Boundary Detection via Deep Learning for Grant-Free Asynchronous Random Access in Control-to-Control Industrial Networks

This paper addresses grant-free, asynchronous control-to-control (C2C) communications over a shared wireless channel. Controllers transmit commands of variable length, encoded using low-density parity-check codes and encapsulated in self-contained packets that include a preamble and a tail sequence. In the absence of a global time reference, each controller operates independently and sends replicas of its packet within a locally defined virtual frame to improve reliability. The receiver has no access to metadata and must detect message boundaries (i.e., preamble and tail sequences) directly from the received signal. We propose a multi-branch, multi-label convolutional neural network that jointly detects preamble and tail sequences, enabling fully data-driven boundary identification. Simulation results show that the proposed receiver enables robust boundary detection under high traffic load, supporting dense and uncoordinated C2C deployments.