Collision Detection and Load Estimation for Massive Random Access in Satellite-Based Internet of Things: A Deep Learning Approach

Satellite communications have been regarded as a promising solution to be incorporated in future Internet of Things (IoT) to support continuous and ubiquitous connectivity services. Constrained by limited contention resources, the conventional random access (RA) scheme will suffer from severe overload problem when applied to the emerging satellite-based IoT. Focusing on improving access efficiency in the massive and concurrent access scenarios, we propose a sample feature enhancement based collision detection and load estimation scheme with the aid of deep learning. Specifically, by leveraging the inherent characteristics of correlation results in case of preamble collision, a feature extraction method is first designed to precisely screen important sample features related to the current load information. Then, a multi-feature enhancement network with an adaptive neuron optimization strategy is further proposed to enable original 1D features mapped to a 2D domain, so as to improve the representation capability of the model while preventing overfitting. Simulation results validate the feasibility of our scheme in large-scale RA collision scenarios and demonstrate its remarkable performance superiority in terms of load estimation accuracy and collision detection probability over the state-of-the-art schemes.