Collision-Aware Decision Fusion in Distributed Detection Using Reliability-Splitting Algorithms

We consider a large, single-hop, wireless sensor network performing distributed detection under time and bandwidth constraints. The fusion center (FC) has a limited time to collect binary local decisions, which are sent through a shared transmission channel. Assume that the allocated collection time is not large enough to collect the local decisions from all sensor nodes. The considered distributed detection applies a sensor selection strategy called the reliability-based splitting algorithm to collect the local decisions. By using this strategy, the sensor nodes are divided into groups according to their observation reliabilities. The sensor nodes in the same group will compete for packet transmissions with each other by exploiting slotted ALOHA. In addition, we propose a modified algorithm called the two-level reliability-based splitting algorithm, where the sensor nodes are divided into groups based on both observation reliabilities and local decisions. Unlike other distributed detection with random access protocols, where the packet collisions are treated as errors and ignored, we derive collision-aware fusion rules, where the numbers of successful and collision time slots are also exploited in making a global decision. The numerical results show that both algorithms outperform TDMA-based distributed detection.