Vibration sensing by laser radar with a pulsed emission is a promising technique for long range target recognition and identification. However, compared to its continuous-wave counterpart, it is impaired by its greater sensitivity to speckle noise and its lack of robustness to multiple independent vibrations of the target surface. Using poly-pulse waveforms for greater velocity resolution, we developed a new estimator to take into account signal statistics, and time frequency representations that can achieve better performance than classical processing methods. Simulations show a 5dB improvement in Signal-to- Noise Ratio (SNR) when speckle noise is dominant, and 4dB improvement when signal is weak in regard to detection noise.