Human Activities Recognition with Amplitude-Phase of Channel State Information using Deep Residual Networks

To address the issue of low accuracy in existing activities recognition schemes with channel state information (CSI), a new scheme is proposed based on collecting CSI time series data and building a deep residual systolic neural network (DRSN) for end-to-end supervised learning feature extraction, which takes into account the amplitude and phase changes of CSI, to achieve high accuracy in activity recognition. The results show that the amplitude and phase analysis based on CSI data in the deep residual network can achieve good activity recognition with an average accuracy of 97.2%.