PARAMETER DRIFT IN LMS ADAPTIVE ALGORITHM

Keywords: LMS algorithm , parameter drift, leakage.
Abstract

This paper examines the conditions under which the LMS adaptive algorithm generates unbounded coefficients .
The drift is not due to the numerical implementation errors or the inadequate choose of the step size parameter. This type of instability may occur due to the unexcited (decaying) excitation and due to a perturbation (e.g. unmodeled components, measure noise).
For a real implementation ( e.g., with a DSP fixed point microcomputer ) this problem must be completed with the stalling phenomena.
Some modified LMS algorithms can be used to remove the parameters drift . A mixed LMS algorithm : the normalized LMS - NLMS combined with the leaky LMS - LLMS can "reject" the parameter drift under the condition that the perturbation has a relative short time duration.

FREQUENCY HOPPING SPREAD SPECTRUM TECHNIQUE FOR WIRELESS COMMUNICATION SYSTEMS

Abstract

The paper presents a wireless communication system that involved the frequency hopping spread spectrum technique M-ary frequency - shift keyed (FH-MFSK) with M = 2. The advantages of the FH are illustrated comparing with a direct sequence (DS) spread spectrum system from the point of view of: interference suppression, multi-user interference suppression and implementation. An implementation with a low cost Digital Signal Processor is presented. The conclusion are that the FH spread spectrum technique is very attractive for wireless communication systems due its resistance to interference, multi -user access and simplicity to implementation with DSP micro controllers.

KEYWORDS: Discrete Wavelet Transform (DWT), dilation, convolution, FIR.

Abstract

This paper presents a compression system witch run successfully at a voice-band rate of 10 kHz, using the DWT.
The paper is organized as follows : section I present a brief introduction in the wavelet theory and emphasized the compression principle, section II illustrated the implementation, in real time, of the compression system using the ADSP2181 microcontroller, section III presents the main results and section IV represents the conclusions.

Abstract

This paper presents the performance evaluation of a genetic algorithm for adaptive IIR filtering, implemented with the floating point DSP processor ADSP21020.
A real time implementation may be obtained using a high speed DSP processor (30 ns instruction cycle and a powerful set of instructions) such as ADSP 21020 microcomputer, for speech applications (e.g., acoustic echo canceller ).

Index terms : adaptive filtering, Infinite Impulse Response (IIR), Least Mean Squares (LMS), genetic algorithm, Digital Signal Processing (DSP) processor.

Abstract

The paper presents an alternative algorithm for transmission of the images on the Internet. The algorithm is fast and permits packet losses. The quality of the reconstructed image is very similar to the original image and the transmission time is decreased comparing with a losses transmission algorithm.
A simulation program, that implements the FLIIT algorithm, was written in order to evaluate the algorithm performance for various compression and loss rates.

DIGITAL SYSTEM FOR ENHANCEMENT OF IMAGES CORRUPTED BY IMPULSIVE NOISE

Abstract

This paper presents a new image enhancement filter that was implemented on a DSP system with ADSP21061 SHARC. The filter is a modified median filter that operates with sub-spaces of the space of neighborhood pixels and involved some shape function in order to improve the image quality. The paper is organized as follows : section I is a briefly introduction in median filtering , section II presents the new filter algorithm, section III illustrates the digital system and implementation considerations (e.g. execution time and complexity) and section IV presents the experimental results, comparing the new enhancement filter with some classical approaches and conclusions.

Abstract

This paper investigates the possibility of transmission of images over a bandlimited channel for multimedia applications (e.g. video conferences). The original image is splitted into several subbands and each subband is transmitted over a bandlimited channel using the CDMA technique.
The paper emphasizes the signal processing algorithm involved in the image transmission. The article is organized as follows: section I represent an introduction , section II illustrated the signal processing algorithm and the global transmission system, section III investigates the implementation in real time using a DSP processor, section IV presents the experimental results and section V represents the conclusions.

Abstract

In a personal communication system two essential requirements are better voice quality and reasonable cost of the communication terminal.
The paper presents a transmission system that improves the speech signal for communication channel corrupted by impulsive noise. The noise cancellation scheme assumes that the speech signal is framed in 5 ms length windows. For each windows the presence of the noise is detected ( using a cyclic redundancy check (CRC) code, overflows detection and the difference between two consecutive speech samples). If a window was determined as a noisy window it is muted by digital attenuation. A real time implementation using the DSP microcomputer ADSP2181 and some experimental results are illustrated.
The conclusions are that the noise cancellation scheme works with good results and the system witch implements it is realizable at low cost.

Abstract

This paper presents a digital system for image compression, in real time, using the wavelet filtering. The system is implemented with a digital signal processor (ADSP2181) and a video codec (ADV601).
The wavelet based compression method processes the whole image in the same time and eliminates the block shaped artifacts that occur when the image is broken in small areas to be processed separately.

Abstract

This paper studies the effect of the LMS coefficients' decorrelation. The main idea is that the adaptive filter uses the coefficient's differences in order to reduce the power dissipation. However this approach may degrade the adaptive filter performance (e.g., speed of convergence and residual error [3] ) due the quantization. The paper is organized as follows : section I - an introduction, section II - model of the LMS decorrelated adaptive filter, section III - the implementation of the LMS decorrelated algorithm and section IV - experimental results.

Abstract

Power control is essential in DS/CDMA to compensate for the differing received power due to the fading and co-channel interference. The conventional feedback power control represents a fixed-step approach determined according to the bang-bang-like control policy. This control scheme suffers from poor system stability, large overshoot and long rise time. An adaptive algorithm can be used in order to tackle this difficulty. This algorithm uses the error between received power and desired power and the error change. This paper describes the adaptive power control algorithm and its implementation using a digital signal processing microcomputer. The paper is organized as follows: section I represents an introduction, section II describes the adaptive power control algorithm, section III illustrated the real-time multi-user implementation of the algorithm, section IV shows the experimental results and section V represents the conclusion.

Abstract

The paper illustrated a real time implementation of an adaptive echo canceller using the joint time-frequency analysis. Comparing with a classical adaptive echo canceller, witch involves an adaptive LMS ( or NLMS) filter the joint time-frequency approach involves a number of adaptive filters. The input of each filter is a subsignal derived from original input signal; these subsignals are the coefficients obtained by a joint time-frequency analysis ( like Gabor expansion). The advantage is that the subsignals are whitened by the joint time-frequency analysis and therefore the adaptive filters work better.

KEYWORDS: voice activity detector (VAD), discontinuous transmission (DTX), average bit rate , digital signal processing (DSP) microcomputer.

Abstract

The paper presents a new and robust voice activity detection algorithm. This algorithm distinguishes between speech and acoustic background noise better than traditional VAD algorithm ( e.g. GSM voice activity algorithm). Moreover, an enhanced VAD is essential in reducing the average bit rate in future generation of digital cellular networks such as Universal Mobile Telecommunications System (UMTS).
The article is organized as follows: section I represent an introduction , section II presents the principle of the new VAD algorithm, section III illustrated the VAD implementation in real time using the ADSP2181 microcomputer, section IV shows the main experimental results and section V represents the conclusions.

KEY WORDS: adaptive noise canceller, adaptive signal enhancement, Digital Signal Processing (DSP), biomedical signal.

Abstract

The biomedical signals are often very small in amplitude. Such signals are affected by large signal contamination ( or artifacts) which make it difficult in distinguish normal biomedical signal from abnormal one. This paper presents a new adaptive filtering technique implementation with Analog Devices ADSP2181 DSP microcomputer. This technique uses both noise cancellation and signal enhancement in a single adaptive system.
The paper is organized as follows: an introduction, a description of the new adaptive technique, the real time multichannel implementation, experimental results and conclusions.
The presented adaptive technique removes the biomedical signals' artifacts very effectively.

Abstract

This paper presents an implementation of an adaptive acoustic echo canceller using a more efficient algorithm - the convex variable step size (CVSS) in order to update the filter coefficients. The CVSS algorithm has better tracking capabilities comparing with other adaptive algorithms (such as LMS or NLMS algorithms). Takes into consideration the fact that the acoustic path may change the adaptive algorithm must be capable to update fast the filter coefficients. On the other hand the CVSS algorithm works better than LMS algorithms with input signals that are not white noise (e.g. speech signals).
The paper has five sections: section I presents the acoustic echo model, section II is a brief presentation of the CVSS algorithm, section III illustrated the implementation of the adaptive cancellation scheme using the ADSP 2189 microcomputer from Analog Devices, section IV presents main results and section V represents the conclusions.

Abstract

This paper presents a digital processing system for image restoration using the ADSP21160 DSP SHARC and ADV601 video codec. The image processing algorithm is a new median filter called progressive switching (PS) median filter.
This algorithm has two parts: an impulse detection part that is used before the second part named filtering part. Both the impulse detection part and the filtering part implement median filters. The algorithm is progressively applied throughout some iterations.
The experiments show that the PS median filter is better than the median filter.

Abstract

This paper evaluates the possibility to use a Digital Signal Processor (DSP) in order to implement an image pattern recognition system based on neural network architecture.
The paper is organized as follows: section I represent a brief introduction in neural networks architectures and how such architecture can be used in pattern recognition, section II present the implementation of the neural network using a very powerful DSP microcomputer (e.g. ADSP 2189 from Analog Devices), section III illustrated the main results for a letters recognition system (execution time and error probability) and section IV represents the conclusions.

Keywords - DSP algorithm, DSP Microcomputer, neural network architecture

Abstract

This paper illustrated a digital system that implements the watermarking technique in real time. The paper is organized in several sections. Section I is a brief introduction witch describes the requirements of watermarking technique and the approaches used to implement it, section II represents the implementation using a digital signal processing (DSP) microcomputer - ADSP21061 from Analog Devices, section III illustrates the main results and section IV represents the conclusion.

PARTIAL UPDATED DISCRETE COSINE TRANSFORM ALGORITHM

Abstract

The paper presents an algorithm based on the discrete cosine transform domain LMS (DCT-LMS) that minimize the computational complexity by partial update of the filter coefficients. The paper compares the full update algorithm (FU-DCTLMS) and the partial update (PU-DCTLMS) algorithm in terms of overall performance (the speed of convergence and the residual error) and the computation time.
The overall performance are very similar for the FU and PU algorithms even the number of updated coefficient is about 30% from the total number of coefficients.
The paper is organized as follows: an introduction, the partial update algorithm description, the evaluation of the computational complexity and the experimental results.

NOISE REDUCTION AND ECHO CANCELLATION SYSTEM

Abstract

The paper presents an algorithm, based on the transform domain approach, used in order to reduce the noise and to eliminate the echo in the telecommunications systems such as hands free telephony. The algorithm has the ability to suppress the noise from the microphone signal as well as the acoustic echo.

SPREAD SPECTRUM WIRELESS COMMUNICATION SYSTEM WITH POWER SAVING
PSEUDO-NOISE CODE

Abstract

Wireless communication systems request power and bandwidth efficient systems. The paper illustrated a spread spectrum system that involved an efficient pseudo noise code (named power saving pseudo noise - PSPN code) that provides both power and bandwidth efficiency. The paper is organized as follows: a short introduction, the PSPN definition and properties, the searching algorithm for PSPN code, experimental results and conclusion.
The SS - PSPN system offer better performances (e.g. power consumption and bandwidth) and it is very suitable to be implemented in CMOS VLSI technology.

DIGITAL SIGNAL PROCESSOR SYSTEM FOR ACTIVE NOISE CONTROL WITH IMPROVED SECONDARY PATH ESTIMATION

Abstract

The paper presents a DSP system for active noise control. The secondary path modeling is improved using insertion of additive noise. The system works in real time (with a DSP microcomputer such as ADSP21060 - Analog Devices). The paper is organized as follows: section I represents a introduction in noise control system, section II presents the approach for modeling the secondary path that improves the system performance, section III illustrated the implementation and the main results and section IV presents the conclusions.

Keywords: Active noise control, secondary path, Digital Signal Processor (DSP)

Abstract

The paper presents a hierarchical least mean square (HLMS) algorithm where the taps of the filter are structured into a hierarchy. The minimization process is performed from the bottom level to top level. The performance in terms of speed of convergence and residual error is better than classical LMS algorithm.
The HLMS algorithm was implemented using a digital signal microcomputer from Analog Devices (ADSP21160).
The paper is organized as follows: section I is a brief introduction, section II presents the HLMS algorithm, section III presents the implementation of the HLMS and section IV illustrated the main results. Section V presents the conclusion.

Key words: Hierarchical filter, Least mean squared (LMS), Hierarchical LMS (HLMS) filter , Digital signal processor (DSP).

DIGITAL SIGNAL PROCESSING IMPLEMENTATION OF AN IMPROVED WEIGHTED MEDIAN FILTER

Abstract

This paper presents an improved weighed median filter that allows negative weights. The improved median filter has spectral characteristics very closed to their corresponding FIR (both low pass and high pass filters) but the weighed median filter are more robust in presence of impulsive noise.
Comparing with linear filters (FIR) the computation time is increased for the weighed median filters but the system may works still in real time.

CDMA SYSTEM WITH ADAPTIVE NONLINEAR COMPENSATION

Abstract

The paper presents a CDMA system with adaptive compensation for nonlinear transmitter.
The CDMA system may be easily implemented using the Analog Devices microcomputer (floating point) ADSP21160.
Two cases are studied: the first one using a linear amplifier and the second one using a non-linear amplifier. The pre-distortion filter compensates both the non-linearity in the transmission path and the channel fading. The adaptation algorithm is a normalized Least Mean Squares (NLMS) algorithm.

Keywords
Digital Signal Processor (DSP), NLMS adaptation algorithm, non-linear, CDMA, FIR.

IMPROVED MEDIAN FILTER FOR IMPULSE NOISE REMOVAL

Abstract

The paper presents a simpler and efficient median filter used for removing the impulsive noise from the images. A real time implementation using a Digital Signal Processor (DSP) is also presented. The improved median filter presented has better performance than the classical median filter with no very great additional computational effort.

Keywords: median filter, corrupted image, DSP, impulse detector.

VARIABLE STEP-SIZE LMS ADAPTIVE FILTER

Abstract

This paper presents a VSS-LMS adaptive algorithm that updates the step-size parameter taking into consideration the input signal power and the average residual error. That is the step-size parameter will have a local component (for each filter coefficient and given by the input signal power and a global component ( given by the average of the residual error). The adaptive filter performance is significantly improved.

Keywords: Least Mean Square (LMS) algorithm, variable step size (VSS), global and local component, DSP processor, real time implementation.

IMPROVED ADAPTIVE IIR FILTER

Abstract

The paper presents an improved IIR adaptive filtering algorithm. This algorithm eliminates the measurement noise at the unknown system output. The IIR filter structure is based on two FIR adaptive filters. The main idea is that the FIR filters are excited with signals close to white noise even the measurement noise is colored.