Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2013, Article ID 601623, 7 pageshttp://dx.doi.org/10.1155/2013/601623
Research ArticleSubband Adaptive Filtering with π1-Norm Constraint forSparse System Identification
Young-Seok Choi
Department of Electronic Engineering, Gangneung-Wonju National University, Gangneung 210-702, Republic of Korea
Correspondence should be addressed to Young-Seok Choi; [email protected]
Received 27 September 2013; Revised 26 November 2013; Accepted 26 November 2013
Academic Editor: Yue Wu
Copyright Β© 2013 Young-Seok Choi. This is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This paper presents a new approach of the normalized subband adaptive filter (NSAF) which directly exploits the sparsity conditionof an underlying system for sparse system identification.The proposed NSAF integrates a weighted π
1-norm constraint into the cost
function of the NSAF algorithm. To get the optimum solution of the weighted π1-norm regularized cost function, a subgradient
calculus is employed, resulting in a stochastic gradient based update recursion of the weighted π1-norm regularized NSAF. The
choice of distinct weighted π1-norm regularization leads to two versions of the π
1-norm regularized NSAF. Numerical results clearly
indicate the superior convergence of the π1-norm regularized NSAFs over the classical NSAF especially when identifying a sparse
system.
1. Introduction
Over the past few decades, the relative simplicity and goodperformance of the normalized least mean square (NLMS)algorithm have made it a popular tool for adaptive filter-ing applications. However, its convergence performance issignificantly deteriorated in case of correlated input signals[1, 2]. As a popular solution, adaptive filtering in the subbandhas been recently developed, which is referred to as subbandadaptive filter (SAF) [3β7]. Its distinct feature is based on theproperty that the LMS-type adaptive filters converge faster forwhite input signals than colored ones [1, 2]. Thus, carryingout a prewhitening on colored input signals, it results in theaccelerated convergence compared to the LMS-type adaptivefilters. Recently, the use of multiple-constraint optimizationcriteria into formulation of a cost function has resulted in thenormalized SAF (NSAF) with its computational complexityclose to that of the NLMS algorithm [6, 7].
In the context of a system identification, the unknownsystem to be identified is sparse in common scenarios, suchas echo paths [8] and digital TV transmission channels [9].Namely, the unknown system consists of many near-zerocoefficients and a small number of large ones. However, theadaptive filtering algorithms suffer from poor convergence
performance in case of identifying the sparse system [8].Indeed, the capability of the NSAF is faded in a sparse systemidentification scenario. To deal with this issue, a variety ofproportionate adaptive algorithms have been presented forNSAF, which utilize proportionate step sizes to distinct filtertaps [10β12]. However, these algorithms have not exploitedthe sparsity condition of an underlying system.
Recently, motivated by compressive sensing framework[13, 14] and the least absolute shrinkage and selection oper-ator (LASSO) [15], a number of adaptive filtering algorithmswhich make use of the sparsity condition of an underlyingsystem have been developed [16β20]. The core idea behindthis approach is to incorporate the sparsity condition ofunderlying system by imposing an a sparsity-inducing con-straint term. Adding the sparsity constraint using π
0or π1-
norm constraint to the cost function makes the least relevantweights of the filter shrink to zeros. However, to the best ofthe authorβs knowledge, adaptive filtering in subband whichexploits the sparsity condition has not been studied yet.
With regard, this paper presents a novel approach of thesparsity-regularized NSAFs, which incorporates the sparsitycondition of the system directly into the cost function viaa sparsity-inducing constraint term. This is carried out byregularizing a weighted π
1-norm of the filter weights estimate
2 Mathematical Problems in Engineering
u(n)
wβ
H0(z)
H1(z)
HNβ1(z)
H0(z)
βN
βN
βN
βN
βN
βN
d0(n) d0(k)οΏ½(n)
d(n)d1(n) d1(k)
H1(z)
......
HNβ1(z)dNβ1(n) dNβ1(k)
u0(n) y0(n) y0(k)
u1(n) y1(n) y1(k)
uNβ1(n) yNβ1(n yNβ1(k))w(k)
w(k)
w(k)...
......
G0(z)e0(n) e0(k)
βN
G1(z)e1(n) e1(k)
βN
GNβ1(z)eNβ1(n) eNβ1(k)
βN
e(n)...
...
+
+
+
+
+
β
β
β
Figure 1: Subband structure with the analysis filters and synthesis filters and the subband desired signals, subband filter outputs, and subbanderror signals.
to the cost function. Considering the two choices of theweighted π
1-norm regularization, two stochastic gradient-
based π1-norm regularized NSAF algorithms are derived.
First, the π1-norm NSAF (π
1-NSAF) is obtained by uti-
lizing the identity matrix as a weighting matrix. Second,the reweighted π
1-norm NSAF (π
1-RNSAF) which uses the
current estimate of the system as a weighted π1-norm is
developed. Through numerical simulations, the resultantsparsity-regularized NSAFs have proven their superiorityover the classical NSAFs, especially when the sparsity of theunderlying system becomes severe.
The remainder of the paper is organized as follows.Section 2 introduces the classical NSAF, followed by thederivation of the proposed sparsity-regularized NSAFs inSection 3. Section 4 illustrates the computer simulationresults and Section 5 concludes this study.
2. Conventional NSAF
Consider a desired signal π(π) that arises from the systemidentification model
π (π) = u (π)wβ + V (π) , (1)
where wβ is a column vector for the impulse response of anunknown system that we wish to estimate, V(π) accounts formeasurement noise with zeromean and varianceπ2V , and u(π)denotes the 1 Γπ input vector,
u (π) = [π’ (π) π’ (π β 1) β β β π’ (π βπ + 1)] . (2)
Figure 1 shows the structure of the NSAF, where thedesired signal π(π) and input signal π’(π) are partitioned into
π subbands by the analysis filtersπ»0(π§),π»
1(π§), . . . , π»
πβ1(π§).
The resulting subband signals, ππ(π) and π¦
π(π) for π =
0, 1, . . . , π β 1, are critically decimated to a lower samplingrate commensurate with their bandwidth. Here, the variableπ to index the original sequences and π to index the decimatedsequences are used for all signals. Then, the decimatedfilter output signal at each subband is defined as π¦
π,π·(π) =
uπ(π)w(π), where u
π(π) is 1 Γπ row vector, such that
uπ(π) = [π’
π(ππ) , π’
π(ππ β 1) , . . . , π’
π(ππ βπ + 1)] , (3)
andw(π) = [π€0(π), π€
1(π), . . . , π€
πβ1(π)]π denotes an estimate
for wβ with length π. Thus the decimated subband errorsignal is given by
ππ,π·
(π) = ππ,π·
(π) β π¦π,π·
(π) = ππ,π·
(π) β uπ(π)w (π) , (4)
where ππ,π·(π) = π
π(ππ) is the decimated desired signal at
each subband.In [6], the authors have formulated the Lagrangian-
based multiple-constraint optimization problem, which isformulated as
π½NSAF (π) = βw (π + 1) β w (π)β2
+
πβ1
βπ=0
ππ[ππ,π·
(π) β uπ(π)w (π + 1)] ,
(5)
where ππfor π = 0, 1, . . . , π β 1 denote the Lagrange multipli-
ers. Solving the cost function (5), the update recursion of theNSAF algorithm is given by [6, 7]. Consider
w (π + 1) = w (π) + ππβ1
β
π=0
uππ(π)
uπ (π)2ππ,π·
(π) , (6)
where π is the step-size parameter.
Mathematical Problems in Engineering 3
3. Weighted π1-Norm Regularized NSAF
3.1. Derivation of the Proposed Algorithm. To reflect thesparsity condition of the true system, that is, wβ, a weightedπ1-norm of the filter weight estimate is regularized on the cost
function of the NSAF, which is given by
π½π1βNSAF (π) = βw (π + 1) β w (π)β
2
+
πβ1
β
π=0
ππ[ππ,π·
(π) β uπ(π)w (π + 1)]
+πΎβΞ w (π + 1)β1,
(7)
where βΞ w(π + 1)β accounts for the weighted π1-norm of the
filter weight vector w(π + 1) and is written as
βΞ w (π + 1)β1 =πβ1
β
π=0
ππ
π€π (π + 1) , (8)
where Ξ is a π Γ π weighting matrix whose diagonalelements are π
πand other elements are equal to zero, and
π€π(π + 1) denotes the πth tap weight of w(π + 1), for π =
0, 1, . . . ,π β 1. In addition, πΎ is a positive value parameterwhich plays a role in compromising the error related term andthe weighted π
1-norm regularization in the right-hand side of
(7).To find the optimal weight vector w(π + 1) which
minimizes the cost function (7), the derivative of (7) withrespect to w(π + 1) is taken and set to zero. Note that theweighted π
1-norm regularization term, that is, βΞ w(π + 1)β
1,
is not differentiable at any point in case π€π(π + 1) = 0. To
address this issue, a subgradient calculus [21] is carried out.Thus, taking the derivative of (7) with respect to the
weight vector w(π + 1) and letting the derivative be equal tozero, it leads to
w (π + 1) = w (π) + 12
πβ1
β
π=0
ππuππ(π) β
πΎ
2βπ
wβΞ w (π + 1)β1,
(9)
where βπ wπ(β ) denotes a subgradient vector of a function π(β )with respect to w(π + 1). An available subgradient vectorβπ
wβΞ w(π + 1)β1 is obtained as [21]. Consider
βπ
wβΞ w (π + 1)β1 = Ξ π sgn (Ξ w (π + 1))
= Ξ sgn (w (π + 1)) ,(10)
sinceΞ is assumed as a diagonal matrix with positive-valuedelements, where sgn(β ) is a componentwise sign functiondefined by
sgn (π₯) ={
{
{
π₯
|π₯|, π₯ ΜΈ= 0
0, elsewhere.(11)
Substituting (10) into (9) and assuming sgn[w(π + 1)] βsgn[w(π)], it is given by
w (π + 1) = w (π) + 12
πβ1
β
π=0
ππuππ(π) β
πΎ
2Ξ sgn (w (π)) . (12)
Substituting (12) into the multiple constraints of theNSAF, that is, π
π,π·(π) = u
π(π)w(π + 1), π = 0, 1, . . . , π β 1
and rewriting as a matrix form, it leads to
Ξ = 2[U (π)Uπ (π)]β1
eπ·(π)
+πΎ[U (π)Uπ (π)]β1
U (π)Ξ sgn (w (π)) ,(13)
where Ξ = [π0, π1, . . . , π
πβ1]π is theπ Γ 1 Lagrange vector,
U (π) = [[[
u0(π)
...uπβ1
(π)
]]
]
, eπ·(π) =
[[
[
π0,π·
(π)
...ππβ1,π·
(π)
]]
]
. (14)
By neglecting the off-diagonal elements ofU(π)Uπ(π) [6],the components of Ξ in (13) can be simplified to
ππ= 2
ππ,π·
(π)
uπ (π)2+ πΎ
uπ(π)
uπ (π)2Ξ sgn (w (π)) , (15)
for π = 0, 1, . . . , π β 1.Consequently, combining (12) and (15), the update recur-
sion of the sparsity-regularized NSAF is given by
w (π + 1) = w (π)
+ π
πβ1
β
π=0
[uππ(π)
uπ (π)2ππ,π·
(π)
+1
2πΎ
uπ(π)
uπ (π)2Ξ sgn (w (π)) uπ
π(π)]
βππΎ
2Ξ sgn (w (π)) ,
(16)
where π is the step-size parameter.
3.2. Determination of the Weighted π1-Norm Regularization.
Here, by choosing the weighting matrix Ξ , two versions ofthe sparsity-regularized NSAF are developed. First, the use ofthe identity matrix as the weighting matrix, that is, Ξ = I
π,
results in the following update recursion:
w (π + 1) = w (π)
+ π
πβ1
β
π=0
[uππ(π)
uπ (π)2ππ,π·
(π)
+1
2πΎ
uπ(π)
uπ (π)2sgn (w (π)) uπ
π(π)]
βππΎ
2sgn (w (π)) ,
(17)
which is referred to as the π1-norm NSAF (π
1-NSAF) as
an unweighted case. The π1-NSAF uniformly attracts the
4 Mathematical Problems in Engineering
Table 1: Computational complexity.
NSAF π1-NSAF π
1-RNSAF
Multiplications 3π + 3ππΏ 6π + 3ππΏ 7π + 3ππΏDivisions 1 2 2 +π/π(π: filter length,π: number of subbands, and πΏ: length of the analysis filtersand synthesis filters).
tap coefficients of w(π) to zero. The zero attraction processleads to the improved convergence of the π
1-NSAF when the
majority of entries of a system are zero; that is, a system issparse.
Second, to approximate the actual sparsity condition of anunderlying system, that is, π
0-norm of the system, the weights
of Ξ are chosen inversely proportional to magnitude of theactual coefficients of the system as given by
ππ={
{
{
1π€π
, π€π
ΜΈ= 0
β, π€π= 0,
(18)
where π€πdenotes the πth coefficients of the system, that
is, wβ. However, since the actual coefficients of the systemare unavailable, the estimate of the current filter weights isutilized instead of the actual weights, which is referred to asthe reweighting scheme [22], as follows:
ππ(π) =
1π€π (π)
+ πfor π = 0, 1, . . . ,π β 1, (19)
where π€π(π) denotes theπth tap weight of the w(π) and π is
a small positive value to avoid singularity when |π€π(π)| = 0.
Then, the weighting matrix Ξ consists of the values of ππ(π)
as theπth diagonal elements and has a time-varying feature.Finally, the update recursion is given by
w (π + 1) = w (π)
+ π
πβ1
β
π=0
[uππ(π)
uπ (π)2ππ,π·
(π)
+1
2πΎ
uπ(π)
uπ (π)2sgn (w (π)) uπ
π(π)]
βππΎ
2
sgn (w (π))|w (π)| + π
,
(20)
where uπ(π) = u
π(π)Ξ and the vector division operation in
last term accounts for a componentwise division. Then, thisrecursion is called the reweighted π
1-normNSAF (π
1-RNSAF)
Table 1 lists the number of multiplications and divisionsof the NSAF [6], π
1-NSAF, and π
1-RNSAF per iteration. As
shown in Table 1, the use of π1-norm constraint leads to an
acceptable increase in computation.
4. Numerical Results
Theperformance of the proposed sparsity-regularizedNSAFsis validated by carrying out computer simulations in a system
identification scenario in which the unknown channel israndomly generated. The lengths of the unknown system areπ = 128 and 512 in experiments where π of them arenonzero. The nonzero filter weights are positioned randomlyand their values are taken from a Gaussian distributionN(0, 1/π). Here, π = 4 is used in the simulations exceptFigure 5 in which various π values are considered. Theadaptive filter and the unknown system are assumed to havethe same number of taps. The input signals are obtainedby filtering a white, zero-mean, Gaussian random sequencethrough a first-order systemπΊ(π§) = 1/(1β0.9π§β1).The signal-to-noise ratio (SNR) is calculated by
SNR = 10 log10(πΈ [π¦2(π)]
πΈ [V2 (π)]) , (21)
where π¦(π) = u(π)wβ. The measurement noise V(π) is addedtoπ¦(π) such that SNR= 10, 20, and 30 dB. In order to comparethe convergence performance, the normalized mean squaredeviation (MSD),
NormalizedMSD = πΈ[wββ w(π)
2
βwββ2] , (22)
is taken and averaged over 50 independent trials.The cosine-modulated filter banks [23] with the subband number ofπ =4 are used in the simulations. The prototype filter of lengthπΏ = 32 is used. For comparison purpose, the proportionateNSAF (PNSAF) [12] is considered, which has been developedfor sparse system identification.The step-size is set to π = 0.5for SAF algorithms except the PNSAFwhere the step sizesπ =0.6 (Figure 2) and π = 0.65 (Figure 6) are chosen to achievesimilar steady-state MSD with the π
1-RNSAF for comparison
purpose. For the π1-RNSAF, π = 0.01 is chosen. In addition,
π = 0.05 is used for the PNSAF. The πΎ values are obtained byrepeated trials to minimize the steady-state MSD.
Figure 2 shows the normalizedMSD curves of the NLMS,NSAF, π
1-NSAF, and π
1-RNSAF, in cases ofπ = 4 and SNR =
30 dB. For the π1-NSAF and π
1-RNSAF, πΎ = 3Γ10β5 is chosen.
As shown in Figure 2, the not only π1-RNSAFoutperforms the
conventional NLMS, NSAF, PNSAF, and π1-NSAF, but also
the π1-NSAF has better performance than other conventional
algorithms, in terms of the convergence rate and the steady-state misalignment.
In Figure 3, to verify the effect of πΎ on convergenceperformance, the normalized MSD curves of the π
1-RNSAF
for different πΎ values are illustrated, in case of π = 4 andSNR = 30 dB. For different πΎ values (πΎ = 1Γ10β4, 1Γ10β5, 5Γ10β5, and 1 Γ 10β6), the π
1-RNSAF is not excessively sensitive
to πΎ.The analysis of an optimal πΎ value remains a future work.Next, the performance of the proposed π
1-norm regular-
izedNSAFs is compared to the original NSAF under differentSNR conditions. Figure 4 depicts the normalizedMSD curvesof the NSAF, π
1-NSAF, and π
1-RNSAF under SNR = 10
and 20 dB, respectively. The πΎ value for the π1-NSAF and π
1-
RNSAF is set to 5 Γ 10β5. It is clear that both the π1-NSAF
and π1-RNSAF are superior to the NSAF under several SNR
cases. Furthermore, the π1-RNSAF performs well compared
to π1-NSAF.
Mathematical Problems in Engineering 5
0 200 400 600 800 1000 1200 1400
β35
β30
β25
β20
β15
β10
β5
0
5
Number of iterations
Nor
mal
ized
mea
n sq
uare
dev
iatio
n (d
B)
NLMSNSAF (N = 4)PNSAF (N = 4)
l1-NSAF (N = 4)l1-RNSAF (N = 4)
Figure 2: Normalized MSD curves of the NLMS, NSAF, PNSAF, π1-
NSAF, and π1-RNSAF (π = 4).
0 200 400 600 800 1000 1200 1400
β35
β30
β25
β20
β15
β10
β5
0
Number of iterations
Nor
mal
ized
mea
n sq
uare
dev
iatio
n (d
B)
πΎ = 1 Γ 10β3
πΎ = 1 Γ 10β4
πΎ = 1 Γ 10β5
πΎ = 5 Γ 10β5
πΎ = 7 Γ 10β5
Figure 3: Normalized MSD curves of the π1-RNSAF for various πΎ
values (π = 4).
In Figure 5, the convergence properties of the NSAF andπ1-RNSAF are compared under various sparsity conditions of
an underlying system. With the same length of the system,that is, π = 128, different sparsity conditions (π = 4, 8, 16,and 32) are considered under SNR = 30 dB. The value of πΎis set to 3 Γ 10β5 for the π
1-RNSAF. Figure 5 shows that the
NSAF is independent of the sparsity condition. On the otherhand, the results indicate that the more sparse the underlyingsystem, the better the π
1-RNSAF.
The comparison of performance of the NSAF, π1-NSAF,
and π1-RNSAF with a long system, here, the filter lengthπ =
0 200 400 600 800 1000 1200 1400β25
β20
β15
β10
β5
0
5
10
Number of iterations
Nor
mal
ized
mea
n sq
uare
dev
iatio
n (d
B)
(a) NSAF (SNR = 10 dB)(b) l1-NSAF (SNR = 10 dB)(c) l1-RNSAF (SNR = 10 dB)
(d) NSAF (SNR = 20 dB)(e) l1-RNSAF (SNR = 20 dB)(f) l1-RNSAF (SNR = 20 dB)
Figure 4: Normalized MSD curves of the NSAF, π1-NSAF, and π
1-
RNSAF under various SNR conditions (SNR = 10, 20, and 30 dB,π = 4).
0 200 400 600 800 1000 1200β35
β30
β25
β20
β15
β10
β5
0
Number of iterations
Nor
mal
ized
mea
n sq
uare
dev
iatio
n (d
B)
NSAF (S = 4)l1-RNSAF (S = 4)NSAF (S = 8)l1-RNSAF (S = 8)
NSAF (S = 16)l1-RNSAF (S = 16)NSAF (S = 32)l1-RNSAF (S = 32)
Figure 5: Normalized MSD curves of the NSAF and π1-RNSAF
under various sparsity conditions (π = 4, 8, 16, and 32,π = 4).
512, is presented in Figure 6. For the π1-NSAF and π
1-RNSAF,
πΎ = 5Γ10β5 is chosen. A similar result of Figure 2 is observed
in Figure 6.Finally, the tracking capabilities of the algorithms of a
sudden change in the system are tested for π = 4 andSNR = 30 dB. Figure 7 shows the results when the unknownsystem is right-shifted for 20 taps. Same value of πΎ in Figure 2is used. As can be seen, the π
1-NSAF and π
1-RNSAF keep track
of weight change without losing the convergence rate nor
6 Mathematical Problems in Engineering
0 1000 2000 3000 4000 5000β40
β35
β30
β25
β20
β15
β10
β5
0
Number of iterations
Nor
mal
ized
mea
n sq
uare
dev
iatio
n (d
B)
NLMSNSAF (N = 4)PNSAF (N = 4)
l1-NSAF (N = 4)l1-RNSAF (N = 4)
Figure 6: Normalized MSD curves of the NLMS, NSAF, PNSAF, π1-
NSAF, and π1-RNSAF for long system ofπ = 512 (π = 4).
0 500 1000 1500 2000 2500 3000β40
β35
β30
β25
β20
β15
β10
β5
0
5
Number of iterations
Nor
mal
ized
mea
n sq
uare
dev
iatio
n (d
B)
NSAFPNSAF
l1-NSAFl1-RNSAF
Figure 7: Normalized MSD curves of the NSAF, PNSAF, π1-NSAF,
and π1-RNSAF in case of a time-varying unknown system (π = 4).
The system is right-shifted for 20 taps at 1500 iterations.
the steady-state misalignment compared to the conventionalNLMS, NSAF, and PNSAF. To be specific, the π
1-RNSAF
achieves better performance than the π1-NSAF in terms of
both convergence rate and steady-state misalignment.
5. Conclusion
A new family of the NSAFs which takes into accountthe sparsity condition of an underlying system has beenpresented by incorporating a weighted π
1-norm constraint of
filter weights in the cost function. The update recursion isobtained by employing subgradient calculus on the weightedπ1-norm constraint term. Subsequently, two sparsity regular-
ized NSAFs, that is, the unweighted π1-NSAF and π
1-RNSAF
have been developed. The numerical results indicate that theproposed π
1-NSAF and π
1-RNSAF achieve highly improved
convergence performance over the conventional algorithmsfor sparse system identification.
Conflict of Interests
The author declares that there is no conflict of interestsregarding the publication of this paper.
Acknowledgment
This research was supported by new faculty research pro-gram funded by Gangneung-Wonju National University(2013100162).
References
[1] S. Haykin, Adaptive Filter Theory, Prentice Hall, Upper SaddleRiver, NJ, USA, 4th edition, 2002.
[2] A. H. Sayed, Fundamentals of Adaptive Filtering, Wiley, NewYork, NY, USA, 2003.
[3] A. Gilloire and M. Vetterli, βAdaptive filtering in subbandswith critical sampling: analysis, experiments, and applicationto acoustic echo cancellation,β IEEE Transactions on SignalProcessing, vol. 40, no. 8, pp. 1862β1875, 1992.
[4] M. De Courville and P. Duhamel, βAdaptive filtering in sub-bands using a weighted criterion,β IEEE Transactions on SignalProcessing, vol. 46, no. 9, pp. 2359β2371, 1998.
[5] S. S. Pradhan and V. U. Reddy, βA new approach to subbandadaptive filtering,β IEEE Transactions on Signal Processing, vol.47, no. 3, pp. 655β664, 1999.
[6] K. A. Lee andW. S. Gan, βImproving convergence of the NLMSalgorithm using constrained subband updates,β IEEE SignalProcessing Letters, vol. 11, no. 9, pp. 736β739, 2004.
[7] K. A. Lee and W. S. Gan, βInherent decorrelating and leastperturbation properties of the normalized subband adaptivefilter,β IEEE Transactions on Signal Processing, vol. 54, no. 11, pp.4475β4480, 2006.
[8] D. L. Duttweiler, βProportionate normalized least-mean-squares adaptation in echo cancelers,β IEEE Transactions onSpeech and Audio Processing, vol. 8, no. 5, pp. 508β518, 2000.
[9] W. F. Schreiber, βAdvanced television systems for terrestrialbroad-casting: some problems and some proposed solutions,βProceedings of IEEE, vol. 83, no. 6, pp. 958β981, 1995.
[10] S. L. Gay, βEfficient, fast converging adaptive filter for networkecho cancellation,β in Proceedings of the 32nd Asilomar Confer-ence on Signals, Systems & Computers, pp. 394β398, November1998.
[11] H. Deng and M. DoroslovacΜki, βImproving convergence of thePNLMS algorithm for sparse impulse response identification,βIEEE Signal Processing Letters, vol. 12, no. 3, pp. 181β184, 2005.
[12] M. S. E. Abadi, βProportionate normalized subband adaptivefilter algorithms for sparse system identification,β Signal Process-ing, vol. 89, no. 7, pp. 1467β1474, 2009.
Mathematical Problems in Engineering 7
[13] D. L. Donoho, βCompressed sensing,β IEEE Transactions onInformation Theory, vol. 52, no. 4, pp. 1289β1306, 2006.
[14] E. J. CandeΜs, J. Romberg, and T. Tao, βRobust uncertaintyprinciples: exact signal reconstruction from highly incompletefrequency information,β IEEE Transactions on InformationThe-ory, vol. 52, no. 2, pp. 489β509, 2006.
[15] R. Tibshirani, βRegression shrinkage and selection via the lasso,βJournal of Royal Statistical Society B, vol. 58, pp. 267β288, 1996.
[16] Y. Chen, Y. Gu, and A. O. Hero, βSparse LMS for systemidentification,β in Proceedings of the International Conference onAcoustics, Speech, and Signal Process (ICASSP β09), pp. 3125β3128, April 2009.
[17] Y. Gu, J. Jin, and S. Mei, βπ1norm constraint LMS algorithm for
sparse system identification,β IEEE Signal Processing Letters, vol.16, no. 9, pp. 774β777, 2009.
[18] J. Jin, Y. Gu, and S. Mei, βA stochastic gradient approach oncompressive sensing signal reconstruction based on adaptivefiltering framework,β IEEE Journal on Selected Topics in SignalProcessing, vol. 4, no. 2, pp. 409β420, 2010.
[19] E. M. Eksioglu and A. K. Tanc, βRLS algorithm with convexregularization,β IEEE Signal Processing Letters, vol. 18, no. 8, pp.470β473, 2011.
[20] N. Kalouptsidis, G. Mileounis, B. Babadi, and V. Tarokh,βAdaptive algorithms for sparse system identification,β SignalProcessing, vol. 91, no. 8, pp. 1910β1919, 2011.
[21] D. P. Bertsekas, Convex Analysis and Optimization, AthenaScientific, Cambridge, Mass, USA, 2003.
[22] E. J. CandeΜs, M. B. Wakin, and S. P. Boyd, βEnhancing sparsityby reweighted π
1minimization,βThe Journal of Fourier Analysis
and Applications, vol. 14, no. 5-6, pp. 877β905, 2008.[23] P. P. Vaidyanathan,Multirate Systems and Filterbanks, Prentice-
Hall, Englewood Cliffs, NJ, 1993.
Submit your manuscripts athttp://www.hindawi.com
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttp://www.hindawi.com
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
CombinatoricsHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
International Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com
Volume 2014 Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Stochastic AnalysisInternational Journal of