Artificial Neural Network for Radar-based Respiration Detection

A continuous wave (CW) radar system that is designed to detect respiration and heart rate of a human, is presented. The radar operates at 5.8 GHz and has two outputs, the in-phase (I) and quadrature (Q) signals. The respiration and heartbeat frequencies can be extracted by processing these outputs using traditional methods, such as Fast Fourier Transform (FFT) and simple peak searches. Emphasis is given to the developed artificial neural network (ANN) used to determine the respiration frequency of the waveform. The motivation is that ANNs are faster than FFT-based approaches. The proposed method consists of training, validating and testing the ANN model with digitally constructed waves emulating respiration waveforms. Real world scenario measurement data from the CW radar system are then fed to the ANN model and the respiration frequency is obtained consistently. Experimental results demonstrate that this approach is on average 91.84% accurate and effective.