Seite - 322 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

x(n)= { s(n)+w(n) forn∈φi, i=1,···I w(n) otherwise (1) where s(n) is the desired signal (inhalation signal),Φi=[mi, qi] is the sample indices corresponding to the ith inhalation segment, andw(n) is the background noise. LetNi denote thenumberof samplescorresponding to the ith inhalation, i.e.Ni=qi−mi−1. First, we estimate the power of the noise-corrupted signal corresponding to each inhalation, aswell as thepowerof thenoise-only signal as follows: Pi= 1 Ni qi ∑ n=mi x(n)2 (2) Pw= 1 |Ω|∑n∈Ω x(n)2 (3) whereΩ=A\Φ1,··· ,ΦI and |Ω| its cardinality. The average energyof the noise-free signal, thatwill be used as a predictor for the flowrateestimation, is calculatedas follows: Es= 1 I I ∑ i=1 (Pi−Pw)ti (4) where ti is the timeduration (inseconds)of the ith inhalation,which isdeterminedusing mi,qi and the sampling frequency. Finally,weestimate the inhalationflowrate usinga simple linear regressionmodel whereEs is thepredictor, i.e. fˆ=β×Es+α. (5) where the parametersα and β are estimated using the available datasets and the least squaremethod. 3. Results Inour experiments, a total of54 inhalationswere recordedandadjusted todifferent test flowrates (15 to90L/minwitha step sizeof5L/min).Theacoustic signalswere stored in .wavfiles.UsingMatlab, the signal is filtered and analysed as described above. The actual value of the inspiratory flow rate (IFR)was extracted during the experiments; it wasdisplayedonthemassflow-meter.Subsequently, theactualvaluesof(IC)and(FIV1) wereeasilyextracted fromthe inhalationsegments. The results showthat thesignal energyEs ishighlycorrelated to the inhalationflow rate and can significantly explain 99%of its variance, i.e. (R2=99%,with p-value< 0.0001).. Wetestedourmethodologyon3acousticsignals thatwerenot includedinour train- ingmodel (Eq. (5)).Basedon theflowrate estimation, estimates of theother inhalation Z.Jeddi etal. /Estimationof InhalationFlowParameters forAsthmaMonitoring322