DESIGN AND ANALYSIS OF A ROBUST CODEC FOR WIRELESS CHANNEL USING MACHINE LEARNING

Abstract

This work presents the design and analysis of adaptive modulation and coding scheme intended for a codec used in short-range wireless communication system; the study aims to investigate and analyse the existing channel prediction parameters for adapting the variation of wireless communication channels. Three modulation and coding schemes are used to develop the codec namely 8QAM, 16QAM with a 1/3 BCH encoder and 4QAM without error correction. MIMO antennas are introduced in the design to increase the spectral efficiency of the codec within the fading channel. To make channel prediction possible a linear discrete channel model is presented based on ARMA processes. The presented scheme is operated at a BER threshold of 10-3 , with a channel correlation of 0.96 and a noise variance of 3 dB. The adaption parameter and the channel model provide the basis of the machine learning model. A conventional neural network (NN) and Long-short term memory (LSTM) predictors are used as prediction models for the adaptation with root means square error (RMSE) of 5.6121 and 3.9157, respectively. a perfect prediction model based on the BER, and adaption parameter is also presented with Zero RMSE as a standard predictor. The result found that the adaptive modulation and codec parameters chosen under the LSTM-NN have a data rate of 15 Mbits/s and 10 Mbit/s for conventional NN predictor. The codec achieved the design requirements, and it can serve users operating within 2.4GHz. However, the design is not a sustainable one as it will soon be outdated due to increasing data rate demands. The data rate of the codec can be improved by using a more robust encoding such as turbo and LDPC codes and transmitting using more antennas to transmit a variety of different data types such as random packets and images.