We propose the use of compressive sensing (CS) in the context of a multi-input multi-output (MIMO) radar system that is implemented by a small scale network. Each receive node compressively samples the incoming signal, and forwards a small number of samples to a fusion center. At the fusion center, all received data are jointly processed to extract information on the potential targets via the CS approach. Since CS-based MIMO radar would require many fewer measurements than conventional MIMO radar for reliable target detection, there would be power savings during the data transmission to the fusion center, which would prolong the life of the wireless network. First, we propose a direction of arrival (DOA)-Doppler estimation approach. Assuming that the targets are sparsely located in the DOA-Doppler space, based on the samples forwarded by the receive nodes, the fusion center formulates an ℓ1-optimization problem, the solution of which yields the target DOA-Doppler information. The proposed approach achieves the superior resolution of MIMO radar with far fewer samples than required by conventional approaches. Second, we propose the use of step frequency to CS-based MIMO radar, which enables high range resolution, while transmitting narrowband pulses. For slowly moving targets, a novel approach is proposed that achieves significant complexity reduction by successively estimating angle-range and Doppler in a decoupled fashion and by employing initial estimates to further reduce the search space. Numerical results show that the achieved complexity reduction does not hurt resolution. Finally, we investigate optimal designs for the measurement matrix that is used to linearly compress the received signal. One optimality criterion amounts to decorrelating the bases that span the sparse space of the incoming signal and simultaneously enhancing signal-to-interference ratio (SIR). Another criterion targets SIR improvement only. It is shown via simulations that, in certain cases, the measurement matrices obtained based on the aforementioned criteria can improve detection accuracy as compared to the typically used Gaussian random measurement matrix.
展开▼
