The sparse scene recovery performance of compressed sensing radar (CSR) requires that the coherence parameter of the sensing matrix should be as small as possible. Based on this thought, a new optimal sensing matrix design method is proposed. To minimize the coherence parameter of the sensing matrix, the waveform and measurement matrix are designed separately and simultaneously using simulated annealing (SA). Simulation results demonstrate the algorithm can improve recovery accuracy, enhance noise immunity and increase the maximum permissible sparsity of CSR, and that the joint optimization algorithm can achieve a better result than the algorithms that optimize waveform or measurement matrix separately.