Learning Based Preamble Collision Detection of Cellular Random Access by Physical Layer Features

This paper proposes an early collision detection scheme based on the physical layer features of user equipment (UE) and convolutional neural networks (CNN). Our scheme can detect preamble collisions at the first stage of the random access procedure of 4G/5G, thus reducing the average RA delay without requiring protocol modification of the UE. We only considered passive detection using the raw signal of the preamble created according to the existing protocol. The simulation results indicate that our proposed scheme can achieve high detection performance under different channels and parameter settings.