Quantization for distributed detection under link outages

Abstract

In this paper, we address the design of binary sensor quantizers for decentralized detection in wireless sensor networks (WSN). The objective is to develop robust quantizers in the events of link outages and/or sensor failures for a K-sensor fusion model (K > 2). To this end, we propose two different design procedures. The first is formulated as a constrained optimization problem and achieves guaranteed performance given any subset of local decisions available at a fusion center. The second approach is a channel-aware design using an explicit erasure channel model, which provides a balanced detection performance under transmission loss. We show that for both designs the optimal quantizers are in the forms of likelihood ratio tests. The corresponding thresholds can be determined through iterative algorithms. We further demonstrate that the channel-aware approach is more practical and suitable for large-scale WSN applications in terms of low implementation complexity.