Axial Attention-Based Transformer Receiver for Robust OFDM-CPM Signal Detection

Orthogonal Frequency Division Multiplexing with Continuous Phase Modulation (OFDM-CPM) is a compelling hybrid signaling scheme that synergistically combines the robustness of OFDM against multipath fading with the power efficiency and superior detection gain of CPM. Despite its potential, practical implementation is hindered by prohibitive detection complexity and high sensitivity to radio frequency (RF) impairments. To date, dedicated deep learning (DL)-based receiver designs specifically optimized for the unique two-dimensional structure of OFDM-CPM signals remain largely unexplored. In this paper, we propose a novel axial attention-based Transformer receiver for robust OFDM-CPM signal detection. By decomposing the attention mechanism into sequential operations along the temporal and spectral axes, the proposed architecture effectively captures the phase memory inherent in CPM and the frequencyselective fading of the OFDM channel with high computational efficiency. Extensive simulations across standardized fading models demonstrate that our receiver consistently outperforms both conventional model-based benchmarks and state-of-the-art deep learning methods. Notably, the proposed method exhibits exceptional robustness to carrier frequency offset (CFO) and inphase/quadrature (IQ) imbalance, representing a favorable tradeoff between detection accuracy and computational complexity.