Active Sensing for Communications by Learning

This paper proposes a deep learning approach to a class of active sensing problems in wireless communications in which an agent sequentially interacts with an environment over a predetermined number of time frames to gather information in order to perform a sensing or actuation task for maximizing some utility function. In such an active learning setting, the agent needs to design an adaptive sensing strategy sequentially based on the observations made so far. To tackle such a challenging problem in which the dimension of historical observations increases over time, we propose to use a long short-term memory (LSTM) network to exploit the temporal correlations in the sequence of observations and to map each observation to a fixed-size state information vector. We then use a deep neural network (DNN) to map the LSTM state at each time frame to the design of the next measurement step. Finally, we employ another DNN to map the final LSTM state to the desired solution. We investigate the performance of the proposed framework for adaptive channel sensing problems in wireless communications. In particular, we consider the adaptive beamforming problem for mmWave beam alignment and the adaptive reconfigurable intelligent surface sensing problem for reflection alignment. Numerical results demonstrate that the proposed deep active sensing strategy outperforms the existing adaptive or nonadaptive sensing schemes.

I. Introduction

The active sequential learning problem naturally arises in many complex inference, sensing, and control settings, e.g., tree-search [1], sequential design of experiments [2] (an example of which is the game of twenty questions [3]), and the multi-armed bandit [4] problems. These problems typically involve an estimation or adaptive control task which is based on sequential sensing of the environment—in the sense that instead of passively collecting all the observations of the environment at once, the agent seeks to actively and sequentially query the environment to build up its knowledge about the problem instance, in order to produce a final optimal solution. In other words, at each time step of the query procedure, the agent can adaptively design its sensing strategy based on the information obtained from all the previous steps. The design of active sensing strategies is a highly nontrivial task. The goal of this paper is to develop a unified deep learning framework to design active sensing procedures in a data-driven way.