In the CS theory framework, the knowledge of the signal sparsity allows signal reconstruction from a small number of measurements. In the array signal processing framework, this means that the knowledge of the spatial signal sparsity allows to achieve a high bearing angle resolution using short sensor array. Our spatial interpretation of the compressive sensing relates to the pioneering results in [27]-[29], where the compressive beamforming was proposed for the problem of direction of arrival estimation. However, these works, similarly to the majority of the published works exploit the temporal sparsity of the received signals.
The paper is organized as follows. Results from the com- pressive sensing theory are summarized in Section II. The addressed problem is formulated in Section III. The spatial compressive sampling approach for the field directionality estimation is presented in Section IV. Estimation performance of the proposed spatial CS-based approach is evaluated via simulations in Section V. Our conclusions are summarized in Section VI.
II. COMPRESSIVE SENSING
This section summarizes notations and that main results from the CS theory [15]-[26]. The CS theory addresses the following underdetermined and noisy problem:
where xJ is a pure J-sparse signal, and where constants c0 and c1 are well behaved and small. Note that this results suggest that when the signal x is J-sparse, the estimation error is bounded by the energy of the noise w only.
This framework provides an opportunity for the sensing matrix Φ design [17]. One should find the sensing matrix that obeys the RIP and allows to recover as many elements of the signal x from M measurements, as possible. The RIP that was proposed in [18] and [19] is closely related to the incoherency between the sparsity and the measurements basses, providing an efficient way to obtain the sensing matrix that satisfy it. In [23] it was shown that the incoherency property allows exact reconstruction of the signal that is sparse in one basis using the sensing matrix from the second incoherent basis. It was shown in [23] that the time-frequency pair of orhtonormal basses that are related via the Fourier transform is highly incoherent. Moreover it was shown that the time-frequency pair of spikes and complex sinusoids yields the most mutually incoherent pair providing the best sparsity conditions. This property was used in [19] and [23] to show that considering that the signal x is sparse in the basis of spikes, the minimal number of the measurements that is required for its recovery, using the partial Fourier matrix of complex sinusoids with M uniformly selected rows, is: M ≥ c2 J/(log N)4 .