In the problem of compressed sensing (CS) successful reconstruction can be achieved by maintaining a low mutual coherence between the columns in the vector space. In this work, a way to increase the mutual incoherence is introduced. This is achieved by replacing certain matrix domain of the sparse random matrix, which is used as the measurement matrix with null space bases. For convenience, this can be replaced even by identity matrices.
The result shows that there is a substantial improvement in Peak Root mean Square deviation (PRD). Many different alternatives have been tried out and relative PRD were plotted. Thresholding is generally adapted in CS in order to reduce the PRD values. It was found that without using thresholding technique, it is possible to obtain reduction in PRD values. The time algorithmic performance was also analyzed and found to be better.