Interference Characterization in Wireless Networks: A Determinantal Learning Approach

Owing to its tractability, stochastic geometry (SG) has recently gained immense popularity for the modeling and analysis of wireless networks. However, given their simplicity, the state-of-the-art SG-based models are not well-equipped to capture complex interactions that exist in actual wireless networks. On the other hand, this problem of modeling interactions in actual data patterns has been explored in machine learning (ML) community from different perspectives. In particular, the idea of determinantal point process (DPP) from SG is used extensively in ML to solve a general class of subset selection problems with diversity. Since subset selection also arises naturally in wireless networks, the DPP learning (DPPL) framework can be leveraged to develop a new class of learning-enabled analytical models using SG. In this paper, we explore this idea through a case study. In particular, we use the DPPL framework for characterizing the interference distribution in a wireless local area network (WLAN) with carrier sense multiple access (CSMA) where a subset of the transmitters (Tx-s) are active in a given time instant. In particular, we model the locations of the active Tx-s as a DPP and learn its distribution from a sequence of network realizations. Numerical results demonstrate that our analytical characterization of interference based on DPP learning (DPPL) is remarkably accurate.