IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. ×, NO.

××, MONTH 2010 1

A Quaternion Widely Linear Adaptive Filter

Clive Cheong Took, Member, IEEE, and Danilo P. Mandic, Senior Member, IEEE

Abstract—A quaternion widely linear (QWL) model for quaternion the work by Vakhania, who defined the concept of Q-properness as
valued mean-squared-error (MSE) estimation is proposed. The aug- the invariance of the distribution of a quaternion valued variable under
mented statistics are first introduced into the field of quaternions, and
some specific rotations around the angle of π/2 [8]. Amblard and Le
it is demonstrated that this allows for capturing the complete second
order statistics available. The QWL model is next incorporated into the Bihan relaxed the conditions of Q-properness to an arbitrary axis and
Quaternion Least Mean Square (QLMS) algorithm to yield the widely angle of rotation ϕ, that is, q , eνϕ q [9] for any pure unit quaternion
linear QLMS (WL-QLMS). This allows for a unified approach to adaptive ν (whose real part vanishes); where symbol , denotes equality in
filtering of both Q-proper and Q-improper signals, leading to improved terms of probability density function (pdf).
accuracies compared to the QLMS class of algorithms. Simulations on
both benchmark and real world data support the analysis. Although these results provide an initial insight into the processing
of general quaternionic signals, they are not straightforward to apply
Index Terms—Quaternion adaptive filtering, Q-properness, quaternion
in the context of adaptive filtering applications. To this end, we first
second order noncircularity, Wiener model, quadrivariate processes,
quaternion LMS (QLMS), widely linear model, widely linear QLMS. propose the quaternion widely linear model, specifically designed for
the unified modelling of the generality of quaternion signals, both Q-
proper and Q-improper. The benefits of such an approach are shown
to be analogous to the benefits that the augmented statistics provides
I. I NTRODUCTION for complex valued data [7]. Next, the QWL model is incorporated
Standard techniques employed in multichannel statistical signal into the quaternion LMS [3] to yield the widely linear QLMS (WL-
processing typically do not fully cater for the ‘coupled’ nature of QLMS), and its theoretical and practical advantages are demonstrated
the available information within the channels. Thus, most practical through analysis and simulations.
approaches operate based on channelwise processing, which is not
optimal for general multivariate signals (where data channels are II. P ROPERTIES OF Q UATERNION R ANDOM V ECTORS
typically correlated). On the other hand, the quaternion domain H A. Quaternion Algebra
allows for the direct modelling of three- and four-dimensional signals, The quaternion domain, a non-commutative extension of the com-
and its algebra naturally accounts for the coupling between the signal plex domain, provides a natural framework for the processing of
components. three- and four-dimensional signals. A quaternion variable q ∈ H
The use of quaternions is rapidly gaining in popularity, as for comprises a real part ℜ{·} and a vector-part, also known as a pure
instance, many multivariate problems based on vector sensors (mo- quaternion ℑ{·}, consisting of three imaginary components, and can
tion body sensors, seismics, wind modelling) can be cast into the be expressed as:
quaternion domain. The recent resurgence of quaternion valued
signal processing stems from the potential advantages that special q = ℜ{q} + ℑ{q}
properties of quaternion algebra offer over real valued vector algebra = ℜ{q} + ıℑi {q} + ℑj {q} + κℑk {q}
in multivariate modelling. Applications of quaternions include those = qa + ıqb + qc + κqd ∈H (1)
in vector sensing [1], machine learning [2], and adaptive filters [3].
Recent advances in complex valued signal processing have been The relationship between the orthogonal unit vectors, ı, , κ are
based on the widely linear model proposed by Picinbono [4]. This given by
model, together with the corresponding augmented complex statistics, ı = κ κ = ı κı = 
has been successfully used to design enhanced algorithms in commu- 2 2 2
nications [5] [6] and adaptive filters [7]. These studies have shown ıκ = ı =  = κ = −1 (2)
that widely linear modelling and the associated augmented statistics Given q1 , q2 ∈ H, the noncommutative quaternion product is com-
offer theoretical and practical advantages over the standard complex puted as
models, and are applicable to the generality of complex signals, both
circular and noncircular. q1 q2 = ℜ{q1 q2 } + ℑ{q1 q2 }
Models suitable for the processing of signals with rotation depen- where ℜ{q1 q2 } = q1,a q2,a + q1,b q2,b + q1,c q2,c + q1,d q2,d
dent distribution (noncircular) are lacking in the quaternion domain, ℑ{q1 q2 } = q1,a ℑ{q2 } + q2,a ℑ{q1 } + ℑ{q1 } × ℑ{q2 } (3)
and their development has recently attracted significant research effort
[3]. Current second order algorithms operate based on only the where the symbol “×” denotes the vector product.
quaternion valued covariance [1]–[3] and thus do not fully exploit the
available statistical information. Advances in this direction include B. Quaternion Involutions
Given a complex number z = za + ızb , its real and imaginary part
Copyright (c) 2010 IEEE. Personal use of this material is permit- 1
can be extracted as za = 12 (z + z ∗ ) and zb = 2ı (z − z ∗ ). However,
ted. However, permission to use this material for any other purposes
must be obtained from the IEEE by sending a request to pubs- such convenient manipulation is not possible in the quaternion do-
permissions@ieee.org. Manuscript received December 4, 2009; revised April main. To circumvent this problem, the three perpendicular quaternion
9, 2010. involutions (self-inverse mappings) given by
C. Cheong Took and D. P. Mandic are with the Dept. of Electrical
and Electronic Engineering, Imperial College London, London SW7 2AZ, qı = −ıqı = qa + ıqb − qc − κqd
U.K. (Email: {c.cheong-took, d.mandic }@imperial.ac.uk). Color versions 
q = −q = qa − ıqb + qc − κqd
of one or more of the figures in this paper are available online at
http://ieeexplore.ieee.org. qκ = −κqκ = qa − ıqb − qc + κqd (4)
2 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. ×, NO. ××, MONTH 2010

can be employed, and the four components of the quaternion variable property states that the power of a quaternion variable is a sum of
q can now be computed as [10] the powers of the signal components. Observe that properties P1 and
1 1 P2 imply properties P3 and P4.
qa = (q + q ∗ ) qb = (q − q ı∗ )
2 2ı
1 1
qc = (q − q ∗ ) qd = (q − q κ∗ ) (5) B. Augmented Statistics of Q-proper variables
2 2κ
Notice that Q-properness also implies that the quaternion vector q
The quaternion conjugate operation (·)∗ is also an involution and is uncorrelated with the vector involutions qı , q , qκ , that is1 ,
can be expressed as a linear function of the three perpendicular
involutions, that is E{qqıH } = 0 E{qqH } = 0 E{qqκH } = 0 (9)
1 ı
q∗ = (q + q  + q κ − q) (6) This simplifies the structure of the augmented covariance matrix Cqa
2
of a Q-proper random vector, as
III. Q UATERNION S TATISTICS    
Cq Pq Pqı Pq 2I −I I I
Picinbono demonstrated that the complete description of the second a
 Pq Cq 0 0   −I
2 2I 0 0 
Cq =  ı
  = 2σ  
order statistics in C can be achieved, provided the real valued bivari- Pq 0 Cq 0  I 0 2I 0 
ate covariance matrices can be calculated from their complex valued Pq 0 0 Cq I 0 0 2I
counterparts (see pp. 118-119 [11]). Following on this result, we next (10)
show that the complete second order statistical description in H is ob- that is, the cross-quasicovariance matrices C˜αβ all vanish.
tained when the real valued quadrivariate covariances are expressed in
terms of their quaternion valued counterparts, as shown in Appendix IV. T HE Q UATERNION W IDELY L INEAR M ODEL
IX-A. However, unlike the complex domain C, where for this purpose
it is sufficient to combine the complex vector z and its conjugate z∗ To account for the complete second order statistics of quaternion
into the augmented complex vector za = [zT zH ]T , in the quaternion valued signals in mean-squared error (MSE) estimation, we need to
domain, we also need to consider the involutions (4). We can therefore introduce a filtering model corresponding to the widely linear model
build an augmented quaternion vector, comprising of any four of the in the complex case [4]. Consider the MSE estimator of a signal y in
five quantities {q, q ∗ , q ı , q  , q κ } or their conjugates. One convenient terms of another observation x, that is, ŷ = E[y|x]. For zero mean,
augmented basis is q a = {q, q ∗ , q ı∗ , q ∗ }, and will be used in jointly normal real y and x, the solution is
this work. Then, the augmented vector qa = [qT qH qıH qH ]T ŷ = hT x (11)
contains all the necessary second order statistical information and its
augmented covariance matrix is given by In the quaternion domain, however, the real estimator (11) applies to
  each component (the real and the three imaginary parts) of quaternion
Cq Pq Pqı Pq
variables, that is
 PqH C˜q C˜qqı C˜qq 
Cqa = E{qa qaH } =   PqıH C˜qq ˜
 (7)
H
ı Cq ı C˜qı q  ŷγ = E[yγ |xa , xb , xc , xd ], γ ∈ {a, b, c, d}
PqH C˜qq H
 C˜qHı q C˜q
and thus
The submatrices in (7) are calculated according to
ŷ = E[ya |xa , xb , xc , xd ] + ıE[yb |xa , xb , xc , xd ]
Cα = E{ααH } C˜α = E{α∗ αT }
+E[yc |xa , xb , xc , xd ] + κE[yd |xa , xb , xc , xd ] (12)
C˜αβ = E{α∗ β T } Pα = E{ααT } α, β ∈ {q, qı , q }
Upon employing the identities (5), it is clear that the quaternion
We refer to C˜α as the quasicovariance and C˜αβ as the cross-
estimator can also be expressed2 as
quasicovariance matrices. Unlike in the complex case [4], the non-
commutativity of the quaternion product results in Cα 6= C˜α
∗
. ŷ = E[y|x, x∗ , xı∗ , x∗ ] + ıE[y ı |x, x∗ , xı∗ , x∗ ]
+E[y  |x, x∗ , xı∗ , x∗ ] + κE[y κ |x, x∗ , xı∗ , x∗ ] (13)
A. Circularity in H and Q-properness
and thus arrive at the widely linear estimator for general quaternion
For a quaternion valued variable to be second order circular (or Q- signals
proper), its probability distribution should be rotation-invariant with
respect to the six pairs of axes: {1, ı}, {1, }, {1, κ}, {ı, }, {κ, } y = hH x + gH x∗ + uH xı∗ + vH x∗ = waH xa (14)
and {κ, ı}, where ‘1’ represents the real axis and ı, , κ denote the
corresponding imaginary axes. In other words, a Q-proper quaternion where wa = [hT gT uT vT ]T and xa = [x xH xıH xH ]T .
random variable should satisfy the following properties [8] Following on the proposed quaternion widely linear model, the
Wiener solution is now derived as the optimal Mean Square Estimator
P1: E{qδ2 } = E{qǫ2 } = σ 2 ∀ δ, ǫ = a, b, c, d (MSE). Consider the standard real valued quadratic cost function, that
P2: E{qδ qǫ } = 0 ∀ δ, ǫ = a, b, c, d and δ 6= ǫ is,
P3: E{qq} = −2E{qδ2 } = −2σ 2 ∀ δ = a, b, c, d J = E{ee∗ } = E{[d − y][d∗ − y ∗ ]}
2
P4: E{|q| } = 4E{qδ2 } = 4σ 2
∀ δ = a, b, c, d (8) = E{dd∗ } + E{yy ∗ } − E{yd∗ } − E{dy ∗ } (15)
The first property, P1, states that all the four-signal components 1 Similarly, for a complex valued random vector z, C-properness means
of a quaternion valued variable have equal variances. The property
that z is uncorrelated with z∗ in the ‘complex sense’, since E{z(z∗ )H } =
P2 implies that the components of q are not correlated. Property E{zzT } = 0, for more detail see [7].
P3 suggests that the pseudocovariance matrix of Q-proper signals 2 Any other augmented basis other than {x, x∗ , xı∗ , x∗ } can be used, as
does not vanish, in contrast to the complex case. Finally, the fourth explained in the Section III.
 SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS
The derivative of the cost function (15) can be expressed as (for full
derivation, see Appendix IX-B)
 

∗ ∗

∗ ∗

∇ w J = E ∇w y y + y ∇w y − ∇ w y d − d ∇w y
a a a a a
   
= E 4[xa y ∗ − xa d∗ ] + E 2[d(n)xa − yxa ] (16)
| {z } | {z } 4D wind field signal [3]. The quantitative performance index was the
I II
To obtain the Wiener solution, the expectations of I and II in (16) are
set to zero. In the complex domain, we can sum up the terms I and
II in (16); however, due to the noncommutativity of the quaternion
product, we need to consider the terms in (16) individually, giving
the solution3
I : wo = E{xa xaH }−1 E{xa d∗ } (17)
II : wo = E{xa∗ xaH }−1 E{xa∗ d∗ } (18)
The first condition for the Wiener solution (17) requires the inversion
of the augmented covariance Cxa = E{xa xaH }. On the other hand,
the second condition (18) also depends on the conjugate of pseudo-
covariance matrix of the augmented vector xa , which conforms with
the observation in [3] that the quaternion domain accounts inherently
for the information contained in pseudocovariance.
V. T HE W IDELY L INEAR Q UATERNION L EAST M EAN S QUARE
A LGORITHM
We now extend some recent results in quaternion adaptive filtering
[3], and employ the quaternion widely linear model (14) within the
stochastic gradient adaptive filtering framework in H, to propose
the Widely Linear Quaternion Least Mean Square (WL-QLMS)
adaptive filtering algorithm. Within the stochastic gradient descent
optimisation, the gradient of the instantaneous cost function (15) is



∇wa J (n) = e(n) ∇wa e∗ (n) + ∇wa e(n) e∗ (n)
= 2e(n)xa (n) − 4xa (n)e∗ (n) (19)
Notice that due to the non-commutativity of the quaternion product,
the two error gradient terms in (19) need to be treated sepa-
rately. Based on the generic stochastic gradient update ∆wa =
−µ∇wa J (n), the update of the weight vector of the WL-QLMS
algorithm can be obtained as4

available.
wa (n + 1) = wa (n) + µ 2xa (n)e∗ (n) − e(n)xa (n) (20)
where the scaling factor of two has been absorbed in the stepsize µ.
Given y(n) = wH (n)x(n), observe the same form of the update in
(20) as that within the QLMS in [3]. Although AQLMS outperformed
QLMS, as shown in [3] in the context of Q-improper signals, its
second order information is derived from only the covariance and the
pseudocovariance, and is therefore still suboptimal for Q-improper
data, as it cannot model the information contained in matrices
such as Cqc (see Appendix IX-A). Also, observe that the real-
valued multichannel LMS does not exploit the interchannel cross-
correlation in the same elegant and intuitive way as the WL-QLMS.
For more details, see p. 132 of [12] and the performance comparisons
with AQLMS in [3]. Finally, a convergence analysis of WL-QLMS
algorithm is provided in Appendix IX-C, and the upper bound on the
stepsize is found to be 0 < µ < 2/λmax , where λmax denotes the
maximum right eigenvalue of Cxα = 2Cxa + Pxa∗ .
3 Similarly to complex-valued case [7], there are several equivalent formu-
lations for the quaternion-valued Wiener solution. For instance, if the filter
output is considered as y(n) = waT (n)xa (n) instead of y(n) = waH xa ,
an alternative solution is obtained.
4 For more detail of the derivation, see [3] and the Appendix IX-B.
3
VI. S IMULATIONS
Three sets of simulations were conducted in an M -step ahead
prediction setting in order to comprehensively assess the performance
of the proposed WL-QLMS algorithm against QLMS and AQLMS.
The datasets used were a Q-proper synthetic AR(4) process [7], the
Q-improper four dimensional (4D) Saito’s signal [2] and real-world
σ2
prediction gain Rp = 10 log σx2 , where σx2 and σe2 denote respectively
e
the estimated variances of the input and error; the filter length is
denoted by L.
1) Q-proper Autoregressive(4) Model: The autoregressive AR(4)
process x(n) = 1.79x(n − 1) − 1.85x(n − 2) + 1.27x(n − 3) −
0.41x(n − 4) was driven by quadruply white Gaussian noise, whose
real and imaginary components were uncorrelated, but had equal
variances. From Fig. 1, initially, QLMS converged faster than WL-
QLMS and AQLMS. This is because 1) it operates based on the
covariance Cx only, which is adequate to describe the complete
statistics of Q-proper signals, due to the deterministic relationship
Cx = −Px /2 = Pxı /2 = Px /2; 2) QLMS has fewer filter parameters
to adapt. However, at steady state the prediction gain for all the
algorithms considered was approximately 30 dB, as they are all suited
to process Q-proper data.
2) Q-improper Four-Dimensional Saito’s Circuit [2]: Fig. 2 com-
pares the performance of the quaternion algorithms [3] over a range
of filter parameters. Conforming with the analysis in Section IV,
the WL-QLMS algorithm outperformed the QLMS and AQLMS on
the Q-improper Saito process, also exhibiting better convergence
properties (see top plot of Fig. 4), as QLMS accounts for the second
order information only from the covariance matrix Cx , and the
AQLMS operates based on Cx and Px ; both are not adequate to
account for the complete statistics of Q-improper signals.
3) Q-improper Wind Forecasting: The 4D quaternion dataset
comprised the 3D wind speed (North-South, East-West, and vertical
directions5 ) as a vector part (pure quaternion) and air temperature
as the scalar part. Fig. 3 shows that the prediction results over a
range of filter parameters are in agreement with theoretical analysis,
illustrating that WL-QLMS outperformed QLMS and AQLMS on
Q-improper nonstationary data. Fig. 4 (bottom plot) illustrates the
improved convergence properties of WL-QLMS over QLMS, due to
WL-QLMS using the complete second order statistical information
VII. C ONCLUSION
We have introduced a quaternion widely linear model (QWL) for
enhanced second order estimation of general quaternionic signals. We
have demonstrated the efficacy of the QWL model, by incorporating
it into the Quaternion Least Mean Square (QLMS) algorithm [3] to
yield the widely linear QLMS (WL-QLMS) algorithm. For rigour, the
convergence analysis includes the stepsize bound and learning curves
for both second order circular and noncircular signals. Experiments
have been conducted for a range of filter parameters and dataset,
illustrating the WL-QLMS outperforming other algorithms of the
kind.
VIII. ACKNOWLEDGEMENT
We would like to thank Gill instruments for kindly providing us
with the 3D WindMaster anemometer.
5 The wind data were sampled at 32 Hz and recorded by the 3D WindMaster
anemometer by Gill Instruments.
 4 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. ×, NO. ××, MONTH 2010

∗
∂y ∂y ∂y ∂y ∂y
R EFERENCES Therefore, ∂h = ∂h a
+ ı ∂hb
+  ∂h c
+ κ ∂h d
= 4x and ∂y∂h
=
[1] S. Miron, N. L. Bihan, and J. Mars, “Quaternion-MUSIC for vector- −2x. Similarly, the derivatives of the terms g x , uH xı∗ , vH x∗
H ∗

sensor array processing,” IEEE Transactions on Signal Processing, and their conjugates in (14) can also be computed respectively as
∗ ∗
∂y ∂y ∂y
vol. 54, no. 4, pp. 1218– 1229, 2006. ∂g
= 4x∗ , ∂u = 4xı∗ , ∂v = 4x∗ ; ∂y ∂g
= −2x∗ , ∂y∂u
=
[2] P. Arena, L. Fortuna, G. Muscato, and M. G. Xibilia, Neural Networks ı∗ ∂y ∗ ∗
in Multidimensional Domains. Lecture Notes in Control and Information
−2x , and ∂v = −2x . As a result, the augmented quaternion
∂y a ∂y ∗ a
Sciences (Springer Verlag), vol. 234, 1998. gradients can be expressed as ∂w a = 4x and ∂w a = −2x .
[3] C. Cheong-Took and D. P. Mandic, “The Quaternion LMS Algorithm
for Adaptive Filtering of Hypercomplex Real World Processes,” IEEE
Transactions on Signal Processing, vol. 57, no. 4, pp. 1316–1327, 2009. C. Convergence Analysis
[4] B. Picinbono and P. Chevalier, “Widely linear estimation with complex To factor out the conjugate error e∗ (n) on the right hand side, we
data,” IEEE Transactions on Signal Processing, vol. 43, no. 8, pp. 2030– start by replacing e(n)xa (n) = xa∗ (n)e∗ (n) − 2ℑ{xa∗ (n)e∗ (n)}
2033, 1995.
[5] H. Gerstacker, R. Schober, and A. Lampe, “Receivers with widely into (20) to yield
linear processing for frequency-selective channels,” IEEE Transactions
 
 
on Communication, vol. 51, no. 9, pp. 1512–1523, 2003. w(n+1) = w(n)+µ 2xa (n)+xa∗ (n) e∗ (n)−2ℑ{xa∗ (n)e∗ (n)}
[6] R. Schober, W. H. Gerstacker, and L. H.-J. Lampe, “ Data-aided and
blind stochastic gradient algorithms for widely linear MMSE MAI
(23)
suppression for DS-CDMA,” IEEE Transactions on Signal Processing, Since e(n) = −[w(n) − wo ]H xa (n) = −vH (n)xa (n) [with wo as
vol. 52, no. 3, pp. 746– 756, 2004. the Wiener solution], (23) becomes
[7] D. P. Mandic and V. S. L. Goh, Complex Valued Nonlinear Adaptive 

Filters: Noncircularity, Widely Linear and Neural Models. John Wiley v(n + 1) = v(n) − µ 2E xa (n)xaH (n)
and Sons Ltd, 2009.
[8] N. N. Vakhania, “Random vectors with values in quaternion Hilbert 

−E xa∗ (n)xaH (n) v(n)
spaces,” Theory of Probability and its Applications, vol. 43, no. 1,  
pp. 99–115, 1999. 
[9] P. O. Amblard and N. LeBihan, “On properness of quaternion valued +2ℑ E xa∗ (n)xaH (n) v(n)
random variables,” in Proceedings of International Conference on Math-  


ematics (IMA) in Signal Processing, Cirencester, pp. 23–26, 2004. v(n + 1) = v(n) − µ 2Cxa − Pua∗ v(n) + 2ℑ Pxa∗ v(n) (24)
[10] T. A. Ell and S. J. Sangwine, “Quaternion Involutions and Anti-
Involutions,” Computers & Mathematics with Applications, vol. 53,
no. 1, pp. 137–143, 2007. The upper bound on the stepsize can then be approximated, by
[11] B. Picinbono, Random Signals and Systems. Prentice Hall, 1993. considering the imaginary part ℑ{Pxa∗ v(n)} as a full quaternion,
[12] Y. Huang and J. Benesty, Audio Signal Processing for Next Generation to give v(n + 1) ≈ v(n) − µ[2Cxa + Pxa∗ ]v(n). By letting Cxα =
Multimedia Communication Systems. Kluwer Academic Publishers, 2Cxa +Pxa∗ and taking the right eigenvalue decomposition of Cxα , with
2004. its maximum eigenvalue λmax , we can obtain the stepsize bound for
2
WL-QLMS as 0 < µ < λmax , a generic form which also applies
IX. A PPENDIX to QLMS and AQLMS. For instance, in the case of AQLMS, the
A. Complete Second Order Statistics augmented covariance matrix Cxa degenerates into its 2 × 2 top-left
submatrix in (7), whereas Cxa becomes Cx for QLMS. This implies
The real valued correlation matrices of single components qa , qb ,
a larger stepsize is required for these algorithms to converge at the
qc and qd of the quaternion random vector q can be expressed in
same rate as WL-QLMS in the case of Q-improper signals (when
terms of the quaternion valued covariance matrices as
every element of Cxa does not vanish).
1 1
Cq a = ℜ{Cq + Pq } Cqb = ℜ{Cq − Pqı } 4
2 2   3.5
1
Cqc = ℜ{Cq − Pq } Cqd = ℜ{Cq } − Cqa + Cqb + Cqc 3 WL−QLMS
Q−proper AR(4) process
2
Absolute Error

2.5
AQLMS
1 1 2

Cqb qa = ℑi {Cq + Pq } Cqc qa = ℑj {Cq + Pq } 1.5

QLMS
2 2 1
1 1
Cqd qa = ℑk {Cq + Pq } Cqc qb = ℑk {Cq − Pqı } 0.5

2 2 0

1 1 500 1000 1500 2000 2500

Cqd qc = ℑi {Cq − Pq } = − ℑj {Cq − Pqκ }

Number of iterations (n)
Cqd qb (21)
2 2
Fig. 1. Learning curves of WL-QLMS, QLMS and AQLMS on the prediction
of the Q-proper AR(4) process x(n) = 1.79x(n − 1) − 1.85x(n − 2) +
B. Derivation of the Stochastic Gradients 1.27x(n − 3) − 0.41x(n − 4).
To calculate the derivatives of the output y(n) and its conjugate
with respect to the augmented weight vector, the terms waH (n)x(n)
and xaH (n)w need to be examined. Consider the first term in (14)
hH x and its conjugate xH h. The partial derivatives of the output
y(n) = waH (n)x(n) with respect to h are
∂y ∂y ∗
= xa + ıxb + xc + κxd = xa − ıxb − xc − κxd
∂ha ∂ha
∂y ∂y ∗
ı = xa + ıxb + xc + κxd ı = −xa + ıxb − xc − κxd
∂hb ∂hb
∂y ∂y ∗
 = xa + ıxb + xc + κxd  = −xa − ıxb + xc − κxd
∂hc ∂hc
∂y ∂y ∗
κ = xa + ıxb + xc + κxd κ = −xa − ıxb − xc + κxd (22)
∂hd ∂hd
 SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS 5

35 WL−QLMS
35
WL−QLMS
AQLMS
30
30 AQLMS

Prediction Gain (dB)

Prediction Gain (dB)

25
25

QLMS
20
20

QLMS

15 15

0 10
10
0 5 20

10 15 4
5 0
15 Filter Length 10 3.5
0.5
Prediction Horizon L WL−QLMS 4D SAITO PROCESS
10 20 Filter Length 5 1 x 10
−4 3

Absolute Error
M L 0 Stepsize µ 2.5
1.5
AQLMS
2

1.5 QLMS
1

Fig. 2. The performance of WL-QLMS , QLMS and AQLMS on the 0.5

prediction of 4D Saito’s process. 0

20 40 60 80 100 120 140
Number of iterations (n)

30

25

WL−QLMS 4D WIND MODEL

Absolute Error

20

15
AQLMS

10
QLMS
5

0
10 20 30 40 50 60 70 80 90 100

40 Number of iterations (n)

40
WL−QLMS WL−QLMS
35
35
Fig. 4. Learning curves of WL-QLMS, QLMS and AQLMS on the prediction
30 of Q-improper processes, i.e. 4D Saito’s process (top) and 4D wind field
Prediction Gain (dB)

Prediction Gain (dB)

30 model (bottom).
25

25 20
AQLMS
15
20
QLMS 10 AQLMS

15
5
10
1
8 QLMS
10 0.8
10
6 8
−5 0.6 8
6 x 10
4 6
0.4
Prediction Horizon 4 Stepsize µ 4
2 Filter Length
M 2 0.2
L 2 Filter Length
0 0 L
0 0

Fig. 3. The performance of WL-QLMS, QLMS and AQLMS on the

prediction of a 4D wind field.