Radar Spectrum Allocation for Vehicular Networks with QMIX-LSTM Network

To ensure driving safety, radar detection is an essential function of autonomous vehicle. However, with the increasing number of automotive radars and limited spectrum resources, the co-channel interference among radars seriously affects the detection performance. Spectrum allocation is a representative method for interference elimination. However, the centralized scheme suffers from long latency and is not applicable to delay-sensitive scenarios. In this paper, we construct a noval signal mutual interference model, and build the Decentralized Partially Observable Markov Decision Process (Dec-POMDP) framework. In particular, the QMIX-LSTM algorithm based on Centralized Training with Decentralized Execution (CTDE) architecture is used for spectrum allocation to mitigate the mutual interference and improve the detection probability. Simulation results show that the proposed scheme achieves a higher radar detection probability compared with myopic scheme.