Decision Fusion for Personalized Recommendation in Context-Aware Wireless Sensor Networks

The over-the-air computation-based distributed decision fusion is studied for context-aware wireless sensor networks (WSNs) under multiple-access channels (MAC) for personalized recommendation. The impact of the multi-routing multi-relay transmission link is investigated. Clearly, over-the-air computation that utilizes the wireless channel coverage characteristics to directly implement the objective function computation is the preferred choice for complex e-commerce systems, which not only fits the low-latency and high-reliability performance, but also saves the bandwidth resources. Lowcomplexity decision fusion rules with only small amount of or no channel state information (CSI) are critical for rapid decision making in resource-constrained WSNs, and considering the effect of relays with unknown channel characteristics is scarce in related research to wireless computing. Precisely, we first establish the optimal decision fusion rule based on the loglikelihood ratio (LLR), assuming that the fusion center is fully aware of the CSI. Secondly, three suboptimal decision fusion rules are corresponding given. In the first scenario, large crossover probability is assumed and the logarithmic operation is successfully avoided in the decision fusion scheme. Then, the Chair-Varshney (CV) rule is achieved, wherein the channel condition is assumed to be good enough. In the third scenario, ideal local channel is set, and the idea from the equal gain combining (EGC) is then successfully borrowed. Finally, the exact expression for the detection error probability (DEP) of the fusion center is given. Our results show that the detection performance under odd number of transmission paths is always better than that under even number. The final simulation results match the performance analysis.