A new algorithm for efficient linear convolution of real signals is presented. It is shown that the circulant required in traditional overlap-and-save (OLS) and overlap-and-add (OLA) methods can be substituted by aj-circulant, that is, a circulant matrix where the shifted elements are multiplied by the imaginary unit. Suchj-circulant can be implemented easily and efficiently with half-length complex Fast Fourier Transforms. The latency remains the same as that of OLS and OLA. This method results in computational savings when compared to OLA and OLS, reducing the total arithmetic operations and particularly the execution time.

Discrete multitone (DMT) systems need the use of a guard interval, known as cyclic prefix (CP), to avoid intersymbol interference (ISI) and interchannel interference (ICI). Unfortunately, this prefix produces a decrease in the achievable bandwidth efficiency. To overcome this problem, we present in this paper an IIR linear precoder that adds memory in the time-domain, avoiding the need of a guard time. It is shown that the received signal with this precoder and the received signal in a DMT system that uses CP are identical, so perfect reconstruction is possible in a noise-free situation. In addition, the behavior of this scheme in presence of noise is the same as in a DMT system with CP, allowing the use of bit and power loading algorithms. It is also proved that the only condition for the precoder stability is that the channel is minimum-phase. As an example, it is shown the good performance of this algorithm in an ADSL environment.

In this paper, we present an algorithm that minimizes the mutual information between the outputs of a multilayer perceptron (MLP). The neural network is then used as separating system in the nonlinear blind source separation problem. It has been reported that MLPs with one hidden layer are sufficient to achieve desirable performance. However, in some cases, we may prefer approximating nonlinear mappings by using networks with several hidden layers. For the sake of simplicity, the present paper is focused on MLPs with two hidden layers. The performance is illustrated by some experiments.

A practical N-user arbiter and its implementation are presented in this correspondence. Because of the asynchronous character of its input variables (request signals), the design proposed is asynchronous and keeps in mind the possibility of metastable operations. The structure of the arbiter is very simple and modular.

This correspondence proposes a generalization of the normal DON'T CARE condition and develops the input and next-state equations for a R-S type M-stable. An example is used to clarify the application of such a device to sequential machine inplementation.

A self-synchronization model is developed. It is based on a general clock function which depends only on the state variables. This function is associated with the flip-flop type to be used and is independent of the flow table. A modular implementation which uses integrated circuits is obtained. With this technique, the synthesis procedures for synchronous machines can be directly applicable to asynchronous circuits without taking care of the critical races and essential hazards.