Compressed sensing (CS) theory relies on sparse representations in order to recover signals from an undersampled set of measurements. The sensing mechanism is described by the projection matrix, which should possess certain properties to guarantee high quality signal recovery, using efficient algorithms. Although the major breakthrough in compressed sensing results is obtained for random matrices, recent efforts have shown that CS performance could be improved with optimized non-random projections. Designing matrices that satisfy CS theoretical requirements is closely related to the construction of equiangular tight frames, a problem that has applications in various scientific fields like sparse approximations, coding, and communications. In this paper, we employ frame theory and propose an algorithm for the optimization of the projection matrix that improves sparse signal recovery.