Fig. 5: Ablation study.
From: Monitoring long-term cardiac activity with contactless radio frequency signals
a We compared three signal selection algorithms (mmHRV24, Mtrack35, and BNR36) with our proposed algorithm while keeping other modules the same as RF-HRV. Our algorithm outperformed the others in terms of real-time inter-beat interval (RT-IBI) error (Ours: 26.1 ms, mmHRV: 36.6 ms, Mtrack: 45.2 ms, BNR: 40.1 ms). b We calculated the ratio of the selected optimal signal to all signals. Our signal selection algorithm achieved a ratio of 79.3%, surpassing mmHRV (62.1%), Mtrack (58.7%), and BNR (60.4%). c We compared the impact of several methods, including band-pass filtering (BPF) and second-order difference (SOD), for obtaining high-frequency components on the system performance. d Five different decomposed components. Imf 1 denotes the direct component of the signal, Imf 2 represents the respiratory signal, and Imf 5 is the high-frequency component, showing clear heartbeat information. e Frequency distribution of raw signal and different components. n = 6,222 outpatient data points in box plots of a–c. Square, mean; minima, smallest non-outlier value; maxima, largest non-outlier value; center line, median; box bounds, upper and lower quartiles; whiskers, 1.5  × interquartile range, in box plots of a, c.
