Doppler Sidelobe Suppression via Quasi-Neural Network for ST-CDMA MIMO Radar

In co-located multiple-input multiple-output (MIMO) radar, slow-time code division multiple access (ST-CDMA) is an essential option of orthogonal waveforms, because the transmit power and bandwidth are well exploited with a relatively low level of hardware complexity. However, the code produces high sidelobes in the Doppler spectrum, which poses a negative impact on weak target detection. To address this issue, we propose a novel sidelobe suppression method that leverages the CLEAN framework in conjunction with a quasi-neural network (Quasi-NN). The novelty lies in the application of Quasi-NN for signal modeling in the target parameter estimation step of CLEAN. Specifically, Quasi-NN is employed to represent the signal after range compression. Its inputs, outputs, and internal weights are determined by the characteristics of antennas, waveforms, and targets. In this way, the estimation of target parameters is transformed into the optimization of weights, which thereby can be solved using a back-propagation (BP) algorithm. Simulation results demonstrate the superior performance of the proposed method under various scenarios. Real-data results using a 77-GHz radar also show that the proposed method achieves lower sidelobes and thus improves the detection of weak targets.