Clearning Phased Array Radar Data

Abstract : Many military and civilian problems can be viewed as pattern recognition: given a set of measured inputs, the task is to predict the corresponding output. Typical examples range from image recognition and classification, to time series prediction and regression. Most modeling assumes that the inputs can be measured exactly, without noise. Building a model then means to construct (or "learn") a mapping from these inputs to the expected values of the outputs. The usually tacit assumption of noise-free inputs is violated in most real-world problems where only a noisy version of the "true" input is observed. This research found that while it was possible for time series proilems even if there is a lot of noise present, to use information from adjacent patterns in time, the problem could be solved for non-time series problems, such as the phase array radar data. The effort lead to several papers. Results are presented on discrete hidden states (Hidden Markov models), and continuous hidden state (state space models). A paper on finding the true inputs using Independent Component Analysis is in preparation. A paper on evaluation methodology using the bootstrap also employs the state space approach.