Table 2.

Respiration rate monitoring using Wi-Fi CSI signals.

Ref. year	Extraction tool	Pre-processing	Detection algorithm	Multi-person	Real-time	Performance summary
[42] 2016	Linux 802.11n CSI Tool	Noise Reduction	Modeling-based	No	No	N/A
[43] 2016	Linux 802.11n CSI Tool	Noise Reduction	Modeling-based	No	No	N/A
[36] 2018	Linux 802.11n CSI Tool	Noise Reduction Signal Extraction	Modeling-based	No	No	Accuracy: good positions 98.8%, bad positions 61.5%
[34] 2019	Linux 802.11n CSI Tool	Noise Reduction Signal Extraction	Modeling-based	No	No	Accuracy: over 99%
[44] 2021	Linux 802.11n CSI Tool	Noise Reduction Signal Transform	Modeling-based	No	No	Max. error $<0.7$ bpm, average errors $\approx 0.15$ bpm
[38] 2020	Linux 802.11n CSI Tool	Signal Extraction	Modeling-based	Yes	No	Error rate of 0.73 bpm (breaths per minute)
[45] 2017	Linux 802.11n CSI Tool	Noise Reduction	Hybrid	Yes	No	Accuracy: 1 person 96% less than 0.5 bpm, 2 and 3 person 93% smaller than 0.5 bpm, and 5 person 62% less than 0.5 bpm
[46] 2017	Linux 802.11n CSI Tool	Noise Reduction	Modeling-based	No	N/A	Median error: 0.09 bpm, 0.15 bpm, 0.06 bpm for three different detectable regions
[30] 2017	Linux 802.11n CSI Tool	Signal Transform	Modeling-based	Yes	Yes	Accuracy: $>95$% (1 person) $>88$% (2 persons)
[51] 2017	Linux 802.11n CSI Tool	Signal Transform	Modeling-based	No	Yes	Mean accuracy: single-person NLOS 99%, dozen people LOS 98.65%, 9 people NLOS 98.07%
[47] 2018	Linux 802.11n CSI Tool	Noise Reduction	Modeling-based	No	Yes	Reported accuracy of nearly 100% in LOS
[58] 2019	Linux 802.11n CSI Tool	Noise Reduction	Modeling-based	No	Yes	Reported overall detection rate of nearly 100%; mean absolute error less than 0.3 bpm for breath rate
[37] 2021	Nexmon CSI Extractor	Noise Reduction Signal Extraction	Modeling-based	No	Yes	N/A
[54] 2021	Nexmon CSI Extractor	Noise Reduction Signal Extraction	Hybrid	No	No	N/A
[59] 2022	Nexmon CSI Extractor	Noise Reduction Signal Extraction	Modeling-based	Yes	No	Accuracy: over 93%
[60] 2022	Wi-ESP	Noise Reduction Signal Extraction	Modeling-based	Yes	No	Accuracy between 91% and 99%.
[61] 2020	Linux 802.11n	Signal Transform	Modeling-based	Yes	N/A	Accuracy: 1 person 98.8%, 2 person 98.4%, 3 person 97.5%.
[62] 2021	Linux 802.11n	Signal Transform	Modeling-based	No	Yes	Phase and amplitude based measurements had median percentage errors of 8.5% and 7.4% respectively
[53] 2021	Linux 802.11n	Noise Reduction Signal Extraction	Learning-based	No	N/A	K-nearest neighbor classifier using relief feature selection techniques: 85.12%.