Parallel ELM-Based Timing Synchronization in OFDM Systems

In orthogonal frequency division multiplexing (OFDM) systems, initial feature extraction (IFE) approaches that rely on either auto-correlation or cross-correlation can only suppress either noise or carrier frequency offset (CFO) interference. This limitation degrades the correctness of extreme learning machine (ELM)-based timing synchronization (TS) methods in practical wireless communication scenarios. To tackle this issue, we propose a parallel ELM architecture consisting of anti-noise and anti-CFO IFE-assisted branches. By training both branches with identical labels, their ELMs converge to the same timing metric (TM) vector, thereby reducing data collection complexity. A Hadamard product-based fusion block combines their outputs, enhancing TS correctness. Simulations show that the proposed method outperforms single-IFE ELM and CS-based methods under multipath and CFO variations.