Académique Documents
Professionnel Documents
Culture Documents
DOI 10.1007/s10586-017-1307-4
123
Cluster Comput
(maximum a posteriori) Block QRD method using quadra- tion and with the variants of QRD algorithms was discussed
ture phase shift keying (QPSK). in [18]. Higher value of K (= 80) was considered by sacri-
The trade-off between performance and complexity still ficing the complexity issues to show a better performance in
exist in the present day networks due to the huge network HetNets. Within the extensive literature on QR decomposi-
deployments. Complexity mainly arises in the channel. Lat- tion, comparatively little research has been done on RBQRD
tice reduction is a method applied to get the reduced basis and multiple QRD algorithms in the MIMO scenario. So uti-
for the channel. It was originally introduced by Lenstra lizing the proposed DQRD algorithm, which uses two stages
Lenstra Lovasz (LLL) in 1982. It is a method for fac- of QRD algorithms, a good performance improvement was
torizing polynomials with rational coefficients [11]. LLL shown in the two MIMO scenario compared to the previous
algorithm is well explained in [12]. Complexity and diag- and existing works. The major contribution of this research
onal variants of LLL algorithm was also proposed and the work is as follows:
performance improvement was made [12,13]. Various other
• We present a novel DQRD algorithm and its use in the
reduction techniques such as Brun’s algorithm, Seysen algo-
UL-LRA–KSD for a MIMO scenario.
rithm, Hermite-Korkine-Zolotareff (HKZ) and Minkowski
• As an extension of the work done in [19], we considered
reduction algorithms are also followed for lattice reduction
complexity analysis of KSD using RBQRD/GQRD for
[14]. Seysen is a lattice reduction algorithm which performs
HetNets in this paper.
well compared to Brun’s algorithm but at the expense of
• Run-time complexity of UL-LRA–KSD in MIMO using
complexity [14].
the novel DQRD was analyzed and compared with KSD.
Lattice reduction aided–K-best sphere detection is a
method applied with lattice reduction to KSD algorithm.
After thoroughly surveying the literature it is observed that The remainder of the paper is organized as follows. In Sect. 2
KSD with lattice reduction has not been addressed for MIMO the system model for MIMO in flat faded and indoor chan-
HetNet scenario and hence in this research work, we eval- nel is introduced. Section 3 gives a review of several QRD
uated the performance of UL-LRA–KSD in MIMO under algorithms with the proposed DQRD algorithm for the KSD.
two different scenarios namely MIMO-general and MIMO The UL-LRA–KSD for MIMO system with emphasis on
HetNet using our novel DQRD algorithm. To have a low com- DQRD and LLL algorithm for lattice reduction is explained
plexity the original LLL algorithm is used in this research in Sect. 4. Simulation results are given in Sect. 5. Conclusion
work for lattice reduction which was used intensively in the is outlined in Sect. 6.
previous literatures. The proposed DQRD algorithm is novel
in the sense that it alternates to MIMO-general and MIMO 2 System model
HetNet scenario using the switching logic (QR) present in
the technique.It is assumed that the switching logic operates We are very much aware of the MIMO model that has been
according to the strength of the signal detected in the two used almost in every MIMO literature. Considering the same
MIMO scenarios. the system model of IoT and HetNet is shown in Fig. 1a and
In MIMO the channel is usually faced with the least b respectively. We have used the combination of the exist-
squares problem before Lattice reduction. Normal equations, ing MIMO-IoT and MIMO-IoT HetNet model for evaluating
QRD and singular value decomposition (SVD) are the meth- the performance analysis of UL-LRA–KSD in a IoT environ-
ods adopted to overcome least squares problem [15]. QRD ment. The scenario shown here is 4 × 4 system, which can
methods have been followed to avoid least squares problem be decomposed in to two 2 × 2 systems using RBQRD for
in a full-rank matrix. Most of the techniques used in the effective MIMO detection. A HetNet, consists of a Macro
literature are based on Gram–Schmidt (GS), modified GS cell and the other sub cells, such as the microcell, picocell,
(MGS), householder QRD (HQRD), Givens rotation-QRD femtocells, etc. In our system, the mobile terminal (MT) in
(GQRD) and recursive block QRD (RBQRD). Utilizing mul- one of the sub cell–picocell is considered for modeling the
tiple QRD algorithms will increase the performance in terms HetNet.eNodeBs (eNBs) each with two antenna in both the
of BER [16]. So QRD is introduced before as well as inside Macrocell (eNB1) and the picocell (eNB2) forms the trans-
the lattice reduction steps in our novel DQRD algorithm. mitter and the MT with four antenna forms the receiver. User
Detection based on QRD algorithm was a drawback in the information from the eNB1 and eNB2 are considered as s1
MIMO–NOMA design for IoT networks [17]. So the prob- and s2 respectively. The desired stream is s2 and the interfer-
lem was taken in to consideration for variants of QRD and the ing stream is s1 . Conversely if the MT is in Macro cell, the
results are compared in the IoT model. The algorithm works interfering stream is s2 and the desired stream is s1 . In the
with GQRD in MIMO-general and RBQRD in MIMO Het- subsequent sections, the system model for HetNet is consid-
Net before and GQRD inside the LLL algorithm used for ered under two separate channel scenarios via flat fading and
lattice reduction. Performance of KSD without lattice reduc- indoor fading.
123
Cluster Comput
where H11 , H12 are the channel matrices due to the desired
signal s2 in the picocell and H11 , H12 are the channel matrix
due to the interfering signal s1 in the macro cell. The objective
of the MIMO detection algorithm is to compute an estimate
such that
y = Hk s + n (4)
123
Cluster Comput
Table 1 Tgac channel D Table 2 Improvement and drawbacks of various QRD algorithms
Parameters D Algorithm Improvement and Complexity
drawbacks
Avg. 1st wall dist. (m) 10
RMS delay spread (ns) 50 MGS [21] Performs well but not 2mn 2
numerically stable
Maximum delay (ns) 390
HQRD [21] Requires fewer n 3 /3
Number of taps 18
operations than GS
Number of clusters 3 and more stable than
K (dB) LOS/NLOS 3/-∞ GS
GQRD [22] Numerical properties are 3mn 2 − n 3
closer to Householder
but less stable
where the elements r, t and s represent receive, transmit and
RBQRD [23] Numerically stable for 2nlog N n + 2n 3 + 19
cross correlation respectively. The estimate ŝ is now com- k
large block size and
puted using the kronecker model of the Eq. (3). So using more number of
Equation (2) the estimate ŝ is given as streams especially in
dense networks
ŝ = arg min y − H k s2 (11)
s∈C
Using the kronecker channel models, the indoor channel compare the performance of UL-LRA–KSD MIMO detec-
models for IEEE 802.11n/ac were obtained under the TGac tion both in flat faded and indoor channel model (channel
amendment [20]. Channel D is defined [21] for different envi- D).
ronment using Rician fading and are shown in Table 1.
3.1 QRD algorithms
123
Cluster Comput
3.2.1 Givens rotation QRD (GQRD) algorithm From the Eq. (15), Q1 R11 = H 1 , R12 = QT1 H 2 , Q2 R22 =
H 2 − Q1 R12 the decomposed channels [H 1 |H 2 ] for the
The GQRD which is used before and inside the LLL algo- RBQRD are obtained.
rithm is a widely used numerical linear algebra algorithm. It is
often used by the wireless communication platformto achieve H 1 = Q1 R11 (16)
the system throughput and stability. In embedded systems, it
H 2 = Q2 R22 + Q1 R12 (17)
is often preferred over the HQRD, due to its high degree of
123
Cluster Comput
4.1 LR algorithm
Consider a lattice L as
L {s : si ∈ A, 1 ≤ i ≤ 2N T } (18)
4.2 Sphere decoding (SD) algorithm
and
An ML solution is to obtain the closest estimate ŝ from the
Z {z : z = Hs, si ∈ A, 1 ≤ i ≤ 2N T } (19) received vector y which is obtained from Eq. (3). SD algo-
rithm can be applied to reduce the complexity by setting the
In Eqs. (18) and (19) L is the transmit lattice, Z is the radius of the sphere r to narrow down the search space [28].
received lattice with noise and N T is the number of transmit It is an algorithm which is transformed from closest-point
antennas. The channel matrix can be reduced in LRA MIMO search to tree-search using QRD. The new-QRD obtained
detectors with the well known algorithms [24,25]. Various from LLL algorithm of lattice reduction is applied to the SD
reduction techniques and LLL algorithms such as original search space. The modified estimate and the input–output
LLL, complex LLL and MLLL had been used for LRA relation after the lattice reduction is given as
MIMO detection. The lattice reduction in MIMO uses basi-
cally a original LLL algorithm for the reduced complexity.
ŝ = arg min T y − H̄x 2 (23)
The original LLL algorithm [26,27] for the UL-LRA–KSD s∈C
uses the GQRD. The pseudocode for the original LLL algo- H
ŷ = R̄s + T Q̄ n (24)
rithm based on the novel DQRD is given in Algorithm 2.
Using the pseudocode we get the reduced channel matrix H
H̄ . It is denoted as H̄ = HT, where T is the uni-modular where ŷ = T Q̄1 |Q̄2 n has been modified with the uni-
matrix. The system model of Eq. (1) can be written for a modular matrix T and has the same statistics as n. Q̄and R̄
general MIMO using GQRD, are the new QRD obtained from the LLL algorithm. Utiliz-
ing the triangle nature of HetNet R matrix Eq. (19) can be
y = H̄x + n (20) rewritten as,
123
Cluster Comput
2
NT
NT LLL algorithm is used for channel splitting into two cells
(i)
ŝ = ŷi − R̄ s (25) (macro and sub-macro). LLL algorithm reduces the basis
i, j j
i=1 j=i from RBQRD output and gives a new-QRD to the KSD. The
KSD principle is explained in Algorithm 3. Higher value of K
Equation (22) can be thought of as a detection problem in a gives better BER performance where as the lower value of K
tree that has a root above the N T layer and leaves on the first reduces the complexity. The last iteration of the K-best algo-
layer. The survived candidate s(i) called as partialsymbol rithm produces the best hard decision output of the decoder.
(i) (i) (i)
vector (PSV) is given by s (i) = si , si+1 , . . . , s NT is used The main advantage of the KSD over the other types is its
to compute the partial Euclidean distance (PED)di = d s (i) fixed throughput.
which is given as,
123
Cluster Comput
Fig. 4 Performance of UL-LRA–KSD in MIMO-IoT for a 16-QAM Fig. 6 Run-time complexity of UL-LRA–KSD in MIMO-IoT for a
(4 × 4) in flat faded channel (4 × 4) in flat faded channel
123
Cluster Comput
Flat faded 45 20 7 17 23 15
Channel D 39 17 4 16 18 9
mal. As a future work, we would like to extend this work 16. Ding, Z., Dai, L., Poor, V.H.: MIMO-NOMA design for small
to analyze the coded BER and run time complexity for the packet transmission in the Internet of Things. In: IEEE Access:
Internet of Things (IoT) in 5G Wireless Communications, vol. 4,
proposed LRA–KSD scheme in HetNets. pp. 1393–1405 (2016)
17. Arar, M., Yongacoglu, A.: Parallel low-complexity MIMO detec-
tion algorithm using QR decomposition and Alamouti space-time
code. In: Proceedings of European Wireless Conference, pp. 141–
References 148 (2010)
18. Syed Moinuddin Bokhari, B., Bhagyaveni, M.A.: Performance
1. Wang, J., Daneshrad, B.: A comparative study of MIMO detec- evaluation of K-best sphere detector using recursive Block QR
tion algorithms for wideband spatial multiplexing systems. In: decomposition. In: Proceedings of International Conference on
Proceedings of IEEE Wireless Communication and Networking Next Generation Computing and Communication Technologies
conference, 2005, vol. 1, pp. 408–413 (2005) (ICNGCCT-15), Dubai, pp. 134–138 (2015)
2. Waters, D.W.: Signal detection strategies and algorithms for 19. Oestges, C.: Validity of the Kronecker model for MIMO correlated
multiple-input multiple-output channels. Doctoral of philosophy channels. In: Proceedings of IEEE Vehicular Technology Confer-
thesis, Georgia Institute of Technology (2005) ence, pp. 2818–2822, Melbourne, VIC (2006)
3. Kim, T.K., Kim, H.M., Im, G.H.: Enhanced QRD-M algorithm 20. Wireless LAN Medium Access Control (MAC) and Physical Layer
for soft-output MIMO detection. In: Proceedings of Signal Pro- (PHY) Specifications, Amendment 4: Enhancement for Very High
cessing for Communications Symposium-Global Communications Throughput for Operations in Bands below 6 GHz. IEEE P802.11ac
Conference (GLOBECOM), pp. 3572–3576. IEEE, Anaehim, CA /D3.0 (2012)
(2012) 21. Chockalingam, A., Sundarrajan, B.: Large Mimo Systems, 1st edn.
4. Zhu, X., Murch, R.D.: Performance analysis of maximum likeli- Cambridge University Press, New York (2014)
hood detection in a MIMO antenna system. IEEE Trans. Commun. 22. Mahmoud, H.A., Arslan, H.: A low-complexity high-speed QR
50(2), 187–191 (2002) decomposition implementation for MIMO receivers. In: Proceed-
5. Hasibbi, B., Vikalo, H.: On the sphere-decoding algorithm I. ings of International symposium on Circuits and Systems, pp.
Expected complexity. IEEE Trans. Commun. 53(8), 2806–2818 33–36. IEEE, Taipei (2009)
(2005) 23. Timothy, S.: Numerical Analysis. Pearson Education Inc., George
6. Huang, C.J., Sung, C.S., Lee, T.S.: A near-ML complex K-best Mason University, Upper Saddle River (2006)
decoder with efficient search design for MIMO systems. EURASIP 24. Hammarling, S., Lucas, C.: Updating the QR factorization and
J. Adv. Signal Process. 2010, 1–18 (2010) the least squares problem, Manchester Institute for Mathematical
7. Piao, C., et al.: Dynamic K-best sphere decoding algorithms for Sciences, School of Mathematics (2008)
MIMO detection. Sci. Res. Publ. 5(38), 103–107 (2013) 25. Singhal, K.A., Datta, T., Chockalingam, A.: Lattice reduction aided
8. Lou, X., et al.: Research on low complexity K-best sphere decoding detection in large-MIMO systems. In: Proceedings of IEEE Sig-
algorithm for MIMO systems. Wirel. Pers. Commun. 84, 53–56 nal Processing Advances in Wireless Communication (SPAWC)
(2015) Darmstadt, pp. 594–598 (2013)
9. Guo, X., Yuan, S.: A new tree pruning SD algorithm to eliminate 26. Seethaler, D., Matz, G., Hlawatsch, F.: Low-complexity MIMO
interference. Int. J. Digit. Content Technol. Appl. JDCTA 7(7), data detection using Seysen’s lattice reduction algorithm. In: Pro-
476–482 (2013) ceedings of IEEE International Conference on Acoustics, Speech
10. Thomas, R., Knopp, R., Maharaj, B.T., Cottatellucci, L.: Detection and Signal Processing (ICASSP 2007), pp. 53-56. Honolulu, HI
using Block QR decomposition for MIMO HetNets. In: Proceed- (2007)
ings of Asilomar Conference on Signals, Systems and Computers, 27. Li, Q., Wang, Z.: Reduced complexity K-best sphere decoder
Pacific Groove, CA (2014) design for MIMO systems. Circuits Syst. Signal Process 27, 491–
11. Lenstra, A.K., Lenstra Jr., H.W., Lováz, L.: Factoring polynomials 505 (2008)
with rational coefficients. Math. Ann. 261, 515–534 (1982) 28. Guo, Z., Nilsson, P.: Algorithm and implementation of the K-best
12. Ling, C., Mow, W.H., Howgrave-Graham, N.: Reduced and fixed- sphere decoding for MIMO detection. In: IEEE Journal on Selected
complexity variants of the LLL algorithm for communications. Areas in Communications, pp. 491–503 (2006)
IEEE Trans. Commun. 61(3), 1040–1050 (2013) 29. Syed Moinuddin Bokhari, B., Bhagyaveni, M.A.: Low complex
13. Zhang, W., Qiao, S., Wei, Y.: A diagonal lattice reduction algorithm lattice reduction aided- zero forcing MIMO detection in Hetero-
for MIMO detection. IEEE Signal Process. Lett. 19(5), 311–314 geneous networks. In: Proceedings of International Journal of
(2012) Engineering Research & Technology (IJERT), pp. 510–514 (2015)
14. Wübben, D., Seethaler, D., Jaldén, J., Matz, G.: Lattice reduction
- a survey with applications in wireless communications. In: IEEE
Signal processing Magazine, pp. 70–91 (2011)
15. Demmel, J.W.: Applied Numerical Algebra, 2nd edn. SIAM Pub-
lications, Prentice-Hall, Englewood Cliffs, NJ (1997)
123
Cluster Comput
M. A. Bhagyaveni received
B. Syed Moinuddin Bokhari her B.E. degree in Electronics
completed his B.E. from Gov- and Communication Engineer-
ernment College of Engineer- ing from GCT, Coimbatore, India
ing, Bargur, INDIA affiliated in 1997, M.E. degree in Opti-
to University of Madras in cal Communication from CEG,
2001, received his M.E. from Gunidy, India in 1999 and Ph.D.
Jayaram College of Engineer- degree from CEG, Guindy, India
ing and Technology, Thuri- in 2006. She is currently working
ayur, INDIA affiliated to Anna as Professor in the department
Univeristy, Chennai INDIA in of Electronics and Communica-
2005.Currently he is a research tion Engineering, CEG Campus,
scholar at the department of Anna University, Chennai, India.
Electronics and Communication Her present research interests
Engineering, College of Engi- include wireless communication,
neering Guindy, Anna Univer- digital communication, MIMO systems, Ad hoc networks, sensor net-
sity,Chennai, INDIA. His field of interests include wireless commu- works, cloud computing, cognitive radio technologies, radio resource
nication and networks, signal processing. His current research interests allocation in LTE and next generation networks. She has published about
are multiuser detection, multiuser scheduling applied to FPGA based 40 papers in Journals, National and International Conferences. She is a
SISO/MIMO test beds. He is a life member of ISTE. member of IEEE and several International association bodies.
123