Convolutional Neural Network-Based MIMO Radar Channel Selection for Improving Robust Remote Heart Rate Estimation Accuracy

In this paper, a solution for automatic MIMO radar channel selection utilizing convolutional neural networks is presented for the task of advancing robust remote heart rate monitoring. A millimeter-wave radar with a high number of transmit-receive channels is used to help mitigate the effects of random body swaying motions through advanced signal processing. It is shown here that deep learning methods can classify radar channels as either good or bad by learning features of time-frequency representations of the corresponding phase variation signals. A classification accuracy of over 80% is achieved and is shown to have the ability to improve heart rate estimation accuracy by over 20% when compared to more traditional channel combination methods.