Abstract
One way to construct adaptive algorithms leads to the so called Stochastic Gradient algorithms which will be the subject of this chapter. The most important algorithm in this family, the Least Mean Square algorithm (LMS), is obtained from the SD algorithm, employing suitable estimators of the correlation matrix and cross correlation vector. Other important algorithms as the Normalized Least Mean Square (NLMS) or the Affine Projection (APA) algorithms are obtained from straightforward generalizations of the LMS algorithm. One of the most useful properties of adaptive algorithms is the ability of tracking variations in the signals statistics. As they are implemented using stochastic signals, the update directions in these adaptive algorithms become subject to random fluctuations called gradient noise. This will lead to the question regarding the performance (in statistical terms) of these systems. In this chapter we will try to give a succinct introduction to this kind of adaptive filter and to its more relevant characteristics.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
The convergence analysis will be properly done and justified in Sect. 4.5.3. Here we just want to give an intuitive result concerning the limiting behavior of the LMS.
- 2.
In the marginal case \(\Vert \mathbf x (n)\Vert =0\), the direction of update will be the null vector so there is no need to compute \(\mu (n)\).
- 3.
The misadjustment will be properly defined in Sect. 4.5 when we analyze the convergence of adaptive filters. For now, it can be seen as the ratio between the steady state EMSE and the MMSE.
- 4.
The notation \(\mathbf A ^{\dag }\) denotes the Moore-Penrose pseudoinverse of matrix \(\mathbf A \) (see Chap. 5 for further details). When \(\mathbf A =\mathbf x ^T(n)\) it can be shown that:
$$\left[\mathbf x ^T(n)\right]^{\dag }=\frac{\mathbf x (n)}{\Vert \mathbf x (n)\Vert ^2}. $$ - 5.
If this subtraction is not done properly (under the control of an adaptive filter) it might lead to an increase of the output noise power.
- 6.
This is not the only possible cost function that could be considered. Another popular solution is to obtain a filter \(\mathbf w \) that completely inverts the channel \(\mathbf h \), without taking into account the noise \(v(n)\). This solution called zero forcing (ZF) [15] completely eliminates the ISI at the cost of possibly increasing the influence of the noise. However, when the noise is sufficiently small and the channel \(\mathbf h \) does not present nulls in its frequency response, ZF offers a good performance.
- 7.
There are adaptive filtering variants for the equalization problem that do not require a training sequence. These are the so called blind adaptive filters [19]. Those filters do not need an exact reference or a reference at all, and can work directly with the channel outputs. There exists numerous algorithms of this kind. The most famous are the Sato algorithm [20] and the Godard algorithm [21]. The first basically works on a decision directed mode from the beginning of the adaptation process, whereas the second uses a modified cost function based solely on the amplitude of the channel outputs (no reference signal is specified). The interested reader on these types of algorithms can see [22].
- 8.
QPSK is a digital constellation composed by four symbols, which can be represented as complex quantities: \(e^{j\pi /4}\), \(e^{j3\pi /4}\), \(e^{j5\pi /4}\) and \(e^{j7\pi /4}\).
- 9.
- 10.
We use \(\mathrm eig _i\left[\mathbf A \right]\) to denote the \(i\)-th eigenvalue of matrix \(\mathbf A \).
- 11.
In (4.85) we used the fact that \(\mathbf A (n,j+1)\) and \(\tilde{\mathbf f }\left(\mathbf x (j)\right)\) are independent and that for two matrices \(\mathbf A \) and \(\mathbf B \) of appropriate dimensions, \(\mathrm tr \left[\mathbf A \mathbf B \right]=\mathrm tr \left[\mathbf B \mathbf A \right]\).
- 12.
The fact that \(N(\varvec{ \mu })\) depends on \(\varvec{ \mu }\) is not relevant from the point of view of the stability, because it has no influence on the asymptotic behavior of \(\mathrm tr \left[\mathbf D (n,k+1)\right]\).
- 13.
We will use the usual partial ordering defined for symmetric positive definite matrices [35].
- 14.
It is in this place where we only keep the sufficiency and lose the necessity.
- 15.
For Gaussian random variables we have the following result [39]:
$$\begin{aligned} E[x_{1}x_2x_3x_4]=E[x_1x_2]E[x_3x_4]+E[x_1x_3]E[x_2x_4]+E[x_1x_4] E[x_2x_3]. \end{aligned}$$ - 16.
For this, we need \(\mathbf R _\mathbf x \) to be strictly positive definite, which was assumed in Sect. 4.5.1.
- 17.
Although in (4.130) we should put \(\approx \), in an abuse of notation we state it as an equality.
- 18.
It should be emphasized that \(\Vert \mathbf e (n)\Vert ^2\) is not the same as the sum of the squares of the last \(K\) output estimation errors, \(\{e(i)\}_{i=n-K+1}^n\). Each component of the vector \(\mathbf e (n)\) is computed using the same filter estimate \(\mathbf w (n-1)\).
- 19.
In Chap. 5 we will provide a deeper discussion about the properties of orthogonal projection operators.
- 20.
Usual matrix inverting algorithms as Gaussian elimination require basically \(K^3\) multiplications [1]. The cost of APA is dominated by the \(K^2L\) multiplications and \(K^2(L-1)\) additions required to compute \(\mathbf X ^T(n) \mathbf X (n)\), since with \(L\gg K\) the cost of matrix inversion becomes less important.
- 21.
Notice the similarity of this with a linear prediction problem from Sect. 2.5!!
References
G.H. Golub, C.F. van Loan, Matrix Computations (The John Hopkins University Press, Baltimore, 1996)
W.W. Hager, Updating the inverse of a matrix. SIAM Review 31, 221–239 (1989)
R. Nitzberg, Application of the normalized LMS algorithm to MSLC. IEEE Trans. Aerosp. Electron. Syst. AES-21, 79–91, (1985)
B. Widrow, S.D. Stearns, Adaptive Signal Processing (Prentice-Hall, Upper Saddle River, 1985)
A. Bhavani Sankar, D. Kumar, K. Seethalakshmi, Performance study of various adaptive filter algorithms for noise cancellation in respiratory signals. Signal Processing: An International Journal (SPIJ) 4, 267–278 (2010)
J. Glover Jr, Adaptive noise canceling applied to sinusoidal interferences. IEEE Trans. Acoust. Speech Signal Process. 25, 484–491 (1977)
R. Quian Quiroga, Dataset #1: Human single-cell recording. http://www.vis.caltech.edu/ rodri/data.htm. (2003)
C.S. Herrmann, T. Demiralp, Human EEG gamma oscillations in neuropsychiatric disorders. Clinical Neurophysiology 116, 2719–2733 (2005)
R.D. Traub, M.A. Whittington, Cortical Oscillations in Health and Disease (Oxford University Press, New York, 2010)
M.H. Costa, J.C. Moreira Bermudez, A noise resilient variable step-size LMS algorithm. Elsevier Signal Process. 88, 733–748 (2008)
J.W. Kelly, J. L. Collinger, A. D. Degenhart, D. P. Siewiorek, A. Smailagic, W. Wang, Frequency tracking and variable bandwidth for line noise filtering without a reference. Proc. of IEEE EMBS, (Boston, 2011), pp. 7908–7911
A. Gersho, Adaptive filtering with binary reinforcement. IEEE Trans. Inform. Theory IT-30, 191–199 (1984)
W. A. Sethares and C. R. Johnson Jr., A comparison of two quantized state adaptive algorithms. IEEE Trans. Acoust. Speech Signal Process. ASSP-37, 138–143 (1989)
E. Eweda, Analysis and design of a signed regressor LMS algorithm for stationary and nonstationary adaptive filtering with correlated Gaussian data. IEEE Trans. Circuits Syst. 37, 1367–1374 (1990)
J. Proakis, Digital Communications, 4th edn. (McGraw-Hill, New York, 2000)
S.U. Qureshi, Adaptive equalization. Proc. IEEE 73, 1349–1387 (1985)
A.R. Bahai, B.R. Saltzberg, M. Ergen, Multi-carrier Digital Communications: Theory And Applications of OFDM, 2nd edn. (Springer, New York, 2004)
J. Liu, X. Lin, Equalization in high-speed communication systems. IEEE Circuits Syst. Mag. 4, 4–17 (2004)
S. Haykin, Adaptive Filter Theory, 4th edn. (Prentice-Hall, Upper Saddle River, 2002)
Y. Sato, A method of self-recovering equalization for multi level amplitude modulation. IEEE Trans. Commun. 23, 679–682 (1975)
D. Goddard, Self-recovering equalization and carrier tracking in two-dimensional data communication systems. IEEE Trans. Commun. 28, 1867–1875 (1980)
R. Johnson, P. Schniter, T.J. Endres, J.D. Behm, D.R. Brown, R.A. Casas, Blind equalization using the constant modulus criterion: a review. Proc. IEEE 86, 1927–1950 (1998)
P. Billingsley, Probability and Measure, 2nd edn. (Wiley-Interscience, New York, 1986)
T. Kailath, Linear Systems (Prentice-Hall, Englewood Cliffs, 1980)
L. Guo, Stability of recursive stochastic tracking algorithms. SIAM J. Control and Opt. 32, 1195–1225 (1994)
J.A. Bucklew, T.G. Kurtz, W.A. Sethares, Weak convergence and local stability properties of fixed step size recursive algorithms. IEEE Trans. Inform. Theory 30, 966–978 (1993)
V. Solo, The stability of LMS. IEEE Trans. Signal Process. 45, 3017–3026 (1997)
L. Guo, L. Ljung, G. Wang, Necessary and sufficient conditions for stability of LMS. IEEE Trans. Autom. Control 42, 761–770 (1997)
L. Guo, L. Ljung, Exponential stability of general tracking algorithms. IEEE Trans. Autom. Control 40, 1376–1387 (1995)
B. Widrow, J. Mc Cool, M.G. Larimore, C.R. Johnson, Stationary and nonstationary learning characteristic of the LMS adaptive filter. Proc. IEEE 64, 1151–1162 (1976)
A. Feuer and E. Weinstein, Convergence analysis of LMS filters with uncorrelated gaussian data. IEEE Trans. Acoust. Speech Signal Process. ASSP-33, 222–230 (1985).
J.E. Mazo, On the independence theory of equalizer convergence. Bell Syst. Techn. J. 58, 963–993 (1979)
P.S.R. Diniz, Adaptive Filtering: Algorithms And Practical Implementation, 3rd edn. (Springer, Boston, 2008)
B. Farhang-Boroujeny, Adaptive Filters: Theory and Applications (John Wiley& Sons, New York, 1998)
R.A. Horn, C.R. Johnson, Matrix Analysis (Cambridge University Press, New York, 1990)
R. Price, A useful theorem for nonlinear devices having Gaussian inputs. IRE Trans. Inform. Theory IT-4, 69–72 (1958)
L. Rey Vega, H. Rey, J. Benesty, Stability analysis of a large family of adaptive filters. Elsevier Signal Process. 91, 2091–2100 (2011)
T. Hu, A. Rosalsky, A. Volodin, On convergence properties of sums of dependent random variables under second moment and covariance restrictions. Stat. and Prob. Lett. 78, 1999–20 (2008)
A. Papoulis, Probability, Random Variables, and Stochastic Processes (McGraw-Hill, New York, 1965)
T. Al-Naffouri, A. Sayed, Transient analysis of data normalized adaptive filters. IEEE Trans. Signal Process. 51, 639–652 (2003)
M. Tarrab, A. Feuer, Convergence and performance analysis of the normalized LMS algorithm with uncorrelated gaussian data. IEEE Trans. Inform. Theory 34, 680–691 (1988)
D.T. Slock, On the convergence behaviour of the LMS and the normalized LMS algorithms. IEEE Trans. Signal Process. 41, 2811–2825 (1993)
M. Rupp, The behaviour of LMS and NLMS algorithms in the presence of spherically invariant processes. IEEE Trans. Signal Process. 41, 1149–1160 (1993)
W. Sethares, I. Mareels, B. Anderson, C. Johnson, R. Bitmead, Excitation conditions for signed regressor least mean squares adaptation. IEEE Trans. Circuits Syst. 35, 613–624 (1988)
A.H. Sayed, Adaptive Filters (John Wiley& Sons, Hoboken, 2008)
L. Rrtveit and J.H. Husy, A new prewhitening-based adaptive filter which converges to the Wiener-solution. Proc. Asilomar Conf. Sig. Syst. Comp. (Pacific Grove, 2009), pp. 1360–1364
C. Breining, P. Dreiscitel, E. Hansler, A. Mader, B. Nitsch, H. Puder, T. Schertler, G. Schmidt, J. Tilp, Acoustic echo control. An application of very-high-order adaptive filters. IEEE Signal Process. Mag. 16, 42–69 (1999)
N. Yousef, A. Sayed, A unified approach to the steady-state and tracking analyses of adaptive filters. IEEE Trans. Signal Process. 49, 314–324 (2001)
K. Ozeki and T. Umeda, An adaptive filtering algorithm using an orthogonal projection to an affine subspace and its properties. Electron. Commun. in Japan 67-A, 19–27 (1984)
J. Apolinário Jr, M.L.R. Campos, P.S.R. Diniz, Convergence analysis of the binormalized data-reusing LMS algorithm. IEEE Trans. Signal Process. 48, 3235–3242 (2000)
S.G. Sankaran, A.A.L. Beex, Convergence behavior of affine projection algorithms. IEEE Trans. Signal Process. 48, 1086–1096 (2000)
S. Gay and S. Tavathia, The fast affine projection algorithm. Proc. IEEE ICASSP, (Detroit, 1995), pp. 3023–3026
H. Ding, Fast affine projection adaptation algorithms with stable and robust symmetric linear system solvers. IEEE Trans. Signal Process. 55, 1730–1740 (2007)
M. Tanaka, S. Makino, J. Kojima, A block exact fast affine projection algorithm. IEEE Trans. Speech Audio Process. 7, 79–86 (1999)
M. Rupp, A.H. Sayed, A time-domain feedback analysis of filtered-error adaptive gradient algorithms. IEEE Trans. on Signal Process. 44, 1428–1439 (1996)
H. Rey, L. Rey Vega, S. Tressens, J. Benesty, Variable explicit regularization in affine projection algorithm: robustness issues and optimal choice. IEEE Trans. Signal Process. 55, 2096–2109 (2007)
G. Meng, T. Elmedyb, S. Jensen, J. Jensen, Analysis of acoustic feedback/echo cancellation in multiple-microphone and single-loudspeaker systems using a power transfer function method. IEEE Trans. Signal Process. 59, 5774–5788 (2011)
M. Honig, M.K. Tsatsanis, Adaptive techniques for multiuser CDMA receivers. IEEE Signal Process. Mag. 17, 49–61 (2000)
R.L. Calcavante, I. Yamada, Multiaccess interference suppression in orthogonal space-time block coded MIMO systems by adaptive projected subgradient method. IEEE Signal Process. Mag. 56, 1028–1042 (2007)
A. Zanella, M. Chiani, M. Win, Statistical analysis of steepest descend and LMS detection algorithms for MIMO systems. IEEE Trans. Veh. Technol. 60, 4667–4672 (2011)
N.V. Thakor, Y.S. Zhu, Applications of adaptive filtering to ECG analysis: noise cancellation and arrhythmia detection. IEEE Trans. Biomed. Eng. 38, 785–794 (1991)
S.M.M. Martens, M. Mischi, S.G. Oei, J.W.M. Bergmans, An improved adaptive power line interference canceller for electrocardiography. IEEE Trans. on Biomed. Eng. 53, 2220–2231 (2006)
M. Bouchard, Multichannel affine and fast affine projection algorithms for active noise control and acoustic equalization systems. IEEE Trans. Speech Audio Process. 11, 54–60 (2003)
E.P. Reddy, D.P. Das, K.M. Prabhu, Fast adaptive algorithms for active control of nonlinear noise processes. IEEE Trans. Signal Process. 56, 4530–4536 (2008)
J.S. Soo, K.K. Pang, New structures for adaptive filtering in subbands with critical sampling. IEEE Trans. Acoust. Speech Signal Process. 38, 373–376 (1990)
M.R. Petraglia, R.G. Alves, P.S. Diniz, Multidelay block frequency domain adaptive filters. IEEE Trans. Signal Process. 48, 3316–3327 (2000)
S.S. Pradhan, V.U. Reddy, A new approach to subband adaptive filtering. IEEE Trans. Signal Process. 48, 655–664 (1999)
J. Benesty, C. Paleologu, S. Ciochina, On regularization in adaptive filtering. IEEE Trans. Audio, Speech, Lang. Process. 19, 1734–1742 (2011)
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
Copyright information
© 2013 The Author(s)
About this chapter
Cite this chapter
Rey Vega, L., Rey, H. (2013). Stochastic Gradient Adaptive Algorithms. In: A Rapid Introduction to Adaptive Filtering. SpringerBriefs in Electrical and Computer Engineering. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-30299-2_4
Download citation
DOI: https://doi.org/10.1007/978-3-642-30299-2_4
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-642-30298-5
Online ISBN: 978-3-642-30299-2
eBook Packages: EngineeringEngineering (R0)