Hardware implementation of Distributed Learning Algorithm for mapping selection for Wireless Physical Layer Network Coding

A wireless relay node employing Wireless Physical Layer Network Coding (WPLNC) must use a specific mapping in order to combine incoming signals. This mapping, however, cannot be selected arbitrarily. Together with the signals from the other network relays, it has to allow the destinations to be able to recover the source data from the available observations. Moreover the mapping should optimize a local relay utility function. This task can be easily solved in centralized networks. In decentralized ones, such as sensor or smart metering networks, a mapping assignment should be derived from mutual node communication, cooperation and/or signaling. In this paper we focus on the practical hardware implementation of such a distributed algorithm called a Distributed Learning Algorithm (DLA). In a two source, two relay and two destination network scenario we have implemented a non-cooperative game-based process that selects the WPLNC mapping of each individual relay node guaranteeing invertibility of WPLNC at the destinations as well as optimizing the relay´s utility function, namely the output modulation cardinality. The implementation testbed is based on Software Defined Radio (SDR) modules and it is used to verify the algorithm properties in real-world conditions.