方位超分辨一直是雷达领域里受到广泛关注的研究课题,针对求解过程中遇到的病态性,深入研究了L1范数正则化方法及其约束方法.在目标稀疏性质的前提下,建立了L1范数正则化与约束L1范数正则化模型.鉴于雷达数据的维数较多,利用梯度投影算法对模型进行求解.在不同信噪比情况下对两个等幅点目标进行了计算机仿真,结果初步表明:随着信噪比的降低,两种算法的分辨效果变差,在相同条件下约束L1范数正则化方法分辨效果更好,在信噪比为0 dB时,仍可分辨相隔1/2半功率波束宽度的两个等幅点目标;同时约束L1范数正则化方法分辨性能优于L1范数正则化方法、迭代反卷积法、维纳逆滤波法和Richardson-Lucy算法(RL算法);此两种范数正则化方法具有较强的噪声适应能力,可用于雷达方位超分辨.

Azimuth super-resolution has always been a hot research topic in radar domain.For the ill-condition encountered in the solving process,a thorough analysis is made of L1 norm regularization method and its constrained method.Under the premise of the target sparse nature,L1 norm regularization model and its constrained model are established,and a gradient projection algorithm is used to solve them for the large dimensions of radar data.A computer simulation for two equal amplitude point targets is made at different Signal-to-Noise Ratios (SNRs).The preliminary findings show that:1) with the decease of SNR,the resolution effect of both algorithms becomes worse,of which,the constrained L1 norm regularization method has a relatively better resolution under the same condition,and can distinguish the two equal amplitude point targets with a interval of 1/2 half power beam width when SNR is 0 dB;2) the constrained method has a better performance in resolution than L1 norm regularization method,constrained iterative deconvolution algorithm,wiener inverse filter algorithm and Richardson-Lucy algorithm (RL algorithm);and 3) these two norm regularization methods have stronger noise adoptability,which can be used for radar azimuth super-resolution.