Finding sparse solutions of underdetermined linear 1 systems commonly requires the solving of L1 regularized least 2 squares minimization problem, which is also known as the basis 3 pursuit denoising (BPDN). They are computationally expensive 4 since they cannot be solved analytically. An emerging technique 5 known as deep unrolling provided a good combination of the 6 descriptive ability of neural networks, explainable, and compu-7 tational efficiency for BPDN. Many unrolled neural networks for 8 BPDN, e.g. learned iterative shrinkage thresholding algorithm 9 and its variants, employ shrinkage functions to prune elements 10 with small magnitude. Through experiments on synthetic aper-11 ture radar tomography (TomoSAR), we discover the shrinkage 12 step leads to unavoidable information loss in the dynamics 13 of networks and degrades the performance of the model. We 14 propose a recurrent neural network (RNN) with novel sparse 15 minimal gated units (SMGUs) to solve the information loss 16 issue. The proposed RNN architecture with SMGUs benefits 17 from incorporating historical information into optimization, and 18 thus effectively preserves full information to the final output. 19 Taking TomoSAR inversion as an example, extensive simulations 20 demonstrated that the proposed RNN outperforms the state-21 of-the-art deep learning-based algorithm in terms of super-22 resolution power as well as generalization ability. It achieved 10% 23 to 20% higher double scatterers detection rate and is less sensitive 24 to phase and amplitude ratio difference between scatterers. Test 25 on real TerraSAR-X spotlight images also shows high-quality 3-D 26 reconstruction of test site. 27 Index Terms-SAR tomography (TomoSAR), basis pursuit de-28 noising (BPDN), recurrent neural network, sparse reconstruction. 29 30 I. INTRODUCTION 31 A. Motivation 32 Sparse solutions are ordinarily desired in a multitude of 33 fields, such as radar imaging, medical imaging and acoustics 34 signal processing. Compressive sensing theory tells that the 35