Professional Documents
Culture Documents
Abstract—A new Multiple Input Multiple Output (MIMO) work MIMO [6], massive MIMO [7] and interference-limited
transmission scheme termed as Generalised Pre-coding aided MIMO [8] concepts.
Spatial Modulation (GPSM) is proposed, where the key idea
is that a particular subset of receive antennas is activated and Despite having a plethora of studies on classic MIMO
the activation pattern itself conveys useful information. This systems, their practical constraints, such as their I/Q imbal-
is in contrast to the previously proposed Spatial Modulation ance, their transmitter and receiver complexity as well as
(SM) schemes, which operated by activating a specific subset the cost of their multiple Radio Frequency (RF) Power Am-
of transmit antennas. We provide both analytical and numerical
results characterizing the GPSM scheme proposed for both con- plifiers (PA), Digital-Analogue / Analogue-Digital (DA/AD)
ventional as well as for large-dimensional MIMO configurations converters etc have received limited attention [9], despite
subject to both realistic imperfect Channel State Information the many-fold benefits of MIMO systems. To circumvent
at the Transmitter (CSIT) and to the low-rank approximation these problems, low complexity alternatives to conventional
invoked for large-dimensional MIMO configurations. We also MIMO transmission schemes have also been proposed, such
design a so-called reinforcement matrix for attaining substantial
performance improvements for our proposed GPSM scheme. Our as the Antenna Selection (AS) [10] and the Spatial Modulation
investigations show that the GPSM scheme constitutes a flexible (SM) [11] philosophies, where AS constitutes an attractive use
alternative to the state-of-the-art MIMO transmission schemes, of MIMOs associated with a low front-end complexity, which
especially because it is capable of achieving a high throughput. is achieved by switching a subset of all available antenna
Moreover, the benefits of mapping information to the spatial elements in both/either the transmitter and/or receiver. This
domain rather than relying on conventional modulation has
substantial benefits in the medium to high Signal to Noise Ratio scheme benefits from a high diversity gain and it is also
(SNR) region. Quantitatively, GPSM is capable of supporting robust to channel estimation errors. Hence it has been adopted
the same throughput as the conventional full-multiplexing gain in the Long Term Evolution (LTE) system’s uplink. On the
based MIMO arrangements at an SNR gain of about 1dB at other hand, SM constitutes a novel MIMO technique, which
the same receiver complexity. Furthermore, the reinforcement is motivated by the goal of providing a higher throughput
matrix aided GPSM scheme attains a further 3-4dB performance
improvement as compared to the conventional GPSM scheme. than a single-antenna aided system, while maintaining both
a lower complexity and a lower cost than the conventional
Index Terms— MIMOs. Since only one transmit antenna is activated at a time,
traditional problems associated with MIMOs, such as the Inter-
Channel-Interference (ICI) and Inter-Antenna-Synchronisation
I. I NTRODUCTION
(IAS) are avoided.
information in the spatial domain. As a specific counterpart modulated symbols mapped to it. More explicitly, in the PSM
of the original SM, PSM benefits from both a low cost and arrangement, only the kth receive antenna is activated by the
complexity at the receiver side, therefore it may be considered spatial symbol k ∈ Z, where Z = {1, . . . , N r } denotes the
to be most suitable for downlink transmission. indices of activated receive antennas.
Hence, in this paper, we extend and generalise the PSM More explicitly, the spatial modulated super-symbol ss ∈
scheme of [22] to offer a full treatment, where we propose CNr ×1 is constituted by an all-zero vector, except for its
the Generalised Pre-coding Aided Spatial Modulation (GPSM) kth entry occupied by the conventional modulated symbol
concept as a new MIMO transmission scheme. We commence bm ∈ A and E[|bm |2 ] = 1, where A denotes the alphabet of
by its analytical study under the idealised simplifying as- a particular M -ary PSK/QAM scheme. Let s km be an explicit
sumption of having perfect Channel State Information at the representation of the super-symbol ss, indicating the transmis-
Transmitter (CSIT). Then two practical issues are investigated, sion of bm by activating the kth antenna. Mathematically, it
namely the detrimental impact of realistic imperfect CSIT can be written as s km = e k bm , where e k is the kth column
followed by a low-rank approximation invoked for large- of an identity matrix of size N r , representing its activation
dimensional MIMOs. Furthermore, we propose a reinforce- by the spatial symbol k. As a result, the total number of bits
ment matrix aided GPSM scheme, which is capable of boosting transmitted by the PSM scheme is k ef f = kssk + kmod , where
the performance of GPSM. kssk = log2 (Nr ) is the extra information conveyed by the
The rest of our paper is organised as follows. In Section II, spatial symbol.
we introduce the concept as well as the detection methods of After transmitter pre-coding, the resultant transmit signal
the GPSM scheme under the idealised simplifying assumption x ∈ CNt ×1 can be written as
of having perfect CSIT, which is complemented by our discus-
sions on the associated detection complexity. This is followed
by our analytical study in Section III under the perfect CSIT x = P s km . (1)
assumption as well as by our investigations of the impact of
realistic imperfect CSIT, of a low-rank approximation and of In order to avoid power amplification during the pre-
a reinforcement matrix aided enhancement. Our simulation processing, i.e. for the sake of maintaining E[||xx ||2 ] = 1, we
results are provided in Section IV, while we conclude in have to impose the constraint Tr{P P P H } = Nr on the pre-
Section V. coding matrix.
Notations: In this paper, we use lower-case bold variable to 2) Design: The signal observed at the N r receive antennas
represent a vector and capital bold fonts to represent a matrix. may be written as
The operator E represents expectation, Tr denotes the trace
of a square matrix, R represents taking the real part and ·
y = H T P s km + w , (2)
indicates flooring operator. In addition, the superscripts () T
and ()∗ represent matrix transpose and matrix conjugate, re-
spectively, while we use ()H to represent the matrix transpose where w ∈ CNr ×1 is the circularly symmetric complex
and conjugate. Gaussian noise vector with each entry having a zero mean
and a variance per dimension of σ 2 = N0 /2, i.e. wj ∼
II. S YSTEM D ESCRIPTION CN (0, N0 ), ∀j, while H ∈ CNt ×Nr represents the MIMO
channel involved. We assume furthermore that each entry of H H
Consider a MIMO system equipped with N t transmit anten- undergoes frequency-flat Rayleigh fading and it is uncorrelated
nas and Nr receive antennas, where we assume N t ≥ Nr . In between different super-symbol transmissions, while remains
this MIMO set-up, a maximum of N s = min(Nt , Nr ) = Nr static within the duration of a super-symbol’s transmission.
parallel data streams may be supported, conveying a total of The super-symbols transmitted are statistically independent
kef f = Nr kmod bits altogether, where k mod = log2 (M ) from the noise.
denotes the number of bits per symbol of a conventional
Hence, the pre-coding matrix designed would have to max-
M -ary PSK/QAM scheme [13], [24]. Transmitter pre-coding
imize the desired decision D 0 associated with super-symbol
relying on the matrix P ∈ CNt ×Nr may be used for pro-
s km and simultaneously minimize the undesired i.e. erroneous
actively mitigating the ICI at the transmitter by preprocessing
decisions D1 associated with all other super-symbols s jn , ∀j =
the source signal before its transmission by exploiting the
k, ∀n = m. Mathematically, the optimum pre-coding matrix
CSIT [25]. As a benefit, a low-complexity single-stream
P o should satisfy
detection scheme may be used by the downlink receiver,
because the ICI is eliminated at the downlink transmitter.
P o = arg max {D0 − D1 } , (3)
P
A. Pre-coding Aided Spatial Modulation
1) Concept: In contrast to the above-mentioned classic where the desired decision D 0 may be written as
full-multiplexing of N r data streams, the concept of PSM
relies on activating one out of N r receive antennas with D0 = −||yy − s km ||2
the aid of transmitter pre-coding, where the particular choice ∝ 2R{yy H s km } − ||sskm ||2
of the receive antenna activated conveys extra information
in addition to the information carried by the conventional ∝ 2R{(sskm )H P H H ∗s km + w H s km } − ||sskm ||2 , (4)
ZHANG et al.: GENERALISED PRE-CODING AIDED SPATIAL MODULATION 3
while the undesired decisions D 1 may be written as 1) Concept: In our GPSM scheme, a total of N a < Nr
receive antennas are activated so as to facilitate the simulta-
D1 = −||yy − s jn ||2 neous transmission of N a data streams. Hence, the number of
j=k n=m
bits in GPSM conveyed by a spatial symbol becomes k ssk =
∝ 2R{yy H s jn } − ||ssjn ||2 log2 (|Ct |), where the set Ct contains all the combinations
j=k n=m associated with choosing N a active receive antennas from N r
receive antennas. Thus the effective number of bits transmitted
∝ 2R{(ssjn )H P H H ∗s jn + w H s jn } − ||ssjn ||2 . (5)
by our GPSM scheme is k ef f = kssk + Na kmod .
j=k n=m
Unlike in standard PSM, where each spatial symbol cor-
To achieve the above objective, the pre-coding matrix has responds to a particular receive antenna, each spatial symbol
to ensure that no energy leaks into the unintended receive in GPSM corresponds to a particular pattern of N a activated
antennas. In other words, we capture all the energy by the receive antennas.
kth receive antenna. Hence, the so-called Minimum-Power
Distortionless Response (MPDR) of [26] design criterion can Example II.1. For example, activating N a = 2 receive
be employed in form of its typical instantiations, namely antennas from N r = 4 receive antennas results in a total
the Zero Forcing (ZF) pre-coding and the Minimum Mean of |Ct | = 6 legitimate activation patterns, i.e. the pat-
Square Error (MMSE) pre-coding. In this paper, we take terns of Ct = {[1, 2], [1, 3], [1, 4], [2, 3], [2, 4], [3, 4]}. This
ZF pre-coding into consideration. More explicitly, the ZF configuration delivers k ssk = 2 bits of information because
pre-coding [27] constitutes the pseudo-inverse of the MIMO log2 (|6|) = 2, albeit only 2 kssk = 4 out of |Ct | = 6 patterns
channel H T , which is written as are required for the actual transmission of two bits. Let us
define the set of selected activation patterns as C ⊂ C t . Upon
H ∗ (H
P = βH H T H ∗ )−1 , (6) selecting for example C = {C t (1), Ct (2), Ct (3), Ct (4)}, we
have the mapping between the spatial symbol and the pattern
where β = Nr /Tr[(H H T H ∗ )√
−1 ] is a normalisation factor.
of activated receive antennas as
Note that letting simply β = Nt also satisfies our power
constraint statistically.
3) Detection: When ZF pre-coding is employed, the re- k = 1 → C(1) = [1, 2] k = 2 → C(2) = [1, 3]
ceived signal can be further written as k = 3 → C(3) = [1, 4] k = 4 → C(4) = [2, 3]
y = βsskm + w , (7)
Note that, the above pattern is not a balanced pattern thus
which may be decomposed into the receive antennas are not uniformly activated. In this paper,
we simply opt for C = {C t (1), . . . , Ct (2kssk )}, while the
yk = βbm + wk ; yj = wj , ∀j = k. (8) optimisation of pattern selection will be left for our future
work. Furthermore, we let C(k) and C(k, i) denote the kth
Hence the joint detection of both the conventional modulated activation pattern and the ith activated receive antenna in
symbol bm and of the spatial symbol k obeys the Maximum the kth activation pattern, respectively. We also let sskm be
Likelihood (ML) criterion. More explicitly, we have an explicit representation of the super-symbol ss, indicating
that the receive antenna pattern k is activated and N a con-
[m̂, k̂] = arg min {||yy /β − s jn ||2 }, (9)
s jn ∈B ventional modulated data streams bbm = [bm1 , . . . , bmNa ] are
transmitted. In other words, we can revise the expression
where B = Z × A stands for the super-alphabet of the of the spatially modulated super-symbol as sskm = Ω kb m ,
super-symbol s jn . Alternatively, decoupled detection may also where b m ∈ CNa ×1 contains Na conventional modulated data
be employed, which treats the detection of the conventional streams of our GPSM scheme and Ω Ωk = I [C(k)] ∈ BNr ×Na
modulated symbol b m and the spatial symbol k separately. In is constituted by the selected columns of an identity matrix
this low-complexity variant, we have of size Nr , where the specific selection is denoted by C(k)
and BNr ×Na represents a (Nr × Na )-element matrix having
k̂ = arg max {|yj |2 } (10)
j∈[1,Nr ] binary entries.
m̂ = arg max {|yk̂ /β − bn |2 }. (11) 2) Detection: When ZF pre-coding is employed, the signal
n∈[1,M]
observed at the N r receive antennas may be written as
Thus, correct detection is declared, when we have k̂ = k and y = βsskm + w . As in the standard PSM, we have two optional
m̂ = m. detection algorithms. When joint detection is considered,
similar to (9), we have
considered, we have since they both require DL pre-processing and CSIT. This
Na constitutes the main difference between our PSM/GPSM and
k̂ = arg max { |yC(j,i) |2 } (13) the classic SM/GSM [12], [20], [21], although they may
j∈[1,|C|] be considered as dual pair of each other. As a result, the
i=1
m̂i = arg max {|yC(k̂,i) /β − bni |2 }, (14) comparisons between PSM/GPSM and SM/GSM cannot be
ni ∈[1,M] readily made under identical circumstances, because
where (14) has to be evaluated for all N a data streams and • the inevitable complexity of pre-processing imposed by
correct detection is declared, when we have k̂ = k and m̂i = PSM/GPSM is not imposed on SM/GSM.
mi , ∀i. • the channel knowledge requirement of PSM/GPSM is
It is thus plausible that the standard PSM scheme constitutes more demanding than that of SM/GSM, since the former
a special case of GPSM for N a = 1 and the detection scheme suffers further from potential feedback impairments in
of (13) is non-coherent, while that of (14) only requires the addition to the channel estimation errors encountered at
knowledge of the scaling factor β. the receiver, that are common for both schemes.
3) Complexity: Let us now discuss the complexity of the
GPSM scheme in comparison to the conventional pre-coding In fact, PSM/GPSM may become beneficial in different ap-
aided MIMO arrangement supporting N r data streams. We plications, for example in relay selection. Furthermore, the
first note that the transmitter pre-coding exhibits the same PSM/GPSM scheme proposed may provide the potential of
complexity in both schemes, hence our focus is on the using efficient power allocation at the transmitter side for
detection complexity at the receiver. the sake of attaining an additional beam-forming gain, which
In conventional MIMO arrangements, when transmitter pre- constitutes an important benefit of our PSM/GPSM scheme, as
coding is capable of perfectly separating the N r parallel data facilitated by the knowledge of the CSIT. More importantly,
streams, such as the ZF pre-coding, low-complexity single- they may be used in a hybrid manner in conjunction with
stream detection is facilitated at the receiver. Hence, the SM/GSM. Hence, in summary, the PSM/GPSM scheme pro-
detection complexity increases linearly with the number of posed in this paper constitutes a complement, rather than a
receive antennas N r , but exponentially with the number of competitor of the classic SM/GSM.
bits per symbol in the M-ary PSK/QAM scheme employed.
Hence, we may characterise the complexity order as O(N r M ). III. S YSTEM A NALYSIS
On the other hand, in the GPSM scheme, the complexity of
the joint detection of (12) is quite high, which is on the order A. Analytical Average Bit Error Probability
determined by the super-alphabet B, i.e. obeys O(|C|M Na ).
Let us now use the union-bound approach for deriving
By contrast, the separate detection of (13) and (14) facili-
the analytical Average Bit Error Probability (ABEP) P e of
tates a substantially reduced complexity. More explicitly, the
our GPSM scheme using the joint detection approach, where
complexity is imposed by detecting N a data streams as in
formally, we have
a conventional MIMO arrangement, plus the complexity (Δ)
imposed by the comparisons invoked for detecting the spatial 1
symbol, which may be written as O(N a M + Δ). Pe ≤ E{ d(ss km , s jn )Pr (sskm → s jn )},
kef f |B| j
To elaborate a little further, in the separate detection regime, sk
m ∈B s n ∈B=s m
k
Proof: Let us now consider the first case, where only the Theorem III.1. The PEP Pr (sskm → sjn ) of GPSM may be
spatial symbol is erroneous and hence we have compactly formulated as
PK (sskm → s jm ) = Pr (||yy /β − s km ||2 > ||yy /β − s jm ||2 ) B−C
Pr (ss m → s n ) = Q β
k j
, (24)
= Pr [|bm |2 − 2R(yk∗ bm /β) > |bm |2 − 2R(yj∗ bm /β)] (D + E)N0
= Pr [R(yj∗ bm /β) − R(yk∗ bm /β) > 0] where we have
= Pr [R(b∗m wj /β) − R(b∗m wk /β) > |bm |2 ]. (19) |bn |2 + |bm |2
B= i i
; C= R(b∗mi bni ); (25)
2
Since wj and wk are independent identical distributed i∈Ψ i∈Ψ1
|b | 2 + |b |2
Hence, the PEP Pr (sskm → s kn ) may be written as in (24). 1
PJ (sskm → sjn ) = Q ⎝β ⎠.
n m
(22)
2N0
B. Robustness Investigations
Proof: Finally, when both the spatial symbol and the conven-
1) Channel Estimation Error: Like in all transmitter pre-
tional modulated symbol are erroneous, we have
coding schemes, an important aspect related to GPSM is its re-
PJ (sskm → s jn ) = Pr (||yy /β − s km ||2 > ||yy /β − s jn ||2 ) silience to CSIT inaccuracies. When considering a Frequency
Division Duplex (FDD) system for example, the cause of this
= Pr [|bm |2 − 2R(yk∗ bm /β) > |bn |2 − 2R(yj∗ bn /β)] inaccuracy may be many-fold, such as for example the channel
= Pr [R(yj∗ bn /β) − R(yk∗ bm /β) > (|bn |2 − |bm |2 )/2] estimation errors at the receiver, quantization errors imposed
= Pr [R(b∗n wj /β) − R(b∗m wk /β) > (|bn |2 + |bm |2 )/2]. on the CSIT feedback and error-infested imperfect feedback
(23) links, which are also outdated. In this paper, we investigate the
impact of inaccurate CSIT on our GPSM scheme by assuming
Now the LHS of (23) is a Gaussian random variable obeying that the transmitter is capable of accurately tracking the long
N (0, (|bn |2 + |bm |2 )N0 /2β 2 ). Hence, the PEP PJ (sskm → s jn ) term average Channel State Information (CSI), but it is subject
may be written as in (22). to instantaneous CSI errors. This consideration is practical for
2) ABEP of GPSM: The above analytic formulae provide example in the LTE system, where usually carefully designed
an insight into the three different error-components of our periodic pilot patterns are embedded into the payload.
PSM scheme. When considering the analytical ABEP of More explicitly, we let H = H a +H H i , where H a represents
GPSM, partitioning the error patterns similar to (17) would the matrix of the average CSI, with each entry obeying the
be obscure. Instead, we derive it with the aid of the three
1 Note that the PEP derivations of Section III.A were carried out under
terms in (19), (21) and (23). The final ABEP of GPSM still
the perfect CSIT assumption, while more involved numerical integrations are
follows (16), while the derivation of PEP obeys the following required for the imperfect CSIT scenario for the sake of averaging the effects
theorem. of CSIT imperfections.
6 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. XX, NO. XX, MONTH 2013
Pr [ R(zj∗i bni ) − R(vk∗i bmi ) > (|bni |2 + |bmi |2 )/2
i∈Ψ i∈Ψ
= Pr [ R(vj∗i bni ) − R(vk∗i bmi ) + R(vk∗i bni ) − R(vk∗i bmi ) > (|bni |2 + |bmi |2 )/2 − R(b∗mi bni ) (28)
i∈Ψ0 i∈Ψ1 i∈Ψ i∈Ψ1
N (0,DN0 /β 2 ) N (0,EN0 /β 2 ) B C
Na
complex Gaussian distribution of h a ∼ CN (0, σa2 ) and H i is 2
i=1 |yC(j,i) | will be higher for the intended antenna pattern
the instantaneous CSI deviation matrix obeying the complex of j = k.
Gaussian distribution of h i ∼ CN (0, σi2 ), where we have σa2 + By contrast, this artificial energy leakage will degrade the
σi2 = 1. As a result, only H a is available at the transmitter, detection regime of (14), since it introduces ICI upon detecting
when forming the ZF pre-coder of P P . There is a plethora a particular conventionally modulated symbol. By exploiting
of CSIT inaccuracy counter-measures conceived for general the fact that a clear decision boundary exists when detecting
transmitter pre-coding schemes in the literature [28], [29], but a PSK modulated symbol, we can thus shape the ICI in
they are beyond the scope of our GPSM discussions in this order to minimize its contamination by beneficially aligning
paper. its phase with that of the desired signal. In this case, the above-
2) Low-rank Approximation: We now discuss another im- mentioned energy leakage will become beneficial to both the
portant aspect associated with GPSM, namely the low-rank regimes of (13) and of (14).
pre-coder scenario, while assuming perfect CSIT for the sake Mathematically, to construct the beneficially co-phased ar-
of avoiding the potentially obfuscating effects of having multi- tificial ICI, we have to rotate the phase of the original ICI
ple combined sources of impairments. Since typical linear pre- relative to the desired signal. Hence, given the super-symbol
coders involve matrix inversion, low-rank pre-coder solutions s and its entry of s v = ω vb m , ∀v ∈ [1, Nr ], where ω v
are preferable, when the matrix dimension is high or when denotes the vth row of Ω k , the relative phase represented by
the matrix is ill-conditioned. Originally, the kth column vector φu,v between sv and su may be shown to become φ u,v =
p k of the transmitter pre-coding matrix P P may be written as sv − (ρu,v su ), where ρu,v is the entry of the Maximal
p k = R −1
D kh ∗
, where we have R D = H ∗ T
H . The low-rank Ratio Transmission (MRT) matrix H T H ∗ . As a result, the
pre-coder design invokes a projection matrix P P k ∈ CNt ×L appropriately shaped energy leakage can be constructed by
for the sake of transforming the original inversion to an L- the reinforcement matrix of R R ∈ CNr ×Nr , having entries of
dimensional subspace of ru,v = φu,v ρu,v . Consequently, our pre-coding matrix of (6)
would have to be modified as
pk = P ∗k (P
P Tk R D P ∗k )−1P Tk h∗k . (29)
P = βr H ∗ (H
H T H ∗ )−1R T , (31)
Typical low-rank pre-coders relying on the first L principal
where we set βr = Nr /Tr[R R ∗ (H
H T H ∗ )−1R T ] for ensuring
components of the eigen-decomposition of R R D achieve a low
appropriate normalisation, which is typically smaller than the
performance [26]. Hence, let us now invoke low-rank pre-
original β.
coders relying on the Taylor Polynomial (TP) approximation
Let us now see how the reinforcement matrix assists in the
detailed in [26], where RRD is expanded into its TP repre-
detection process of GPSM. Assuming that the vth receive
sentation and the first L terms are then retained to form the
antenna is activated, namely we have v ∈ C(k), then the
projection matrix. According to [26], we have
received signal becomes:
P k = [II N , R ∗D , . . . , R ∗L−1
D ](II L ⊗ h k ), (30) yv = βr ρv,v sv + βr φu,v ρu,v su + wv , (32)
where ⊗ stands for the Kronecker product. u∈C(k)/v
ρ∗u,v s∗u
= βr ρv,v sv + βr sv ρu,v su + wv , (33)
|ρu,v |
C. Further Improvements u∈C(k)/v
⎛ ⎞
Let us now conceive further improvements for detect-
= βr ⎝ρv,v + |ρu,v |⎠ sv + wv . (34)
ing the proposed GPSM without incurring the performance
u∈C(k)/v
degradation effects discussed in Section III-B. When separate
detection is considered, we can see that (13) is all about deter- This is in contrast to the conventional GPSM design, where
mining the correct antenna pattern / spatial symbol by simply yv = βsv + wv . Our empirical results provided in the next
sorting the received power accumulated over each receive section will show that the reinforcement matrix arrangement
antenna pattern. The above-mentioned ZF design criterion attains a substantially improved BER performance, thanks to
aims for avoiding any energy leakage to other antenna indices. the artificial appropriately shaped energy leakage.
However, we may be able to improve the above design by
allowing energy leakage to the specific antenna indices of the IV. N UMERICAL R ESULTS
activated antenna pattern. More explicitly, this will naturally Let us now characterize the Bit Error Ratio (BER) per-
improve the detection philosophy of (13), since the quantity formance of our GPSM scheme in comparison to that of the
ZHANG et al.: GENERALISED PRE-CODING AIDED SPATIAL MODULATION 7
TABLE I
E FFECTIVE THROUGHPUT kef f AND DETECTION COMPLEXITY χ OF 0
10
0
10
GPSM AND OF THE CONVENTIONAL MIMO BENCH - MAKER N =1
a
N =1
a
CORRESPONDING TO Na = Nr ( HIGHLIGHTED IN •) FOR DIFFERENT Na = 2 N =2
a
SETTINGS , WHERE kef f = kssk + Na kmod , χ = Nr + M + Na M FOR −1 −1 N =3
GPSM, χ = M + Nr M FOR CONVENTIONAL MIMO. 10
hollow legends: seperate
10 a
Na = 4
filled legends: joint
dotted lines: analytical
Nt Nr Na M kef f χ −2 −2
10 10
4/8 2 1 4 3 = 1+1×2 10 = 2+4+1×4
•
BER
BER
4/8 2 2 4 4 = 0+2×2 12 = 0+4+2×4
4/8 2 1 8 4 = 1+1×3 18 = 2+8+1×8
−3 −3
8/16 4 1 4 4 = 2+1×2 12 = 4+4+1×4 10 10
8/16 4 2 4 6 = 2+2×2 16 = 4+4+2×4
8/16 4 3 4 8 = 2+3×2 20 = 4+4+3×4
• 8/16 4 4 4 8 = 0+4×2 20 = 0+4+4×4 −4
10
−4
10
8/16 4 1 64 8 = 2+1×6 132 = 4+64+1×64
8/16 4 2 8 8 = 2+2×3 28 = 4+8+2×8 N =8;N =2 N =8;N =4
t r t r
16 8 1 2 5 = 3+1×2 16 = 8+4+1×4 −5
10
−5
10
16 8 2 2 8 = 4+2×2 20 = 8+4+2×4 0 5 10 15 0 5 10 15
SNR [dB] SNR [dB]
16 8 3 2 11 = 5+3×2 24 = 8+4+3×4 b b
16 8 4 2 14 = 6+4×2 28 = 8+4+4×4
16 8 5 2 15 = 5+5×2 32 = 8+4+5×4 Fig. 1. BER performance of both the GPSM scheme and of the conventional
16 8 6 2 16 = 4+6×2 36 = 8+4+6×4 MIMO scheme for both {Nt , Nr } = {[8, 2]} (left plot) and for {Nt , Nr } =
• 16 8 8 2 16 = 0+8×2 36 = 0+4+8×4 {[8, 4]} (right plot).
16 8 2 64 16 = 4+2×6 200 = 8+64+2×64
16 8 3 16 17 = 5+3×4 72 = 8+16+3×16
0 0
10 10
dotted lines: N = 8; N = 2 Na = 1, M = 8 dotted lines: no error Na = 1
t r
solid lines: Nt = 8; Nr = 4 N = 2, M = 4 solid lines: erroneous N =2
a a
−1 Na = 1, M = 64 −1 Na = 3
10 10
Na = 2, M = 8 Na = 4
N = 3, M = 4 Na = 5
a
−2 N = 4, M = 4 −2 N =6
10 a 10 a
Na = 8
BER
BER
−3 −3
10 10
−4 −4
10 10
N =16
t
N =8
−5 −5 r
10 10
−4 −2 0 2 4 6 8 10 12 14 −6 −4 −2 0 2 4 6 8 10 12
SNR [dB] SNR [dB]
b b
Fig. 2. BER performance of our GPSM scheme in comparison to the Fig. 4. Impact of CSIT error on the BER performance of both the GPSM
identical-rate full multiplexing MIMO bench-mark scheme using QPSK for scheme and of the conventional MIMO scheme for {Nt , Nr } = {[16, 8]},
{Nt , Nr } = {[8, 2], [8, 4]}. where the CSI error was σi = 0.1.
0
10 for the standard MIMO settings. Explicitly, the GPSM using
dotted lines: N = 16; N = 4 Na = 1, M = 64
t r
solid lines: Nt = 16; Nr = 8 Na = 2, M = 8 Na = 6 achieves in excess of 1dB SNR gain at BER of
−1
10
Na = 3, M = 4 Pe = 10−5 , when compared to the conventional MIMO bench-
N = 4, M = 4
a mark scheme having the same throughput and complexity, as
N = 2, M = 64
a highlighted by the symbol in Table I.
Na = 3, M = 16
When a low CSIT error of σ i = 0.1 is considered, Fig
−2
10
Na = 6, M = 4
4 shows that the GPSM scheme associated with various
BER
N = 8, M = 4
a
−3 configurations is less adversely affected by the CSIT error in
10
comparison to the conventional MIMO bench-mark scheme.
As expected, the higher the GPSM throughput, the more grave
−4
10 the BER performance erosion becomes.
When a severe CSIT error of σ i = 0.4 is encountered, it can
−5
be seen in Fig 5 that a substantially higher BER performance
10
−5 0 5 10 15 degradation is observed in comparison to that found in Fig 4.
SNRb [dB]
When considering various GPSM throughputs having diverse
configurations, those supporting a lower throughput exhibit
Fig. 3. BER performance of our GPSM scheme in comparison to the
identical-rate full multiplexing MIMO bench-mark scheme using QPSK for
a better resilience against CSIT errors than those supporting
{Nt , Nr } = {[16, 4], [16, 8]}. a higher throughput, which is in line with our expectations.
Although, the BER performance of the GPSM scheme using
Na = 6 becomes worse than that of the conventional MIMO
impose a higher detection complexity than those that prefer scheme for σi = 0.4 because the spatial symbol errors
mapping their information to spatial symbols, namely those associated with the antenna pattern detection inflict error-
corresponding to {N t , Nr , Na } = {[16, 4, 3], [16, 8, 6]}. Since precipitation upon the QPSK modem used. However, the BER
the increased complexity only brings about BER performance performance of both schemes remains simply unacceptable
improvements in the low to medium SNR region, it might be for the σi = 0.4 scenarios. Hence, more sophisticated joint
a better design option to allocate information to the spatial iterative channel estimation and data detection methods have
symbols right across the entire SNR region, rather than using to be conceived in conjunction with channel coding.
an SNR-dependent mapping.
D. Low-rank Approximation
C. Imperfect CSIT Fig 6 shows the effect of using TP based low-rank pre-
Fig 4 and Fig 5 show the BER performance of both the coders on both the GPSM scheme and on the conventional
GPSM scheme using separate detection and of the conven- MIMO bench-mark scheme. As expected, the lower the rank,
tional MIMO benchmark scheme, when either a relatively the worse the performance. Nonetheless, supporting a lower
low CSIT error of σ i = 0.1 or a severe CSIT error of GPSM throughput corresponding to N a = 2, the low-rank
σi = 0.4 is encountered by the large-dimensional MIMO approximation remains more robust than in case of a higher
settings, respectively. Focusing our attention on the perfect GPSM throughput corresponding to N a = 3 and Na = 6.
CSIT based scenarios shown in dotted lines in both figures, When considering our GPSM associated with N a = 6
they demonstrate even more explicitly the observations made and the conventional bench-mark MIMO, both having the
ZHANG et al.: GENERALISED PRE-CODING AIDED SPATIAL MODULATION 9
BER
BER
−3 −3 −3
10 10 10
−4 −4 −4
10 10 10
Fig. 5. Impact of CSIT error on the BER performance of both the GPSM Fig. 7. BER performance of conventional GPSM scheme and reinforcement
scheme and of the conventional MIMO scheme for {Nt , Nr } = {[16, 8]}, matrix aided GPSM scheme under {Nt , Nr } = {[16, 8]}.
where the CSI error was σi = 0.4.
0
V. C ONCLUSIONS
10
L=4 f In this paper, we introduced the concept and design of PSM
L=5 h
L=6 and its generalisation as a new MIMO transmission scheme.
−1
10 Na = 2 Both numerical and analytical results were provided. Further-
N =3
a more, the detection complexity of GPSM and its robustness
Na = 6
−2
10 Na = 8
against CSIT error and against a low-rank approximation were
also discussed. Our simulation results demonstrated that the
BER
[10] A. Molisch and M. Win, “MIMO systems with antenna selection,” IEEE Rong Zhang (M’09) received his PhD (Jun 09)
Microwave Mag., vol. 5, no. 1, pp. 46–56, Mar. 2004. from Southampton University, UK and his BSc
[11] R. Mesleh, H. Haas, S. Sinanovic, C. W. Ahn, and S. Yun, “Spatial (Jun 03) from Southeast University, China. Before
modulation,” IEEE Trans. Veh. Technol., vol. 57, no. 4, pp. 2228–2241, doctorate, he was an engineer (Aug 03-July 04) at
July 2008. China Telecom and a research assistant (Jan 06-
[12] M. Renzo, H. Haas, and P. Grant, “Spatial modulation for multiple- May 09) at Mobile Virtual Center of Excellence
antenna wireless systems: A survey,” IEEE Commun. Mag., vol. 49, (MVCE), UK. After being a post-doctoral researcher
no. 12, pp. 182–191, Dec. 2011. (Aug 09-July 12) at Southampton University, he
[13] M. D. Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, took industrial consulting leave (Aug 12-Jan 13)
“Spatial modulation for generalized MIMO: Challenges, opportunities for Huawei Sweden R& D as a system algorithms
and implementation,” Proc. IEEE. [Online]. Available: http://eprints. specialist. Since Feb 13, he has been appointed as
soton.ac.uk/354175/ a lecturer at CSPC group of ECS, Southampton University. He has 25+
[14] M. Di Renzo and H. Haas, “Bit error probability of SM-MIMO over journals in prestigious publication avenues (e.g. IEEE, OSA) and many more
generalized fading channels,” IEEE Trans. Veh. Technol., vol. 61, no. 3, in major conference proceedings. He regularly serves as reviewer for IEEE
pp. 1124–1144, Mar. 2012. transactions/journals and has been several times as TPC member/invited
[15] R. Rajashekar, K. Hari, and L. Hanzo, “Antenna selection in spatial session chair of major conferences. He is the recipient of joint funding of
modulation systems,” IEEE Commun. Lett., vol. 17, no. 3, pp. 521–524, MVCE and EPSRC and is also a visiting researcher of Nanjing University
2013. under Worldwide University Network (WUN). More details can be found at
[16] R. Mesleh, M. D. Renzo, H. Haas, and P. M. Grant, “Trellis coded http://www.ecs.soton.ac.uk/people/rz
spatial modulation,” IEEE Trans. Wireless Commun., vol. 9, no. 7, pp.
2349–2361, July 2010.
[17] E. Basar, U. Aygolu, E. Panayirci, and H. Poor, “Space-time block coded
spatial modulation,” IEEE Trans. Commun., vol. 59, no. 3, pp. 823–832,
Mar. 2011. Lie-Liang Yang (M’98, SM’02) received his BEng
[18] J. Jeganathan, A. Ghrayeb, L. Szczecinski, and A. Ceron, “Space degree in communications engineering from Shang-
shift keying modulation for MIMO channels,” IEEE Trans. Wireless hai TieDao University, Shanghai, China in 1988,
Commun., vol. 8, no. 7, pp. 3692–3703, July 2009. and his MEng and PhD degrees in communications
[19] S. Sugiura, S. Chen, and L. Hanzo, “Generalized space-time shift keying and electronics from Northern (Beijing) Jiaotong
designed for flexible diversity-, multiplexing- and complexity-tradeoffs,” University, Beijing, China in 1991 and 1997, re-
IEEE Trans. Wireless Commun., vol. 10, no. 4, pp. 1144–1153, Apr. spectively. From June 1997 to December 1997 he
2011. was a visiting scientist of the Institute of Radio
[20] A. Younis, N. Serafimovski, R. Mesleh, and H. Haas, “Generalised Engineering and Electronics, Academy of Sciences
spatial modulation,” in 2010 Conf. Record Forty Fourth Asilomar Conf. of the Czech Republic. Since December 1997, he has
Signals, Syst., Comput. (ASILOMAR), Nov. 2010, pp. 1498–1502. been with the University of Southampton, United
[21] J. Wang, S. Jia, and J. Song, “Generalised spatial modulation system Kingdom, where he is the professor of wireless communications in the School
with multiple active transmit antennas and low complexity detection of Electronics and Computer Science. Dr. Yang’s research has covered a
scheme,” IEEE Trans. Wireless Commun., vol. 11, no. 4, pp. 1605–1615, wide range of topics in wireless communications, networking and signal
Apr. 2012. processing. He has published over 290 research papers in journals and
[22] L.-L. Yang, “Transmitter preprocessing aided spatial modulation for conference proceedings, authored/co-authored three books and also published
multiple-input multiple-output systems,” in IEEE 73rd Veh. Technol. several book chapters. The details about his publications can be found at
Conf. (VTC Spring), May 2011, pp. 1–5. http://www-mobile.ecs.soton.ac.uk/lly/. He is a fellow of the IET, served as
[23] A. Stavridis, S. Sinanovic, M. D. Renzo, and H. Haas, “Transmit precod- an associate editor to the IEEE Trans. on Vehicular Technology and Journal
ing for receive spatial modulation using imperfect channel knowledge,” of Communications and Networks (JCN), and is currently an associate editor
in IEEE 75th Veh. Technol. Conf. (VTC Spring), May 2012, pp. 1–5. to the IEEE Access and the Security and Communication Networks (SCN)
[24] S. Sugiura, S. Chen, and L. Hanzo, “A universal space-time architecture Journal.
for multiple-antenna aided systems,” IEEE Commun. Surveys Tuts.,
vol. 14, pp. 401–420, Second Quarter 2012.
[25] A. Goldsmith, Wireless Communications. Cambridge University Press,
2005.
[26] L.-L. Yang, “Capacity and error performance of reduced-rank trans- Lajos Hanzo (FREng, FIEEE, FIET, Fellow of
mitter multiuser preprocessing based on minimum power distortionless EURASIP, DSc) received his degree in electronics
response,” IEEE Trans. Wireless Commun., vol. 7, no. 11, pp. 4646– in 1976 and his doctorate in 1983. In 2009 he was
4655, Nov. 2008. awarded the honorary doctorate “Doctor Honoris
[27] Q. Spencer, A. Swindlehurst, and M. Haardt, “Zero-forcing methods Causa” by the Technical University of Budapest.
for downlink spatial multiplexing in multiuser MIMO channels,” IEEE During his 37-year career in telecommunications he
Trans. Signal Process., vol. 52, no. 2, pp. 461–471, Feb. 2004. has held various research and academic posts in
[28] D. Gesbert, “Robust linear MIMO receivers: A minimum error-rate Hungary, Germany and the UK. Since 1986 he has
approach,” IEEE Trans. Signal Process., vol. 51, pp. 2863–2871, Nov. been with the School of Electronics and Computer
2003. Science, University of Southampton, UK, where
[29] X. Zhang, D. Palomar, and B. Ottersten, “Statistically robust design of he holds the chair in telecommunications. He has
linear MIMO transceivers,” IEEE Trans. Signal Process., vol. 56, pp. successfully supervised 80+ PhD students, co-authored 20 John Wiley/IEEE
3678–3689, Aug. 2008. Press books on mobile radio communications totalling in excess of 10 000
pages, published 1300+ research entries at IEEE Xplore, acted both as TPC
and General Chair of IEEE conferences, presented keynote lectures and has
been awarded a number of distinctions. Currently he is directing a 100-strong
academic research team, working on a range of research projects in the field of
wireless multimedia communications sponsored by industry, the Engineering
and Physical Sciences Research Council (EPSRC) UK, the European Research
Council’s Advanced Fellow Grant and the Royal Society’s Wolfson Research
Merit Award. He is an enthusiastic supporter of industrial and academic
liaison and he offers a range of industrial courses. He is also a Governor
of the IEEE VTS. During 2008 - 2012 he was the Editor-in-Chief of the
IEEE Press and a Chaired Professor also at Tsinghua University, Beijing.
His research is funded by the European Research Council’s Senior Research
Fellow Grant. For further information on research in progress and associated
publications please refer to http://www-mobile.ecs.soton.ac.uk. Lajos has 17
000+ citations.