Optimal Quantization in Decentralized Detection by Maximizing the Average Entropy of the Sensors

Abstract

In a wireless sensor network the sensor outputs are required to be quantized because of energy and bandwidth requirements. We propose such a distributed detection scheme for a point source which is based on Neyman-Pearson criterion where sensor outputs are quantized maximizing the average output entropy of the sensors under both hypotheses. The quantized local outputs are transmitted to a fusion center (FC) where they are used to make a global decision. The performance of the proposed maximum average entropy (MAE) method in quantizing sensor outputs was tested for binary, ternary and quarternary quantization. The effects of the channel from the sensors to the FC is also addressed by simplified channel models. The simulation studies show the success of the MAE method.

Maximum average entropy-based quantization of local observations for distributed detection

“…We consider scenarios with non-equivalently important hypotheses; that is why the NP criterion is considered more suitable than the probability of error criterion in this work. This paper extends its preliminary version [48] in the following aspects. We compare the proposed method to DC [32,33], CI [33], MI [34,35], CMI [35] and JD [30,31] based methods, demonstrate its positively proportional relation with JD, include increased quantization levels resulting in a similar performance to non-quantized signaling.…”