Neural network quantifier for solving the mixture problem and its implementation by systolic arrays

In this paper we present the application of a method, based on the orthogon al transformation, to develop an optimal neural network for solving the Mix ture Problem and a linear systolic to design it is provided. We use a back- propagation neural model for determining and quantifying the components in a composite spectrum obtained from a given mixture of elements. The spectra of the possible components are used as the training patterns. The orthogon al transformation used in the present work are the singular value decomposi tion (SVD) and the QR with column pivoting factorization (QRcp). An interes ting property of the proposed method is related to the possibility of reduc ing the input and hidden nodes at least to the number class. This reduction allows us to obtain an optimum VLSI implementation by a linear systolic. ( C) 1998 Elsevier Science Ltd. All rights reserved.