3:30, NEURAL-L1.1
MINIMAL RESOURCE ALLOCATION NETWORK (MRAN) FOR MAGNETIC RECORDING CHANNEL EQUALIZATION
J. DENG, N. SUNDARARAJAN, P. SARATCHANDRAN
This paper presents the performance results of a recently developed minimal radial basis function neural network referred to as Minimal Resource Allocation Network (MRAN) for equalization of a highly nonlinear magnetic recording data storage channel. Using a realistic magnetic channel model, MRAN equalizer's performance has been studied in the presence of channel impairments like partial erasure, additive white gaussian noise and jitter and width variance. Compared with the earlier neural equalizers, MRAN equalizer has better performance in terms of higher Signal to Distortion Ratios (SDR).

3:50, NEURAL-L1.2
A RECURRENT RBF NETWORK FOR NON-LINEAR CHANNEL
M. MIMURA, T. FURUKAWA
On the conventional method to design the recurrent RBF networks for channel equalizer, firstly, the impulse response of channel is estimated with an adaptive FIR filter. Secondly, all noise free received signals are estimated with the estimate of impulse response. However, the performance of the network designed with this method is degraded down if the channel would be nonlinear. In order to overcome this drawback, we apply Miyake's method to the conventional training method of recurrent RBF networks. In the proposed method, we estimate the received signals of noise free with the relation between the training signals and received signal. Then we design the recurrent RBF networks with noise free received signals.

4:10, NEURAL-L1.3
COMPARISON OF NEURAL NETWORK NATURAL AND ORDINARY GRADIENT ALGORITHMS FOR SATELLITE DOWN LINK IDENTIFICATION
F. LANGLET, H. ABDULKADER, D. ROVIRAS, L. LAPIERRE, F. CASTANIÉ
In this paper, we present a neural network architecture that belongs to the multi-layer perceptron family, associated with two different algorithms: the ordinary gradient and the natural gradient, we compare performances of those algorithms. The identification of a non-normalized power amplifier yielded to the introduction of an additional weight in the classical multi-layer perceptron structure. The application of this network is space telecommunications: identification of satellite communication channels, and especially the down link. This link is made up with two elements. The first one is a high power amplifier (non-linearity). The second one is a filter (memory).

4:30, NEURAL-L1.4
FUZZY RECURSIVE SYMBOL-BY-SYMBOL DETECTOR FOR SINGLE USER CDMA RECEIVERS
J. BAS, A. NEIRA, M. LAGUNAS
The Bayesian or maximum a posteriori (MAP) symbol-by-symbol detector allows minimum BER single user detection in CDMA systems with less memory requirements than the multiuser Maximum Likelihood Sequence Estimator. This work proposes a further complexity reduction by developing a fuzzy recursive implementation of the Bayesian detector. Simulation studies demonstrate the almost optimal performance of the developed fuzzy detector. The resulting system offers a performance v.s.complexity trade-off very appealing for detection in a 3rd generation mobile handset.

4:50, NEURAL-L1.5
AUTOMATIC CLASSIFICATION OF QAM SIGNALS BY NEURAL NETWORKS
S. TAIRA
In this paper, automatic classification of QAM signals including 64-state QAM and 256-state QAM is discussed. Three layer neural networks whose input data is the histogram distribution of instantaneous amplitude at symbol points is used for the classification. The evaluations of classification performance are carried out for both cases in which the synchronization of symbol timing is assured at the receiver and not assured. Good classification results are obtained by the computer simulations at SNR�†10dB. The influence of the number of symbol points which are used for the calculation of histogram is also discussed.

5:10, NEURAL-L1.6
ON-LINE ORDER SELECTION FOR COMMUNICATIONS
T. ADALI, H. NI
We address the problem of on-line order determination for communications and show that penalized partial likelihood criterion provides a suitable likelihood framework for the problem by allowing correlations among samples and on-line processing ability. An on-line, efficient order selection scheme is developed assuming that the observations can be modeled by a finite normal mixture model without imposing any additional conditions on the unknown system, such as linearity. Channel equalization by finite normal mixtures is considered as an example for which correct order determination is critical and examples are presented to show the application and effectiveness of the approach.