A Multitask 1D Convolutional Neural Network for Joint Occupancy, Movement, and Heart Rate Extraction via Wi-Fi Channel State Information

Wi-Fi Channel State Information (CSI) has emerged as a robust, non-invasive modality for human activity recognition and vital sign monitoring. However, decoupling macroscopic physical movement from micro-Doppler vital signs, such as resting heart rate, remains a significant digital signal processing challenge. This study proposes a multitask 1D Convolutional Neural Network (CNN) framework designed to simultaneously classify room occupancy, detect physical movement, and estimate heart rate from CSI streams. Utilizing a comprehensive synthetic dataset of 17.28 million samples captured at an aggressive sampling rate of 200 Hz across 15 subcarriers, the raw CSI amplitude and phase shifts are preprocessed using a mathematically rigorous pipeline involving linear phase detrending, Principal Component Analysis (PCA) for spatial diversity extraction, and a 3rd-order Butterworth bandpass filter to isolate the 0.1–10.0 Hz frequency band. Comprehensive ablation studies validate the efficacy of the preprocessing mechanisms and the multitask architecture. The results demonstrate that the proposed framework achieves an occupancy classification accuracy of 98.2% and estimates resting heart rate with a Mean Absolute Error (MAE) of 1.45 BPM. Furthermore, the shared representation learning inherently regularizes the network, allowing the micro-task regression head to successfully isolate micro-variations associated with cardiac activity despite the presence of ambient multipath noise.