Deep Learning-Based Grant-Free Massive Access with Age-of-Information Minimization

Existing studies in grant-free random access (GF-RA) for massive machine-type communications focus mainly on improving the accuracy of user detection and data recovery, treating all devices as equally important. However, in many emerging Internet of Things applications like smart city and smart agriculture, the information freshness becomes a major concern. In this paper, we concentrate on minimizing the system average age of information (AoI) in GF-RA. Specifically, we first analyze the Age-based Random Access (ARA) scheme, in which each IoT device accesses the channel with a certain probability only when its instantaneous AoI exceeds a predetermined threshold, and optimize the access parameters according to the derived average AoI (AAoI). Then, we design an autoencoder to jointly minimize the system AAoI, together with learned pilots used in GF-RA. After deriving the optimal threshold, the decoder named Age-based Learned Iterative Shrinkage Thresholding Algorithm (LISTA-AGE) is proposed to utilize the AoI of all devices as prior information to enhance active user detection. Experiments demonstrate the advantage of the proposed method in minimizing the system AAoI.