Skip to Main Content
In the original paper (see ibid., vol.150, no.1, p.33-6, Feb. 2003) a novel technique based on hidden Markov models (HMM) for Doppler classification was presented and performance was reported for a 3-class problem. It was proposed that a HMM model using a cyclic topology would be suited to representing the time varying Doppler signatures observed for moving targets. The paper concluded that the problem was adequately modelled with a HMM algorithm using a cyclic topology with 9 states. At the time of the production of the paper it was believed that the HMMs were using all 9 states to explain the data within each data file that was being classified. However, results obtained subsequent to the publication of the paper show that the models were, for the most part, fitting any given test data file to a single state. The 9 states were being used to explain different data files and thus the system was behaving effectively as a Gaussian mixture based classifier.