W.B. Mikhael

Optimum block adaptive algorithm for gradient based independent component analysis(OBA/ICA) for time varying wireless channels

The fixed-point Independent Component Analysis (ICA) algorithm is widely used because of its fast convergence under static conditions. However, in a highly dynamic environment, it lacks the ability to adapt to the time variation of the mixing matrix. This paper develops an Optimum Block Adaptation ICA algorithm (OBA/ICA) that is capable of tracking time variation. Simulation results for mobile telecommunication applications indicate that the resulting performance, particularly with respect to convergence properties, is superior to Fast-ICA under dynamic channel conditions.

Fidelity enhancement of transform based image coding using non-orthogonal basis images

Existing mixed transform based image coders been formulated for use with separable two-dimensional (2-D) orthogonal transforms such as the discrete cosine transform (DCT) and Haar transform. In this paper, we demonstrate that the mixed transform technique may also be applied to more general collections of non-orthogonal and non-separable basis images. To illustrate the effectiveness of the technique, image representations are constructed from an over-complete set which includes the DCT basis images plus basis images specifically designed to efficiently represent edges. The resulting representations are shown to provide much better visual quality and energy compaction than those constructed from the DCT alone when an equal number of basis images are employed.

Formal development and convergence analysis of the parallel adaptive mixed transform algorithm

Mixed transform techniques represent signals using combinations of basis functions, chosen from two or more transform domains simultaneously, to achieve higher energy compaction than can be achieved using a single transform. The parallel adaptive mixed transform (PAMT) technique has been shown to produce excellent energy compaction and greatly reduced computational burden compared with previous adaptive mixed transform techniques. In this paper, the PAMT algorithm is formally developed and its convergence properties examined. It is shown that convergence of the algorithm can be guaranteed independent of the transforms chosen, as long as those transforms are orthonormal in their own domains.

Two-dimensional, transform-domain, least-square techniques for accurate representation of dominant transform components and noise cancellation

Accurate and efficient representation of 2D signals in the transform domain is very useful for image processing and coding. The authors present a least-square transform domain technique to accurately represent the dominant complex transform components of 2D signals and 2D systems. The algorithm derived here is based on the equationary model (recursive-like structure) which can accurately match the location, magnitude, and phase of complex transform components in the least square sense for the unknown signal. The technique is applied successfully in noise cancellation for image restoration. Examples are given to illustrate the performance of this method in both spectral representation and noise cancellation applications.<<ETX>>

Generation of multidimensional variable step size sequential adaptive gradient algorithms with identification and noise cancellation applications

The development of two-dimensional, gradient-based sequential algorithms with applications in 2-D system identification and noise cancellation is addressed. Existing gradient sequential algorithms use a convergence factor which is used to adjust the two-dimensional adaptive filter coefficients at each iteration. The performance of the algorithm depends entirely on the accuracy of the estimated convergence factor. The objective of the present work is to derive the optimality criterion governing the choice of the convergence factor in the case of 2-D gradient-based sequential algorithms. The 2-D variable step-size sequential algorithms meeting the above constraint are proposed and investigated: the 2-D individual adaptive (TDIA) and the 2-D homogeneous adaptive (TDHA) algorithms. The TDIA algorithm uses optimal convergence factors tailored for each 2-D adaptive filter coefficient at each iteration. The TDHA algorithm uses the same convergence factor for all coefficients but is optimally updated at each iteration.<<ETX>>

An efficient quadratic correlation filter for automatic target recognition

Quadratic Correlation Filters have recently been used for Automatic Target Recognition (ATR). Among these, the Rayleigh Quotient Quadratic Correlation Filter (RQQCF) was found to give excellent performance when tested extensively with Infrared imagery. In the RQQCF method, the filter coefficients are obtained, from a set of training images, such that the response to the filter is large when the input is a target and small when the input is clutter. The method explicitly maximizes a class separation metric to obtain optimal performance. In this paper, a novel transform domain approach is presented for ATR using the RQQCF. The proposed approach, called the Transform Domain RQQCF (TDRQQCF) considerably reduces the computational complexity and storage requirements, by compressing the target and clutter data used in designing the QCF. Since the dimensionality of the data points is reduced, this method also overcomes the common problem of dealing with low rank matrices arising from the lack of large training sets in practice. This is achieved while retaining the high recognition accuracy of the original RQQCF technique. The proposed method is tested using IR imagery, and sample results are presented which confirm its excellent properties.

Speaker recognition employing waveform based signal representation in nonorthogonal multiple transform domains

Automatic speaker recognition (ASR) technique employing split vector quantized speech representation in multiple transform domains is presented. In this approach, a set of appropriate transform domains are selected and a vector quantized codebook is generated in each of these selected transform domains for the signal waveform. For each speaker, each signal vector is represented from the codebooks that yield the highest accuracy of representation. The algorithm is given and a performance measure is developed and used to evaluate the algorithm performance. Improved speech recognition accuracy was consistently obtained employing the proposed technique in comparison with vector quantization employing single transform VQ representations. Sample results for 10 speakers are presented to illustrate the considerable performance improvement for ASR.

Two new model order selection approaches for ARMA system modeling using the two-dimensional frequency domain least square algorithm

The linear algorithm for two-dimensional least square approximation in the frequency domain (2D-FD-LS) is well-established and has been applied successfully to 2D signal representation and image noise cancellation. But it does not give any explicit model order selection method. In this paper, two approaches for model order selection for the 2D-FD-LS algorithm are proposed. One is a rank-checking method, and the other is a grouping measure based method. Analysis and simulation results demonstrate the applicability of these new approaches. The first approach is recommended for deterministic (known spectrum) applications, while the second approach is more suitable for the random case.

Adaptive step-size digital controller for switching frequency auto-tuning

A switching frequency auto-tuning digital controller for a power converter with adaptive step-size is presented in this paper. The controller automatically tracks and finds the switching frequency of a power converter in order to achieve highest power conversion efficiency under variable conditions and load. The adaptive controller utilizes a proposed adaptive variable-step-size function to improve the convergence speed and convergences error. The auto-tuning controller with the proposed adaptive step-size function is discussed and verified in this paper.

A performance comparison of the transform domain Rayleigh quotient quadratic correlation filter (TDRQQCF) approach to the regularized RQQCF

The Rayleigh Quotient Quadratic Correlation Filter (RQQCF) has been used to achieve very good performance for Automatic Target Detection/Recognition. The filter coefficients are obtained as the solution that maximizes a class separation metric, thus resulting in optimal performance. Recently, a transform domain approach was presented for ATR using the RQQCF called the Transform Domain RQQCF (TDRQQCF). The TDRQQCF considerably reduced the computational complexity and storage requirements, by compressing the target and clutter data used in designing the QCF. In addition, the TDRQQCF approach was able to produce larger responses when the filter was correlated with target and clutter images. This was achieved while maintaining the excellent recognition accuracy of the original spatial domain RQQCF algorithm. The computation of the RQQCF and the TDRQQCF involve the inverse of the term A1 = Rx + Ry where Rx and Ry are the sample autocorrelation matrices for targets and clutter respectively. It can be conjectured that the TDRQQCF approach is equivalent to regularizing A1. A common regularization approach involves performing the Eigenvalue Decomposition (EVD) of A1, setting some small eigenvalues to zero, and then reconstructing A1, which is now expected to be better conditioned. In this paper, this regularization approach is investigated, and compared to the TDRQQCF.