Chair: Douglas Jones, University of Illinois (USA)
Ashwin Rao, University of Rhode Island (USA)
Ramdas Kumaresan, University of Rhode Island (USA)
A procedure to decompose a signal consisting of non-stationary sinusoidal components based on the principles of Residual Signal Analysis [1,2] is proposed. A tracking unit consisting of an all zero filter (AZF) in cascade with a dynamic tracking filter (DTF) is assigned to each component. While the adaptively varying zeros of the AZF suppresses all interfering neighbors, the DTF captures the slowly varying instantaneous frequency (IF) of the desired component. Our earlier methods improved upon Costa's estimator-predictor filter bank [2] by using a better interfering signal predictor. The alternative procedure described in this paper avoids the use of prediction and is not overly restricted by the number of components. We also show that by using simple feedback loops (a loop-filter as in [2] is thus avoided) the tracking information is ensured to be in phase. Finally, the algorithms' ability to decompose synthetic signals, speech signals into harmonic partials as well as tracking the formants present in voiced speech segments is illustrated.
Balasubramaniam Santhanam, Georgia Institute of Technology (USA)
J.H. McClellan, Georgia Institute of Technology (USA)
The Continuous-time Angular Fourier Transformation (AFT) represents a rotation in Continuous time-frequency space and also serves as an orthonormal signal representation for chirp signals. In this paper we present a discrete version of the AFT (DRFT) that represents a rotation in Discrete time-frequency space and some properties of the transform that support its interpretation as a rotation. The transform is a generalization of the DFT. The Eigenvalue structure of the DFT is then exploited to develop an efficient algorithm for the computation of this transform.
H. Oehlmann, CRAN-CNRS-URA (FRANCE)
D. Brie, CRAN-CNRS-URA (FRANCE)
V. Begotto, CRAN-CNRS-URA (FRANCE)
M. Tomczak, CRAN-CNRS-URA (FRANCE)
Nonstationary gearbox vibration signals are analysed using a time-frequency (TF) representation which is chosen with respect to its interpretability. With its help, a crack transient is examined in detail and decomposed into three physical parts. Following this analysis, a time-domain signal is synthesized. Its good phase fitting proves on one hand, the validity of the analysis and on the other hand, the good accuracy of the TF representation chosen.
Tulay Adali, University of Maryland (USA)
M. Kemal Sonmez, University of Maryland (USA)
Kartik Patel, University of Maryland (USA)
We present the general formulation for the adaptive equalization by distribution learning introduced in [Adal 94]. In this framework, adaptive equalization can be viewed as a parametrized conditional distribution estimation problem where the parameter estimation is achieved by learning on a multilayer perceptron (MLP). Depending on the definition of the conditioning event set either supervised or unsupervised (blind) algorithms in either recurrent or feedforward networks result. We derive the least relative entropy (LRE) algorithm for binary data communications and analyze its statistical and dynamical properties. Particularly, we show that LRE learning is consistent and asymptotically normal by working in the partial likelihood estimation framework, and that the algorithm can always recover from convergence at the wrong extreme as opposed to the MSE based MLP's by working within an extension of the well-formed cost functions framework of Wittner and Denker [Wittner 88]. We present simulation examples to demonstrate this fact.
M. Kajiki, Keio University (JAPAN)
J. Xin, Keio University (JAPAN)
H. Ohmori, Keio University (JAPAN)
A. Sano, Keio University (JAPAN)
This paper proposes robust adaptive algorithms for adjusting coefficients of an adaptive filter which is used for feedforward control together with a feedback compensator. The filtered-x algorithm which is widely employed in adaptive signal processing cannot always assure the stability. In this paper, stability-guaranteed adaptive algorithms are given in time domain and frequency domain on a basis of the strictly positive real property of adaptive systems in the presence of unknown disturbances. Time domain algorithms can be applied to an IIR or FIR adaptive filter, while frequency domain approaches use a structure of adaptive frequency sampling filter. Numerical simulations and experimentally obtained results exhibited significant improvement on convergency and stability of the proposed adaptive algorithms in application to adaptive active noise control.
M. Alper Kutay, Bilkent University (TURKEY)
Haldun M. Ozaktas, Bilkent University (TURKEY)
Levent Onural, Bilkent University (TURKEY)
Orhan Arikan, Bilkent University (TURKEY)
The ordinary Fourier transform is suited best for analysis and processing of time-invariant signals and systems. When we are dealing with time- varying signals and systems, filtering in fractional Fourier domains might allow us to estimate signals with smaller minimum- mean-square error (MSE). We derive the optional fractional Fourier domain filter that minimizes the MSE for given non- stationary signal and noise statistics, and time-varying distortion kernel. We present an example for which the MSE is reduced by a factor of 50 as a result of filtering in the fractional Fourier domain, as compared to filtering in the conventional Fourier or time domains. We also discuss how the fractional Fourier transformation can be computed in O(NlogN) time, so that the improvement in performance is achieved with little or no increase in computational complexity.
Hiroshi Kanai, Tokohu University (JAPAN)
Noriyoshi Chubachi, Tokohu University (JAPAN)
Yoshiro Koiwa, Tokohu University (JAPAN)
In this paper, we present a new method to estimate spectrum transition of a nonstationary signals in low signal-to-noise ratio (SNR) cases. If the spectrum transition pattern is complex and/or there are large differences in the transition patterns among the individual nonstationary signals, it is difficult to estimate the transition pattern stably by the previously proposed time-varying AR modeling because the results are considerably dependent on the choice of the basic functions to be used. We propose a new approach of modeling to estimate the spectrum transition of the nonstationary signals by using a linear algorithm without assuming any basic functions. Instead of basic functions we use the spectrum transition constraint. By applying this method to the analysis of vibration signals on the interventricular septum of the heart, noninvasively measured by the method developed in our laboratory using ultrasonic, spectrum transition pattern is clearly obtained during one beat period. The proposed method will serve a tool for the noninvasive acoustic diagnosis of heart diseases in near future.
Masashi Tanaka, NTT Human Interface Laboratories (JAPAN)
Yutaka Kaneda, NTT Human Interface Laboratories (JAPAN)
Shoji Makino, NTT Human Interface Laboratories (JAPAN)
Junji Kojima, NTT Human Interface Laboratories (JAPAN)
This paper reported a fast version of a Projection algorithm whose computational complexity is $2L+20p$ ($L$ is the filter length and $p$ is the projection order), which is much smaller than the $(p+1)L+O(p^3)$ of the conventional algorithm and is comparable to NLMS of $2L$. We also described a step size control method that gives the same steady-state excess MSE for various projection orders. Computer simulations for colored noise and speech input signal confirm the effectiveness of the Projection algorithm and the step size control.
Babak Hassibi, Stanford University (USA)
Thomas Kailath, Stanford University (USA)
H-infinty optimal estimators guarantee the smallest possible estimation error energy over all possible disturbances of fixed energy, and are therefore robust with respect to model uncertainties and lack of statistical information on the exogenous signals. We have recently shown that if prediction error is considered, then the celebrated LMS adaptive filtering algorithm is H-infinty optimal. In this paper we consider prediction of the filter weight vector itself, and for the purpose of coping with time-variations, exponentially weighted, finite-memory and time-varying adaptive filtering. This results in some new adaptive filtering algorithms that may be useful in uncertain and non-stationary environments. Simulation results are given to demonstrate the feasibility of the algorithms and to compare them with well-known H-2 (or least-squares based) adaptive filters.
T. F. Haddad, Jordan University of Science & Technology (JORDAN)
M. A. Khasawneh, Jordan University of Science & Technology (JORDAN)
In this paper, a new fast LMS-based adaptive algorithm is proposed. It is derived by incorporating a damping force in the LMS update recursion in analogy with the force acting upon a damped planar pendulum. An expression for the evolution and the steady-state behaviour for the mean weight vector is developed. This expression provides a mathematical bound which constrains the parameter that controls the maximum contribution of the introduced damping force. Simulation results show an improved robust performance for the new algorithm as compared with the conventional LMS algorithm in smoothly tracking the optimal solution in correlated and nonstationary power environments, especially, in the presence of plant noise.