Signal detection in passive wireless sensor networks based on back-propagation neural networks

Passive wireless sensors with no onboard power source, enable a myriad of embedded sensor applications in hard to reach areas, rotating or moving objects and harsh environments. Structural monitoring, weight-limited aerospace vehicles, smart homes and smart grids are some examples in which usage of wireless sensors are ideal. However, power consumption requirements for active wireless sensors is a limiting factor in wide usage of wireless sensing applications. Passive wireless sensors aim at addressing this shortcoming by eliminating the power source and shifting the complexity to the reader side. In most embedded sensor applications, several sensors are placed in close proximity of each other posing new challenges in the multiuser interference domain. In this study, neural networks are used for signals detection in a multiple-access passive wireless sensor network. In a multiuser system, conventional receivers are based on matched filters suffering severe performance degradation as the relative powers of the interfering signals become large. Despite prior research works in active sensor networks, where channel, code and power control is easy to perform, in this study, passive wireless sensors are considered where there is no control at the node level making signal detection a challenging problem. A novel method for passive wireless sensing employing back-propagation neural network is proposed. A conventional receiver is used as a benchmark to compare the performance of the proposed method. The proposed algorithm demonstrates near-optimum performance in side-lobe suppression, while requiring a much lower implementation complexity compared to conventional receivers.