A complex-domain CSI processing framework for WiFi-based respiration monitoring

Reliable respiration rate estimation from WiFi channel state information (CSI) remains challenging due to multipath interference and orientation-dependent signal degradation. This study aimed to provide a two-stage mathematical framework that exploits the statistical properties of complex-valued CSI for robust respiration estimation. First, a binary frequency-domain mask was applied to the complex CSI ratio to isolate respiration-induced components. Motivated by the zero-mean Gaussian characteristics of ambient noise, the proposed spectral reconstruction separated periodic respiratory signals from stochastic interference through inverse fast Fourier transform reconstruction. Second, a dual principal component analysis (PCA) scheme was introduced. The first-stage PCA aggregated complementary respiration-related information across multiple subcarriers in the complex domain to mitigate multipath effects, whereas the second-stage PCA extracted a real-valued respiration waveform by projecting the real-imaginary representation of the fused complex time series onto the principal-component direction of maximum variance. Experiments involving 11 participants across four sleep postures and three antenna deployment configurations yielded a mean absolute error of 0.10 breaths per minute, outperforming representative existing methods. The results indicate that preserving amplitude-phase coupling in the complex domain improves the robustness of respiration signal extraction.