Deep Learning-based MIMO Radar Waveform Design under PSL Constraint

In multiple-input multiple-output (MIMO) radar waveform design, imposing a peak sidelobe level (PSL) constraint proves particularly effective for controlling both auto-correlation and cross-correlation sidelobes of the designed waveform, which is a critical requirement for enhancing the radar detection performance. Compared with the similarity constraint or integrated sidelobe level (ISL) constraint, the PSL constraint offers superior performance by directly suppressing the highest sidelobes, thereby providing more stringent and precise control over the sidelobes level. However, the implementation of the PSL constraint presents significant computational challenges due to the inherently non-convex and NP-hard nature of the associated optimization problem. To overcome this challenge, this work proposes a deep learning-based optimization method for MIMO radar waveform design under the PSL constraint, which formulates the waveform synthesis as a deep neural network (DNN)-driven optimization process. Specifically, the waveform is progressively optimized through the nonlinear mapping of the DNN, while the mapping function dynamically adapts in accordance with an elaborately designed loss function. Simulation results demonstrate the effectiveness of the proposed method in achieving the PSL constraint.