Cloud Processing versus Independent Processing of Independent Data Sets for Distributed Detection

Abstract

A distributed detection problem where sensors are deployed to observe a common source of interest is studied. For centralized processing, decision is made by utilizing all the data collected at the sensors, which takes more resources of transmission and computation. For independent processing, it takes less resource at the cost of some performance loss. Motivated by the recently proposed cloud radio access network (C-RAN) and cloud radar, this paper proposes the cloud processing, where each sensor directly compresses (quantizes) its data and the central processor makes decision through all the compressed data. To model the quantization effects, the additive quantization noise model (AQNM) is adopted. Then, the performances of the generalized likelihood ratio test (GLRT), independent GLRT (IGLRT) and cloud GLRT (CGLRT) through deflection coefficients are analyzed. We especially focus on the performance comparison of cloud processing and independent processing, which depends on the number of sensors M, the variances of the additive quantization noise $\sigma _{\text{q}}^2$ and the additive noise σ2. Numerical results are conducted to verify the analysis.