Parameter Estimation For Bursty-Intermittent Observations

Parameter estimation problem is examined in the setting where the noise power is allowed to change from sample to sample. Parameters of the noise source is assumed to be generated by a Markov chain whose state sequence is not known by the observation system. Expectation-maximization algorithm is applied for the estimation of desired parameter with the inclusion of unknown state vector of the Markov chain realization as a latent variable. The suggested scheme can be utilized in applications with bursty noise and/or intermittent signals.