Project DetailsIn this work we provide an alteration to the multichannel compressed sensing signal model prevalent in literature.
We propose to send a rank-1 matrix uncompressed instead of a single vector as is done in many works dealing with resembling problems. We suggest that this matrix be either the first SVD component of the data of the L matrix of the RPCA decomposition.
We provide mathematical derivation to obtain these matrices, and why they should work well on highly correlated sensor data. We generate a dataset in accordance to the new signal model and perform an analysis of the sparsifying step of the compression and overall reconstruction error.
We show that both the RPCA and SVD methods achieve results better than the current method by orders of magnitude in terms of mean-squared reconstruction error. We obtain that using RPCA is preferable, but sometimes computationally infeasible, in which case the SVD method still provides ample improvement over current methodology.
