Multiterminal Source Coding for Many Sensors with Entropy Coding and Gaussian Process Regression

Abstract

Summary form only given. In this paper, we take a different approach from the coding community. Instead of taking the usual route of quantization plus Slepian-Wolf coding, we do not perform any Slepian-Wolf coding on the transmitter side. We simply perform quantization on the sensor readings, compress the quantization indexes with conventional entropy coding, and send the compressed indexes to the receiver. On the decoder side, we simply perform entropy decoding and Gaussian process regression to reconstruct the joint source. To reduce the sum rate over all sensors, some sensors are censored and do not transmit anything to the decoder.