Robust adaptive filtering algorithms based on (inverse)hyperbolic sine function

Sihai Guan; Qing Cheng; Yong Zhao; Bharat Biswal

doi:10.1371/journal.pone.0258155

Abstract

Recently, adaptive filtering algorithms were designed using hyperbolic functions, such as hyperbolic cosine and tangent function. However, most of those algorithms have few parameters that need to be set, and the adaptive estimation accuracy and convergence performance can be improved further. More importantly, the hyperbolic sine function has not been discussed. In this paper, a family of adaptive filtering algorithms is proposed using hyperbolic sine function (HSF) and inverse hyperbolic sine function (IHSF) function. Specifically, development of a robust adaptive filtering algorithm based on HSF, and extend the HSF algorithm to another novel adaptive filtering algorithm based on IHSF; then continue to analyze the computational complexity for HSF and IHSF; finally, validation of the analyses and superiority of the proposed algorithm via simulations. The HSF and IHSF algorithms can attain superior steady-state performance and stronger robustness in impulsive interference than several existing algorithms for different system identification scenarios, under Gaussian noise and impulsive interference, demonstrate the superior performance achieved by HSF and IHSF over existing adaptive filtering algorithms with different hyperbolic functions.

Citation: Guan S, Cheng Q, Zhao Y, Biswal B (2021) Robust adaptive filtering algorithms based on (inverse)hyperbolic sine function. PLoS ONE 16(10): e0258155. https://doi.org/10.1371/journal.pone.0258155

Editor: Qichun Zhang, University of Bradford, UNITED KINGDOM

Received: July 12, 2021; Accepted: September 18, 2021; Published: October 11, 2021

Copyright: © 2021 Guan et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data are within the paper and its Supporting Information files.

Funding: This work was supported by the Fundamental Research Funds for the Central Universities Southwest Minzu University (Grant Number: 2021NQNCZ04) and National Cultural Science Foundation of China (Grant number: 61871420).

Competing interests: The authors have declared that no competing interests exist.

1. Introduction

Adaptive filter (AF) algorithms are frequently employed in linear systems [1–3], nonlinear systems [4], and distributed network systems [5] and have been used in many fields, including biomedical engineering [6,7]. Among adaptive filter algorithms, the least mean square (LMS) algorithm has probably become the most popular adaptive filtering algorithm for its simple configuration, low computational complexity, sufficient tracking capability, and easiness of implementation [2,3,5,7–13]. However, in actual engineering, non-Gaussian distribution measurement noise with a heavy-tailed pdf, e.g., a Laplace or α-stable noise, is everywhere [8,12,14]. However, those measurement noises significantly affect the adaptive estimation accuracy of estimation, and most of the adaptive estimation algorithms are highly susceptible to impulsive interference [8,12]. In the presence of non-Gaussian noise, adaptive filters are required to be less sensitive to large noises and be more sensitive to small noises [15]. If the measurement noise involves impulse interference, the adaptive filtering algorithm based on the mean square error criterion may severely reduce the convergence performance and even cause divergence problems [16–18]. In such cases, the commonly used cost functions include high order error power (HOEP) conditions [19–24], the mixed-norm cost function [12,25], the logarithmic and exponential cost functions [15,26], the correntropy cost function [27], and the kernel risk-sensitive cost function [28]. The most common HOEP algorithm is the least mean absolute-third [20,21] and least mean fourth (LMF) [22,24,29,30] algorithms. The LMF algorithm outperforms the LMS algorithm and achieves a better trade-off between the LMS’s transient and steady-state performances [22,24,29,30]; the computational complexity of LMF is very high. Besides, to fully take advantage of both LMS and LMF, Lim and colleagues proposed a combined LMS/F algorithm [31], and its simplicity and stability developed in [32,33]. Besides, based on the mean absolute error (MAE) [34], the sign-error-type algorithm generally exhibits a slow convergence rate in practice. Based on information-theoretic learning [35], Chen and colleagues proposed the generalized maximum correntropy criterion (GMCC) based adaptive filtering robust under non-Gaussian/impulsive noise [27]. However, the correntropy’s performance surface is sharp around the optimal solution but flat far away from the optimal solution [28]. In robust adaptive filters, the robust LMS algorithm [36] utilizes a linear combination of bounded hyperbolic tangent basis functions to approximate the optimal score function. The correntropy or the logarithmic hyperbolic function is considered a candidate for the cost function in the switching algorithm for super-Gaussian noises due to their robustness [37]. In an endeavor to achieve lower steady-state misalignment, a generalized hyperbolic secant function as a robust norm and derive the generalized hyperbolic secant adaptive filter was proposed [38]. To address both Gaussian and non-Gaussian noises with a uniform expression, Liu and colleagues [39] proposed a novel HTCC algorithm by combining nonlinear function and mapping mode. And hyperbolic tangent function was also constructed in a drought disaster losses model, which was used to analyze the characteristics of a drought disaster losses curve with multiple inflection points and nonlinearity [40]. In addition, to reduce the interference of impulse noise on the traditional spline adaptive filter algorithm for identifying Wiener-type non-systems, an arctangent function that is insensitive to large outliers is used to construct a cost function, so that the spline adaptive filter is robustness to impulse noise [41]. Since it is difficult to obtain an analytical result of the expectation of the hyperbolic tangent function under the Gaussian assumption, the hyperbolic tangent-based robust filter and its performance analysis are not considered in both robust LMS and the switching algorithm. Wang and colleagues proposed a logarithmic hyperbolic cosine adaptive filter (LHCAF) [42] by using only the logarithmic hyperbolic cosine based cost function. LHCAF was shown to provide better convergence performance compared to the well-known GMCC [27], it may not provide optimal performance as it does not take into account the sparse nature of the system. Then, a sparsity-aware zero attraction LHCAF and a reweighted zero attraction LHCAF were proposed [43]. Besides, Liang and colleagues [44] developed a recursive constrained least lncosh adaptive filtering algorithm to suppress impulsive interference. Also for dealing with non-Gaussian distribution noise signal, an adaptive filtering algorithm, namely least lncosh (Llncosh) algorithm was proposed, and VLlncosh scheme PLlncosh algorithm were also extend [45]. Inspired by the hyperbolic secant cost function, Lu and colleague [46] improved performance of the adaptive filtering algorithm in white Gaussian noise and uniform noise environments. Recently, Tao and colleagues [47] proposed a constrained least lncosh adaptive filtering algorithm under non-Gaussian noise environment.

Although a class of adaptive filtering algorithms was designed using the hyperbolic function, most of those algorithms have few parameters that need to be set. The estimation accuracy and convergence performance of those adaptive algorithms can be improved further. More importantly, the hyperbolic sine function has not yet been discussed. This paper proposed a family of hyperbolic sine functions, including joint hyperbolic sine function (HSF) and inverse hyperbolic sine function (IHSF), as cost functions. Compared with several existing adaptive filtering algorithms, superior steady-state performance and stronger robustness can be attained. Extensive simulation studies for different system identification scenarios under Gaussian and non-Gaussian disturbances demonstrate the superior performance achieved by HSF and IHSF over existing robust adaptive filtering algorithms.

This paper further studies and improves how to design an adaptive filtering algorithm using hyperbolic sine function as the cost function. To summarize, the main contributions of this paper are: (1) development of a robust adaptive filtering algorithm based on hyperbolic sine function (HSF); (2) then extend the HSF algorithm to another novel adaptive filtering algorithm based on inverse hyperbolic sine function (IHSF); (3) analyses of the computational complexity for the HSF and IHSF algorithms; (4) validation of the analyses and superiority of the proposed algorithm via simulations. Moreover, the first step in this paper’s schematic diagram is to list current research results, then point out which ones can be further studied. The rest of this paper is arranged as follows. In Section 2, the HSF and IHSF algorithms are derived based on the hyperbolic sine function and the inverse hyperbolic sine function. In Section 3, the computational complexity of HSF and IHSF algorithms is analyzed. Section 4 provides several experiments simulation results. Finally, Section 5 gives a conclusion. Note: Bold type refers to vectors, [∙]^T denotes for the transpose, and [∙]⁻¹ denotes the inverse operation.

2. Hyperbolic sine function

One of the cost functions is given by (1) where sinh(∙) is the hyperbolic sine function which is expressed as (2)

And then, the score function can be given by (3) where cosh(∙) is the hyperbolic cosine function which is expressed as (4)

Another cost function is given by (5)

The score function can be given by (6)

According to the definition of the hyperbolic sine function, strongly convex at e(n)∈(−∞,+∞) are presented as follows

Proof of convex function

The second derivative of J_HSF(e(n)) concerning e(n) is (7)

Based on Eq (7) (8)

So, , i.e., the hyperbolic sine function is strongly convex at e(n)∈(−∞,+∞).

Proof of convex function

The second derivative of J_HSF(e(n)) concerning e(n) is (9)

So, the inverse hyperbolic sine function is strongly convex at e(n)∈(−∞,+∞).

Overall, this means that the adaptive filtering algorithm designed based on these two cost functions can be the better estimation result.

3. Proposed algorithm based on the hyperbolic sine function

We consider a system identification problem in which the desired signal is generated by (10) where ρ(n) is a stationary additive noise with zero mean and variance of . Also, {ρ(n)} is a stationary sequence of independent zero-mean random variables with a finite variance . Furthermore, it is zero odd-order moments and is assumed to be uncorrelated with any other signal. W_OϵR^L×1 is the unknown L-dimensional system parameter vector. X(n) is also stationary with zero-mean, a variance of , and X(n) is Gaussian with a positive-definite autocorrelation matrix R_XX = E[X(n)X^T(n)].

The system estimation error signal can be expressed as (11)

The corresponding filter output is (12)

Assuming (13) where W_O denotes the weight vector of the unknown system with the length of L.

So (14)

3.1 Least hyperbolic sine function

Based on the error signal e(n), various error optimization criteria have been developed in search of an optimal solution. For tractability, using the instantaneous error instead of the expectation in Eq (1) as usual, so in this work, one of the cost function used for obtaining the proposed adaptive filtering algorithm is given by (15)

The gradient term for the optimization algorithm is given as: (16) where cosh(∙) is the hyperbolic cosine function which is expressed as (17)

The iterative scheme is given by (18) where μ is the step-size (or learning-ratio).

3.2 Least inverse hyperbolic sine function

Based on the error signal e(n), various error optimization criteria have been developed in search of an optimal solution. For tractability, using the instantaneous error instead of the expectation in Eq (5) as usual, in this work, another cost function used for obtaining the proposed adaptive filtering algorithm is given by (19)

The gradient term for the optimization algorithm is given as: (20)

The iterative scheme is given by (21) where μ is the step-size (or learning-ratio).

3.3 Computational complexity

Computational complexity is an important property that influences an adaptive filtering algorithm’s performance. The adaptive filtering algorithm’s computational complexity is called the number of arithmetic operations per iteration of the weight vector or coefficient vector. That is the number of multiplications, additions, and et.al. The multiplication operation’s time-consuming operation is far greater than the addition operation’s time-consuming operation, so the multiplication operation occupies a large proportion of the adaptive filtering algorithm’s computational complexity. Therefore, each implementation’s total number of multiplications provided reasonably accurate comparative estimates of their overall complexity. In our proposed HSF and IHSF algorithms, there is no updated step-size formula compared to the VSS-LHCAF [42] and VSS-Llncosh [45] algorithms, meaning that the computational complexity of HSF and IHSF algorithms is smaller than that of the VSS-LHCAF [42] and VSS-Llncosh[45] algorithms. There is also no normalized required to compute, as is the case with the adaptive filtering algorithm with arctangent cost [48]. The HSF and IHSF algorithms also reduce the number of parameters that need to be set. Besides, through comparison, it is also found that there is no λ or λ(n) in HSF and IHSF; that is, there are no artificially set parameters in the algorithms of HSF and IHSF, which further reduces the complexity of the algorithm. For convenience, the weight vector iterative scheme of the VSS-LHCAF [42], VSS-Llncosh [45], adaptive filter with secant cost [46], with arctangent cost [48], the proposed HSF and IHSF algorithms are listed in Table 1.

Download:

Table 1. The weight vector iterative scheme of adaptive filters by using the hyperbolic function.

https://doi.org/10.1371/journal.pone.0258155.t001

4. Simulation results

This section presents the simulation experiments in system identification using various distributions noise of an unknown system, then using those experiments simulation results to see how robust the proposed algorithms are and illustrate the proposed algorithms’ adaptive estimation accuracy. The misadjustment performance of the VSS-LHCAF [42], VSS-Llncosh [45], adaptive filter with secant cost [46], with arctangent cost [48], HSF, and IHSF algorithms are compared. Besides, the input signal X_UC(n) is Gaussian white noise with zero mean and . The correlated input signal X_C(n) is calculated by using X_C(n) = 0.95X_C(n)+X_UC(n). In all of our experiments, the coefficient vectors are initialized as zero vectors. Impulsive distribution noise and Gaussian noise as a measurement noise are used in those experiments. is used to measure the steady-state performance of the adaptive fitering algorithm. The results are obtained via Monte Carlo simulation using 10 independent run sets and an iteration number of 6000.

Then using Figs 1–4 to see how the proposed algorithms outperform the VSS-LHCAF, VSS-Llncosh, adaptive filter with secant cost, with arctangent cost algorithms with the uncorrelated input signal X_UC(n) and the correlated input signal X_C(n) for Gaussian noise and impulsive interference when the correlated and uncorrelated input signal, respectively. Figs 1–4 shows that the proposed algorithms’ convergence rate is faster than that of the VSS-LHCAF, VSS-Llncosh, adaptive filter with secant cost, with arctangent cost algorithms. Moreover, the proposed HSF and IHSF algorithms have a minor steady-state misalignment (i.e., steady-state estimation error equal to SNR). Compared with the VSS-LHCAF, VSS-Llncosh, adaptive filter with secant cost, with arctangent cost algorithms, the proposed algorithms are more robust to Gaussian noise, impulsive interference, and various types of input signals.

Download:

Fig 1. MSEs for the adaptive filtering algorithms by using hyperbolic functions.

(left) with the uncorrelated input signal X_UC(n); (right) with the correlated input signal X_C(n).

https://doi.org/10.1371/journal.pone.0258155.g001

Download:

Fig 2. MSEs for the adaptive filtering algorithms by using hyperbolic functions.

(left) with the uncorrelated input signal X_UC(n); (right) with the correlated input signal X_C(n).

https://doi.org/10.1371/journal.pone.0258155.g002

Download:

Fig 3. MSEs for the adaptive filtering algorithms by using hyperbolic functions.

(left) with the uncorrelated input signal X_UC(n); (right) with the correlated input signal X_C(n).

https://doi.org/10.1371/journal.pone.0258155.g003

Download:

Fig 4. MSEs for the adaptive filtering algorithms by using hyperbolic functions.

(left) with the uncorrelated input signal X_UC(n); (right) with the correlated input signal X_C(n).

https://doi.org/10.1371/journal.pone.0258155.g004

(1). Effect of impulsive interference for a time-invariant system

The system noise contains Gaussian white noise and impulsive interference with SNR = 20dB. The length of the unknown coefficient vector W_O = [0.6,−0.4,0.25,−0.15,0.1,−0.05,0.001]^T is L = 7. Parameters are set as the VSS-LHCAF {μ = 0.0005}, VSS-Llncosh {A₀ = 0.1,h = 0.1,λ₀ = 5,μ = 0.0005}, adaptive filter with secant cost {λ₀ = 0.1,μ = 0.0005}, with arctangent cost {μ = 0.0005}, and the proposed HSF and IHSF algorithms {μ = 0.0005}.

(2). Effect of impulsive interference for a time-varying system

The system noise contains Gaussian white noise and impulsive interference with SNR = 20dB. The length of the unknown coefficient vector L = 7, when iteration number smaller than 3000: W_O = [0.6,−0.4,0.25,−0.15,0.1,−0.05,0.001]^T, when iteration number larger than 3000: W_O = 2×[0.6,−0.4,0.25,−0.15,0.1,−0.05,0.001]^T. Parameters are set as the VSS-LHCAF {μ = 0.0005}, VSS-Llncosh {α₀ = 0.1,h = 0.1,λ₀ = 5,μ = 0.0005}, adaptive filter with secant cost {λ₀ = 0.1,μ = 0.0005}, with arctangent cost {μ = 0.0005}, and the proposed HSF and IHSF algorithms {μ = 0.0005}.

(3). Effect of the input signal

The system noise contains Gaussian white noise with SNR = 20dB. The length of the unknown coefficient vector W_O = [0.6,−0.4,0.25,−0.15,0.1,−0.05,0.001]^T is L = 7. Parameters are set as the VSS-LHCAF, VSS-Llncosh, adaptive filter with secant cost, arctangent cost, and the proposed HSF and IHSF algorithms.

(4). Effect of a more general model

To verify the generalization and evaluate the computation time of the proposed algorithm, the system to be identified adopts the Back and Tsoi NARMA model in [41,49], which is . For a time varying system, when iteration number smaller than 3000, the unknown coefficient vector is W_O, when iteration number is larger than 3000, the unknown coefficient vector is 2W_O. The system noise contains Gaussian white noise with SNR = 10dB. Parameters are set as the VSS-LHCAF {μ = 0.0005}, VSS-Llncosh {α₀ = 0.1,h = 0.1,λ₀ = 5,μ = 0.0005}, adaptive filter with secant cost {λ₀ = 0.1,μ = 0.0005}, with arctangent cost {μ = 0.0005}, and the proposed HSF and IHSF algorithms {μ = 0.0005}.

5. Conclusions

In the context of system identification using the hyperbolic-type function algorithm under Gaussian white noise and impulsive interference, most of them have few parameters that need to be set, thereby increasing the complexity considerably. Besides, the adaptive estimation accuracy and convergence performance can be improved further; the hyperbolic sine function has not been discussed. So, in this paper, a family of the hyperbolic sine function is proposed, including a joint hyperbolic sine function (HSF) and inverse hyperbolic sine function (IHSF) as the cost function. Theoretically, it proves that both of the cost functions of HSF and IHSF are strongly convex at estimation error, and the computational complexity of HSF or IHSF is relatively low. Moreover, multi-types of experimental simulations show that compared with the VSS-LHCAF, VSS-Llncosh, adaptive filter with secant cost, with arctangent cost algorithms, the proposed algorithms are more robust Gaussian noise, impulsive interference, and various types of input signals. In short, that is, our methods could achieve a significantly lower steady-state error theoretically (i.e., MSE equal to SNR) and a faster convergence rate than prior methods (the VSS-LHCAF [42], VSS-Llncosh [45], adaptive filter with secant cost [46], with arctangent cost [48] algorithms) under different scenarios involving both uncorrelated and correlated input. Both theoretical analysis and simulations provided corroborated results. Although the HSF and IHSF algorithms have superior performance, the environment in actual engineering applications is complex nonlinear [50] and time-varying [51]and needs to be adjusted accordingly for different application scenarios.

Supporting information

S1 File. Code.

https://doi.org/10.1371/journal.pone.0258155.s001

(ZIP)

Acknowledgments

The author would like to thank Dr. Chun Meng (The Clinical Hospital of Chengdu Brain Science Institute, MOE Key Laboratory for Neuroinformation, Center for Information in Medicine, School of Life Science and Technology, University of Electronic Science and Technology of China, Chengdu, China.) and Dr. Yang Qiu (Key Laboratory of State Ethnic Affairs Commission for Electronic and Information Engineering, College of Electrical & Information Engineering, Southwest Minzu University, Chengdu, China.) for thoughtful comments on a draft manuscript.

References

1. Cheng S, Wei Y, Sheng D, Wang Y. Identification for Hammerstein nonlinear systems based on universal spline fractional order LMS algorithm. Communications in Nonlinear Science and Numerical Simulation. 2019;79.
- View Article
- Google Scholar
2. Sayed AH. Adaptive Filters: WILEY-INTERSCIENCE: IEEE PRESS; 2008.
3. Haykin SS. Adaptive filter theory. Fifth edition. ed. Upper Saddle River, New Jersey: Pearson; 2014. xvii, 889 pages p.
4. Scarpiniti M, Comminiello D, Scarano G, Parisi R, Uncini A. Steady-State Performance of Spline Adaptive Filters. Ieee Transactions on Signal Processing. 2016;64(4):816–28. WOS:000369438500001.
- View Article
- Google Scholar
5. Sayed AH. Adaptive Networks. Proceedings of the IEEE. 2014;102(4):460–97.
- View Article
- Google Scholar
6. Chen H, Chen J, Muir LA, Ronquist S, Meixner W, Ljungman M, et al. Functional organization of the human 4D Nucleome. Proceedings of the National Academy of Sciences. 2015;112(26):8002–7. pmid:26080430
- View Article
- PubMed/NCBI
- Google Scholar
7. Hitziger S, Clerc M, Saillet S, Benar C, Papadopoulo T. Adaptive Waveform Learning: A Framework for Modeling Variability in Neurophysiological Signals. IEEE Transactions on Signal Processing. 2017;65(16):4324–38.
- View Article
- Google Scholar
8. Guan S, Meng C, Biswal B. Diffusion-Probabilistic Least Mean Square Algorithm. Circuits, Systems, and Signal Processing. 2021;40(3):1295–313.
- View Article
- Google Scholar
9. Diniz PSR. Adaptive filtering: algorithms and practical implementation. 4th ed. New York: Springer; 2013. xxi, 652 p. p.
10. Scarpiniti M, Comminiello D, Parisi R, Uncini A. Nonlinear spline adaptive filtering. Signal Processing. 2013;93(4):772–83. WOS:000315316400015.
- View Article
- Google Scholar
11. Liu W, Prncipe JC, Haykin S. Kernel Adaptive Filtering: A Comprehensive Introduction: WILEY; 2010.
12. Li Z, Guan S. Diffusion normalized Huber adaptive filtering algorithm. Journal of the Franklin Institute. 2018;355(8):3812–25.
- View Article
- Google Scholar
13. Chen J, Richard C, Sayed AH. Diffusion LMS Over Multitask Networks. IEEE Transactions on Signal Processing. 2015;63(11):2733–48.
- View Article
- Google Scholar
14. Zhang Q. Performance enhanced Kalman filter design for non-Gaussian stochastic systems with data-based minimum entropy optimisation. AIMS Electronics and Electrical Engineering. 2019;3(4):382–96.
- View Article
- Google Scholar
15. Sayin MO, Vanli ND, Kozat SS. A Novel Family of Adaptive Filtering Algorithms Based on the Logarithmic Cost. Ieee Transactions on Signal Processing. 2014;62(17):4411–24. WOS:000340846100006.
- View Article
- Google Scholar
16. Wen P, Zhang J. Variable Step-Size Diffusion Normalized Sign-Error Algorithm. Circuits, Systems, and Signal Processing. 2018;37(11):4993–5004.
- View Article
- Google Scholar
17. Ming L, Ming-Jiang W, Bo-Yang S. An efficient architecture of the sign-error LMS adaptive filter. 2016 13th IEEE International Conference on Solid-State and Integrated Circuit Technology (ICSICT)2016. p. 753–5.
18. Ni J, Chen J, Chen X. Diffusion sign-error LMS algorithm: Formulation and stochastic behavior analysis. Signal Processing. 2016;128:142–9.
- View Article
- Google Scholar
19. Zheng Z, Liu Z, Huang M. Diffusion least mean square/fourth algorithm for distributed estimation. Signal Processing. 2017;134:268–74.
- View Article
- Google Scholar
20. Guan S, Li Z. Nonparametric Variable Step-Size LMAT Algorithm. Circuits, Systems, and Signal Processing. 2016;36(3):1322–39.
- View Article
- Google Scholar
21. Guan S, Li Z. Optimal step size of least mean absolute third algorithm. Signal, Image and Video Processing. 2017;11(6):1105–13.
- View Article
- Google Scholar
22. Walach E, Widrow B. The least mean fourth (LMF) adaptive algorithm and its family. IEEE Transactions on Information Theory. 1984;30(2):275–83.
- View Article
- Google Scholar
23. Eweda E. Global Stabilization of the Least Mean Fourth Algorithm. IEEE Transactions on Signal Processing. 2012;60(3):1473–7.
- View Article
- Google Scholar
24. Nascimento VH, Bermudez JCM. Probability of divergence for the least-mean fourth algorithm. IEEE Transactions on Signal Processing. 2006;54(4):1376–85.
- View Article
- Google Scholar
25. Petrus P. Robust Huber adaptive filter. IEEE Transactions on Signal Processing. 1999;47(4):1129–33.
- View Article
- Google Scholar
26. Chen B, Li Y, Dong J, Lu N, Qin J. Common Spatial Patterns Based on the Quantized Minimum Error Entropy Criterion. IEEE Transactions on Systems, Man, and Cybernetics: Systems. 2020;50(11):4557–68.
- View Article
- Google Scholar
27. Chen B, Xing L, Zhao H, Zheng N, Principe JC. Generalized Correntropy for RobustAdaptive Filtering. IEEE Transactions on Signal Processing. 2016;64(13):3376–87.
- View Article
- Google Scholar
28. Chen B, Xing L, Xu B, Zhao H, Zheng N, Principe JC. Kernel Risk-Sensitive Loss: Definition, Properties and Application to Robust Adaptive Filtering. IEEE Transactions on Signal Processing. 2017;65(11):2888–901.
- View Article
- Google Scholar
29. Inacio Hubscher P, Bermudez JCM. An improved statistical analysis of the least mean fourth (LMF) adaptive algorithm. IEEE Transactions on Signal Processing. 2003;51(3):664–71.
- View Article
- Google Scholar
30. Eweda E. Dependence of the Stability of the Least Mean Fourth Algorithm on Target Weights Non-Stationarity. IEEE Transactions on Signal Processing. 2014;62(7):1634–43.
- View Article
- Google Scholar
31. Lim S-J, Harris JG. Combined LMS/F algorithm. Electronics Letters. 1997;33(6).
- View Article
- Google Scholar
32. Gui G, Peng W, Adachi F. Adaptive system identification using robust LMS/F algorithm. Int J Commun Syst. 2013:n/a-n/a.
- View Article
- Google Scholar
33. Gui G, Mehbodniya A, Adachi F. Sparse LMS/F algorithms with application to adaptive system identification. Wireless Communications and Mobile Computing. 2015;15(12):1649–58.
- View Article
- Google Scholar
34. Tiange S, Zheng YR, Benesty J. An Affine Projection Sign Algorithm Robust Against Impulsive Interferences. IEEE Signal Processing Letters. 2010;17(4):327–30.
- View Article
- Google Scholar
35. Principe JC. Information Theoretic Learning: Springer New York; 2010.
36. Al-Sayed S, Zoubir AM, Sayed AH. Robust Adaptation in Impulsive Noise. Ieee Transactions on Signal Processing. 2016;64(11):2851–65. WOS:000374993300009.
- View Article
- Google Scholar
37. Wang G, Xue R, Zhao J. Switching criterion for sub-and super-Gaussian additive noise in adaptive filtering. Signal Processing. 2018;150:166–70.
- View Article
- Google Scholar
38. Bhattacharjee SS, Kumar K, George NV. Nearest Kronecker Product Decomposition Based Generalized Maximum Correntropy and Generalized Hyperbolic Secant Robust Adaptive Filters. IEEE Signal Processing Letters. 2020;27:1525–9.
- View Article
- Google Scholar
39. Liu T, Qiu T, Luan S. Hyperbolic-tangent-function-based cyclic correlation: Definition and theory. Signal Processing. 2019;164:206–16.
- View Article
- Google Scholar
40. Hou W, Chen Z-Q, Zuo D-D, Feng G-l. Drought loss assessment model for southwest China based on a hyperbolic tangent function. International Journal of Disaster Risk Reduction. 2019;33:477–84.
- View Article
- Google Scholar
41. Yang L, Liu J, Yan R, Chen X. Spline adaptive filter with arctangent-momentum strategy for nonlinear system identification. Signal Processing. 2019;164:99–109.
- View Article
- Google Scholar
42. Wang S, Wang W, Xiong K, Iu HHC, Tse CK. Logarithmic Hyperbolic Cosine Adaptive Filter and Its Performance Analysis. IEEE Transactions on Systems, Man, and Cybernetics: Systems. 2019:1–13.
- View Article
- Google Scholar
43. Kumar K, Bhattacharjee SS, George NV. Joint Logarithmic Hyperbolic Cosine Robust Sparse Adaptive Algorithms. IEEE Transactions on Circuits and Systems II: Express Briefs. 2021;68(1):526–30.
- View Article
- Google Scholar
44. Liang T, Li Y, Xue W, Li Y, Jiang T. Performance and Analysis of Recursive Constrained Least Lncosh Algorithm under Impulsive Noises. IEEE Transactions on Circuits and Systems II: Express Briefs. 2020:1-.
- View Article
- Google Scholar
45. Liu C, Jiang M. Robust adaptive filter with lncosh cost. Signal Processing. 2020;168.
- View Article
- Google Scholar
46. Lu L, Chen L, Zheng Z, Yu Y, Yang X. Behavior of the LMS algorithm with hyperbolic secant cost. Journal of the Franklin Institute. 2020;357(3):1943–60.
- View Article
- Google Scholar
47. Liang T, Li Y, Zakharov YV, Xue W, Qi J. Constrained least lncosh adaptive filtering algorithm. Signal Processing. 2021;183.
- View Article
- Google Scholar
48. Yang LD, Liu JX, Yan RQ, Chen XF. Spline adaptive filter with arctangent-momentum strategy for nonlinear system identification. Signal Processing. 2019;164:99–109. WOS:000480665700011.
- View Article
- Google Scholar
49. Back AD, Tsoi AC. A Simplified Gradient Algorithm for IIR Synapse Multilayer Perceptrons. Neural Computation. 1993;5(3):456–62.
- View Article
- Google Scholar
50. Yin X, Zhang Q, Wang H, Ding Z. RBFNN-Based Minimum Entropy Filtering for a Class of Stochastic Nonlinear Systems. IEEE Transactions on Automatic Control. 2020;65(1):376–81.
- View Article
- Google Scholar
51. Guan S, Jiang R, Bian H, Yuan J, Xu P, Meng C, et al. The profiles of non-stationarity and non-linearity in the time series of resting-state brain networks. Front Neurosci-Switz. 2020;14. pmid:32595440
- View Article
- PubMed/NCBI
- Google Scholar

[ref1] 1. Cheng S, Wei Y, Sheng D, Wang Y. Identification for Hammerstein nonlinear systems based on universal spline fractional order LMS algorithm. Communications in Nonlinear Science and Numerical Simulation. 2019;79.
View Article
Google Scholar

[2] View Article

[3] Google Scholar

[ref2] 2. Sayed AH. Adaptive Filters: WILEY-INTERSCIENCE: IEEE PRESS; 2008.

[ref3] 3. Haykin SS. Adaptive filter theory. Fifth edition. ed. Upper Saddle River, New Jersey: Pearson; 2014. xvii, 889 pages p.

[ref4] 4. Scarpiniti M, Comminiello D, Scarano G, Parisi R, Uncini A. Steady-State Performance of Spline Adaptive Filters. Ieee Transactions on Signal Processing. 2016;64(4):816–28. WOS:000369438500001.
View Article
Google Scholar

[7] View Article

[8] Google Scholar

[ref5] 5. Sayed AH. Adaptive Networks. Proceedings of the IEEE. 2014;102(4):460–97.
View Article
Google Scholar

[10] View Article

[11] Google Scholar

[ref6] 6. Chen H, Chen J, Muir LA, Ronquist S, Meixner W, Ljungman M, et al. Functional organization of the human 4D Nucleome. Proceedings of the National Academy of Sciences. 2015;112(26):8002–7. pmid:26080430
View Article
PubMed/NCBI
Google Scholar

[13] View Article

[14] PubMed/NCBI

[15] Google Scholar

[ref7] 7. Hitziger S, Clerc M, Saillet S, Benar C, Papadopoulo T. Adaptive Waveform Learning: A Framework for Modeling Variability in Neurophysiological Signals. IEEE Transactions on Signal Processing. 2017;65(16):4324–38.
View Article
Google Scholar

[17] View Article

[18] Google Scholar

[ref8] 8. Guan S, Meng C, Biswal B. Diffusion-Probabilistic Least Mean Square Algorithm. Circuits, Systems, and Signal Processing. 2021;40(3):1295–313.
View Article
Google Scholar

[20] View Article

[21] Google Scholar

[ref9] 9. Diniz PSR. Adaptive filtering: algorithms and practical implementation. 4th ed. New York: Springer; 2013. xxi, 652 p. p.

[ref10] 10. Scarpiniti M, Comminiello D, Parisi R, Uncini A. Nonlinear spline adaptive filtering. Signal Processing. 2013;93(4):772–83. WOS:000315316400015.
View Article
Google Scholar

[24] View Article

[25] Google Scholar

[ref11] 11. Liu W, Prncipe JC, Haykin S. Kernel Adaptive Filtering: A Comprehensive Introduction: WILEY; 2010.

[ref12] 12. Li Z, Guan S. Diffusion normalized Huber adaptive filtering algorithm. Journal of the Franklin Institute. 2018;355(8):3812–25.
View Article
Google Scholar

[28] View Article

[29] Google Scholar

[ref13] 13. Chen J, Richard C, Sayed AH. Diffusion LMS Over Multitask Networks. IEEE Transactions on Signal Processing. 2015;63(11):2733–48.
View Article
Google Scholar

[31] View Article

[32] Google Scholar

[ref14] 14. Zhang Q. Performance enhanced Kalman filter design for non-Gaussian stochastic systems with data-based minimum entropy optimisation. AIMS Electronics and Electrical Engineering. 2019;3(4):382–96.
View Article
Google Scholar

[34] View Article

[35] Google Scholar

[ref15] 15. Sayin MO, Vanli ND, Kozat SS. A Novel Family of Adaptive Filtering Algorithms Based on the Logarithmic Cost. Ieee Transactions on Signal Processing. 2014;62(17):4411–24. WOS:000340846100006.
View Article
Google Scholar

[37] View Article

[38] Google Scholar

[ref16] 16. Wen P, Zhang J. Variable Step-Size Diffusion Normalized Sign-Error Algorithm. Circuits, Systems, and Signal Processing. 2018;37(11):4993–5004.
View Article
Google Scholar

[40] View Article

[41] Google Scholar

[ref17] 17. Ming L, Ming-Jiang W, Bo-Yang S. An efficient architecture of the sign-error LMS adaptive filter. 2016 13th IEEE International Conference on Solid-State and Integrated Circuit Technology (ICSICT)2016. p. 753–5.

[ref18] 18. Ni J, Chen J, Chen X. Diffusion sign-error LMS algorithm: Formulation and stochastic behavior analysis. Signal Processing. 2016;128:142–9.
View Article
Google Scholar

[44] View Article

[45] Google Scholar

[ref19] 19. Zheng Z, Liu Z, Huang M. Diffusion least mean square/fourth algorithm for distributed estimation. Signal Processing. 2017;134:268–74.
View Article
Google Scholar

[47] View Article

[48] Google Scholar

[ref20] 20. Guan S, Li Z. Nonparametric Variable Step-Size LMAT Algorithm. Circuits, Systems, and Signal Processing. 2016;36(3):1322–39.
View Article
Google Scholar

[50] View Article

[51] Google Scholar

[ref21] 21. Guan S, Li Z. Optimal step size of least mean absolute third algorithm. Signal, Image and Video Processing. 2017;11(6):1105–13.
View Article
Google Scholar

[53] View Article

[54] Google Scholar

[ref22] 22. Walach E, Widrow B. The least mean fourth (LMF) adaptive algorithm and its family. IEEE Transactions on Information Theory. 1984;30(2):275–83.
View Article
Google Scholar

[56] View Article

[57] Google Scholar

[ref23] 23. Eweda E. Global Stabilization of the Least Mean Fourth Algorithm. IEEE Transactions on Signal Processing. 2012;60(3):1473–7.
View Article
Google Scholar

[59] View Article

[60] Google Scholar

[ref24] 24. Nascimento VH, Bermudez JCM. Probability of divergence for the least-mean fourth algorithm. IEEE Transactions on Signal Processing. 2006;54(4):1376–85.
View Article
Google Scholar

[62] View Article

[63] Google Scholar

[ref25] 25. Petrus P. Robust Huber adaptive filter. IEEE Transactions on Signal Processing. 1999;47(4):1129–33.
View Article
Google Scholar

[65] View Article

[66] Google Scholar

[ref26] 26. Chen B, Li Y, Dong J, Lu N, Qin J. Common Spatial Patterns Based on the Quantized Minimum Error Entropy Criterion. IEEE Transactions on Systems, Man, and Cybernetics: Systems. 2020;50(11):4557–68.
View Article
Google Scholar

[68] View Article

[69] Google Scholar

[ref27] 27. Chen B, Xing L, Zhao H, Zheng N, Principe JC. Generalized Correntropy for RobustAdaptive Filtering. IEEE Transactions on Signal Processing. 2016;64(13):3376–87.
View Article
Google Scholar

[71] View Article

[72] Google Scholar

[ref28] 28. Chen B, Xing L, Xu B, Zhao H, Zheng N, Principe JC. Kernel Risk-Sensitive Loss: Definition, Properties and Application to Robust Adaptive Filtering. IEEE Transactions on Signal Processing. 2017;65(11):2888–901.
View Article
Google Scholar

[74] View Article

[75] Google Scholar

[ref29] 29. Inacio Hubscher P, Bermudez JCM. An improved statistical analysis of the least mean fourth (LMF) adaptive algorithm. IEEE Transactions on Signal Processing. 2003;51(3):664–71.
View Article
Google Scholar

[77] View Article

[78] Google Scholar

[ref30] 30. Eweda E. Dependence of the Stability of the Least Mean Fourth Algorithm on Target Weights Non-Stationarity. IEEE Transactions on Signal Processing. 2014;62(7):1634–43.
View Article
Google Scholar

[80] View Article

[81] Google Scholar

[ref31] 31. Lim S-J, Harris JG. Combined LMS/F algorithm. Electronics Letters. 1997;33(6).
View Article
Google Scholar

[83] View Article

[84] Google Scholar

[ref32] 32. Gui G, Peng W, Adachi F. Adaptive system identification using robust LMS/F algorithm. Int J Commun Syst. 2013:n/a-n/a.
View Article
Google Scholar

[86] View Article

[87] Google Scholar

[ref33] 33. Gui G, Mehbodniya A, Adachi F. Sparse LMS/F algorithms with application to adaptive system identification. Wireless Communications and Mobile Computing. 2015;15(12):1649–58.
View Article
Google Scholar

[89] View Article

[90] Google Scholar

[ref34] 34. Tiange S, Zheng YR, Benesty J. An Affine Projection Sign Algorithm Robust Against Impulsive Interferences. IEEE Signal Processing Letters. 2010;17(4):327–30.
View Article
Google Scholar

[92] View Article

[93] Google Scholar

[ref35] 35. Principe JC. Information Theoretic Learning: Springer New York; 2010.

[ref36] 36. Al-Sayed S, Zoubir AM, Sayed AH. Robust Adaptation in Impulsive Noise. Ieee Transactions on Signal Processing. 2016;64(11):2851–65. WOS:000374993300009.
View Article
Google Scholar

[96] View Article

[97] Google Scholar

[ref37] 37. Wang G, Xue R, Zhao J. Switching criterion for sub-and super-Gaussian additive noise in adaptive filtering. Signal Processing. 2018;150:166–70.
View Article
Google Scholar

[99] View Article

[100] Google Scholar

[ref38] 38. Bhattacharjee SS, Kumar K, George NV. Nearest Kronecker Product Decomposition Based Generalized Maximum Correntropy and Generalized Hyperbolic Secant Robust Adaptive Filters. IEEE Signal Processing Letters. 2020;27:1525–9.
View Article
Google Scholar

[102] View Article

[103] Google Scholar

[ref39] 39. Liu T, Qiu T, Luan S. Hyperbolic-tangent-function-based cyclic correlation: Definition and theory. Signal Processing. 2019;164:206–16.
View Article
Google Scholar

[105] View Article

[106] Google Scholar

[ref40] 40. Hou W, Chen Z-Q, Zuo D-D, Feng G-l. Drought loss assessment model for southwest China based on a hyperbolic tangent function. International Journal of Disaster Risk Reduction. 2019;33:477–84.
View Article
Google Scholar

[108] View Article

[109] Google Scholar

[ref41] 41. Yang L, Liu J, Yan R, Chen X. Spline adaptive filter with arctangent-momentum strategy for nonlinear system identification. Signal Processing. 2019;164:99–109.
View Article
Google Scholar

[111] View Article

[112] Google Scholar

[ref42] 42. Wang S, Wang W, Xiong K, Iu HHC, Tse CK. Logarithmic Hyperbolic Cosine Adaptive Filter and Its Performance Analysis. IEEE Transactions on Systems, Man, and Cybernetics: Systems. 2019:1–13.
View Article
Google Scholar

[114] View Article

[115] Google Scholar

[ref43] 43. Kumar K, Bhattacharjee SS, George NV. Joint Logarithmic Hyperbolic Cosine Robust Sparse Adaptive Algorithms. IEEE Transactions on Circuits and Systems II: Express Briefs. 2021;68(1):526–30.
View Article
Google Scholar

[117] View Article

[118] Google Scholar

[ref44] 44. Liang T, Li Y, Xue W, Li Y, Jiang T. Performance and Analysis of Recursive Constrained Least Lncosh Algorithm under Impulsive Noises. IEEE Transactions on Circuits and Systems II: Express Briefs. 2020:1-.
View Article
Google Scholar

[120] View Article

[121] Google Scholar

[ref45] 45. Liu C, Jiang M. Robust adaptive filter with lncosh cost. Signal Processing. 2020;168.
View Article
Google Scholar

[123] View Article

[124] Google Scholar

[ref46] 46. Lu L, Chen L, Zheng Z, Yu Y, Yang X. Behavior of the LMS algorithm with hyperbolic secant cost. Journal of the Franklin Institute. 2020;357(3):1943–60.
View Article
Google Scholar

[126] View Article

[127] Google Scholar

[ref47] 47. Liang T, Li Y, Zakharov YV, Xue W, Qi J. Constrained least lncosh adaptive filtering algorithm. Signal Processing. 2021;183.
View Article
Google Scholar

[129] View Article

[130] Google Scholar

[ref48] 48. Yang LD, Liu JX, Yan RQ, Chen XF. Spline adaptive filter with arctangent-momentum strategy for nonlinear system identification. Signal Processing. 2019;164:99–109. WOS:000480665700011.
View Article
Google Scholar

[132] View Article

[133] Google Scholar

[ref49] 49. Back AD, Tsoi AC. A Simplified Gradient Algorithm for IIR Synapse Multilayer Perceptrons. Neural Computation. 1993;5(3):456–62.
View Article
Google Scholar

[135] View Article

[136] Google Scholar

[ref50] 50. Yin X, Zhang Q, Wang H, Ding Z. RBFNN-Based Minimum Entropy Filtering for a Class of Stochastic Nonlinear Systems. IEEE Transactions on Automatic Control. 2020;65(1):376–81.
View Article
Google Scholar

[138] View Article

[139] Google Scholar

[ref51] 51. Guan S, Jiang R, Bian H, Yuan J, Xu P, Meng C, et al. The profiles of non-stationarity and non-linearity in the time series of resting-state brain networks. Front Neurosci-Switz. 2020;14. pmid:32595440
View Article
PubMed/NCBI
Google Scholar

[141] View Article

[142] PubMed/NCBI

[143] Google Scholar

Figures

Abstract

1. Introduction

2. Hyperbolic sine function

Proof of convex function

Proof of convex function

3. Proposed algorithm based on the hyperbolic sine function

3.1 Least hyperbolic sine function

3.2 Least inverse hyperbolic sine function

3.3 Computational complexity

4. Simulation results

(1). Effect of impulsive interference for a time-invariant system

(2). Effect of impulsive interference for a time-varying system

(3). Effect of the input signal

(4). Effect of a more general model

5. Conclusions

Supporting information

S1 File. Code.

Acknowledgments

References