research article comparison of semidefinite relaxation ...research article comparison of...
TRANSCRIPT
Research ArticleComparison of Semidefinite Relaxation Detectors forHigh-Order Modulation MIMO Systems
Z Y Shao1 S W Cheung2 and T I Yuk2
1 School of Mechanical amp Electric Engineering Guangzhou University Guangzhou Higher Education Mega CenterGuangzhou 510006 China
2Department of Electrical amp Electronic Engineering University of Hong Kong Pokfulam Road Hong Kong
Correspondence should be addressed to Z Y Shao zyshaoeeehkuhk
Received 30 June 2014 Revised 26 October 2014 Accepted 30 October 2014 Published 27 November 2014
Academic Editor Stefano Selleri
Copyright copy 2014 Z Y Shao et alThis is an open access article distributed under theCreativeCommonsAttributionLicensewhichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Multiple-input multiple-output (MIMO) system is considered to be one of the key technologies of LTE since it achievesrequirements of high throughput and spectral efficiency The semidefinite relaxation (SDR) detection for MIMO systems is anattractive alternative to the optimummaximum likelihood (ML) decoding because it is very computationally efficient We proposea new SDR detector for 256-QAM MIMO system and compare its performance with two other SDR detectors namely BC-SDRdetector and VA-SDR detector The tightness and complexity of these three SDR detectors are analyzed Both theoretical analysisand simulation results demonstrate that the proposed SDR can provide the best BLER performance among the three detectorswhile the BC-SDR detector and the VA-SDR detector provide identical BLER performance Moreover the BC-SDR has the lowestcomputational complexity and the VA-SDR has the highest computational complexity while the proposed SDR is in between
1 Introduction
Multiple-input multiple-output (MIMO) system has beenconsidered as a promising solution to provide high datarate and good quality of future wireless communicationsIn MIMO systems detection algorithm is one of the majorchallenges due to its limitations of either unsatisfactoryperformance or high complexity The maximum likelihood(ML) detection can provide the best block-error-rate (BLER)performance but its computational complexity is extremelyhigh since it searches the vectors in the entire latticespace of the transmitted signals Although equalization-baseddetectors such as zero-forcing (ZF) decoding have very lowcomplexity they suffer from unacceptable degradations inBLER performance Sphere decoding (SD) is able to providethe BLER performance of ML detection with less complexityby searching only a subset of the entire lattice space Never-theless it has been proven that its expected complexity is stillexponential [1]Thus it becomes impractical when the systemorder is high and the signal-to-noise ratio (SNR) is low
The decoding algorithms based on semidefinite relax-ation (SDR) approach have becomemore andmore attractive
simply because of the fact that semidefinite programming(SDP) problems can be efficiently solved in polynomial time[2ndash5] The SDR approach was firstly applied to detect binaryphase-shift keying (BPSK) and quadrature amplitude modu-lation (4-QAM) signals [6 7]Then the extensions to differentSDR techniques for 16-QAM signals had been proposedsuch as polynomial-inspired SDR (PI-SDR) [8] bound-constrained SDR (BC-SDR) [9] and virtually antipodal SDR(VA-SDR) [10] all exhibiting acceptable BLER performanceand relatively low complexity In [11] it has been proved thatthere exists an equivalence among PI-SDR BC-SDR andVA-SDR for 16-QAM The extension of BC-SDR and VA-SDRto 256-QAM has also been investigated and both of themprovide the same BLER performance However due to itshigh complexity the PI-SDR is not suitable for extension to256-QAM system
In this paper a new SDR detector is proposed for256-QAM system [12] Then a comprehensive comparisonbetween the proposed method and the previous SDR detec-tors is made The results show that the proposed SDR canprovide better BLER performance than both the BC-SDRand VA-SDR Moreover the complexity of the proposed
Hindawi Publishing CorporationInternational Journal of Antennas and PropagationVolume 2014 Article ID 614876 8 pageshttpdxdoiorg1011552014614876
2 International Journal of Antennas and Propagation
method is higher than that of the BC-SDR but lower thanthat of the VA-SDR This paper is organized as follows InSection 2 the BC-SDR and VA-SDR are reviewed and theproposed SDR is introduced In Sections 3 and 4 the tightnessand the complexity of these methods are analyzed Thenthe simulation results are shown in Section 5 Finally weconclude the paper in Section 6
2 Semidefinite Relaxation Detectors for256-QAM MIMO System
21 System Model The system model for the MIMO trans-mission using 119872-QAM is considered in this paper TheMIMO system is modeled as
r = Hx + n (1)
where r is the119873119877-dimensional received signal vector and x is
the 119873119879-dimensional signal vector in the transmit lattice H
denotes the channel matrix with elements ℎ119894119895representing
the transfer function from the 119895th transmit antenna to 119894threceive antenna n is the119873
119877-dimensional additive noise The
signal vector x is assumed to be a statistically independentvariable with zero mean and variance 1205902 = 119873
02 where 119873
0
is noise power spectral density Perfect channel knowledgeis assumed to be known to the receiver In addition thechannel matrix is assumed to be a flat fading channel and allits entries are complex Gaussian and independent The noiseis an independently and identically distributed (iid) zero-mean Gaussian noise vector with elements having a fixedvariance
The complex transmission in (1) can be equivalentlyrepresented in real matrix form as
[Re (r)Im (r)] = [
Re (H) minus Im (H)Im (H) Re (H)
] [Re (x)Im (x)] + [
Re (n)Im (n)]
r = Hx + n(2)
with Re(∙) and Im(∙) being the real and imaginary partsof (∙) respectively The dimension of x and r is 119872
119877=
2119873119877and 119872
119879= 2119873
119879 respectively H becomes an 119872
119877times
119872119879matrix And the noise n is an 119872
119877-dimensional vector
ML decoding calculates the Euclidean distances betweenthe possible transmit signal vectors and the received signalvector and then chooses the one with shortest distanceas the solution For MIMO system using 256-QAM theML detection aims at finding the solution of the followingoptimization problem
min r minusHx2 (3a)
st x isin plusmn1 plusmn3 plusmn5 plusmn7 plusmn9 plusmn11 plusmn13 plusmn15119872119879 (3b)
where sdot represents the vector 2-norm It is well knownthat ML detection can provide the best BLER performanceHowever its computation requirement is too complicated tobe implemented especially for the cases with large number ofantennas or high level modulation
22 Review of BC-SDR Define a rank-1 semidefinite matrixΩ which is given by
Ω = [x119879 1]119879
[x119879 1] (4)
It is easy to find that the ML detection problem given in(3a) and (3b) can be rewritten as
min TrΩ [H119879H minusH119879r
minusr119879H r119879r ] (5a)
st Ω = Ω119879 isin R(119872119879+1)times(119872119879+1) (5b)
Ω11= Ω12Ω21isin R119872119879times119872119879 (5c)
Ω12isin plusmn1 plusmn3 plusmn5 plusmn7 plusmn9 plusmn11 plusmn13 plusmn15
119872119879 (5d)
Ω22= 1 (5e)
It can be observed that the high complexity of the MLdetection is due to the presence of the two nonconvexconstraints (5c) and (5d)Thus relaxation of these constraintswill be engaged to transform the original problem into asemidefinite problem which can then be efficiently solvedin polynomial time First constraint (5d) implies 1 le 1199092
119894le
225 where 119909119894denotes the 119894th component of x Then (5b)
along with (5c) can be relaxed into Y≻0 A new symbol Yis introduced here to distinguish Y from the aforementionedΩ since they are actually different matrixes after relaxationConsequently the BC-SDR problem is obtained as
min TrY [H119879H minusH119879r
minusr119879H r119879r ] (6a)
st Y isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (6b)
I≺Diag Y11 ≺ 255I I isin R119872119879times119872119879 (6c)
Y22= 1 (6d)
The BC-SDR problem (6a) (6b) (6c) and (6d) can thenbe solved by any of the SDP solvers such as Sedumi [3] basedon interior point methods Although (6b) could be deducedfrom (5b) and (5c) and (6c) can also be deduced from (5d)problem (6a) (6b) (6c) and (6d) is however not exactlyequivalent to problem (5a) (5b) (5c) (5d) and (5e) Thusthe solution obtained by solving (6a) (6b) (6c) and (6d) hasmore errors than ML solution in (5a) (5b) (5c) (5d) and(5e) Once the solution is found the randomization approachis applied to quantize the resulting Y
12till constraint (3b) is
satisfied
23 Review of VA-SDR It is worth noting that when con-straint (3b) is expected to be satisfied the signal x could beexpressed as
x = Up119879 (7)
where U = [I 2I 4I 8I] p = [p1 p2p3p4] I isin R119872119879times119872119879
and p119894isin plusmn1
119872119879 119894 = 1 2 3 4
International Journal of Antennas and Propagation 3
Table 1 The value of x
q2
q1
minus3 minus1 1 3
minus3 minus15 minus7 1 9
minus1 minus13 minus5 3 11
1 minus11 minus3 5 13
3 minus9 minus1 7 15
Substituting (7) into (4) and defining a matrix Z which isgiven by
Z = [p 1]119879 [p 1] (8)
the objective function (5a) can be equivalently transformedinto (9a) and constraints (5d) and (5e) are equal to (9b)Similarly (5b) along with (5c) can be relaxed into (9c) Thuswe obtain the VA-SDR problem given by
min TrZ [U119879H119879HU minusU119879H119879rminusr119879HU r119879r ] (9a)
st Z isin R(4119872119879+1)times(4119872119879+1)≻ 0 (9b)
Diag (Z) = 1I (9c)
Since the first 4119872119879elements of the last row in the solution
Z can be considered as p = [p1 p2p3p4] thus the
optimum solution x is reconstructed by using (7) Finally thesolution is quantized by using randomization
24 Proposed SDR Considering constraint (3b) the signal xcould also be expressed as
x = Vq119879 (10)
where V = [I 4I] q = [q1 q2] I isin R119872119879times119872119879 and q
1 q2isin
plusmn1 plusmn3119872119879
Table 1 gives the value of 119909119895for the possible combinations
of 1199021119895and 119902
2119895 where 119909
119895 1199021119895 and 119902
2119895denote the 119895th element
of x q1 and q
2 respectively
By substituting (10) into (4) and defining a matrix Wwhich is given by
W = [q 1]119879 [q 1] (11)
the objective function (5a) can be equivalently transformedinto (12a) Moreover it can be known from (10) that con-straint (5d) is equivalent to q
1 q2isin plusmn1 plusmn3
119872119879 which areessentially thewell-known indices used to characterize the 16-QAM constellation Herein the set operation method [13] isengaged to formulate the alphabet constraint (5d) into (12b)and (12c) Similarly (5b) along with (5c) can be relaxed into
(12d) Also (5e) can be reformulated as (12e) Thus we obtainthe proposed SDR problem given by
min TrW [V119879H119879HV minusV119879H119879rminusr119879HV r119879r ] (12a)
st 1I≺Diag W11 ≺ 9I I isin R2119872119879times2119872119879 (12b)
Diag W11 plusmn 4Diag W
12 + 3I≻ 0 I isin R2119872119879times2119872119879
(12c)
W isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (12d)
W22= 1 (12e)
Similar to the VA-SDR method the first 2119872119879elements
of the last row in the solution W can be considered asq = [q1 q
2] thus the optimum solution x is reconstructedby using (10) Finally the solution is quantized by usingrandomization
3 Comparison of Tightness
As mentioned in Section 2 these three SDR problems areall relaxed from the original problem (3a) and (3b) andtheir objective functions are equivalent which are actuallycalculating the Euclidean distance given by (3a) Thus thetightness of the constraints of each SDR algorithm implieshow close it is to the ML decoding In what follows we willcompare their tightness
31 Equivalence of BC-SDR and VA-SDR Firstly we willdemonstrate that the constraints of the BC-SDR problem areequivalent to those of the VA-SDR problemThis equivalencehas been considered in [11] and for completeness of under-standing we will reorganize the proof in the following twosteps
Step 1 For each matrix Z isin R(4119872119879+1)times(4119872119879+1) that satisfiesconstraints (9b) and (9c) there must be a matrix Y isin
R(119872119879+1)times(119872119879+1) which satisfies constraints (6b)ndash(6d)
Proof For any matrix Z which satisfies constraints (9b) and(9c) it has the following form
Z = [Z11 isin R4119872119879times4119872119879 Z
12isin R4119872119879times1
Z21isin R1times4119872119879 1
] (13)
Since Z≻ 0 there should be a reversible matrix
Λ = [12058211205822 120582
119872119905120582119872119905+1
120582119872119905+2
12058221198721199051205822119872119905+1
1205822119872119905+2
12058231198721199051205823119872119905+1
1205823119872119905+2
12058241198721199051205824119872119905+1
]
(14)
4 International Journal of Antennas and Propagation
that satisfies
Z = Λ119879Λ (15)
where 120582119898isin R(4119872119879+1)times1119898 = 1 2 4119872
119879+ 1
Now we construct a semidefinite matrix Y isin
R(119872119879+1)times(119872119879+1) which is defined as
Y = GZG119879 = (ΛG119879)119879
(ΛG119879) (16)
where
G = [U 00 1
] = [I isin R119872119879times119872119879 2I isin R119872119879times119872119879 4I isin R119872119879times119872119879 8I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 0 isin R1times119872119879 0 isin R1times119872119879 1
] (17)
From (13)ndash(15) and (9c) we get
10038171003817100381710038171205821198991003817100381710038171003817 = (120582
119879
119899120582119899)12
= 1 119899 = 1 2 4119872119879 (18)
120582119879
4119872119879+11205824119872119879+1
= 1 (19)
Substituting (14) into (16) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879(Λ1+ 2Λ2+ 4Λ3+ 8Λ4) (Λ1+ 2Λ2+ 4Λ3+ 8Λ4)1198791205824119872119879+1
120582119879
4119872119879+1(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879
120582119879
4119872119879+11205824119872119879+1
]
(20)
where Λ1= [12058211205822 120582
119873119905] isin R(4119872119879+1)times119872119879 Λ
2= [120582119872119879+1
120582119872119879+2
1205822119872119879] isin R(4119872119879+1)times119872119879 Λ
3= [1205822119873119905+1
1205822119873119905+2
1205823119873119905] isin R(4119872119879+1)times119872119879 and Λ
4= [1205823119873119905+1
1205823119873119905+2
1205824119873119905] isin
R(4119872119879+1)times119872119879 From (19) and (20) it can be known thatY
22= 1 and (6d)
is satisfied Moreover we have
Y11= (Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)
(21)
The element located in the 119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120582119895+ 2120582119872119879+119895
+ 41205822119872119879+119895
+ 81205823119872119879+119895
10038171003817100381710038171003817
2
(22)
From the perspective of geometry and (18) it is easy to knowthat
1 =1003817100381710038171003817100381781205823119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120582119895
10038171003817100381710038171003817minus100381710038171003817100381710038172120582119872119879+119895
10038171003817100381710038171003817minus1003817100381710038171003817100381741205822119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120582119895+ 2120582119872119879+119895
+ 41205822119872119879+119895
+ 81205823119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120582119895
10038171003817100381710038171003817+100381710038171003817100381710038172120582119872119879+119895
10038171003817100381710038171003817+1003817100381710038171003817100381741205822119872119879+119895
10038171003817100381710038171003817+1003817100381710038171003817100381781205823119872119879+119895
10038171003817100381710038171003817= 15
(23)
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(24)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix Z isin
R(4119872119879+1)times(4119872119879+1) which satisfies constraints (9b) and (9c)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (25)
Since Y≻0 there must be a reversible matrix Γ = [1205741 1205742
1205743 120574
119872119879 120574119872119879+1
] that satisfies
Y = Γ119879Γ (26)
where 120574119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (25) and (26) and (6c) it can be obtained that10038171003817100381710038171003817120574119872119879+1
10038171003817100381710038171003817= (120574119879
119872119879+1120574119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120574119895
10038171003817100381710038171003817= (120574119879
119895120574119895)12
le 15 119895 = 1 2 119872119879
(27)
From the perspective of geometry it is easy to know thatthere should be vectors 120574
1198951 1205741198952 1205741198953 1205741198954isin R(119872119879+1)times1 which
satisfy
120574119895= 1205741198951+ 21205741198952+ 41205741198953+ 81205741198954 (28)
100381710038171003817100381710038171205741198951
10038171003817100381710038171003817=100381710038171003817100381710038171205741198952
10038171003817100381710038171003817=100381710038171003817100381710038171205741198953
10038171003817100381710038171003817=100381710038171003817100381710038171205741198954
10038171003817100381710038171003817= 1 (29)
Now we construct another matrix on the basis of (26) and(28) which is given by
Γ1015840= [12057411 12057412 12057413 12057414 12057421 12057422 12057423 12057424 120574
1198721199051
1205741198721199052 1205741198721199053 1205741198721199054 120574119872119905+1
]
isin R(119872119879+1)times(4119872119879+1)
(30)
International Journal of Antennas and Propagation 5
Next we construct a semidefinite matrix Z isin
R(4119872119879+1)times(4119872119879+1) which is defined as
Z = (Γ1015840)119879
Γ1015840 (31)
Substituting (30) into (31) the element located in the (119895 times 119894)throw and the (119895 times 119894)th column of Z is given as
119911(119895119894)(119895119894)
= (120574119879
119895119894120574119895119894) =10038171003817100381710038171003817120574119895119894
10038171003817100381710038171003817
2
= 1 (32)
where 119895 = 1 2 119872119879 119894 = 1 2 3 4 Moreover the (4119872
119879+
1)th row and the (4119872119879+ 1)th column of Z are
119911(4119872119879+1)(4119872119879+1)
= (120574119879
119872119879+1120574119872119879+1
) = 1 (33)
From (32) and (33) it can be obtained that
Diag (Z) = 1I (34)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDR andVA-SDR can be concluded
32 Proposed SDR Tighter Than BC-SDR and VA-SDR Sec-ondly we will demonstrate that the constraints of proposedSDR problem are tighter than those of the BC-SDR problemand also tighter than those of the VA-SDR due to theaforementioned equivalence For this purpose a new SDRproblem is constructed given by
min TrW1015840 [V119879H119879HV minusV119879H119879rminusr119879HV r119879r ] (35a)
st 1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (35b)
W1015840 isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (35c)
W101584022= 1 (35d)
By comparing (35a) (35b) (35c) and (35d) and (12a)(12b) (12c) (12d) and (12e) it is apparent that the proposedSDR is tighter than the new SDR since the proposed SDR isnearly the same as the new SDR except that the proposed SDRhas an extra constraint (12c) In what follows we will provethat the new SDR given by (35a) (35b) (35c) and (35d) isequivalent to BC-SDR given by (6a) (6b) (6c) and (6d) andtherefore the proposed SDR being tighter than BC-SDR canbe proven
Step 1 For each matrix W1015840 isin R(2119872119879+1)times(2119872119879+1) that satis-fies constraints (35b)ndash(35d) there must be a matrix Y isin
R(119872119879+1)times(119872119879+1) which satisfies constraints (6b)ndash(6d)
Proof For any matrix W1015840 that satisfies constraints (35b)ndash(35d) it has the following form
W1015840 = [W1015840
11isin R2119872119879times2119872119879 W1015840
12isin R2119872119879times1
W101584021isin R1times2119872119879 1
] (36)
SinceW1015840≻0 there should be a reversible matrix
Ψ
= [12059511205952 120595
119872119879120595119872119879+1
120595119872119879+2
12059521198721198791205952119872119879+1
]
(37)
that satisfies
W1015840 = Ψ119879Ψ (38)
where 120595119904isin R(2119872119879+1)times1 119904 = 1 2 2119872
119879+ 1
From (36)ndash(38) (35b) and (35d) give
1 le10038171003817100381710038171205951198961003817100381710038171003817 = (120595
119879
119896120595119896)12
le 3 119896 = 1 2 2119872119879 (39)
120595119879
2119872119879+11205952119872119879+1
= 1 (40)
Now we construct a semidefinite matrix Y isin R(119872119879+1)times(119872119879+1)
which is defined as
Y = EW1015840E119879 = (ΨE119879)119879
(ΨE119879) (41)
where
E = [V 00 1
]
= [I isin R119872119879times119872119879 4I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 1
]
(42)
Substituting (38) into (41) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (Ψ1+ 4Ψ2)1198791205952119872119879+1
120595119879
2119872119879+1(Ψ1+ 4Ψ2)119879120595119879
2119872119879+11205952119872119879+1
]
(43)
where
Ψ1= [12059511205952 120595
119873119905] isin R
(2119872119879+1)times119872119879
Ψ2= [120595119872119879+1
120595119872119879+2
1205952119872119879] isin R
(2119872119879+1)times119872119879
(44)
From (40) and (43) it can be known that Y22= 1 and
(6d) is satisfied Moreover we have
Y11= (Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (45)
and the element located in the119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120595119895+ 4120595119873119905+119895
10038171003817100381710038171003817
2
119895 = 1 2 119872119879 (46)
From the perspective of geometry and (39) it is easy to knowthat
1 le100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120595119895
10038171003817100381710038171003817le10038171003817100381710038171003817120595119895+ 4120595119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120595119895
10038171003817100381710038171003817+100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817le 15
(47)
6 International Journal of Antennas and Propagation
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(48)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix W1015840 isinR(2119872119879+1)times(2119872119879+1) which satisfies constraints (35b)ndash(35d)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (49)
Since Y≻0 there must be a reversible matrix Δ =
[120590112059021205903 120590
119872119879120590119872119879+1
] that satisfies
Y = Δ119879Δ (50)
where 120590119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (49) (50) and (6c) it can be obtained that10038171003817100381710038171003817120590119872119879+1
10038171003817100381710038171003817= (120590119879
119872119879+1120590119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120590119895
10038171003817100381710038171003817= (120590119879
119895120590119895)12
le 15 119895 = 1 2 119872119879
(51)
From the perspective of geometry it is easy to know that thereshould be vectors 120590
11989511205901198952isin R(119872119879+1)times1 which satisfy
120590119895= 1205901198951+ 41205901198952 (52)
1 le10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817= (120590119879
119895119894120590119895119894)12
le 3 119894 = 1 2 (53)
Now we construct another matrix on the basis of (50) and(52) which is given by
Δ1015840= [12059011120590121205902112059022 120590
11987211990511205901198721199052120590119872119905+1
]
isin R(119872119879+1)times(2119872119879+1)
(54)
Next we construct a semidefinite matrix W1015840 isin
R(2119872119879+1)times(2119872119879+1) which is defined as
W1015840 = (Δ1015840)119879
Δ1015840 (55)
Substituting (54) into (55) the element located in the (119895times 119894)throw and the (119895 times 119894)th column ofW1015840 is given as
1199081015840
(119895119894)(119895119894)= (120590119879
119895119894120590119895119894) =10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817
2
(56)
where 119895 = 1 2 119872119879 119894 = 1 2
Then from (53) and (56) it can be obtained that
1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (57)
Moreover the (2119872119879+ 1)th row and the (2119872
119879+ 1)th column
ofW1015840 are
W101584022= 1199081015840
(2119872119879+1)(2119872119879+1)= (120590119879
119872119879+1120590119872119879+1
) = 1 (58)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDRand new SDR given by (35a) (35b) (35c) and (35d) can beconcluded Therefore the constraints of the proposed SDRare tighter than those of the BC-SDR and also tighter thanthose of the VA-SDR due to the aforementioned equivalence
4 Comparison of Complexity
Firstly the BC-SDR given in (6a)ndash(6d) consists of a (119872119879+1)times
(119872119879+ 1)matrix variable Y and119874(119872
119879+ 1) linear constraints
Since constraints (6c) and (6d) are separable the complexityof the BC-SDR detector is 119874((119872
119879+ 1)35) [9] Secondly the
VA-SDR given in (9a)ndash(9c) involves a (4119872119879+ 1) times (4119872
119879+ 1)
matrix variable Z and 119874(4119872119879+ 1) The constraint (9c) is
also separable thus the complexity of the VA-SDR detectoris119874((4119872
119879+1)35) Finally the proposed SDR shown in (12a)ndash
(12e) consists of a (2119872119879+ 1) times (2119872
119879+ 1) matrix variable
W and two119874(4119872119879) linear constraints According to [13 14]
it can be known that the complexity of the proposed SDRdetector is 119874(4(2119872
119879+ 1)45+ 8(2119872
119879+ 1)35)
5 Simulation Results
Computer simulations were conducted to evaluate the per-formance of these three SDR detectors An uncoded MIMOsystem with independent Rayleigh fading channel was takeninto account and the Sedumi toolbox within Matlab softwarewas used to implement the SDRdetection algorithms Figures1 and 2 show the BLER performances of the three SDRdetectors for 4 times 4 and 8 times 8 256-QAM systems respectivelyIt can be observed that the BC-SDR detector and the VA-SDR detector provide exactly the same BLER performancewhile the proposed SDR detector can provide the best BLERperformance among these three detectors This agrees wellwith the analysis presented in Section 3 What is more itcan be found that the improvement concerning the BLERperformance provided by the proposed SDR detector in thecase of 8 times 8 MIMO system is larger than that in the caseof 4 times 4 MIMO system The relaxation of the alphabet setengaged in theBC-SDRand theVA-SDRwill cause increasingerrors with the increase of the number of the antennas Thatis because the inaccurate detection of any dimension of thetransmitted signal will result in the failure of the detection ofthe whole signal vector and the higher the dimension of theproblem is the bigger the possibility of failure will be Themain merit of the proposed SDR over the BC-SDR and theVA-SDR is that it can offer more accurate relaxation of thealphabet set Moreover the computational times for solvingone signal vector of these SDR detectors are illustrated inFigures 3 and 4 These figures demonstrate that the BC-SDRhas the lowest computational complexity and the VA-SDRhas the highest computational complexity while the proposedSDR is in between the two detection methods It is alsofound that the complexity of SDR detectors is independentof SNR This is a distinct advantage over the SD whosecomplexity varies as a function of SNR making SD difficultto be implemented in practice
International Journal of Antennas and Propagation 7
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 1The BLER performance of the SDR detectors for 4times4 256-QAMMIMO systems
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 2 The BLER performance of the SDR detectors for 8 times 8256-QAMMIMO systems
6 Conclusion
In this paper we proposed a SDR detector for 256-QAMMIMO system and also reviewed two other SDR detectorsnamely BC-SDR detector and VA-SDR detector Then weanalyzed the tightness and the complexity of these three SDRdetectors Both theoretical analysis and simulation resultsdemonstrate that the proposed SDR can provide the bestBLER performance among these three detectors while theBC-SDR detector and the VA-SDR detector provide exactlythe same BLER performance Moreover the BC-SDR has thelowest computational complexity and the VA-SDR has the
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 3 The computational time of the SDR detectors for 4 times 4256-QAMMIMO systems
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
102
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 4 The computational time of the SDR detectors for 8 times 8256-QAMMIMO systems
highest computational complexity while the proposed SDRis in between
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
8 International Journal of Antennas and Propagation
References
[1] J Jalden and B Ottersten ldquoAn exponential lower bound on theexpected complexity of sphere decodingrdquo in Proceedings of theIEEE International Conference on Acoustics Speech and SignalProcessing (ICASSP rsquo04) vol 4 pp iv-393ndashiv-396 MontrealCanada May 2004
[2] A Ben-Tal and A Nemirovski Lectures on Modern ConvexOptimization MPS-SIAM Series on Optimization 2001
[3] J F Sturm ldquoUsing SEDUMI 102 A Matlab toolbox for opti-mization over symmetric conesrdquo Optimization Methods andSoftware vol 11-12 pp 1ndash30 1999
[4] B Steingrimsson Z Q Luo and K M Wong ldquoSoft quasi-maximum-likelihood detection for multiple-antenna wirelesschannelsrdquo IEEE Transactions on Signal Processing vol 51 no11 pp 2710ndash2719 2003
[5] Z Y Shao SWCheung andT I Yuk ldquoLattice-reduction-aidedsemidefinite relaxation detection algorithms for multiple-inputmultiple-output systemsrdquo IETCommunications vol 8 no 4 pp448ndash454 2014
[6] W K Ma T N Davidson K M Wong Z Q Luo andP C Ching ldquoQuasi-maximum-likelihood multiuser detectionusing semi-definite relaxation with application to synchronousCDMArdquo IEEE Transactions on Signal Processing vol 50 no 4pp 912ndash922 2002
[7] A Wiesel Y C Eldar and S Shamai ldquoSemidefinite relaxationfor detection of 16-QAM signaling in MIMO channelsrdquo IEEESignal Processing Letters vol 12 no 9 pp 653ndash656 2005
[8] P H Tan and L K Rasmussen ldquoThe application of semidefiniteprogramming for detection inCDMArdquo IEEE Journal on SelectedAreas in Communications vol 19 no 8 pp 1442ndash1449 2001
[9] N D Sidiropoulos and Z-Q Luo ldquoA semidefinite relaxationapproach to MIMO detection for high-order QAM constella-tionsrdquo IEEE Signal Processing Letters vol 13 no 9 pp 525ndash5282006
[10] Z Mao X Wang and X Wang ldquoSemidefinite programmingrelaxation approach for multiuser detection of QAM signalsrdquoIEEE Transactions on Wireless Communications vol 6 no 12pp 4275ndash4279 2007
[11] W-K Ma C-C Su J Jalden T-H Chang and C-Y ChildquoThe equivalence of semidefinite relaxation MIMO detectorsfor higher-order QAMrdquo IEEE Journal on Selected Topics inSignal Processing vol 3 no 6 pp 1038ndash1052 2009
[12] Z Y Shao SWCheung andT I Yuk ldquoSemi-definite relaxationdecoder for 256-QAM MIMO systemrdquo Electronics Letters vol46 no 11 pp 796ndash797 2010
[13] Y Yang C Zhao P Zhou and W Xu ldquoMIMO detection of 16-QAM signaling based on semidefinite relaxationrdquo IEEE SignalProcessing Letters vol 14 no 11 pp 797ndash800 2007
[14] C Helmberg F Rendl R J Vanderbei and H Wolkowicz ldquoAninterior-point method for semidefinite programmingrdquo SIAMJournal on Optimization vol 6 no 2 pp 342ndash361 1996
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
2 International Journal of Antennas and Propagation
method is higher than that of the BC-SDR but lower thanthat of the VA-SDR This paper is organized as follows InSection 2 the BC-SDR and VA-SDR are reviewed and theproposed SDR is introduced In Sections 3 and 4 the tightnessand the complexity of these methods are analyzed Thenthe simulation results are shown in Section 5 Finally weconclude the paper in Section 6
2 Semidefinite Relaxation Detectors for256-QAM MIMO System
21 System Model The system model for the MIMO trans-mission using 119872-QAM is considered in this paper TheMIMO system is modeled as
r = Hx + n (1)
where r is the119873119877-dimensional received signal vector and x is
the 119873119879-dimensional signal vector in the transmit lattice H
denotes the channel matrix with elements ℎ119894119895representing
the transfer function from the 119895th transmit antenna to 119894threceive antenna n is the119873
119877-dimensional additive noise The
signal vector x is assumed to be a statistically independentvariable with zero mean and variance 1205902 = 119873
02 where 119873
0
is noise power spectral density Perfect channel knowledgeis assumed to be known to the receiver In addition thechannel matrix is assumed to be a flat fading channel and allits entries are complex Gaussian and independent The noiseis an independently and identically distributed (iid) zero-mean Gaussian noise vector with elements having a fixedvariance
The complex transmission in (1) can be equivalentlyrepresented in real matrix form as
[Re (r)Im (r)] = [
Re (H) minus Im (H)Im (H) Re (H)
] [Re (x)Im (x)] + [
Re (n)Im (n)]
r = Hx + n(2)
with Re(∙) and Im(∙) being the real and imaginary partsof (∙) respectively The dimension of x and r is 119872
119877=
2119873119877and 119872
119879= 2119873
119879 respectively H becomes an 119872
119877times
119872119879matrix And the noise n is an 119872
119877-dimensional vector
ML decoding calculates the Euclidean distances betweenthe possible transmit signal vectors and the received signalvector and then chooses the one with shortest distanceas the solution For MIMO system using 256-QAM theML detection aims at finding the solution of the followingoptimization problem
min r minusHx2 (3a)
st x isin plusmn1 plusmn3 plusmn5 plusmn7 plusmn9 plusmn11 plusmn13 plusmn15119872119879 (3b)
where sdot represents the vector 2-norm It is well knownthat ML detection can provide the best BLER performanceHowever its computation requirement is too complicated tobe implemented especially for the cases with large number ofantennas or high level modulation
22 Review of BC-SDR Define a rank-1 semidefinite matrixΩ which is given by
Ω = [x119879 1]119879
[x119879 1] (4)
It is easy to find that the ML detection problem given in(3a) and (3b) can be rewritten as
min TrΩ [H119879H minusH119879r
minusr119879H r119879r ] (5a)
st Ω = Ω119879 isin R(119872119879+1)times(119872119879+1) (5b)
Ω11= Ω12Ω21isin R119872119879times119872119879 (5c)
Ω12isin plusmn1 plusmn3 plusmn5 plusmn7 plusmn9 plusmn11 plusmn13 plusmn15
119872119879 (5d)
Ω22= 1 (5e)
It can be observed that the high complexity of the MLdetection is due to the presence of the two nonconvexconstraints (5c) and (5d)Thus relaxation of these constraintswill be engaged to transform the original problem into asemidefinite problem which can then be efficiently solvedin polynomial time First constraint (5d) implies 1 le 1199092
119894le
225 where 119909119894denotes the 119894th component of x Then (5b)
along with (5c) can be relaxed into Y≻0 A new symbol Yis introduced here to distinguish Y from the aforementionedΩ since they are actually different matrixes after relaxationConsequently the BC-SDR problem is obtained as
min TrY [H119879H minusH119879r
minusr119879H r119879r ] (6a)
st Y isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (6b)
I≺Diag Y11 ≺ 255I I isin R119872119879times119872119879 (6c)
Y22= 1 (6d)
The BC-SDR problem (6a) (6b) (6c) and (6d) can thenbe solved by any of the SDP solvers such as Sedumi [3] basedon interior point methods Although (6b) could be deducedfrom (5b) and (5c) and (6c) can also be deduced from (5d)problem (6a) (6b) (6c) and (6d) is however not exactlyequivalent to problem (5a) (5b) (5c) (5d) and (5e) Thusthe solution obtained by solving (6a) (6b) (6c) and (6d) hasmore errors than ML solution in (5a) (5b) (5c) (5d) and(5e) Once the solution is found the randomization approachis applied to quantize the resulting Y
12till constraint (3b) is
satisfied
23 Review of VA-SDR It is worth noting that when con-straint (3b) is expected to be satisfied the signal x could beexpressed as
x = Up119879 (7)
where U = [I 2I 4I 8I] p = [p1 p2p3p4] I isin R119872119879times119872119879
and p119894isin plusmn1
119872119879 119894 = 1 2 3 4
International Journal of Antennas and Propagation 3
Table 1 The value of x
q2
q1
minus3 minus1 1 3
minus3 minus15 minus7 1 9
minus1 minus13 minus5 3 11
1 minus11 minus3 5 13
3 minus9 minus1 7 15
Substituting (7) into (4) and defining a matrix Z which isgiven by
Z = [p 1]119879 [p 1] (8)
the objective function (5a) can be equivalently transformedinto (9a) and constraints (5d) and (5e) are equal to (9b)Similarly (5b) along with (5c) can be relaxed into (9c) Thuswe obtain the VA-SDR problem given by
min TrZ [U119879H119879HU minusU119879H119879rminusr119879HU r119879r ] (9a)
st Z isin R(4119872119879+1)times(4119872119879+1)≻ 0 (9b)
Diag (Z) = 1I (9c)
Since the first 4119872119879elements of the last row in the solution
Z can be considered as p = [p1 p2p3p4] thus the
optimum solution x is reconstructed by using (7) Finally thesolution is quantized by using randomization
24 Proposed SDR Considering constraint (3b) the signal xcould also be expressed as
x = Vq119879 (10)
where V = [I 4I] q = [q1 q2] I isin R119872119879times119872119879 and q
1 q2isin
plusmn1 plusmn3119872119879
Table 1 gives the value of 119909119895for the possible combinations
of 1199021119895and 119902
2119895 where 119909
119895 1199021119895 and 119902
2119895denote the 119895th element
of x q1 and q
2 respectively
By substituting (10) into (4) and defining a matrix Wwhich is given by
W = [q 1]119879 [q 1] (11)
the objective function (5a) can be equivalently transformedinto (12a) Moreover it can be known from (10) that con-straint (5d) is equivalent to q
1 q2isin plusmn1 plusmn3
119872119879 which areessentially thewell-known indices used to characterize the 16-QAM constellation Herein the set operation method [13] isengaged to formulate the alphabet constraint (5d) into (12b)and (12c) Similarly (5b) along with (5c) can be relaxed into
(12d) Also (5e) can be reformulated as (12e) Thus we obtainthe proposed SDR problem given by
min TrW [V119879H119879HV minusV119879H119879rminusr119879HV r119879r ] (12a)
st 1I≺Diag W11 ≺ 9I I isin R2119872119879times2119872119879 (12b)
Diag W11 plusmn 4Diag W
12 + 3I≻ 0 I isin R2119872119879times2119872119879
(12c)
W isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (12d)
W22= 1 (12e)
Similar to the VA-SDR method the first 2119872119879elements
of the last row in the solution W can be considered asq = [q1 q
2] thus the optimum solution x is reconstructedby using (10) Finally the solution is quantized by usingrandomization
3 Comparison of Tightness
As mentioned in Section 2 these three SDR problems areall relaxed from the original problem (3a) and (3b) andtheir objective functions are equivalent which are actuallycalculating the Euclidean distance given by (3a) Thus thetightness of the constraints of each SDR algorithm implieshow close it is to the ML decoding In what follows we willcompare their tightness
31 Equivalence of BC-SDR and VA-SDR Firstly we willdemonstrate that the constraints of the BC-SDR problem areequivalent to those of the VA-SDR problemThis equivalencehas been considered in [11] and for completeness of under-standing we will reorganize the proof in the following twosteps
Step 1 For each matrix Z isin R(4119872119879+1)times(4119872119879+1) that satisfiesconstraints (9b) and (9c) there must be a matrix Y isin
R(119872119879+1)times(119872119879+1) which satisfies constraints (6b)ndash(6d)
Proof For any matrix Z which satisfies constraints (9b) and(9c) it has the following form
Z = [Z11 isin R4119872119879times4119872119879 Z
12isin R4119872119879times1
Z21isin R1times4119872119879 1
] (13)
Since Z≻ 0 there should be a reversible matrix
Λ = [12058211205822 120582
119872119905120582119872119905+1
120582119872119905+2
12058221198721199051205822119872119905+1
1205822119872119905+2
12058231198721199051205823119872119905+1
1205823119872119905+2
12058241198721199051205824119872119905+1
]
(14)
4 International Journal of Antennas and Propagation
that satisfies
Z = Λ119879Λ (15)
where 120582119898isin R(4119872119879+1)times1119898 = 1 2 4119872
119879+ 1
Now we construct a semidefinite matrix Y isin
R(119872119879+1)times(119872119879+1) which is defined as
Y = GZG119879 = (ΛG119879)119879
(ΛG119879) (16)
where
G = [U 00 1
] = [I isin R119872119879times119872119879 2I isin R119872119879times119872119879 4I isin R119872119879times119872119879 8I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 0 isin R1times119872119879 0 isin R1times119872119879 1
] (17)
From (13)ndash(15) and (9c) we get
10038171003817100381710038171205821198991003817100381710038171003817 = (120582
119879
119899120582119899)12
= 1 119899 = 1 2 4119872119879 (18)
120582119879
4119872119879+11205824119872119879+1
= 1 (19)
Substituting (14) into (16) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879(Λ1+ 2Λ2+ 4Λ3+ 8Λ4) (Λ1+ 2Λ2+ 4Λ3+ 8Λ4)1198791205824119872119879+1
120582119879
4119872119879+1(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879
120582119879
4119872119879+11205824119872119879+1
]
(20)
where Λ1= [12058211205822 120582
119873119905] isin R(4119872119879+1)times119872119879 Λ
2= [120582119872119879+1
120582119872119879+2
1205822119872119879] isin R(4119872119879+1)times119872119879 Λ
3= [1205822119873119905+1
1205822119873119905+2
1205823119873119905] isin R(4119872119879+1)times119872119879 and Λ
4= [1205823119873119905+1
1205823119873119905+2
1205824119873119905] isin
R(4119872119879+1)times119872119879 From (19) and (20) it can be known thatY
22= 1 and (6d)
is satisfied Moreover we have
Y11= (Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)
(21)
The element located in the 119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120582119895+ 2120582119872119879+119895
+ 41205822119872119879+119895
+ 81205823119872119879+119895
10038171003817100381710038171003817
2
(22)
From the perspective of geometry and (18) it is easy to knowthat
1 =1003817100381710038171003817100381781205823119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120582119895
10038171003817100381710038171003817minus100381710038171003817100381710038172120582119872119879+119895
10038171003817100381710038171003817minus1003817100381710038171003817100381741205822119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120582119895+ 2120582119872119879+119895
+ 41205822119872119879+119895
+ 81205823119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120582119895
10038171003817100381710038171003817+100381710038171003817100381710038172120582119872119879+119895
10038171003817100381710038171003817+1003817100381710038171003817100381741205822119872119879+119895
10038171003817100381710038171003817+1003817100381710038171003817100381781205823119872119879+119895
10038171003817100381710038171003817= 15
(23)
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(24)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix Z isin
R(4119872119879+1)times(4119872119879+1) which satisfies constraints (9b) and (9c)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (25)
Since Y≻0 there must be a reversible matrix Γ = [1205741 1205742
1205743 120574
119872119879 120574119872119879+1
] that satisfies
Y = Γ119879Γ (26)
where 120574119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (25) and (26) and (6c) it can be obtained that10038171003817100381710038171003817120574119872119879+1
10038171003817100381710038171003817= (120574119879
119872119879+1120574119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120574119895
10038171003817100381710038171003817= (120574119879
119895120574119895)12
le 15 119895 = 1 2 119872119879
(27)
From the perspective of geometry it is easy to know thatthere should be vectors 120574
1198951 1205741198952 1205741198953 1205741198954isin R(119872119879+1)times1 which
satisfy
120574119895= 1205741198951+ 21205741198952+ 41205741198953+ 81205741198954 (28)
100381710038171003817100381710038171205741198951
10038171003817100381710038171003817=100381710038171003817100381710038171205741198952
10038171003817100381710038171003817=100381710038171003817100381710038171205741198953
10038171003817100381710038171003817=100381710038171003817100381710038171205741198954
10038171003817100381710038171003817= 1 (29)
Now we construct another matrix on the basis of (26) and(28) which is given by
Γ1015840= [12057411 12057412 12057413 12057414 12057421 12057422 12057423 12057424 120574
1198721199051
1205741198721199052 1205741198721199053 1205741198721199054 120574119872119905+1
]
isin R(119872119879+1)times(4119872119879+1)
(30)
International Journal of Antennas and Propagation 5
Next we construct a semidefinite matrix Z isin
R(4119872119879+1)times(4119872119879+1) which is defined as
Z = (Γ1015840)119879
Γ1015840 (31)
Substituting (30) into (31) the element located in the (119895 times 119894)throw and the (119895 times 119894)th column of Z is given as
119911(119895119894)(119895119894)
= (120574119879
119895119894120574119895119894) =10038171003817100381710038171003817120574119895119894
10038171003817100381710038171003817
2
= 1 (32)
where 119895 = 1 2 119872119879 119894 = 1 2 3 4 Moreover the (4119872
119879+
1)th row and the (4119872119879+ 1)th column of Z are
119911(4119872119879+1)(4119872119879+1)
= (120574119879
119872119879+1120574119872119879+1
) = 1 (33)
From (32) and (33) it can be obtained that
Diag (Z) = 1I (34)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDR andVA-SDR can be concluded
32 Proposed SDR Tighter Than BC-SDR and VA-SDR Sec-ondly we will demonstrate that the constraints of proposedSDR problem are tighter than those of the BC-SDR problemand also tighter than those of the VA-SDR due to theaforementioned equivalence For this purpose a new SDRproblem is constructed given by
min TrW1015840 [V119879H119879HV minusV119879H119879rminusr119879HV r119879r ] (35a)
st 1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (35b)
W1015840 isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (35c)
W101584022= 1 (35d)
By comparing (35a) (35b) (35c) and (35d) and (12a)(12b) (12c) (12d) and (12e) it is apparent that the proposedSDR is tighter than the new SDR since the proposed SDR isnearly the same as the new SDR except that the proposed SDRhas an extra constraint (12c) In what follows we will provethat the new SDR given by (35a) (35b) (35c) and (35d) isequivalent to BC-SDR given by (6a) (6b) (6c) and (6d) andtherefore the proposed SDR being tighter than BC-SDR canbe proven
Step 1 For each matrix W1015840 isin R(2119872119879+1)times(2119872119879+1) that satis-fies constraints (35b)ndash(35d) there must be a matrix Y isin
R(119872119879+1)times(119872119879+1) which satisfies constraints (6b)ndash(6d)
Proof For any matrix W1015840 that satisfies constraints (35b)ndash(35d) it has the following form
W1015840 = [W1015840
11isin R2119872119879times2119872119879 W1015840
12isin R2119872119879times1
W101584021isin R1times2119872119879 1
] (36)
SinceW1015840≻0 there should be a reversible matrix
Ψ
= [12059511205952 120595
119872119879120595119872119879+1
120595119872119879+2
12059521198721198791205952119872119879+1
]
(37)
that satisfies
W1015840 = Ψ119879Ψ (38)
where 120595119904isin R(2119872119879+1)times1 119904 = 1 2 2119872
119879+ 1
From (36)ndash(38) (35b) and (35d) give
1 le10038171003817100381710038171205951198961003817100381710038171003817 = (120595
119879
119896120595119896)12
le 3 119896 = 1 2 2119872119879 (39)
120595119879
2119872119879+11205952119872119879+1
= 1 (40)
Now we construct a semidefinite matrix Y isin R(119872119879+1)times(119872119879+1)
which is defined as
Y = EW1015840E119879 = (ΨE119879)119879
(ΨE119879) (41)
where
E = [V 00 1
]
= [I isin R119872119879times119872119879 4I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 1
]
(42)
Substituting (38) into (41) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (Ψ1+ 4Ψ2)1198791205952119872119879+1
120595119879
2119872119879+1(Ψ1+ 4Ψ2)119879120595119879
2119872119879+11205952119872119879+1
]
(43)
where
Ψ1= [12059511205952 120595
119873119905] isin R
(2119872119879+1)times119872119879
Ψ2= [120595119872119879+1
120595119872119879+2
1205952119872119879] isin R
(2119872119879+1)times119872119879
(44)
From (40) and (43) it can be known that Y22= 1 and
(6d) is satisfied Moreover we have
Y11= (Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (45)
and the element located in the119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120595119895+ 4120595119873119905+119895
10038171003817100381710038171003817
2
119895 = 1 2 119872119879 (46)
From the perspective of geometry and (39) it is easy to knowthat
1 le100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120595119895
10038171003817100381710038171003817le10038171003817100381710038171003817120595119895+ 4120595119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120595119895
10038171003817100381710038171003817+100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817le 15
(47)
6 International Journal of Antennas and Propagation
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(48)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix W1015840 isinR(2119872119879+1)times(2119872119879+1) which satisfies constraints (35b)ndash(35d)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (49)
Since Y≻0 there must be a reversible matrix Δ =
[120590112059021205903 120590
119872119879120590119872119879+1
] that satisfies
Y = Δ119879Δ (50)
where 120590119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (49) (50) and (6c) it can be obtained that10038171003817100381710038171003817120590119872119879+1
10038171003817100381710038171003817= (120590119879
119872119879+1120590119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120590119895
10038171003817100381710038171003817= (120590119879
119895120590119895)12
le 15 119895 = 1 2 119872119879
(51)
From the perspective of geometry it is easy to know that thereshould be vectors 120590
11989511205901198952isin R(119872119879+1)times1 which satisfy
120590119895= 1205901198951+ 41205901198952 (52)
1 le10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817= (120590119879
119895119894120590119895119894)12
le 3 119894 = 1 2 (53)
Now we construct another matrix on the basis of (50) and(52) which is given by
Δ1015840= [12059011120590121205902112059022 120590
11987211990511205901198721199052120590119872119905+1
]
isin R(119872119879+1)times(2119872119879+1)
(54)
Next we construct a semidefinite matrix W1015840 isin
R(2119872119879+1)times(2119872119879+1) which is defined as
W1015840 = (Δ1015840)119879
Δ1015840 (55)
Substituting (54) into (55) the element located in the (119895times 119894)throw and the (119895 times 119894)th column ofW1015840 is given as
1199081015840
(119895119894)(119895119894)= (120590119879
119895119894120590119895119894) =10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817
2
(56)
where 119895 = 1 2 119872119879 119894 = 1 2
Then from (53) and (56) it can be obtained that
1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (57)
Moreover the (2119872119879+ 1)th row and the (2119872
119879+ 1)th column
ofW1015840 are
W101584022= 1199081015840
(2119872119879+1)(2119872119879+1)= (120590119879
119872119879+1120590119872119879+1
) = 1 (58)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDRand new SDR given by (35a) (35b) (35c) and (35d) can beconcluded Therefore the constraints of the proposed SDRare tighter than those of the BC-SDR and also tighter thanthose of the VA-SDR due to the aforementioned equivalence
4 Comparison of Complexity
Firstly the BC-SDR given in (6a)ndash(6d) consists of a (119872119879+1)times
(119872119879+ 1)matrix variable Y and119874(119872
119879+ 1) linear constraints
Since constraints (6c) and (6d) are separable the complexityof the BC-SDR detector is 119874((119872
119879+ 1)35) [9] Secondly the
VA-SDR given in (9a)ndash(9c) involves a (4119872119879+ 1) times (4119872
119879+ 1)
matrix variable Z and 119874(4119872119879+ 1) The constraint (9c) is
also separable thus the complexity of the VA-SDR detectoris119874((4119872
119879+1)35) Finally the proposed SDR shown in (12a)ndash
(12e) consists of a (2119872119879+ 1) times (2119872
119879+ 1) matrix variable
W and two119874(4119872119879) linear constraints According to [13 14]
it can be known that the complexity of the proposed SDRdetector is 119874(4(2119872
119879+ 1)45+ 8(2119872
119879+ 1)35)
5 Simulation Results
Computer simulations were conducted to evaluate the per-formance of these three SDR detectors An uncoded MIMOsystem with independent Rayleigh fading channel was takeninto account and the Sedumi toolbox within Matlab softwarewas used to implement the SDRdetection algorithms Figures1 and 2 show the BLER performances of the three SDRdetectors for 4 times 4 and 8 times 8 256-QAM systems respectivelyIt can be observed that the BC-SDR detector and the VA-SDR detector provide exactly the same BLER performancewhile the proposed SDR detector can provide the best BLERperformance among these three detectors This agrees wellwith the analysis presented in Section 3 What is more itcan be found that the improvement concerning the BLERperformance provided by the proposed SDR detector in thecase of 8 times 8 MIMO system is larger than that in the caseof 4 times 4 MIMO system The relaxation of the alphabet setengaged in theBC-SDRand theVA-SDRwill cause increasingerrors with the increase of the number of the antennas Thatis because the inaccurate detection of any dimension of thetransmitted signal will result in the failure of the detection ofthe whole signal vector and the higher the dimension of theproblem is the bigger the possibility of failure will be Themain merit of the proposed SDR over the BC-SDR and theVA-SDR is that it can offer more accurate relaxation of thealphabet set Moreover the computational times for solvingone signal vector of these SDR detectors are illustrated inFigures 3 and 4 These figures demonstrate that the BC-SDRhas the lowest computational complexity and the VA-SDRhas the highest computational complexity while the proposedSDR is in between the two detection methods It is alsofound that the complexity of SDR detectors is independentof SNR This is a distinct advantage over the SD whosecomplexity varies as a function of SNR making SD difficultto be implemented in practice
International Journal of Antennas and Propagation 7
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 1The BLER performance of the SDR detectors for 4times4 256-QAMMIMO systems
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 2 The BLER performance of the SDR detectors for 8 times 8256-QAMMIMO systems
6 Conclusion
In this paper we proposed a SDR detector for 256-QAMMIMO system and also reviewed two other SDR detectorsnamely BC-SDR detector and VA-SDR detector Then weanalyzed the tightness and the complexity of these three SDRdetectors Both theoretical analysis and simulation resultsdemonstrate that the proposed SDR can provide the bestBLER performance among these three detectors while theBC-SDR detector and the VA-SDR detector provide exactlythe same BLER performance Moreover the BC-SDR has thelowest computational complexity and the VA-SDR has the
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 3 The computational time of the SDR detectors for 4 times 4256-QAMMIMO systems
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
102
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 4 The computational time of the SDR detectors for 8 times 8256-QAMMIMO systems
highest computational complexity while the proposed SDRis in between
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
8 International Journal of Antennas and Propagation
References
[1] J Jalden and B Ottersten ldquoAn exponential lower bound on theexpected complexity of sphere decodingrdquo in Proceedings of theIEEE International Conference on Acoustics Speech and SignalProcessing (ICASSP rsquo04) vol 4 pp iv-393ndashiv-396 MontrealCanada May 2004
[2] A Ben-Tal and A Nemirovski Lectures on Modern ConvexOptimization MPS-SIAM Series on Optimization 2001
[3] J F Sturm ldquoUsing SEDUMI 102 A Matlab toolbox for opti-mization over symmetric conesrdquo Optimization Methods andSoftware vol 11-12 pp 1ndash30 1999
[4] B Steingrimsson Z Q Luo and K M Wong ldquoSoft quasi-maximum-likelihood detection for multiple-antenna wirelesschannelsrdquo IEEE Transactions on Signal Processing vol 51 no11 pp 2710ndash2719 2003
[5] Z Y Shao SWCheung andT I Yuk ldquoLattice-reduction-aidedsemidefinite relaxation detection algorithms for multiple-inputmultiple-output systemsrdquo IETCommunications vol 8 no 4 pp448ndash454 2014
[6] W K Ma T N Davidson K M Wong Z Q Luo andP C Ching ldquoQuasi-maximum-likelihood multiuser detectionusing semi-definite relaxation with application to synchronousCDMArdquo IEEE Transactions on Signal Processing vol 50 no 4pp 912ndash922 2002
[7] A Wiesel Y C Eldar and S Shamai ldquoSemidefinite relaxationfor detection of 16-QAM signaling in MIMO channelsrdquo IEEESignal Processing Letters vol 12 no 9 pp 653ndash656 2005
[8] P H Tan and L K Rasmussen ldquoThe application of semidefiniteprogramming for detection inCDMArdquo IEEE Journal on SelectedAreas in Communications vol 19 no 8 pp 1442ndash1449 2001
[9] N D Sidiropoulos and Z-Q Luo ldquoA semidefinite relaxationapproach to MIMO detection for high-order QAM constella-tionsrdquo IEEE Signal Processing Letters vol 13 no 9 pp 525ndash5282006
[10] Z Mao X Wang and X Wang ldquoSemidefinite programmingrelaxation approach for multiuser detection of QAM signalsrdquoIEEE Transactions on Wireless Communications vol 6 no 12pp 4275ndash4279 2007
[11] W-K Ma C-C Su J Jalden T-H Chang and C-Y ChildquoThe equivalence of semidefinite relaxation MIMO detectorsfor higher-order QAMrdquo IEEE Journal on Selected Topics inSignal Processing vol 3 no 6 pp 1038ndash1052 2009
[12] Z Y Shao SWCheung andT I Yuk ldquoSemi-definite relaxationdecoder for 256-QAM MIMO systemrdquo Electronics Letters vol46 no 11 pp 796ndash797 2010
[13] Y Yang C Zhao P Zhou and W Xu ldquoMIMO detection of 16-QAM signaling based on semidefinite relaxationrdquo IEEE SignalProcessing Letters vol 14 no 11 pp 797ndash800 2007
[14] C Helmberg F Rendl R J Vanderbei and H Wolkowicz ldquoAninterior-point method for semidefinite programmingrdquo SIAMJournal on Optimization vol 6 no 2 pp 342ndash361 1996
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
International Journal of Antennas and Propagation 3
Table 1 The value of x
q2
q1
minus3 minus1 1 3
minus3 minus15 minus7 1 9
minus1 minus13 minus5 3 11
1 minus11 minus3 5 13
3 minus9 minus1 7 15
Substituting (7) into (4) and defining a matrix Z which isgiven by
Z = [p 1]119879 [p 1] (8)
the objective function (5a) can be equivalently transformedinto (9a) and constraints (5d) and (5e) are equal to (9b)Similarly (5b) along with (5c) can be relaxed into (9c) Thuswe obtain the VA-SDR problem given by
min TrZ [U119879H119879HU minusU119879H119879rminusr119879HU r119879r ] (9a)
st Z isin R(4119872119879+1)times(4119872119879+1)≻ 0 (9b)
Diag (Z) = 1I (9c)
Since the first 4119872119879elements of the last row in the solution
Z can be considered as p = [p1 p2p3p4] thus the
optimum solution x is reconstructed by using (7) Finally thesolution is quantized by using randomization
24 Proposed SDR Considering constraint (3b) the signal xcould also be expressed as
x = Vq119879 (10)
where V = [I 4I] q = [q1 q2] I isin R119872119879times119872119879 and q
1 q2isin
plusmn1 plusmn3119872119879
Table 1 gives the value of 119909119895for the possible combinations
of 1199021119895and 119902
2119895 where 119909
119895 1199021119895 and 119902
2119895denote the 119895th element
of x q1 and q
2 respectively
By substituting (10) into (4) and defining a matrix Wwhich is given by
W = [q 1]119879 [q 1] (11)
the objective function (5a) can be equivalently transformedinto (12a) Moreover it can be known from (10) that con-straint (5d) is equivalent to q
1 q2isin plusmn1 plusmn3
119872119879 which areessentially thewell-known indices used to characterize the 16-QAM constellation Herein the set operation method [13] isengaged to formulate the alphabet constraint (5d) into (12b)and (12c) Similarly (5b) along with (5c) can be relaxed into
(12d) Also (5e) can be reformulated as (12e) Thus we obtainthe proposed SDR problem given by
min TrW [V119879H119879HV minusV119879H119879rminusr119879HV r119879r ] (12a)
st 1I≺Diag W11 ≺ 9I I isin R2119872119879times2119872119879 (12b)
Diag W11 plusmn 4Diag W
12 + 3I≻ 0 I isin R2119872119879times2119872119879
(12c)
W isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (12d)
W22= 1 (12e)
Similar to the VA-SDR method the first 2119872119879elements
of the last row in the solution W can be considered asq = [q1 q
2] thus the optimum solution x is reconstructedby using (10) Finally the solution is quantized by usingrandomization
3 Comparison of Tightness
As mentioned in Section 2 these three SDR problems areall relaxed from the original problem (3a) and (3b) andtheir objective functions are equivalent which are actuallycalculating the Euclidean distance given by (3a) Thus thetightness of the constraints of each SDR algorithm implieshow close it is to the ML decoding In what follows we willcompare their tightness
31 Equivalence of BC-SDR and VA-SDR Firstly we willdemonstrate that the constraints of the BC-SDR problem areequivalent to those of the VA-SDR problemThis equivalencehas been considered in [11] and for completeness of under-standing we will reorganize the proof in the following twosteps
Step 1 For each matrix Z isin R(4119872119879+1)times(4119872119879+1) that satisfiesconstraints (9b) and (9c) there must be a matrix Y isin
R(119872119879+1)times(119872119879+1) which satisfies constraints (6b)ndash(6d)
Proof For any matrix Z which satisfies constraints (9b) and(9c) it has the following form
Z = [Z11 isin R4119872119879times4119872119879 Z
12isin R4119872119879times1
Z21isin R1times4119872119879 1
] (13)
Since Z≻ 0 there should be a reversible matrix
Λ = [12058211205822 120582
119872119905120582119872119905+1
120582119872119905+2
12058221198721199051205822119872119905+1
1205822119872119905+2
12058231198721199051205823119872119905+1
1205823119872119905+2
12058241198721199051205824119872119905+1
]
(14)
4 International Journal of Antennas and Propagation
that satisfies
Z = Λ119879Λ (15)
where 120582119898isin R(4119872119879+1)times1119898 = 1 2 4119872
119879+ 1
Now we construct a semidefinite matrix Y isin
R(119872119879+1)times(119872119879+1) which is defined as
Y = GZG119879 = (ΛG119879)119879
(ΛG119879) (16)
where
G = [U 00 1
] = [I isin R119872119879times119872119879 2I isin R119872119879times119872119879 4I isin R119872119879times119872119879 8I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 0 isin R1times119872119879 0 isin R1times119872119879 1
] (17)
From (13)ndash(15) and (9c) we get
10038171003817100381710038171205821198991003817100381710038171003817 = (120582
119879
119899120582119899)12
= 1 119899 = 1 2 4119872119879 (18)
120582119879
4119872119879+11205824119872119879+1
= 1 (19)
Substituting (14) into (16) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879(Λ1+ 2Λ2+ 4Λ3+ 8Λ4) (Λ1+ 2Λ2+ 4Λ3+ 8Λ4)1198791205824119872119879+1
120582119879
4119872119879+1(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879
120582119879
4119872119879+11205824119872119879+1
]
(20)
where Λ1= [12058211205822 120582
119873119905] isin R(4119872119879+1)times119872119879 Λ
2= [120582119872119879+1
120582119872119879+2
1205822119872119879] isin R(4119872119879+1)times119872119879 Λ
3= [1205822119873119905+1
1205822119873119905+2
1205823119873119905] isin R(4119872119879+1)times119872119879 and Λ
4= [1205823119873119905+1
1205823119873119905+2
1205824119873119905] isin
R(4119872119879+1)times119872119879 From (19) and (20) it can be known thatY
22= 1 and (6d)
is satisfied Moreover we have
Y11= (Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)
(21)
The element located in the 119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120582119895+ 2120582119872119879+119895
+ 41205822119872119879+119895
+ 81205823119872119879+119895
10038171003817100381710038171003817
2
(22)
From the perspective of geometry and (18) it is easy to knowthat
1 =1003817100381710038171003817100381781205823119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120582119895
10038171003817100381710038171003817minus100381710038171003817100381710038172120582119872119879+119895
10038171003817100381710038171003817minus1003817100381710038171003817100381741205822119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120582119895+ 2120582119872119879+119895
+ 41205822119872119879+119895
+ 81205823119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120582119895
10038171003817100381710038171003817+100381710038171003817100381710038172120582119872119879+119895
10038171003817100381710038171003817+1003817100381710038171003817100381741205822119872119879+119895
10038171003817100381710038171003817+1003817100381710038171003817100381781205823119872119879+119895
10038171003817100381710038171003817= 15
(23)
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(24)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix Z isin
R(4119872119879+1)times(4119872119879+1) which satisfies constraints (9b) and (9c)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (25)
Since Y≻0 there must be a reversible matrix Γ = [1205741 1205742
1205743 120574
119872119879 120574119872119879+1
] that satisfies
Y = Γ119879Γ (26)
where 120574119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (25) and (26) and (6c) it can be obtained that10038171003817100381710038171003817120574119872119879+1
10038171003817100381710038171003817= (120574119879
119872119879+1120574119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120574119895
10038171003817100381710038171003817= (120574119879
119895120574119895)12
le 15 119895 = 1 2 119872119879
(27)
From the perspective of geometry it is easy to know thatthere should be vectors 120574
1198951 1205741198952 1205741198953 1205741198954isin R(119872119879+1)times1 which
satisfy
120574119895= 1205741198951+ 21205741198952+ 41205741198953+ 81205741198954 (28)
100381710038171003817100381710038171205741198951
10038171003817100381710038171003817=100381710038171003817100381710038171205741198952
10038171003817100381710038171003817=100381710038171003817100381710038171205741198953
10038171003817100381710038171003817=100381710038171003817100381710038171205741198954
10038171003817100381710038171003817= 1 (29)
Now we construct another matrix on the basis of (26) and(28) which is given by
Γ1015840= [12057411 12057412 12057413 12057414 12057421 12057422 12057423 12057424 120574
1198721199051
1205741198721199052 1205741198721199053 1205741198721199054 120574119872119905+1
]
isin R(119872119879+1)times(4119872119879+1)
(30)
International Journal of Antennas and Propagation 5
Next we construct a semidefinite matrix Z isin
R(4119872119879+1)times(4119872119879+1) which is defined as
Z = (Γ1015840)119879
Γ1015840 (31)
Substituting (30) into (31) the element located in the (119895 times 119894)throw and the (119895 times 119894)th column of Z is given as
119911(119895119894)(119895119894)
= (120574119879
119895119894120574119895119894) =10038171003817100381710038171003817120574119895119894
10038171003817100381710038171003817
2
= 1 (32)
where 119895 = 1 2 119872119879 119894 = 1 2 3 4 Moreover the (4119872
119879+
1)th row and the (4119872119879+ 1)th column of Z are
119911(4119872119879+1)(4119872119879+1)
= (120574119879
119872119879+1120574119872119879+1
) = 1 (33)
From (32) and (33) it can be obtained that
Diag (Z) = 1I (34)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDR andVA-SDR can be concluded
32 Proposed SDR Tighter Than BC-SDR and VA-SDR Sec-ondly we will demonstrate that the constraints of proposedSDR problem are tighter than those of the BC-SDR problemand also tighter than those of the VA-SDR due to theaforementioned equivalence For this purpose a new SDRproblem is constructed given by
min TrW1015840 [V119879H119879HV minusV119879H119879rminusr119879HV r119879r ] (35a)
st 1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (35b)
W1015840 isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (35c)
W101584022= 1 (35d)
By comparing (35a) (35b) (35c) and (35d) and (12a)(12b) (12c) (12d) and (12e) it is apparent that the proposedSDR is tighter than the new SDR since the proposed SDR isnearly the same as the new SDR except that the proposed SDRhas an extra constraint (12c) In what follows we will provethat the new SDR given by (35a) (35b) (35c) and (35d) isequivalent to BC-SDR given by (6a) (6b) (6c) and (6d) andtherefore the proposed SDR being tighter than BC-SDR canbe proven
Step 1 For each matrix W1015840 isin R(2119872119879+1)times(2119872119879+1) that satis-fies constraints (35b)ndash(35d) there must be a matrix Y isin
R(119872119879+1)times(119872119879+1) which satisfies constraints (6b)ndash(6d)
Proof For any matrix W1015840 that satisfies constraints (35b)ndash(35d) it has the following form
W1015840 = [W1015840
11isin R2119872119879times2119872119879 W1015840
12isin R2119872119879times1
W101584021isin R1times2119872119879 1
] (36)
SinceW1015840≻0 there should be a reversible matrix
Ψ
= [12059511205952 120595
119872119879120595119872119879+1
120595119872119879+2
12059521198721198791205952119872119879+1
]
(37)
that satisfies
W1015840 = Ψ119879Ψ (38)
where 120595119904isin R(2119872119879+1)times1 119904 = 1 2 2119872
119879+ 1
From (36)ndash(38) (35b) and (35d) give
1 le10038171003817100381710038171205951198961003817100381710038171003817 = (120595
119879
119896120595119896)12
le 3 119896 = 1 2 2119872119879 (39)
120595119879
2119872119879+11205952119872119879+1
= 1 (40)
Now we construct a semidefinite matrix Y isin R(119872119879+1)times(119872119879+1)
which is defined as
Y = EW1015840E119879 = (ΨE119879)119879
(ΨE119879) (41)
where
E = [V 00 1
]
= [I isin R119872119879times119872119879 4I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 1
]
(42)
Substituting (38) into (41) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (Ψ1+ 4Ψ2)1198791205952119872119879+1
120595119879
2119872119879+1(Ψ1+ 4Ψ2)119879120595119879
2119872119879+11205952119872119879+1
]
(43)
where
Ψ1= [12059511205952 120595
119873119905] isin R
(2119872119879+1)times119872119879
Ψ2= [120595119872119879+1
120595119872119879+2
1205952119872119879] isin R
(2119872119879+1)times119872119879
(44)
From (40) and (43) it can be known that Y22= 1 and
(6d) is satisfied Moreover we have
Y11= (Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (45)
and the element located in the119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120595119895+ 4120595119873119905+119895
10038171003817100381710038171003817
2
119895 = 1 2 119872119879 (46)
From the perspective of geometry and (39) it is easy to knowthat
1 le100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120595119895
10038171003817100381710038171003817le10038171003817100381710038171003817120595119895+ 4120595119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120595119895
10038171003817100381710038171003817+100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817le 15
(47)
6 International Journal of Antennas and Propagation
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(48)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix W1015840 isinR(2119872119879+1)times(2119872119879+1) which satisfies constraints (35b)ndash(35d)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (49)
Since Y≻0 there must be a reversible matrix Δ =
[120590112059021205903 120590
119872119879120590119872119879+1
] that satisfies
Y = Δ119879Δ (50)
where 120590119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (49) (50) and (6c) it can be obtained that10038171003817100381710038171003817120590119872119879+1
10038171003817100381710038171003817= (120590119879
119872119879+1120590119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120590119895
10038171003817100381710038171003817= (120590119879
119895120590119895)12
le 15 119895 = 1 2 119872119879
(51)
From the perspective of geometry it is easy to know that thereshould be vectors 120590
11989511205901198952isin R(119872119879+1)times1 which satisfy
120590119895= 1205901198951+ 41205901198952 (52)
1 le10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817= (120590119879
119895119894120590119895119894)12
le 3 119894 = 1 2 (53)
Now we construct another matrix on the basis of (50) and(52) which is given by
Δ1015840= [12059011120590121205902112059022 120590
11987211990511205901198721199052120590119872119905+1
]
isin R(119872119879+1)times(2119872119879+1)
(54)
Next we construct a semidefinite matrix W1015840 isin
R(2119872119879+1)times(2119872119879+1) which is defined as
W1015840 = (Δ1015840)119879
Δ1015840 (55)
Substituting (54) into (55) the element located in the (119895times 119894)throw and the (119895 times 119894)th column ofW1015840 is given as
1199081015840
(119895119894)(119895119894)= (120590119879
119895119894120590119895119894) =10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817
2
(56)
where 119895 = 1 2 119872119879 119894 = 1 2
Then from (53) and (56) it can be obtained that
1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (57)
Moreover the (2119872119879+ 1)th row and the (2119872
119879+ 1)th column
ofW1015840 are
W101584022= 1199081015840
(2119872119879+1)(2119872119879+1)= (120590119879
119872119879+1120590119872119879+1
) = 1 (58)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDRand new SDR given by (35a) (35b) (35c) and (35d) can beconcluded Therefore the constraints of the proposed SDRare tighter than those of the BC-SDR and also tighter thanthose of the VA-SDR due to the aforementioned equivalence
4 Comparison of Complexity
Firstly the BC-SDR given in (6a)ndash(6d) consists of a (119872119879+1)times
(119872119879+ 1)matrix variable Y and119874(119872
119879+ 1) linear constraints
Since constraints (6c) and (6d) are separable the complexityof the BC-SDR detector is 119874((119872
119879+ 1)35) [9] Secondly the
VA-SDR given in (9a)ndash(9c) involves a (4119872119879+ 1) times (4119872
119879+ 1)
matrix variable Z and 119874(4119872119879+ 1) The constraint (9c) is
also separable thus the complexity of the VA-SDR detectoris119874((4119872
119879+1)35) Finally the proposed SDR shown in (12a)ndash
(12e) consists of a (2119872119879+ 1) times (2119872
119879+ 1) matrix variable
W and two119874(4119872119879) linear constraints According to [13 14]
it can be known that the complexity of the proposed SDRdetector is 119874(4(2119872
119879+ 1)45+ 8(2119872
119879+ 1)35)
5 Simulation Results
Computer simulations were conducted to evaluate the per-formance of these three SDR detectors An uncoded MIMOsystem with independent Rayleigh fading channel was takeninto account and the Sedumi toolbox within Matlab softwarewas used to implement the SDRdetection algorithms Figures1 and 2 show the BLER performances of the three SDRdetectors for 4 times 4 and 8 times 8 256-QAM systems respectivelyIt can be observed that the BC-SDR detector and the VA-SDR detector provide exactly the same BLER performancewhile the proposed SDR detector can provide the best BLERperformance among these three detectors This agrees wellwith the analysis presented in Section 3 What is more itcan be found that the improvement concerning the BLERperformance provided by the proposed SDR detector in thecase of 8 times 8 MIMO system is larger than that in the caseof 4 times 4 MIMO system The relaxation of the alphabet setengaged in theBC-SDRand theVA-SDRwill cause increasingerrors with the increase of the number of the antennas Thatis because the inaccurate detection of any dimension of thetransmitted signal will result in the failure of the detection ofthe whole signal vector and the higher the dimension of theproblem is the bigger the possibility of failure will be Themain merit of the proposed SDR over the BC-SDR and theVA-SDR is that it can offer more accurate relaxation of thealphabet set Moreover the computational times for solvingone signal vector of these SDR detectors are illustrated inFigures 3 and 4 These figures demonstrate that the BC-SDRhas the lowest computational complexity and the VA-SDRhas the highest computational complexity while the proposedSDR is in between the two detection methods It is alsofound that the complexity of SDR detectors is independentof SNR This is a distinct advantage over the SD whosecomplexity varies as a function of SNR making SD difficultto be implemented in practice
International Journal of Antennas and Propagation 7
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 1The BLER performance of the SDR detectors for 4times4 256-QAMMIMO systems
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 2 The BLER performance of the SDR detectors for 8 times 8256-QAMMIMO systems
6 Conclusion
In this paper we proposed a SDR detector for 256-QAMMIMO system and also reviewed two other SDR detectorsnamely BC-SDR detector and VA-SDR detector Then weanalyzed the tightness and the complexity of these three SDRdetectors Both theoretical analysis and simulation resultsdemonstrate that the proposed SDR can provide the bestBLER performance among these three detectors while theBC-SDR detector and the VA-SDR detector provide exactlythe same BLER performance Moreover the BC-SDR has thelowest computational complexity and the VA-SDR has the
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 3 The computational time of the SDR detectors for 4 times 4256-QAMMIMO systems
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
102
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 4 The computational time of the SDR detectors for 8 times 8256-QAMMIMO systems
highest computational complexity while the proposed SDRis in between
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
8 International Journal of Antennas and Propagation
References
[1] J Jalden and B Ottersten ldquoAn exponential lower bound on theexpected complexity of sphere decodingrdquo in Proceedings of theIEEE International Conference on Acoustics Speech and SignalProcessing (ICASSP rsquo04) vol 4 pp iv-393ndashiv-396 MontrealCanada May 2004
[2] A Ben-Tal and A Nemirovski Lectures on Modern ConvexOptimization MPS-SIAM Series on Optimization 2001
[3] J F Sturm ldquoUsing SEDUMI 102 A Matlab toolbox for opti-mization over symmetric conesrdquo Optimization Methods andSoftware vol 11-12 pp 1ndash30 1999
[4] B Steingrimsson Z Q Luo and K M Wong ldquoSoft quasi-maximum-likelihood detection for multiple-antenna wirelesschannelsrdquo IEEE Transactions on Signal Processing vol 51 no11 pp 2710ndash2719 2003
[5] Z Y Shao SWCheung andT I Yuk ldquoLattice-reduction-aidedsemidefinite relaxation detection algorithms for multiple-inputmultiple-output systemsrdquo IETCommunications vol 8 no 4 pp448ndash454 2014
[6] W K Ma T N Davidson K M Wong Z Q Luo andP C Ching ldquoQuasi-maximum-likelihood multiuser detectionusing semi-definite relaxation with application to synchronousCDMArdquo IEEE Transactions on Signal Processing vol 50 no 4pp 912ndash922 2002
[7] A Wiesel Y C Eldar and S Shamai ldquoSemidefinite relaxationfor detection of 16-QAM signaling in MIMO channelsrdquo IEEESignal Processing Letters vol 12 no 9 pp 653ndash656 2005
[8] P H Tan and L K Rasmussen ldquoThe application of semidefiniteprogramming for detection inCDMArdquo IEEE Journal on SelectedAreas in Communications vol 19 no 8 pp 1442ndash1449 2001
[9] N D Sidiropoulos and Z-Q Luo ldquoA semidefinite relaxationapproach to MIMO detection for high-order QAM constella-tionsrdquo IEEE Signal Processing Letters vol 13 no 9 pp 525ndash5282006
[10] Z Mao X Wang and X Wang ldquoSemidefinite programmingrelaxation approach for multiuser detection of QAM signalsrdquoIEEE Transactions on Wireless Communications vol 6 no 12pp 4275ndash4279 2007
[11] W-K Ma C-C Su J Jalden T-H Chang and C-Y ChildquoThe equivalence of semidefinite relaxation MIMO detectorsfor higher-order QAMrdquo IEEE Journal on Selected Topics inSignal Processing vol 3 no 6 pp 1038ndash1052 2009
[12] Z Y Shao SWCheung andT I Yuk ldquoSemi-definite relaxationdecoder for 256-QAM MIMO systemrdquo Electronics Letters vol46 no 11 pp 796ndash797 2010
[13] Y Yang C Zhao P Zhou and W Xu ldquoMIMO detection of 16-QAM signaling based on semidefinite relaxationrdquo IEEE SignalProcessing Letters vol 14 no 11 pp 797ndash800 2007
[14] C Helmberg F Rendl R J Vanderbei and H Wolkowicz ldquoAninterior-point method for semidefinite programmingrdquo SIAMJournal on Optimization vol 6 no 2 pp 342ndash361 1996
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
4 International Journal of Antennas and Propagation
that satisfies
Z = Λ119879Λ (15)
where 120582119898isin R(4119872119879+1)times1119898 = 1 2 4119872
119879+ 1
Now we construct a semidefinite matrix Y isin
R(119872119879+1)times(119872119879+1) which is defined as
Y = GZG119879 = (ΛG119879)119879
(ΛG119879) (16)
where
G = [U 00 1
] = [I isin R119872119879times119872119879 2I isin R119872119879times119872119879 4I isin R119872119879times119872119879 8I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 0 isin R1times119872119879 0 isin R1times119872119879 1
] (17)
From (13)ndash(15) and (9c) we get
10038171003817100381710038171205821198991003817100381710038171003817 = (120582
119879
119899120582119899)12
= 1 119899 = 1 2 4119872119879 (18)
120582119879
4119872119879+11205824119872119879+1
= 1 (19)
Substituting (14) into (16) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879(Λ1+ 2Λ2+ 4Λ3+ 8Λ4) (Λ1+ 2Λ2+ 4Λ3+ 8Λ4)1198791205824119872119879+1
120582119879
4119872119879+1(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879
120582119879
4119872119879+11205824119872119879+1
]
(20)
where Λ1= [12058211205822 120582
119873119905] isin R(4119872119879+1)times119872119879 Λ
2= [120582119872119879+1
120582119872119879+2
1205822119872119879] isin R(4119872119879+1)times119872119879 Λ
3= [1205822119873119905+1
1205822119873119905+2
1205823119873119905] isin R(4119872119879+1)times119872119879 and Λ
4= [1205823119873119905+1
1205823119873119905+2
1205824119873119905] isin
R(4119872119879+1)times119872119879 From (19) and (20) it can be known thatY
22= 1 and (6d)
is satisfied Moreover we have
Y11= (Λ1+ 2Λ2+ 4Λ3+ 8Λ4)119879(Λ1+ 2Λ2+ 4Λ3+ 8Λ4)
(21)
The element located in the 119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120582119895+ 2120582119872119879+119895
+ 41205822119872119879+119895
+ 81205823119872119879+119895
10038171003817100381710038171003817
2
(22)
From the perspective of geometry and (18) it is easy to knowthat
1 =1003817100381710038171003817100381781205823119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120582119895
10038171003817100381710038171003817minus100381710038171003817100381710038172120582119872119879+119895
10038171003817100381710038171003817minus1003817100381710038171003817100381741205822119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120582119895+ 2120582119872119879+119895
+ 41205822119872119879+119895
+ 81205823119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120582119895
10038171003817100381710038171003817+100381710038171003817100381710038172120582119872119879+119895
10038171003817100381710038171003817+1003817100381710038171003817100381741205822119872119879+119895
10038171003817100381710038171003817+1003817100381710038171003817100381781205823119872119879+119895
10038171003817100381710038171003817= 15
(23)
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(24)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix Z isin
R(4119872119879+1)times(4119872119879+1) which satisfies constraints (9b) and (9c)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (25)
Since Y≻0 there must be a reversible matrix Γ = [1205741 1205742
1205743 120574
119872119879 120574119872119879+1
] that satisfies
Y = Γ119879Γ (26)
where 120574119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (25) and (26) and (6c) it can be obtained that10038171003817100381710038171003817120574119872119879+1
10038171003817100381710038171003817= (120574119879
119872119879+1120574119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120574119895
10038171003817100381710038171003817= (120574119879
119895120574119895)12
le 15 119895 = 1 2 119872119879
(27)
From the perspective of geometry it is easy to know thatthere should be vectors 120574
1198951 1205741198952 1205741198953 1205741198954isin R(119872119879+1)times1 which
satisfy
120574119895= 1205741198951+ 21205741198952+ 41205741198953+ 81205741198954 (28)
100381710038171003817100381710038171205741198951
10038171003817100381710038171003817=100381710038171003817100381710038171205741198952
10038171003817100381710038171003817=100381710038171003817100381710038171205741198953
10038171003817100381710038171003817=100381710038171003817100381710038171205741198954
10038171003817100381710038171003817= 1 (29)
Now we construct another matrix on the basis of (26) and(28) which is given by
Γ1015840= [12057411 12057412 12057413 12057414 12057421 12057422 12057423 12057424 120574
1198721199051
1205741198721199052 1205741198721199053 1205741198721199054 120574119872119905+1
]
isin R(119872119879+1)times(4119872119879+1)
(30)
International Journal of Antennas and Propagation 5
Next we construct a semidefinite matrix Z isin
R(4119872119879+1)times(4119872119879+1) which is defined as
Z = (Γ1015840)119879
Γ1015840 (31)
Substituting (30) into (31) the element located in the (119895 times 119894)throw and the (119895 times 119894)th column of Z is given as
119911(119895119894)(119895119894)
= (120574119879
119895119894120574119895119894) =10038171003817100381710038171003817120574119895119894
10038171003817100381710038171003817
2
= 1 (32)
where 119895 = 1 2 119872119879 119894 = 1 2 3 4 Moreover the (4119872
119879+
1)th row and the (4119872119879+ 1)th column of Z are
119911(4119872119879+1)(4119872119879+1)
= (120574119879
119872119879+1120574119872119879+1
) = 1 (33)
From (32) and (33) it can be obtained that
Diag (Z) = 1I (34)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDR andVA-SDR can be concluded
32 Proposed SDR Tighter Than BC-SDR and VA-SDR Sec-ondly we will demonstrate that the constraints of proposedSDR problem are tighter than those of the BC-SDR problemand also tighter than those of the VA-SDR due to theaforementioned equivalence For this purpose a new SDRproblem is constructed given by
min TrW1015840 [V119879H119879HV minusV119879H119879rminusr119879HV r119879r ] (35a)
st 1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (35b)
W1015840 isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (35c)
W101584022= 1 (35d)
By comparing (35a) (35b) (35c) and (35d) and (12a)(12b) (12c) (12d) and (12e) it is apparent that the proposedSDR is tighter than the new SDR since the proposed SDR isnearly the same as the new SDR except that the proposed SDRhas an extra constraint (12c) In what follows we will provethat the new SDR given by (35a) (35b) (35c) and (35d) isequivalent to BC-SDR given by (6a) (6b) (6c) and (6d) andtherefore the proposed SDR being tighter than BC-SDR canbe proven
Step 1 For each matrix W1015840 isin R(2119872119879+1)times(2119872119879+1) that satis-fies constraints (35b)ndash(35d) there must be a matrix Y isin
R(119872119879+1)times(119872119879+1) which satisfies constraints (6b)ndash(6d)
Proof For any matrix W1015840 that satisfies constraints (35b)ndash(35d) it has the following form
W1015840 = [W1015840
11isin R2119872119879times2119872119879 W1015840
12isin R2119872119879times1
W101584021isin R1times2119872119879 1
] (36)
SinceW1015840≻0 there should be a reversible matrix
Ψ
= [12059511205952 120595
119872119879120595119872119879+1
120595119872119879+2
12059521198721198791205952119872119879+1
]
(37)
that satisfies
W1015840 = Ψ119879Ψ (38)
where 120595119904isin R(2119872119879+1)times1 119904 = 1 2 2119872
119879+ 1
From (36)ndash(38) (35b) and (35d) give
1 le10038171003817100381710038171205951198961003817100381710038171003817 = (120595
119879
119896120595119896)12
le 3 119896 = 1 2 2119872119879 (39)
120595119879
2119872119879+11205952119872119879+1
= 1 (40)
Now we construct a semidefinite matrix Y isin R(119872119879+1)times(119872119879+1)
which is defined as
Y = EW1015840E119879 = (ΨE119879)119879
(ΨE119879) (41)
where
E = [V 00 1
]
= [I isin R119872119879times119872119879 4I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 1
]
(42)
Substituting (38) into (41) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (Ψ1+ 4Ψ2)1198791205952119872119879+1
120595119879
2119872119879+1(Ψ1+ 4Ψ2)119879120595119879
2119872119879+11205952119872119879+1
]
(43)
where
Ψ1= [12059511205952 120595
119873119905] isin R
(2119872119879+1)times119872119879
Ψ2= [120595119872119879+1
120595119872119879+2
1205952119872119879] isin R
(2119872119879+1)times119872119879
(44)
From (40) and (43) it can be known that Y22= 1 and
(6d) is satisfied Moreover we have
Y11= (Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (45)
and the element located in the119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120595119895+ 4120595119873119905+119895
10038171003817100381710038171003817
2
119895 = 1 2 119872119879 (46)
From the perspective of geometry and (39) it is easy to knowthat
1 le100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120595119895
10038171003817100381710038171003817le10038171003817100381710038171003817120595119895+ 4120595119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120595119895
10038171003817100381710038171003817+100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817le 15
(47)
6 International Journal of Antennas and Propagation
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(48)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix W1015840 isinR(2119872119879+1)times(2119872119879+1) which satisfies constraints (35b)ndash(35d)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (49)
Since Y≻0 there must be a reversible matrix Δ =
[120590112059021205903 120590
119872119879120590119872119879+1
] that satisfies
Y = Δ119879Δ (50)
where 120590119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (49) (50) and (6c) it can be obtained that10038171003817100381710038171003817120590119872119879+1
10038171003817100381710038171003817= (120590119879
119872119879+1120590119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120590119895
10038171003817100381710038171003817= (120590119879
119895120590119895)12
le 15 119895 = 1 2 119872119879
(51)
From the perspective of geometry it is easy to know that thereshould be vectors 120590
11989511205901198952isin R(119872119879+1)times1 which satisfy
120590119895= 1205901198951+ 41205901198952 (52)
1 le10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817= (120590119879
119895119894120590119895119894)12
le 3 119894 = 1 2 (53)
Now we construct another matrix on the basis of (50) and(52) which is given by
Δ1015840= [12059011120590121205902112059022 120590
11987211990511205901198721199052120590119872119905+1
]
isin R(119872119879+1)times(2119872119879+1)
(54)
Next we construct a semidefinite matrix W1015840 isin
R(2119872119879+1)times(2119872119879+1) which is defined as
W1015840 = (Δ1015840)119879
Δ1015840 (55)
Substituting (54) into (55) the element located in the (119895times 119894)throw and the (119895 times 119894)th column ofW1015840 is given as
1199081015840
(119895119894)(119895119894)= (120590119879
119895119894120590119895119894) =10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817
2
(56)
where 119895 = 1 2 119872119879 119894 = 1 2
Then from (53) and (56) it can be obtained that
1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (57)
Moreover the (2119872119879+ 1)th row and the (2119872
119879+ 1)th column
ofW1015840 are
W101584022= 1199081015840
(2119872119879+1)(2119872119879+1)= (120590119879
119872119879+1120590119872119879+1
) = 1 (58)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDRand new SDR given by (35a) (35b) (35c) and (35d) can beconcluded Therefore the constraints of the proposed SDRare tighter than those of the BC-SDR and also tighter thanthose of the VA-SDR due to the aforementioned equivalence
4 Comparison of Complexity
Firstly the BC-SDR given in (6a)ndash(6d) consists of a (119872119879+1)times
(119872119879+ 1)matrix variable Y and119874(119872
119879+ 1) linear constraints
Since constraints (6c) and (6d) are separable the complexityof the BC-SDR detector is 119874((119872
119879+ 1)35) [9] Secondly the
VA-SDR given in (9a)ndash(9c) involves a (4119872119879+ 1) times (4119872
119879+ 1)
matrix variable Z and 119874(4119872119879+ 1) The constraint (9c) is
also separable thus the complexity of the VA-SDR detectoris119874((4119872
119879+1)35) Finally the proposed SDR shown in (12a)ndash
(12e) consists of a (2119872119879+ 1) times (2119872
119879+ 1) matrix variable
W and two119874(4119872119879) linear constraints According to [13 14]
it can be known that the complexity of the proposed SDRdetector is 119874(4(2119872
119879+ 1)45+ 8(2119872
119879+ 1)35)
5 Simulation Results
Computer simulations were conducted to evaluate the per-formance of these three SDR detectors An uncoded MIMOsystem with independent Rayleigh fading channel was takeninto account and the Sedumi toolbox within Matlab softwarewas used to implement the SDRdetection algorithms Figures1 and 2 show the BLER performances of the three SDRdetectors for 4 times 4 and 8 times 8 256-QAM systems respectivelyIt can be observed that the BC-SDR detector and the VA-SDR detector provide exactly the same BLER performancewhile the proposed SDR detector can provide the best BLERperformance among these three detectors This agrees wellwith the analysis presented in Section 3 What is more itcan be found that the improvement concerning the BLERperformance provided by the proposed SDR detector in thecase of 8 times 8 MIMO system is larger than that in the caseof 4 times 4 MIMO system The relaxation of the alphabet setengaged in theBC-SDRand theVA-SDRwill cause increasingerrors with the increase of the number of the antennas Thatis because the inaccurate detection of any dimension of thetransmitted signal will result in the failure of the detection ofthe whole signal vector and the higher the dimension of theproblem is the bigger the possibility of failure will be Themain merit of the proposed SDR over the BC-SDR and theVA-SDR is that it can offer more accurate relaxation of thealphabet set Moreover the computational times for solvingone signal vector of these SDR detectors are illustrated inFigures 3 and 4 These figures demonstrate that the BC-SDRhas the lowest computational complexity and the VA-SDRhas the highest computational complexity while the proposedSDR is in between the two detection methods It is alsofound that the complexity of SDR detectors is independentof SNR This is a distinct advantage over the SD whosecomplexity varies as a function of SNR making SD difficultto be implemented in practice
International Journal of Antennas and Propagation 7
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 1The BLER performance of the SDR detectors for 4times4 256-QAMMIMO systems
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 2 The BLER performance of the SDR detectors for 8 times 8256-QAMMIMO systems
6 Conclusion
In this paper we proposed a SDR detector for 256-QAMMIMO system and also reviewed two other SDR detectorsnamely BC-SDR detector and VA-SDR detector Then weanalyzed the tightness and the complexity of these three SDRdetectors Both theoretical analysis and simulation resultsdemonstrate that the proposed SDR can provide the bestBLER performance among these three detectors while theBC-SDR detector and the VA-SDR detector provide exactlythe same BLER performance Moreover the BC-SDR has thelowest computational complexity and the VA-SDR has the
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 3 The computational time of the SDR detectors for 4 times 4256-QAMMIMO systems
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
102
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 4 The computational time of the SDR detectors for 8 times 8256-QAMMIMO systems
highest computational complexity while the proposed SDRis in between
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
8 International Journal of Antennas and Propagation
References
[1] J Jalden and B Ottersten ldquoAn exponential lower bound on theexpected complexity of sphere decodingrdquo in Proceedings of theIEEE International Conference on Acoustics Speech and SignalProcessing (ICASSP rsquo04) vol 4 pp iv-393ndashiv-396 MontrealCanada May 2004
[2] A Ben-Tal and A Nemirovski Lectures on Modern ConvexOptimization MPS-SIAM Series on Optimization 2001
[3] J F Sturm ldquoUsing SEDUMI 102 A Matlab toolbox for opti-mization over symmetric conesrdquo Optimization Methods andSoftware vol 11-12 pp 1ndash30 1999
[4] B Steingrimsson Z Q Luo and K M Wong ldquoSoft quasi-maximum-likelihood detection for multiple-antenna wirelesschannelsrdquo IEEE Transactions on Signal Processing vol 51 no11 pp 2710ndash2719 2003
[5] Z Y Shao SWCheung andT I Yuk ldquoLattice-reduction-aidedsemidefinite relaxation detection algorithms for multiple-inputmultiple-output systemsrdquo IETCommunications vol 8 no 4 pp448ndash454 2014
[6] W K Ma T N Davidson K M Wong Z Q Luo andP C Ching ldquoQuasi-maximum-likelihood multiuser detectionusing semi-definite relaxation with application to synchronousCDMArdquo IEEE Transactions on Signal Processing vol 50 no 4pp 912ndash922 2002
[7] A Wiesel Y C Eldar and S Shamai ldquoSemidefinite relaxationfor detection of 16-QAM signaling in MIMO channelsrdquo IEEESignal Processing Letters vol 12 no 9 pp 653ndash656 2005
[8] P H Tan and L K Rasmussen ldquoThe application of semidefiniteprogramming for detection inCDMArdquo IEEE Journal on SelectedAreas in Communications vol 19 no 8 pp 1442ndash1449 2001
[9] N D Sidiropoulos and Z-Q Luo ldquoA semidefinite relaxationapproach to MIMO detection for high-order QAM constella-tionsrdquo IEEE Signal Processing Letters vol 13 no 9 pp 525ndash5282006
[10] Z Mao X Wang and X Wang ldquoSemidefinite programmingrelaxation approach for multiuser detection of QAM signalsrdquoIEEE Transactions on Wireless Communications vol 6 no 12pp 4275ndash4279 2007
[11] W-K Ma C-C Su J Jalden T-H Chang and C-Y ChildquoThe equivalence of semidefinite relaxation MIMO detectorsfor higher-order QAMrdquo IEEE Journal on Selected Topics inSignal Processing vol 3 no 6 pp 1038ndash1052 2009
[12] Z Y Shao SWCheung andT I Yuk ldquoSemi-definite relaxationdecoder for 256-QAM MIMO systemrdquo Electronics Letters vol46 no 11 pp 796ndash797 2010
[13] Y Yang C Zhao P Zhou and W Xu ldquoMIMO detection of 16-QAM signaling based on semidefinite relaxationrdquo IEEE SignalProcessing Letters vol 14 no 11 pp 797ndash800 2007
[14] C Helmberg F Rendl R J Vanderbei and H Wolkowicz ldquoAninterior-point method for semidefinite programmingrdquo SIAMJournal on Optimization vol 6 no 2 pp 342ndash361 1996
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
International Journal of Antennas and Propagation 5
Next we construct a semidefinite matrix Z isin
R(4119872119879+1)times(4119872119879+1) which is defined as
Z = (Γ1015840)119879
Γ1015840 (31)
Substituting (30) into (31) the element located in the (119895 times 119894)throw and the (119895 times 119894)th column of Z is given as
119911(119895119894)(119895119894)
= (120574119879
119895119894120574119895119894) =10038171003817100381710038171003817120574119895119894
10038171003817100381710038171003817
2
= 1 (32)
where 119895 = 1 2 119872119879 119894 = 1 2 3 4 Moreover the (4119872
119879+
1)th row and the (4119872119879+ 1)th column of Z are
119911(4119872119879+1)(4119872119879+1)
= (120574119879
119872119879+1120574119872119879+1
) = 1 (33)
From (32) and (33) it can be obtained that
Diag (Z) = 1I (34)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDR andVA-SDR can be concluded
32 Proposed SDR Tighter Than BC-SDR and VA-SDR Sec-ondly we will demonstrate that the constraints of proposedSDR problem are tighter than those of the BC-SDR problemand also tighter than those of the VA-SDR due to theaforementioned equivalence For this purpose a new SDRproblem is constructed given by
min TrW1015840 [V119879H119879HV minusV119879H119879rminusr119879HV r119879r ] (35a)
st 1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (35b)
W1015840 isin R(2119872119879+1)times(2119872119879+1) ≻ 0 (35c)
W101584022= 1 (35d)
By comparing (35a) (35b) (35c) and (35d) and (12a)(12b) (12c) (12d) and (12e) it is apparent that the proposedSDR is tighter than the new SDR since the proposed SDR isnearly the same as the new SDR except that the proposed SDRhas an extra constraint (12c) In what follows we will provethat the new SDR given by (35a) (35b) (35c) and (35d) isequivalent to BC-SDR given by (6a) (6b) (6c) and (6d) andtherefore the proposed SDR being tighter than BC-SDR canbe proven
Step 1 For each matrix W1015840 isin R(2119872119879+1)times(2119872119879+1) that satis-fies constraints (35b)ndash(35d) there must be a matrix Y isin
R(119872119879+1)times(119872119879+1) which satisfies constraints (6b)ndash(6d)
Proof For any matrix W1015840 that satisfies constraints (35b)ndash(35d) it has the following form
W1015840 = [W1015840
11isin R2119872119879times2119872119879 W1015840
12isin R2119872119879times1
W101584021isin R1times2119872119879 1
] (36)
SinceW1015840≻0 there should be a reversible matrix
Ψ
= [12059511205952 120595
119872119879120595119872119879+1
120595119872119879+2
12059521198721198791205952119872119879+1
]
(37)
that satisfies
W1015840 = Ψ119879Ψ (38)
where 120595119904isin R(2119872119879+1)times1 119904 = 1 2 2119872
119879+ 1
From (36)ndash(38) (35b) and (35d) give
1 le10038171003817100381710038171205951198961003817100381710038171003817 = (120595
119879
119896120595119896)12
le 3 119896 = 1 2 2119872119879 (39)
120595119879
2119872119879+11205952119872119879+1
= 1 (40)
Now we construct a semidefinite matrix Y isin R(119872119879+1)times(119872119879+1)
which is defined as
Y = EW1015840E119879 = (ΨE119879)119879
(ΨE119879) (41)
where
E = [V 00 1
]
= [I isin R119872119879times119872119879 4I isin R119872119879times119872119879 0 isin R119872119879times10 isin R1times119872119879 0 isin R1times119872119879 1
]
(42)
Substituting (38) into (41) gives
Y = [Y11 Y12
Y21
Y22
]
= [(Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (Ψ1+ 4Ψ2)1198791205952119872119879+1
120595119879
2119872119879+1(Ψ1+ 4Ψ2)119879120595119879
2119872119879+11205952119872119879+1
]
(43)
where
Ψ1= [12059511205952 120595
119873119905] isin R
(2119872119879+1)times119872119879
Ψ2= [120595119872119879+1
120595119872119879+2
1205952119872119879] isin R
(2119872119879+1)times119872119879
(44)
From (40) and (43) it can be known that Y22= 1 and
(6d) is satisfied Moreover we have
Y11= (Ψ1+ 4Ψ2)119879(Ψ1+ 4Ψ2) (45)
and the element located in the119895th row and the 119895th column ofY11
is
119910119895119895=10038171003817100381710038171003817120595119895+ 4120595119873119905+119895
10038171003817100381710038171003817
2
119895 = 1 2 119872119879 (46)
From the perspective of geometry and (39) it is easy to knowthat
1 le100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817minus10038171003817100381710038171003817120595119895
10038171003817100381710038171003817le10038171003817100381710038171003817120595119895+ 4120595119872119879+119895
10038171003817100381710038171003817
le10038171003817100381710038171003817120595119895
10038171003817100381710038171003817+100381710038171003817100381710038174120595119872119879+119895
10038171003817100381710038171003817le 15
(47)
6 International Journal of Antennas and Propagation
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(48)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix W1015840 isinR(2119872119879+1)times(2119872119879+1) which satisfies constraints (35b)ndash(35d)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (49)
Since Y≻0 there must be a reversible matrix Δ =
[120590112059021205903 120590
119872119879120590119872119879+1
] that satisfies
Y = Δ119879Δ (50)
where 120590119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (49) (50) and (6c) it can be obtained that10038171003817100381710038171003817120590119872119879+1
10038171003817100381710038171003817= (120590119879
119872119879+1120590119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120590119895
10038171003817100381710038171003817= (120590119879
119895120590119895)12
le 15 119895 = 1 2 119872119879
(51)
From the perspective of geometry it is easy to know that thereshould be vectors 120590
11989511205901198952isin R(119872119879+1)times1 which satisfy
120590119895= 1205901198951+ 41205901198952 (52)
1 le10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817= (120590119879
119895119894120590119895119894)12
le 3 119894 = 1 2 (53)
Now we construct another matrix on the basis of (50) and(52) which is given by
Δ1015840= [12059011120590121205902112059022 120590
11987211990511205901198721199052120590119872119905+1
]
isin R(119872119879+1)times(2119872119879+1)
(54)
Next we construct a semidefinite matrix W1015840 isin
R(2119872119879+1)times(2119872119879+1) which is defined as
W1015840 = (Δ1015840)119879
Δ1015840 (55)
Substituting (54) into (55) the element located in the (119895times 119894)throw and the (119895 times 119894)th column ofW1015840 is given as
1199081015840
(119895119894)(119895119894)= (120590119879
119895119894120590119895119894) =10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817
2
(56)
where 119895 = 1 2 119872119879 119894 = 1 2
Then from (53) and (56) it can be obtained that
1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (57)
Moreover the (2119872119879+ 1)th row and the (2119872
119879+ 1)th column
ofW1015840 are
W101584022= 1199081015840
(2119872119879+1)(2119872119879+1)= (120590119879
119872119879+1120590119872119879+1
) = 1 (58)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDRand new SDR given by (35a) (35b) (35c) and (35d) can beconcluded Therefore the constraints of the proposed SDRare tighter than those of the BC-SDR and also tighter thanthose of the VA-SDR due to the aforementioned equivalence
4 Comparison of Complexity
Firstly the BC-SDR given in (6a)ndash(6d) consists of a (119872119879+1)times
(119872119879+ 1)matrix variable Y and119874(119872
119879+ 1) linear constraints
Since constraints (6c) and (6d) are separable the complexityof the BC-SDR detector is 119874((119872
119879+ 1)35) [9] Secondly the
VA-SDR given in (9a)ndash(9c) involves a (4119872119879+ 1) times (4119872
119879+ 1)
matrix variable Z and 119874(4119872119879+ 1) The constraint (9c) is
also separable thus the complexity of the VA-SDR detectoris119874((4119872
119879+1)35) Finally the proposed SDR shown in (12a)ndash
(12e) consists of a (2119872119879+ 1) times (2119872
119879+ 1) matrix variable
W and two119874(4119872119879) linear constraints According to [13 14]
it can be known that the complexity of the proposed SDRdetector is 119874(4(2119872
119879+ 1)45+ 8(2119872
119879+ 1)35)
5 Simulation Results
Computer simulations were conducted to evaluate the per-formance of these three SDR detectors An uncoded MIMOsystem with independent Rayleigh fading channel was takeninto account and the Sedumi toolbox within Matlab softwarewas used to implement the SDRdetection algorithms Figures1 and 2 show the BLER performances of the three SDRdetectors for 4 times 4 and 8 times 8 256-QAM systems respectivelyIt can be observed that the BC-SDR detector and the VA-SDR detector provide exactly the same BLER performancewhile the proposed SDR detector can provide the best BLERperformance among these three detectors This agrees wellwith the analysis presented in Section 3 What is more itcan be found that the improvement concerning the BLERperformance provided by the proposed SDR detector in thecase of 8 times 8 MIMO system is larger than that in the caseof 4 times 4 MIMO system The relaxation of the alphabet setengaged in theBC-SDRand theVA-SDRwill cause increasingerrors with the increase of the number of the antennas Thatis because the inaccurate detection of any dimension of thetransmitted signal will result in the failure of the detection ofthe whole signal vector and the higher the dimension of theproblem is the bigger the possibility of failure will be Themain merit of the proposed SDR over the BC-SDR and theVA-SDR is that it can offer more accurate relaxation of thealphabet set Moreover the computational times for solvingone signal vector of these SDR detectors are illustrated inFigures 3 and 4 These figures demonstrate that the BC-SDRhas the lowest computational complexity and the VA-SDRhas the highest computational complexity while the proposedSDR is in between the two detection methods It is alsofound that the complexity of SDR detectors is independentof SNR This is a distinct advantage over the SD whosecomplexity varies as a function of SNR making SD difficultto be implemented in practice
International Journal of Antennas and Propagation 7
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 1The BLER performance of the SDR detectors for 4times4 256-QAMMIMO systems
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 2 The BLER performance of the SDR detectors for 8 times 8256-QAMMIMO systems
6 Conclusion
In this paper we proposed a SDR detector for 256-QAMMIMO system and also reviewed two other SDR detectorsnamely BC-SDR detector and VA-SDR detector Then weanalyzed the tightness and the complexity of these three SDRdetectors Both theoretical analysis and simulation resultsdemonstrate that the proposed SDR can provide the bestBLER performance among these three detectors while theBC-SDR detector and the VA-SDR detector provide exactlythe same BLER performance Moreover the BC-SDR has thelowest computational complexity and the VA-SDR has the
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 3 The computational time of the SDR detectors for 4 times 4256-QAMMIMO systems
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
102
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 4 The computational time of the SDR detectors for 8 times 8256-QAMMIMO systems
highest computational complexity while the proposed SDRis in between
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
8 International Journal of Antennas and Propagation
References
[1] J Jalden and B Ottersten ldquoAn exponential lower bound on theexpected complexity of sphere decodingrdquo in Proceedings of theIEEE International Conference on Acoustics Speech and SignalProcessing (ICASSP rsquo04) vol 4 pp iv-393ndashiv-396 MontrealCanada May 2004
[2] A Ben-Tal and A Nemirovski Lectures on Modern ConvexOptimization MPS-SIAM Series on Optimization 2001
[3] J F Sturm ldquoUsing SEDUMI 102 A Matlab toolbox for opti-mization over symmetric conesrdquo Optimization Methods andSoftware vol 11-12 pp 1ndash30 1999
[4] B Steingrimsson Z Q Luo and K M Wong ldquoSoft quasi-maximum-likelihood detection for multiple-antenna wirelesschannelsrdquo IEEE Transactions on Signal Processing vol 51 no11 pp 2710ndash2719 2003
[5] Z Y Shao SWCheung andT I Yuk ldquoLattice-reduction-aidedsemidefinite relaxation detection algorithms for multiple-inputmultiple-output systemsrdquo IETCommunications vol 8 no 4 pp448ndash454 2014
[6] W K Ma T N Davidson K M Wong Z Q Luo andP C Ching ldquoQuasi-maximum-likelihood multiuser detectionusing semi-definite relaxation with application to synchronousCDMArdquo IEEE Transactions on Signal Processing vol 50 no 4pp 912ndash922 2002
[7] A Wiesel Y C Eldar and S Shamai ldquoSemidefinite relaxationfor detection of 16-QAM signaling in MIMO channelsrdquo IEEESignal Processing Letters vol 12 no 9 pp 653ndash656 2005
[8] P H Tan and L K Rasmussen ldquoThe application of semidefiniteprogramming for detection inCDMArdquo IEEE Journal on SelectedAreas in Communications vol 19 no 8 pp 1442ndash1449 2001
[9] N D Sidiropoulos and Z-Q Luo ldquoA semidefinite relaxationapproach to MIMO detection for high-order QAM constella-tionsrdquo IEEE Signal Processing Letters vol 13 no 9 pp 525ndash5282006
[10] Z Mao X Wang and X Wang ldquoSemidefinite programmingrelaxation approach for multiuser detection of QAM signalsrdquoIEEE Transactions on Wireless Communications vol 6 no 12pp 4275ndash4279 2007
[11] W-K Ma C-C Su J Jalden T-H Chang and C-Y ChildquoThe equivalence of semidefinite relaxation MIMO detectorsfor higher-order QAMrdquo IEEE Journal on Selected Topics inSignal Processing vol 3 no 6 pp 1038ndash1052 2009
[12] Z Y Shao SWCheung andT I Yuk ldquoSemi-definite relaxationdecoder for 256-QAM MIMO systemrdquo Electronics Letters vol46 no 11 pp 796ndash797 2010
[13] Y Yang C Zhao P Zhou and W Xu ldquoMIMO detection of 16-QAM signaling based on semidefinite relaxationrdquo IEEE SignalProcessing Letters vol 14 no 11 pp 797ndash800 2007
[14] C Helmberg F Rendl R J Vanderbei and H Wolkowicz ldquoAninterior-point method for semidefinite programmingrdquo SIAMJournal on Optimization vol 6 no 2 pp 342ndash361 1996
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
6 International Journal of Antennas and Propagation
Thus we have1 le 119910119895119895le 225
1I≺Diag Y11 ≺ 225I
(48)
The proof is complete
Step 2 For each matrix Y isin R(119872119879+1)times(119872119879+1) that satis-fies constraints (6b)ndash(6d) there must be a matrix W1015840 isinR(2119872119879+1)times(2119872119879+1) which satisfies constraints (35b)ndash(35d)
Proof For anymatrixY that satisfies constraints (6b)ndash(6d) ithas the following form
Y = [Y11 isin R119872119879times119872119879 Y
12isin R119872119879times1
Y21isin R1times119872119879 1
] (49)
Since Y≻0 there must be a reversible matrix Δ =
[120590112059021205903 120590
119872119879120590119872119879+1
] that satisfies
Y = Δ119879Δ (50)
where 120590119897isin R(119872119879+1)times1 119897 = 1 2 119872
119879+ 1
From (49) (50) and (6c) it can be obtained that10038171003817100381710038171003817120590119872119879+1
10038171003817100381710038171003817= (120590119879
119872119879+1120590119872119879+1
)12
= 1
1 le10038171003817100381710038171003817120590119895
10038171003817100381710038171003817= (120590119879
119895120590119895)12
le 15 119895 = 1 2 119872119879
(51)
From the perspective of geometry it is easy to know that thereshould be vectors 120590
11989511205901198952isin R(119872119879+1)times1 which satisfy
120590119895= 1205901198951+ 41205901198952 (52)
1 le10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817= (120590119879
119895119894120590119895119894)12
le 3 119894 = 1 2 (53)
Now we construct another matrix on the basis of (50) and(52) which is given by
Δ1015840= [12059011120590121205902112059022 120590
11987211990511205901198721199052120590119872119905+1
]
isin R(119872119879+1)times(2119872119879+1)
(54)
Next we construct a semidefinite matrix W1015840 isin
R(2119872119879+1)times(2119872119879+1) which is defined as
W1015840 = (Δ1015840)119879
Δ1015840 (55)
Substituting (54) into (55) the element located in the (119895times 119894)throw and the (119895 times 119894)th column ofW1015840 is given as
1199081015840
(119895119894)(119895119894)= (120590119879
119895119894120590119895119894) =10038171003817100381710038171003817120590119895119894
10038171003817100381710038171003817
2
(56)
where 119895 = 1 2 119872119879 119894 = 1 2
Then from (53) and (56) it can be obtained that
1I≺Diag W101584011 ≺ 9I I isin R2119872119879times2119872119879 (57)
Moreover the (2119872119879+ 1)th row and the (2119872
119879+ 1)th column
ofW1015840 are
W101584022= 1199081015840
(2119872119879+1)(2119872119879+1)= (120590119879
119872119879+1120590119872119879+1
) = 1 (58)
The proof is complete
From both Steps 1 and 2 the equivalence of BC-SDRand new SDR given by (35a) (35b) (35c) and (35d) can beconcluded Therefore the constraints of the proposed SDRare tighter than those of the BC-SDR and also tighter thanthose of the VA-SDR due to the aforementioned equivalence
4 Comparison of Complexity
Firstly the BC-SDR given in (6a)ndash(6d) consists of a (119872119879+1)times
(119872119879+ 1)matrix variable Y and119874(119872
119879+ 1) linear constraints
Since constraints (6c) and (6d) are separable the complexityof the BC-SDR detector is 119874((119872
119879+ 1)35) [9] Secondly the
VA-SDR given in (9a)ndash(9c) involves a (4119872119879+ 1) times (4119872
119879+ 1)
matrix variable Z and 119874(4119872119879+ 1) The constraint (9c) is
also separable thus the complexity of the VA-SDR detectoris119874((4119872
119879+1)35) Finally the proposed SDR shown in (12a)ndash
(12e) consists of a (2119872119879+ 1) times (2119872
119879+ 1) matrix variable
W and two119874(4119872119879) linear constraints According to [13 14]
it can be known that the complexity of the proposed SDRdetector is 119874(4(2119872
119879+ 1)45+ 8(2119872
119879+ 1)35)
5 Simulation Results
Computer simulations were conducted to evaluate the per-formance of these three SDR detectors An uncoded MIMOsystem with independent Rayleigh fading channel was takeninto account and the Sedumi toolbox within Matlab softwarewas used to implement the SDRdetection algorithms Figures1 and 2 show the BLER performances of the three SDRdetectors for 4 times 4 and 8 times 8 256-QAM systems respectivelyIt can be observed that the BC-SDR detector and the VA-SDR detector provide exactly the same BLER performancewhile the proposed SDR detector can provide the best BLERperformance among these three detectors This agrees wellwith the analysis presented in Section 3 What is more itcan be found that the improvement concerning the BLERperformance provided by the proposed SDR detector in thecase of 8 times 8 MIMO system is larger than that in the caseof 4 times 4 MIMO system The relaxation of the alphabet setengaged in theBC-SDRand theVA-SDRwill cause increasingerrors with the increase of the number of the antennas Thatis because the inaccurate detection of any dimension of thetransmitted signal will result in the failure of the detection ofthe whole signal vector and the higher the dimension of theproblem is the bigger the possibility of failure will be Themain merit of the proposed SDR over the BC-SDR and theVA-SDR is that it can offer more accurate relaxation of thealphabet set Moreover the computational times for solvingone signal vector of these SDR detectors are illustrated inFigures 3 and 4 These figures demonstrate that the BC-SDRhas the lowest computational complexity and the VA-SDRhas the highest computational complexity while the proposedSDR is in between the two detection methods It is alsofound that the complexity of SDR detectors is independentof SNR This is a distinct advantage over the SD whosecomplexity varies as a function of SNR making SD difficultto be implemented in practice
International Journal of Antennas and Propagation 7
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 1The BLER performance of the SDR detectors for 4times4 256-QAMMIMO systems
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 2 The BLER performance of the SDR detectors for 8 times 8256-QAMMIMO systems
6 Conclusion
In this paper we proposed a SDR detector for 256-QAMMIMO system and also reviewed two other SDR detectorsnamely BC-SDR detector and VA-SDR detector Then weanalyzed the tightness and the complexity of these three SDRdetectors Both theoretical analysis and simulation resultsdemonstrate that the proposed SDR can provide the bestBLER performance among these three detectors while theBC-SDR detector and the VA-SDR detector provide exactlythe same BLER performance Moreover the BC-SDR has thelowest computational complexity and the VA-SDR has the
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 3 The computational time of the SDR detectors for 4 times 4256-QAMMIMO systems
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
102
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 4 The computational time of the SDR detectors for 8 times 8256-QAMMIMO systems
highest computational complexity while the proposed SDRis in between
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
8 International Journal of Antennas and Propagation
References
[1] J Jalden and B Ottersten ldquoAn exponential lower bound on theexpected complexity of sphere decodingrdquo in Proceedings of theIEEE International Conference on Acoustics Speech and SignalProcessing (ICASSP rsquo04) vol 4 pp iv-393ndashiv-396 MontrealCanada May 2004
[2] A Ben-Tal and A Nemirovski Lectures on Modern ConvexOptimization MPS-SIAM Series on Optimization 2001
[3] J F Sturm ldquoUsing SEDUMI 102 A Matlab toolbox for opti-mization over symmetric conesrdquo Optimization Methods andSoftware vol 11-12 pp 1ndash30 1999
[4] B Steingrimsson Z Q Luo and K M Wong ldquoSoft quasi-maximum-likelihood detection for multiple-antenna wirelesschannelsrdquo IEEE Transactions on Signal Processing vol 51 no11 pp 2710ndash2719 2003
[5] Z Y Shao SWCheung andT I Yuk ldquoLattice-reduction-aidedsemidefinite relaxation detection algorithms for multiple-inputmultiple-output systemsrdquo IETCommunications vol 8 no 4 pp448ndash454 2014
[6] W K Ma T N Davidson K M Wong Z Q Luo andP C Ching ldquoQuasi-maximum-likelihood multiuser detectionusing semi-definite relaxation with application to synchronousCDMArdquo IEEE Transactions on Signal Processing vol 50 no 4pp 912ndash922 2002
[7] A Wiesel Y C Eldar and S Shamai ldquoSemidefinite relaxationfor detection of 16-QAM signaling in MIMO channelsrdquo IEEESignal Processing Letters vol 12 no 9 pp 653ndash656 2005
[8] P H Tan and L K Rasmussen ldquoThe application of semidefiniteprogramming for detection inCDMArdquo IEEE Journal on SelectedAreas in Communications vol 19 no 8 pp 1442ndash1449 2001
[9] N D Sidiropoulos and Z-Q Luo ldquoA semidefinite relaxationapproach to MIMO detection for high-order QAM constella-tionsrdquo IEEE Signal Processing Letters vol 13 no 9 pp 525ndash5282006
[10] Z Mao X Wang and X Wang ldquoSemidefinite programmingrelaxation approach for multiuser detection of QAM signalsrdquoIEEE Transactions on Wireless Communications vol 6 no 12pp 4275ndash4279 2007
[11] W-K Ma C-C Su J Jalden T-H Chang and C-Y ChildquoThe equivalence of semidefinite relaxation MIMO detectorsfor higher-order QAMrdquo IEEE Journal on Selected Topics inSignal Processing vol 3 no 6 pp 1038ndash1052 2009
[12] Z Y Shao SWCheung andT I Yuk ldquoSemi-definite relaxationdecoder for 256-QAM MIMO systemrdquo Electronics Letters vol46 no 11 pp 796ndash797 2010
[13] Y Yang C Zhao P Zhou and W Xu ldquoMIMO detection of 16-QAM signaling based on semidefinite relaxationrdquo IEEE SignalProcessing Letters vol 14 no 11 pp 797ndash800 2007
[14] C Helmberg F Rendl R J Vanderbei and H Wolkowicz ldquoAninterior-point method for semidefinite programmingrdquo SIAMJournal on Optimization vol 6 no 2 pp 342ndash361 1996
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
International Journal of Antennas and Propagation 7
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 1The BLER performance of the SDR detectors for 4times4 256-QAMMIMO systems
0 5 10 15 20 25
BLER
100
10minus1
10minus2
10minus3
BC-SDRVA-SDR
Proposed SDRML decoding
EbN0 (dB)
Figure 2 The BLER performance of the SDR detectors for 8 times 8256-QAMMIMO systems
6 Conclusion
In this paper we proposed a SDR detector for 256-QAMMIMO system and also reviewed two other SDR detectorsnamely BC-SDR detector and VA-SDR detector Then weanalyzed the tightness and the complexity of these three SDRdetectors Both theoretical analysis and simulation resultsdemonstrate that the proposed SDR can provide the bestBLER performance among these three detectors while theBC-SDR detector and the VA-SDR detector provide exactlythe same BLER performance Moreover the BC-SDR has thelowest computational complexity and the VA-SDR has the
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 3 The computational time of the SDR detectors for 4 times 4256-QAMMIMO systems
0 5 10 15 20 25
Com
puta
tiona
l tim
e (s)
100
101
102
10minus1
BC-SDRVA-SDRProposed SDR
EbN0 (dB)
Figure 4 The computational time of the SDR detectors for 8 times 8256-QAMMIMO systems
highest computational complexity while the proposed SDRis in between
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
8 International Journal of Antennas and Propagation
References
[1] J Jalden and B Ottersten ldquoAn exponential lower bound on theexpected complexity of sphere decodingrdquo in Proceedings of theIEEE International Conference on Acoustics Speech and SignalProcessing (ICASSP rsquo04) vol 4 pp iv-393ndashiv-396 MontrealCanada May 2004
[2] A Ben-Tal and A Nemirovski Lectures on Modern ConvexOptimization MPS-SIAM Series on Optimization 2001
[3] J F Sturm ldquoUsing SEDUMI 102 A Matlab toolbox for opti-mization over symmetric conesrdquo Optimization Methods andSoftware vol 11-12 pp 1ndash30 1999
[4] B Steingrimsson Z Q Luo and K M Wong ldquoSoft quasi-maximum-likelihood detection for multiple-antenna wirelesschannelsrdquo IEEE Transactions on Signal Processing vol 51 no11 pp 2710ndash2719 2003
[5] Z Y Shao SWCheung andT I Yuk ldquoLattice-reduction-aidedsemidefinite relaxation detection algorithms for multiple-inputmultiple-output systemsrdquo IETCommunications vol 8 no 4 pp448ndash454 2014
[6] W K Ma T N Davidson K M Wong Z Q Luo andP C Ching ldquoQuasi-maximum-likelihood multiuser detectionusing semi-definite relaxation with application to synchronousCDMArdquo IEEE Transactions on Signal Processing vol 50 no 4pp 912ndash922 2002
[7] A Wiesel Y C Eldar and S Shamai ldquoSemidefinite relaxationfor detection of 16-QAM signaling in MIMO channelsrdquo IEEESignal Processing Letters vol 12 no 9 pp 653ndash656 2005
[8] P H Tan and L K Rasmussen ldquoThe application of semidefiniteprogramming for detection inCDMArdquo IEEE Journal on SelectedAreas in Communications vol 19 no 8 pp 1442ndash1449 2001
[9] N D Sidiropoulos and Z-Q Luo ldquoA semidefinite relaxationapproach to MIMO detection for high-order QAM constella-tionsrdquo IEEE Signal Processing Letters vol 13 no 9 pp 525ndash5282006
[10] Z Mao X Wang and X Wang ldquoSemidefinite programmingrelaxation approach for multiuser detection of QAM signalsrdquoIEEE Transactions on Wireless Communications vol 6 no 12pp 4275ndash4279 2007
[11] W-K Ma C-C Su J Jalden T-H Chang and C-Y ChildquoThe equivalence of semidefinite relaxation MIMO detectorsfor higher-order QAMrdquo IEEE Journal on Selected Topics inSignal Processing vol 3 no 6 pp 1038ndash1052 2009
[12] Z Y Shao SWCheung andT I Yuk ldquoSemi-definite relaxationdecoder for 256-QAM MIMO systemrdquo Electronics Letters vol46 no 11 pp 796ndash797 2010
[13] Y Yang C Zhao P Zhou and W Xu ldquoMIMO detection of 16-QAM signaling based on semidefinite relaxationrdquo IEEE SignalProcessing Letters vol 14 no 11 pp 797ndash800 2007
[14] C Helmberg F Rendl R J Vanderbei and H Wolkowicz ldquoAninterior-point method for semidefinite programmingrdquo SIAMJournal on Optimization vol 6 no 2 pp 342ndash361 1996
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
8 International Journal of Antennas and Propagation
References
[1] J Jalden and B Ottersten ldquoAn exponential lower bound on theexpected complexity of sphere decodingrdquo in Proceedings of theIEEE International Conference on Acoustics Speech and SignalProcessing (ICASSP rsquo04) vol 4 pp iv-393ndashiv-396 MontrealCanada May 2004
[2] A Ben-Tal and A Nemirovski Lectures on Modern ConvexOptimization MPS-SIAM Series on Optimization 2001
[3] J F Sturm ldquoUsing SEDUMI 102 A Matlab toolbox for opti-mization over symmetric conesrdquo Optimization Methods andSoftware vol 11-12 pp 1ndash30 1999
[4] B Steingrimsson Z Q Luo and K M Wong ldquoSoft quasi-maximum-likelihood detection for multiple-antenna wirelesschannelsrdquo IEEE Transactions on Signal Processing vol 51 no11 pp 2710ndash2719 2003
[5] Z Y Shao SWCheung andT I Yuk ldquoLattice-reduction-aidedsemidefinite relaxation detection algorithms for multiple-inputmultiple-output systemsrdquo IETCommunications vol 8 no 4 pp448ndash454 2014
[6] W K Ma T N Davidson K M Wong Z Q Luo andP C Ching ldquoQuasi-maximum-likelihood multiuser detectionusing semi-definite relaxation with application to synchronousCDMArdquo IEEE Transactions on Signal Processing vol 50 no 4pp 912ndash922 2002
[7] A Wiesel Y C Eldar and S Shamai ldquoSemidefinite relaxationfor detection of 16-QAM signaling in MIMO channelsrdquo IEEESignal Processing Letters vol 12 no 9 pp 653ndash656 2005
[8] P H Tan and L K Rasmussen ldquoThe application of semidefiniteprogramming for detection inCDMArdquo IEEE Journal on SelectedAreas in Communications vol 19 no 8 pp 1442ndash1449 2001
[9] N D Sidiropoulos and Z-Q Luo ldquoA semidefinite relaxationapproach to MIMO detection for high-order QAM constella-tionsrdquo IEEE Signal Processing Letters vol 13 no 9 pp 525ndash5282006
[10] Z Mao X Wang and X Wang ldquoSemidefinite programmingrelaxation approach for multiuser detection of QAM signalsrdquoIEEE Transactions on Wireless Communications vol 6 no 12pp 4275ndash4279 2007
[11] W-K Ma C-C Su J Jalden T-H Chang and C-Y ChildquoThe equivalence of semidefinite relaxation MIMO detectorsfor higher-order QAMrdquo IEEE Journal on Selected Topics inSignal Processing vol 3 no 6 pp 1038ndash1052 2009
[12] Z Y Shao SWCheung andT I Yuk ldquoSemi-definite relaxationdecoder for 256-QAM MIMO systemrdquo Electronics Letters vol46 no 11 pp 796ndash797 2010
[13] Y Yang C Zhao P Zhou and W Xu ldquoMIMO detection of 16-QAM signaling based on semidefinite relaxationrdquo IEEE SignalProcessing Letters vol 14 no 11 pp 797ndash800 2007
[14] C Helmberg F Rendl R J Vanderbei and H Wolkowicz ldquoAninterior-point method for semidefinite programmingrdquo SIAMJournal on Optimization vol 6 no 2 pp 342ndash361 1996
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal ofEngineeringVolume 2014
Submit your manuscripts athttpwwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of