Knowledge-Based Target Classification Using Baseband Doppler Audio Frequency

In analysis of basebanded Doppler radar signals, it has been noted that trained operators can classify target types by aural analysis of the baseband signals. This implies that there is sufficient structure in the signal so that automated analysis using speech processing techniques can perform classification. Standard analysis techniques assume that the data is stationary and well modeled using Fourier transformations of the data. The resulting spectrum is used to determine signal components that are then used for classification. Unfortunately, the audio signals of interest are highly nonstationary and can even contain sequences of pulses. The standard Fourier spectral analysis techniques cannot determine the time-varying structure wihtin the data and thus do not provide sufficient features for classification. It is this time-varying strucyure that gives the operator the clues he needs for classification. ORINCON proposes to develop and demonstrate an automated system that uses high-resolution time/frequency representations to characterize the data. The feature extractor is coupled with a neural network to solve the classification problems. Neural networks have been used extensivvely and successfully at ORICON to solve detection/classification problems in underwater acoustics, as well as for fault detection/classification problems in rotating machinery. This technology is directly aplicable to the classification problem for baseband Doppler radar return signals.