Codificação analógica conjunta fonte-canal com conjuntos não-gaussianos de amostras de sensores

Abstract

This work proposes an analog joint source-channel coding scheme as an alternative for machine-type communications that require low communications delay. The analog coding scheme is based on mapping the symbols generated by an information source into a curve. The mapping curve can be defined by geometric mappings or by numerical optimizations, denoted by parametric and non-parametric coding, respectively. In this work, we have implemented these two types of coding for the non-Gaussians sets of samples collected by sensors. Considering that the parametric scheme is optimal for Gaussian sources only, the Box-Cox transformation has been used to improve the performance. On the other hand, the non-parametric scheme can easily adapt to different sample sets, at the cost of more complex implementation. The numerical results show that the non-parametric scheme achieves better performance, in general. Also, we can observe that Box-Cox transformation is a simple technique that improves significantly the performance of the parametric scheme. Moreover, for certain values of the channel signal-to-noise ratio, the performance of the parametric scheme is similar, which justifies its choice due to the lower complexity.