IDENTIFICATION OF HELICOPTER NOISE USING A NEURAL NETWORK

An artificial neural network (ANN) has been trained to distinguish between the noise of two helicopters. The performance of the ANN is compared with that of a conventional recognition system. The conventional system uses the ratio of the main-rotor blade passage frequency (bpf) to the tail-rotor bpf. The ANN was trained to use similar main/tail-rotor information, in addition to information describing the distribution of spectral peaks of the main rotor. It is shown that this additional information allows the ANN to distinguish between the helicopters when tail-rotor noise is removed from the spectrum. The performance of the two methods is given as a function of signal-to-noise strength, and propagation distance, using a model of atmospheric sound propagation. The conventional method outperforms the ANN when main- and tail-rotor noise are present, but the conventional method cannot identify helicopters when tail-rotor noise is removed. At 20-dB signal-to-noise ratio (SNR), when tail-rotor noise is not present in the spectrum, the ANN correctly identifies the helicopters 100% of the time, compared to 50% for the conventional method. The performance of the ANN drops as signal strength decreases. At 8-dB SNR, the ANN is correct 77% of the time, while at 0 dB it is correct 58% of the time. Similar results are obtained for the performance when the signal is propagated through the model of the atmosphere.