El Houssaïn Baghious

Overlap-Save and Overlap-Add Filters: Optimal Design and Comparison

Overlap-save (OLS) and overlap-add (OLA) are two techniques widely used in digital filtering. In traditional OLS and OLA implementations, the system is compelled to be time-invariant and conventional filter synthesis techniques are used for designing the block filter. In this paper, based on the OLS and the OLA structures, we develop a fast algorithm for designing the optimal OLS and OLA block filters using a quadratic criterion. Comparing OLA to OLS optimal design, we demonstrate that, as in classical design approaches, they show no difference when the filters are time-invariant. However, when aliasing is not zero, although the global aliasing is the same, its components with respect to frequency are different. This conclusion is supported by simulation results, and a comparison between the optimal approach and some other standard approaches is also provided.

Linear predictive speech coding using fermat number transform

This paper is about the reduction of the computational complexity of a speech codec. A linear predictive coding procedure is developed to allow its implementation with number theoretic transforms. The use of fermat number transform can reduce, in a significant way, the cost of linear predictive algorithm implantation on digital signal processor.

Realization of Multi-Delay Filter Using Fermat Number Transforms

This paper is about an efficient implementation of adaptive filtering for echo cancelers. The first objective of this paper is to propose a simplified method of the flexible block Multi-Delay Filter (MDF) algorithm in the time-domain. Then, we will derive a new method for the step-size adaptation coefficient. The second objective is about the realization of a Block Proportionate Normalized Least Mean Squares (BPNLMS++) with the simplified MDF (SMDF) implementation. Using the new step-size method and the smaller block dimension proposed by SMDF, we achieve a faster convergence of the adaptive process with a limited computational cost. Then, an efficient implementation of the new procedure (SMDF-BPNLMS++) block filtering is proposed using Fermat Number Transform, which can significantly reduce the computation complexity of filter implantation on Digital Signal Processor.

Fast Algorithm for Optimal Design of Block Digital Filters Based on Circulant Matrices

Block digital filtering has been suggested to reduce the computational complexity and to increase the parallelism of computation in digital filtering systems. In this letter, a fast algorithm for optimal design of block digital filters (BDFs) is developed. This algorithm, based on circulant, Toeplitz and shift cyclic matrices, not only reduces the computational complexity of the design process but also decreases the memory requirements.

Fast pitch modelling for CS-ACELP coder using fermat number transforms

This paper presents a double improvements to reduce the speech coding complexity of the pitch prediction in a code-excited linear prediction (CELP) coder. First, the pitch analysis structure is modified. A new fast pitch modelling by linear-filtering (PMF) procedure will determine the adaptive and stochastic codebook contributions of the excitation signal. Afterwards, an efficient implementation of the PMF-CELP coding is proposed by using number theoretic transforms which can significantly reduce the algorithm computation complexity.

Optimal generalized design of transform-based block digital filters

Transform-based block implementation of digital filters is useful for high throughput filtering due to inherent parallelism and complexity reduction provided by using the fast transforms. In basic form, for example the overlap-save implementation, the block digital filter (BDF) is represented by a vector. In this paper, the basic form of block filtering and the optimal design of BDF are described. Therefore, we propose a generalization of the block digital filtering where the BDF is represented by a matrix. This generalized form and its corresponding optimal BDF design are developed. The generalized BDF allows reducing the global distortion of the block filtering.

Design of a variable rate algorithm for CS-ACELP coder

This paper is about the reduction of the computational complexity of the CS-ACELP codec, described in ITU recommendation G.729, and used for the transmission of voice over IP. A voice activity detection module is proposed to replace the G.729 Annex B algorithm. The new procedure was developed to allow its implementation with number theoretic transforms. The use of Fermat number transforms can reduce the cost of variable rate algorithm implementation on digital signal processor (DSP).