On Side-Informed Coding of Noisy Sensor Observations
Chao Yu, Gaurav Sharma · 2018 52nd Asilomar Conference on Signals, Systems, and Computers · 2007
In this paper, we consider the problem of side-informed coding in applications where the data available at the encoder consists of indirect noisy observations of the signal desired at the decoder. In these scenarios, under Gaussian statistics we show that for a mean-squared distortion metric, the side-informed encoding problem can be decomposed into a "side-informed" minimum mean-squared error (MMSE) estimation followed by side-informed coding of the MMSE estimate, without incurring any rate-distortion penally. By recursively exploiting this decomposition, we develop a sequential framework for side-informed coding in multi-sensor networks, where each sensor observes linear noise-corrupted measurements. We construct a practical realization of this encoder using a Karhunen-Loeve transform with 1 -D scalar coset codes. Simulations demonstrate that simple code constructions based on the estimate-then-code partitioned structure provide improvements over their counterparts that perform the encoding directly without a pre-processing estimation step.