
1862 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 2, FEBRUARY 2017

detectors [8]. Under the framework of SCS theory, the solution to (3)
can be achieved by solving the following optimization problem:
 G 1
  q
min x (t) q
p
supp(x(t) )∈A t=1
   
s.t. y(t) = H x(t) , supp x(t) = supp x(1) ∀ t.
(t)
(4)
In this paper, based on the classical subspace pursuit (SP) algorithm
[8], we propose an SSP algorithm by utilizing the structured sparsity
to solve the optimization problem (4) in a greedy way, where p = 0
and q = 2 are advocated [8].
The proposed SSP algorithm is described in Algorithm 1. Specifi-
cally, Lines 1∼3 perform the initialization. In the kth iteration, Line 5
performs the correlation between the MIMO channels and the residual
in the previous iteration; Line 6 obtains the potential true indexes
according to Line 5; Line 7 merges the estimated indexes obtained in
Lines 8∼9 in the previous iteration and the estimated indexes in Line 6
in the current iteration; after the least squares in Line 8, Line 9 removes
wrong indexes and selects Na most likely indexes; Line 10 estimates
SM signal according to Ωk ; and Line 11 acquires the residue. The iter-
ation stops when k > Na . Compared with the classical SP algorithm
that only reconstructs one sparse signal from one received signal, the
proposed SSP algorithm can jointly recover multiple sparse signals
with the structured sparsity but having different measurement matri-
ces, where the structured sparsity of multiple sparse signals can be
leveraged for improved signal detection performance. Therefore, the
classical SP algorithm can be regarded as a special case of the proposed
SSP algorithm when G = 1, and more details will be discussed in
Fig. 2. Proposed SCS-based signal detector, where Nt = 4, Nr = 2, Na = 1,
G = 2, and QPSK are considered. Note that the white dot block in MIMO
Section IV-A. Another difference should be pointed out that in the
channels denotes the deep channel fading. steps of Lines 6 and 9 in Algorithm 1, the selected support set should
belong to the predefined spatial constellation set A for enhanced signal
allows more reliable signal detection performance and eventually detection performance. However, the classical SP algorithm and exist-
could even improve the BER performance of the whole system without ing CS-based signal detectors do not exploit such priori information
the reduction of the total bit per channel use (bpcu). of the expected support set [11], [12]. By using the proposed SSP
On the other hand, due to the temporal channel correlation, channels algorithm, we can acquire the estimation of the spatial constellation
in several consecutive time slots can be considered to be quasi-static, symbol according to supp(x̂(t) )’s and the rough estimation of signal
i.e., H(1) = H(2) = · · · = H(G) , where H(t) for 1 ≤ t ≤ G is the constellation symbols. By searching for the minimum Euclidean dis-
channel associated with the tth SM signal in the group. This implies tance between the rough estimation of signal constellation symbols
that if channels used for SM fall into the deep fading, such deep fading and legitimate constellation symbols, we can finally estimate signal
usually remains unchanged during G time slots, and the corresponding constellation symbols.
signal detection performance will be poor. To solve this issue, we fur-
ther propose the SM signal interleaving at the transmitter. Specifically, Algorithm 1 Proposed SSP Algorithm.
after the original SM signals x(t) ’s are generated, the actually trans-
mitted signals are given by Π(t) x(t) ’s, where each column and row of Input: Received signal y(t) , the channel matrix H (t) , and the
Π(t) ∈ CNt ×Nt only has one nonzero element with the value of one, number of active antennas Na , where 1 ≤ t ≤ G.
and Π(t) can permutate the entries in x(t) . We consider that Π(t) ’s for Output: Estimated SM signal x̂(t) for 1 ≤ t ≤ G.
1 ≤ t ≤ G are different in different time slots, and they are predefined 1: Ω0 = ∅;
and known by both the transmitter and receiver. In this way, the active 2: r(t) = y(t) ∀ t;
antennas vary in different time slots from the same transmission group, 3: k = 1;
although x(t) ’s share the common spatial constellation symbol. Hence, 4: while k ≤ Na do
∗
the channel diversity can be appropriately exploited to improve the 5: a(t) = (H (t) ) r(t) ∀ t;
 (t) 2
signal detection at the receiver. In Section IV-B, such diversity gain 6: Γ = arg max{ G t=1aΓ̃ 2, Γ̃ ∈ A,|Γ̃| = min{2Na, Nr }
Γ̃
will be further discussed.
if k = 1 or |Γ̃| = min{Na , Nr − Na } if k > 1};
7: Ξ = Ωk−1 ∪ Γ;
B. SCS-Based Signal Detector at the Receiver (t) †
bΞ = (H Ξ ) y(t) ∀ t;
(t)
8:
At the receiver, the received signal in the tth time slot is  (t) 2
9: Ωk = arg max{ G t=1 bΩ̃ 2 , Ω̃ ∈ A and |Ω̃| = Na };
y(t) = H(t) Π(t) x(t) + w(t) = H
(t) Ω̃
x(t) + w(t) (3) (t)  (t)
†
10: cΩk = (H Ωk ) y(t) ∀ t;
where H (t) = H(t) Π(t) is the deinterleaving processing. 11: r = y − H
(t) (t)  (t) (t)
c ∀ t;
From (3), we observe that x(t) ’s share the structured sparsity, but 12: k = k + 1;
they have different nonzero values. According to SCS theory, the 13: end while
structured sparsity of x(t) ’s can be exploited to improve the signal de- 14: x̂(t) = c(t) ∀ t;
tection performance compared with the conventional CS-based signal
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 2, FEBRUARY 2017 1863

IV. P ERFORMANCE A NALYSIS (pdfs) of Cm and Cl with l = m are denoted by f1 (x) and
[2] [2]
In this section, we will provide the performance analysis. f2 (x), respectively. The pdf of Cl with l = m is f2 (x) = (Nt −
Na )!/(Nt − Na − 2)!)(F2 (x))Nt −Na −2 (1 − F2 (x))f2 (x), where
A. Comparison of SCS-Based and CS-Based Signal Detectors F2 (x) is the cumulative density function of f2 (x). In this way, we have


[2]
Typically, existing CS-based signal detectors utilize one received PGMMV Cm − Cl > 0|l = m
signal vector to recover one sparse SM signal vector, which is a typical
single measurement vector (SMV) problem, i.e., y = Hx + w. ∞ ∞
If multiple sparse signals share the common support set and =
[2]
f (x)f2 (x − z)dxdz. (6)
identical measurement matrix, i.e., [y(1) , y(2) , . . . , y(G) ] =
0 −∞
H[x(1) , x(2) , . . . , x(G) ] + w, the reconstruction of x(t) ’s from
y(t) ’s for 1 ≤ t ≤ G can be considered as the multiple-measurement- For the conventional MMV problem with identical channel ma-
vector (MMV) problem in SCS theory [8]. The SCS theory has proven trices, similar to the previous analysis, we have Cm ∼ Gσ22 χ21 +
that with the same size of the measurement vector, the recovery Gσ12 χ21 and Cl ∼ Gσ32 χ22 with l = m. Similarly, we can also get
[2]
performance of SCS algorithms is superior to that of conventional PMMV (Cm − Cl > 0|l = m).
CS algorithms [8]. This implies that with the same number of receive To intuitively compare the signal detection probability, we compare
antennas Nr , the proposed SCS-based signal detector can outperform PMMV (Cm − Cl > 0|l = m) and PGMMV (Cm − Cl > 0|l = m)
conventional CS-based signal detectors. when σs2 /σw 2
→ ∞ and G are sufficient large. In this case,
Compared with the conventional MMV problem, our formulated Cm − Cl can be approximated to the Gaussian distribution
problem (4) is to solve multiple sparse signals with the common N (μ4 , σ42 ) with μ4 = G(μ21 + μ22 − 2μ23 + σ12 + σ22 − 2σ32 ), σ42 =
support set but having different measurement matrices. Hence, both G 3i=1 2σi4 + 4μ2i σi2 . In this way, we can obtain that PGMMV (Cm −
conventional SMV problem and MMV problem can be considered the Cl > 0|l = m) ≈ Q(−μ4 /σ4 ), where Q-function is the tail proba-
special cases of our problem. If Π(t) ’s are identical, (4) becomes the bility of the standard normal distribution [14]. By contrast, for the
conventional MMV problem, and furthermore, if G = 1, it reduces to conventional MMV √ case, we can obtain that PMMV (Cm − Cl > 0|l =
the SMV problem. Therefore, our formulated problem can be regarded m) ≈ Q(−μ4 /( Gσ4 )). Clearly, PMMV is larger than PGMMV due
as a generalized MMV (GMMV) problem. to μ4 > 0 and G > 1, which implies that an appropriate SM signal
interleaving will lead to the improved signal detection performance.
To achieve the goal that H l ’s ∀ l, are mutually independent, we
(t)
B. Performance Gain From SM Signal Interleaving
consider the pseudorandom permutation matrix Π(t) .
We discuss the performance gain from the SM signal interleaving In Section V, simulation results confirm the good channel diversity
by comparing the detection probability of the proposed SSP algorithm gain from interleaving, whose performance gain approaches that of the
with and without SM signal interleaving. Here, we consider a case of mutually independent channel matrices in the same group.
simplified scenario with Na = 1 and uncorrelated Rayleigh-fading
MIMO channels. Let m be the index of the active antenna, and for any
given l, H l ’s for 1 ≤ t ≤ G are mutually independent, where 1 ≤
(t) C. Computational Complexity
m, l ≤ Nt . Based on these assumptions, the received signal is given The optimal ML signal detector has the complexity of
by y(t) = α(t) H (t)
m +w
(t)
, for 1 ≤ t ≤ G, where α(t) ∈ B denotes log
Nt

O(M Na 2 2 (Na ) ), which is high for large Na , Nt , and/or M .
the signal constellation symbol carried by the active antenna in the tth
The conventional signal detectors [1], [7], [12] have the complexity
time slot. To identify the active antenna, the proposed SSP algorithm
of O(Nt3 ), which is still high in massive SM-MIMO systems with
relies on the correlation operation in Line 5 of Algorithm 1, i.e.,
large Nt . By contrast, for the proposed signal detector, the main
G
 G
∗ 2
 (t) ∗  (t) 2  (t)  (t) (t)
computational burden comes from the step of least squares with
Cl  y Hl = α H m + w(t) H l the complexity of O(G(2Nr Na2 + Na3 )) [8], or equivalently
t=1 t=1 O(2Nr Na2 + Na3 ) per SM signal in each time slot. This indicates
that the proposed SCS-based signal detector enjoys the same order of
G

(t) 2 complexity with the CS-based signal detector [11].
= Fm,l (5)
t=1

∗ V. S IMULATION R ESULTS
where Fm,l = (α(t) H (t) ) H l
(t) (t)
m +w
(t)
for 1 ≤ l ≤ Nt . Due to
(t)
large Nr in practice, we have Re{Fm,m } ∼ N (μ1 , σ12 ) with A simulation study was carried out to compare the performance of
μ1 = 0, σ12 = ((Nr2 + Nr )σs2 /(2 − δ(M = 2)) + (Nr σw 2
/2), and the proposed SCS-based signal detector with that of the conventional
(t) LMMSE-based signal detector [1] and the CS-based signal detector
Im{Fm,m } ∼ N (μ2 , σ2 ) with μ2 = 0, σ2 = ((1 − δ(M = 2))
2 2
[12]. The performance of the optimal ML detector [6] is also provided
(Nr2 + Nr )σs2 /2) + (Nr σw 2
/2) according to central limit theorem
(t) (t) as the benchmark for comparison.
[14]. Similarly, both Re{Fm,l } and Im{Fm,l } follow the distribution
Fig. 3 compares the simulated and analytical spatial constellation
N (μ3 , σ3 ) with l = m, μ3 = 0, and σ3 = (Nr σs2 /2)+(Nr σw
2 2 2
/2). Note
symbol error rate (SCSER) of the SCS-based signal detector in
(t) T (t) (t)
that σs = Tr{E{x (x ) }}, and Re{Fm,l } and Im{Fm,l } ∀ l are
2 (t)
different cases over uncorrelated Rayleigh-fading MIMO channels,
mutually independent. Moreover, we can have Cm ∼ σ22 χ2G + σ12 χ2G where Nt = 64, Nr = 16, Na = 1, and 8-phase-shift keying (PSK) are
and Cl ∼ σ32 χ22G with l = m, where χ2n is the central chi-squared considered. For the GMMV case, “i.i.d.” denotes the case that H (t) =
distribution with the degrees of freedom n [14]. Since Algorithm 1 H(t) ∀ t and H(t) ’s are independently generated, whereas “interleav-
only has one iteration and |Γ| = |Ξ| = 2 in the iteration for Na = 1, ing” denotes the case that H(1) = H(2) = · · · = H(G) and H (t) =
[2]
we consider PGMMV (Cm − Cl > 0|l = m) as the correct active H(t) Π(t) with different permutation matrices Π(t) ’s. Clearly, the
[1] [2] [N −N ]
antenna detection probability, where Cl > Cl > · · · > Cl t a analytical SCSER derived in Section IV-B have the good tightness
with l = m are sequential statistics. The probability density functions with the simulation results. In addition, the proposed SCS-based signal
1864 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 2, FEBRUARY 2017

Fig. 3. Comparison of the simulated and analytical SCSER of the SCS-based Fig. 5. BER comparison between the traditional CS-based signal detector and
signal detector in different cases over uncorrelated Rayleigh-fading MIMO the proposed SCS-based signal detector over correlated Rayleigh-fading MIMO
channels, where Nt = 64, Nr = 16, Na = 1, and 8-PSK are considered. channels, where rt = rr = 0.4 and Nr = 16 are considered.

Fig. 4. SCSER of different signal detectors over correlated Rayleigh-fading

Fig. 6. BER performance comparison between the proposed SCS-based signal
MIMO channels, where rt = rr = 0.4, Nt = 64, Nr = 16, Na = 1, and
detector and the optimal ML signal detector, where rt = rr = 0.4, Nt = 65,
8-PSK are considered.
Nr = 16, Na = 2, and 8-PSK are considered.

detector outperforms the conventional CS-based signal detector since

the structured sparsity of multiple sparse SM signals is exploited. with interleaving over correlated Rayleigh-fading MIMO channels
Moreover, since the channel diversity can be also exploited, the SCS- with rt = rr = 0.4 and Nr = 16. The existing scheme adopts two
based signal detector with mutually independent channel matrices is transmission modes: 1) Nt = 64, Na = 1, and binary PSK with
superior to that with identical channel matrices by more than 4 dB 7 bpcu; and 2) Nt = 65, Na = 2, and no signal constellation symbol
if the SCSER of 10−3 is considered. Finally, the performance of the with 11 bpcu. In contrast, the SCS-based signal detector with Nt = 65,
SCS-based signal detector with SM signal interleaving approaches that Na = 2, and G = 2 adopts quadrature PSK and 8-PSK, respectively,
with mutually independent channel matrices, which indicates that the and the corresponding data rates are 9.5 and 11.5 bpcu. In Fig. 5, it can
proposed SM signal interleaving can fully exploit the channel diversity. be observed that the proposed SCS-based signal detector with even
Fig. 4 provides SCSER comparison of different signal detectors over higher bpcu achieves better BER performance than the conventional
correlated Rayleigh-fading MIMO channels, where both the channel CS-based signal detector.
correlation coefficients at the transmitter and receiver are rt = rr = Fig. 6 compares the performance of the proposed SCS-based sig-
0.4 [13], Nt = 64, Nr = 16, Na = 1, and 8-PSK are considered. nal detector with interleaving and the optimal ML signal detector,
The conventional LMMSE-based signal detector works poorly due to where rt = rr = 0.4, Nt = 65, Nr = 16, Na = 2, and 8-PSK are
Nr  Nt . The SCS-based signal detector with interleaving outper- considered. We find that with the increasing G, the BER performance
forms the conventional CS-based signal detector and SCS-based signal gap between the SCS-based signal detector and the optimal ML
detector without interleaving. Moreover, it has the similar performance signal detector becomes smaller. When G ≥ 2, the SCS-based signal
with that with mutually independent channel matrices (i.e., H (t) = detector approaches the optimal ML signal detector with a small
H(t) ∀ t and H(t) ’s are independently generated), which indicates performance loss. For example, if the BER of 10−4 is considered, the
the good channel diversity gain from interleaving, even in correlated performance gap between the SCS-based signal detector with G = 3
MIMO channels. and the optimal ML detector is less than 0.2 dB. Thus, the near-
Fig. 5 provides the BER performance comparison of the existing optimal performance of the proposed SCS-based signal detector can
CS-based signal detector and the proposed SCS-based signal detector be verified.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 2, FEBRUARY 2017 1865

VI. C ONCLUSION Performance Analysis of Neighbor Discovery Process in

This paper has proposed a near-optimal SCS-based signal detector Bluetooth Low-Energy Networks
with low complexity for the massive SM-MIMO. First, the grouped
Wha Sook Jeon, Senior Member, IEEE, Made Harta Dwijaksara,
transmission scheme can introduce the desired structured sparsity of
and Dong Geun Jeong, Senior Member, IEEE
multiple SM signals in the same transmission group for improved signal
detection performance. Second, the SSP algorithm can jointly detect
multiple SM signals with low complexity. Third, by using SM signal in-
terleaving, we can fully exploit the channel diversity to further improve Abstract—To support various Internet of Things (IoT) applications, the
the signal detection performance, and the gain from SM signal inter- Bluetooth Low Energy (BLE) standard specifies a wide range of parameter
leaving can approach that of the ideal case of mutually independent values for the neighbor discovery process (NDP). The parameter values
channel matrices in the same transmission group. Moreover, we have used during neighbor discovery directly affect the performance of the NDP.
Therefore, an optimal parameter setting is essential to achieve the best
quantified the gain from SM signal interleaving. Simulation results
tradeoff between discovery latency and energy consumption. An analytical
have confirmed the near-optimal performance of the proposed scheme. model can offer a beneficial guideline for such a parameter selection. In
this paper, we propose a general model for analyzing the performance of
R EFERENCES NDP in BLE networks. In the model, the operations of the scanner and the
advertiser, which are two main components of NDP, are expressed on the
[1] M. Di Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, “Spatial discrete-time axis. Based on the Chinese Remainder Theorem (CRT),
modulation for generalized MIMO: Challenges, opportunities and imple- the discovery latency and energy consumption of advertiser are derived.
mentation,” Proc. IEEE, vol. 102, no. 1, pp. 56–103, Jan. 2014. The numerical results from our model are almost the same as the simu-
[2] P. Yang, M. Di Renzo, Y. Xiao, S. Li, and L. Hanzo, “Design guidelines lation results, for any parameter values specified by the standard. When
for spatial modulation,” IEEE Commun. Surveys Tuts., vol. 17, no. 1, considering that BLE is one of candidate communication technologies for
pp. 6–26, 1st Quart. 2015. IoT, the proposed model is expected to be very useful in setting the default
[3] J. Zheng, “Signal vector based list detection for spatial modulation,” IEEE or initial values of NDP parameters for various IoT applications.
Wireless Commun. Lett., vol. 1, no. 4, pp. 265–267, Aug. 2012.
[4] J. Wang, S. Jia, and J. Song, “Generalised spatial modulation system with Index Terms—Bluetooth Low Energy (BLE), discovery latency, energy
multiple active transmit antennas and low complexity detection scheme,” consumption, Internet of Things (IoT), neighbor discovery.
IEEE Trans. Wireless Commun., vol. 11, no. 4, pp. 1605–1615, Apr. 2012.
[5] R. M. Legnain, R. H. M. Hafez, and A. M. Legnain, “Improved spatial
modulation for high spectral efficiency,” Int. J. Distrib. Parallel Syst., I. I NTRODUCTION
vol. 3, no. 2, pp. 1–7, Mar. 2012.
[6] R. M. Legnain, R. H. M. Hafez, I. D. Marsland, and A. M. Legnain, Bluetooth low energy (BLE), which is a complementary technology
“A novel spatial modulation using MIMO spatial multiplexing,” in Proc. of classic Bluetooth and is targeting ultra-low-power and low-cost
ICCSPA, Feb. 2013, pp. 1–4. communication, has attracted much attention recently as one of the
[7] J. A. Cal-Braz and R. Sampaio-Neto, “Low-complexity sphere decoding
detector for generalized spatial modulation systems,” IEEE Commun.
key enabling technologies for the Internet of Things (IoT) [1].
Lett., vol. 18, no. 6, pp. 949–952, Jun. 2014. To design an ultra-low-power BLE system, some major modifica-
[8] M. Duarte and Y. Eldar, “Structured compressed sensing: From theory to tions were made to the classic Bluetooth, one of which was for the
applications,” IEEE Trans. Signal Process., vol. 59, no. 9, pp. 4053–4085, neighbor discovery process (NDP) [2]. All communications in BLE
Sep. 2009. networks must involve NDP in the first place since it enables a BLE
[9] B. Shim, S. Kwon, and B. Song, “Sparse detection with integer constraint
using multipath matching pursuit,” IEEE Commun. Lett., vol. 18, no. 10,
device to set up a connection or exchange information with its neigh-
pp. 1851–1854, Oct. 2014. bors. Therefore, it is very desirable to have a fast and energy-efficient
[10] A. Garcia-Rodriguez and C. Masouros, “Low-complexity compressive NDP. To this end, unlike classic Bluetooth which may use all the
sensing detection for spatial modulation in large-scale multiple ac- available channels for discovery purposes, BLE dedicates only three
cess channels,” IEEE Trans. Commun., vol. 63, no. 7, pp. 2565–2579, special channels called advertising channels to neighbor discovery.
Jul. 2015.
[11] C. Yu et al., “Compressed sensing detector design for space shift keying Note that fewer advertising channels lead to fast discovery and this
in MIMO systems,” IEEE Commun. Lett., vol. 16, no. 10, pp. 1556–1559, can give the devices more chances to put their transmitter/receiver
Oct. 2012. electronics into sleep. In addition, BLE adopts more relaxed timing
[12] W. Liu, N. Wang, M. Jin, and H. Xu, “Denoising detection for the gener- for NDP so that a device can flexibly control its duty cycle during
alized spatial modulation system using sparse property,” IEEE Commun. NDP. The timing is determined by several discovery parameters of
Lett., vol. 18, no. 1, pp. 22–25, Jan. 2014.
[13] X. Wu, H. Claussen, M. D. Renzo, and H. Haas, “Channel estima- which values directly affect the NDP performance. The BLE standard
tion for spatial modulation,” IEEE Trans. Commun., vol. 62, no. 12, specifies a wide range of feasible parameter values for NDP, to support
pp. 4362–4372, Dec. 2014.
[14] S. M. Kay, Fundamentals of Statistical Signal Processing, Volume II:
Detection Theory. Upper Saddle River, NJ, USA: Prentice-Hall, 1998. Manuscript received February 1, 2015; revised August 2, 2015 and
December 11, 2015; accepted April 13, 2016. Date of publication April 25,
2016; date of current version February 10, 2017. This work was supported by
the National Research Foundation of Korea funded by the Korean Government
(MSIP) under Grant 2015R1A5A7037372. The review of this paper was
coordinated by Dr. M. Dianata.
W. S. Jeon and M. H. Dwijaksara are with the Department of Computer Sci-
ence and Engineering, Seoul National University, Seoul 151-742, South Korea
(e-mail: [email protected]; [email protected]).
D. G. Jeong is with the Department of Electronics Engineering, Hankuk
University of Foreign Studies, Yongin 449-791, Korea (e-mail: dgjeong@hufs.
ac.kr).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TVT.2016.2558194

0018-9545 © 2016 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.