Multiterminal source coding of Bernoulli-Gaussian correlated sources

This paper presents a practical coding scheme for the direct symmetric multiterminal source coding problem with remote source, when the noise between the remote source and the observations is Gaussian-Bernoulli-Gaussian. The idea behind the design is to take advantage from the observed symbols being in the real field, in order to perform low-dimensional compressed sensing. The coding scheme is based on channel coding techniques involving BCH codes over the real field. Simulations with respect to the rate distortion performance are provided, along with insights about optimization issues. Robustness to variations of the probability of impulse is also investigated. Perspectives about the application to the CEO problem in presence of Gaussian-Bernoulli-Gaussian correlation noise between the remote source and the observations conclude the paper.