Autoencoder-Based OFDM Radar Waveform Design with Robustness to Nonlinear Power Amplifier Distortion

Abstract

Orthogonal Frequency Division Multiplexing (OFDM) radar systems offer high Doppler resolution and waveform flexibility, making them well-suited for automotive applications. However, their high peak-to-average power ratio (PAPR) induces nonlinear distortion in power amplifiers (PAs), degrading range resolution. This issue is further exacerbated by the introduction of guard band (GB) subcarriers, which disrupt the ideal properties of constant amplitude zero autocorrelation (CAZAC) sequences. In this work, we propose an autoencoder (AE)-based waveform design that jointly reduces PAPR and suppresses range sidelobes while accounting for PA nonlinearity and GB effects. To enhance decoding, the AE architecture integrates side information (SI) by directly feeding the encoder output alongside the received signal into the decoder, either by simple concatenation or through a learned fusion strategy. Simulation results show that the direct concatenation method outperforms the learned fusion strategy and significantly improves both PAPR reduction and range sidelobe suppression under nonlinear PA distortion. The proposed method achieves a dynamic range exceeding 100 dB in range profiles, even with 0 dB input back-off, meeting the stringent requirements of modern automotive radar systems.