Development of a neuroinference engine for ADSL modem applications in telecommunications using an ANN with fast computational ability

Abstract

Artificial neural networks (ANNs) can be attractive means to achieve smart management efforts in modern telecommunications. Proposed in this paper, is an ANN with fast computational abilities compatible for almost real-time efforts as needed in allocating subschannels to the incoming down-stream of information in the so-called asymmetric digital subscriber line (ADSL) modems. The neuroinference engine facilitated in the proposed ADSL modem is a feedforward, backpropagation-based ANN in which the training and prediction phases are carried out in the entropy (information-theoretic) plane. The ANN is trained by means of an algorithm so as to judiciously allocate the incoming down-stream of data (at the modem) into subchannels of the physical media (on the subscriber access side). The underlying training performed on the ANN enables recognizing the input pattern of the data flow rate (at the ADSL modem) and matching it with the output pattern of access-line subchannel capacity specified by the signal-to-noise characteristics of the physical media. Simulated results are presented and discussed.