Compressed ELM-Based Frame Synchronization

The extreme learning machine (ELM)-based frame synchronization (FS) improves the correctness of classical FS methods to a certain extent. However, it still faces challenges associated with the high complexity of the hidden layer and the difficulty in identifying the first-arriving resolvable path. To tackle the issue of high complexity, this paper leverages the compressibility of FS metrics to construct a compressed ELM network, drawing inspiration from the compression and reconstruction mechanism of compressed sensing (CS). To alleviate the challenge of identifying the first-arriving path in ELM-based FS, a detection threshold is designed to perform a backward search in the output of ELM. Simulation results indicate that the proposed method significantly improves the FS error probability and demonstrates the robustness to parameter variations.