Boundary Multiple Measurement Vectors for Multi-Coset Sampler

As a classical sub-Nyquist sampling technique, multi-coset sampler (MCS) has recently encountered great challenges due to rapid increase in the bandwidth of radio signals. To reduce the sampling cost of MCS, multiple measurement vectors (MMV) exploiting joint-sparsity of signal has been widely used. However, it exhibits poor performance when the signal support is scattered over multiple rows. To overcome this issue, we propose a boundary MMV (bMMV) model by extending the classic MMV model via: i) column-partitioning and ii) row-extraction on the target signal matrix. We show that our model can achieve the theoretical minimum sampling rate of MCS. Moreover, under the bMMV model, we develop the side-information-aided simultaneous subspace pursuit (SSSP) algorithm to recover MCS signals. Iteratively, it utilizes side information of the previously estimated supports of signal sub-matrices to promote recovery accuracy. Analysis shows that SSSP can stably recover MCS signals under a mild condition on the sampling matrix. Experiments shows that SSSP has higher recovery accuracy than existing ones, especially when the sampling rate is low.