M. Ksouri

Equalization with decision delay estimation using recurrent neural networks

In this paper, we propose a new recursive classifier based on a recurrent neural network. A supervised algorithm is employed to estimate the classifier parameters. The proposed classifier is used to form a non-linear Decision Feedback Equalizer (DFE) for communication channels. A new procedure allowing the estimation of the decision delay is also presented so that the classifier parameters and the decision delay are estimated at the same time. This new DFE leads to suitable equalization performances even in presence of non-linear and non-minimum phase channels.

Application of Fuzzy Logic to the On-Line Adjustment of the Parameters of a Generalized Predictive Controller

ABSTRACTTo achieve the desired performance and fully exploit the capabilities of generalized predictive control, we propose in this paper a fuzzy supervisory scheme that adjusts on-line the design parameters. This fuzzy logic based strategy is applicable to both regulation and tracking problems. Moreover, it guarantees a suitable degree of closed-loop stability while considering numerical problems (computational burden and matrix inversion). The results of this paper are satisfactory and bring out the remarkable abilities of the fuzzy supervision system. In fact, in all cases, the fuzzy supervisor achieved optimal performance in light of the design objectives which was not the case when it was off-line.

Automatic tuning of GPC synthesis parameters based on multi-objective optimization

In this paper, a strategy for automatic tuning of predictive controller synthesis parameters based on multi-objective optimization (MOO) is proposed. This strategy integrates the genetic algorithm to generate the synthesis parameters (the prediction horizon, the control horizon and the cost weighting factor) making a compromise between closed loop performances (the overshoot, the variance of the control and the settling time). A simulation example is presented to illustrate the performance of this strategy in the on-line adjustment of generalized predictive control parameters.

A numerical approach for performance analysis of higher-order statistics based identification methods

This paper addresses the problem of performance analysis of linear algebra methods for higher-order statistics based identification. We propose another approach which can be used to evaluate the performance of these methods. This approach is based on the use of the condition number of system of equations. To illustrate the effectiveness of our approach, several simulation examples are presented.

New neural Bussgang solutions applied to decision feedback blind equalizer

In this paper, we propose new neural network-based Bussgang solutions reducing the number of equalizer tap delays. These solutions are directed to decision feedback equalizer structures. The two solutions perform decision feedback equalization well for linear and non-linear non-minimum phase channels if a suitable delay is retained. Modified Elman and Jordan neural network-based equalizers lead to the best performances.

Identification of non-minimum phase finite impulse response systems using the fourth-order cumulants

We have presented two contributions for identification of LTI NMP FIR systems. They use the fourth order cumulants of the noisy observations of the system output and consequently yield consistent parameters estimation in the presence of additive Gaussian noise. Both recursive closed-form and batch least squares solutions of the parameters estimation are proposed for each contribution. The second contribution allows to reduce the redundancy of the vector of unknown parameters is developed. Finally, the simulation results showed that the performance of our contributions is better than the other methods.

GGP approach to solve non convex min-max robust model predictive controller for a class of constrained MIMO systems

This paper proposes a new mathematical method to solve min-max predictive controller for a class of constrained linear Multi Input Multi Output (MIMO) systems. A parametric uncertainty state space model is adopted to describe the dynamic behavior of the real process. Since the resulting optimization problem is non convex, a deterministic global optimization technique is adopted to solve it which is the Generalized Geometric Programming (GGP). The key idea of this method is to transform the initial non convex optimization problem to a convex one by means of variable transformations. The main achievement is that the optimal control value found with the GGP shows successful set point tracking and constraints satisfaction. Moreover, an efficient implementation of this approach will lead to an algorithm with a low computational burden. The main features of the new algorithm are illustrated through a MIMO system.

Neural networks and genetic algorithms for dynamic systems control

This paper &ak with the predictive control strategv of non linear dynamic systems based on Artijiiial Neural Networks and Genetic Algorithms (GAS). The Feed Forward Neural Networks ( F F W is used to obtdn the model of the process. The control action is provided by minimiring a control objective which is function of the future prediction oufput and the future control actions. The optimization is carried out using GAS. The proposed control scheme is applied to numerical problems and the simulation results are included

Neural network structures applied for both channel modeling and blind equalization

We propose an approach for blind equalization in digital communication systems. This approach uses a clustering algorithm based on a neural structure. A channel modeling stage is combined to this last approach in order to validate the performance of the equalizer and to guide the choice of its parameters. A neural structure used for the channel modeling task is the modified Elman network. The whole equalization scheme leads to optimal equalization performances especially with linear and nonlinear nonminimum-phase channels.

Constrained nonlinear model-based predictive control using ARX-plus Volterra models

In this paper a nonlinear adaptive constrained model predictive control scheme based on models identified from input-output data is proposed. We consider single input-single output (SISO) nonlinear systems described by ARX-plus Volterra models. The proposed control action is obtained by solving a fourth order nonlinear programming problem online subject to linear constraints on the input signal. The adaptive nonlinear control strategy is obtained by augmenting the non-adaptive controller with an indirect parameter estimation scheme which accounts for unknown and/or slowly time-varying parameters. Simulation case study is used to demonstrate the practical utility of the proposed control scheme and to evaluate its performance.