Sparse Range-Doppler Image Construction with Neural Networks

The principles outlined by compressed sensing can permit a sensor to collect reduced amount of data and still reconstruct an exact outcome. This can for example be used to generate super-resolution sparse range-Doppler radar images while emitting a reduced number of pulses within a coherent processing interval. In this paper, we investigate the use of neural networks as a mean to solve the sparse reconstruction problem with specific emphasis towards range-Doppler images. The neural networks are trained to generate a sparse Doppler profile from incomplete time domain data in line with traditional sparse $L_{1}$-norm minimization. We show that this approach is viable through fully connected feed forwarding networks and the results closely mimic sparse recovered range-Doppler maps.