Recovery of a high shock probability process using blind deconvolution

We intend in this paper to recover shocks occurring in the situation where they are highly probable using blind deconvolution methods: order 2 (Yule-Walker), higher order (normalized cumulants), and mutual information. We define a shock probability process derived from a Bernoulli process and we show that, in opposite to the other two methods, the normalized cumulants are highly dependent on the shock probability P. This has consequences on performance versus P, which are studied applying a Kurtosis maximization algorithm on simulations. We finally apply the three blind deconvolution methods to a real signal recorded on a rope transportation line. Results comparison with an a priori shock model of the mechanical system confirms that normalized cumulants are less adapted for recovering a high shock probability process.