For example, these two cases have the same respiratory rate, but I cannot estimate them consistently from the spectrums.
Therefore I want to remove the noise from the spectrum directly or from signal indirectly.
By the way, there might be a baseline problem, but breathing sometimes truly causes baseline problem, so I cannot remove or filter the low frequency part directly.
Dear Yufu Liu . Interesting question. Use Fourier domain filter band or band-pass filter. https://fr.mathworks.com/help/dsp/ref/dsp.frequencydomainfirfilter-system-object.html#d120e232713
@
Thanks for your responses !!
I had already tried the frequency domain filter, and the most important part I want to analyze is 0.05-0.75 Hz.
If I filter the signal with a band-pass filter, the respiratory information will be filtered as well.
For example, there are many peaks in the spectrum so that I have a wrong estimation.
By the way, the peaks usually locate at 0.1~0.3 Hz, and they become a problem when the respiratory information is in the band.
Dear Yufu Liu , what is your useful bandwidth and what kind of noise you want to reduce?
Ait Mansour El Houssain
The useful bandwidth is 0.05-0.75 Hz, and the range of respiratory rate is 3-45 breaths/min correspondingly.
Therefore, after bandpass-filtering the signal from 0.05 to 0.75Hz, I still need to find the exact meaningful part of respiratory rate.
The problem shows up when the noise is large, and makes me hard to separate different respiratory cases.
For example, the cases RR = 10 with large noise and RR = 20 sometimes look the same from the Fourier spectrum.
If I pick the highest peak (around 0.4 Hz) from two cases, RR = 10 case will be wrong.
Consequently, although I found possible meaningful parts on the Fourier spectrum, it is still important to find a feature or filter the noise from the RR = 10 case to separate these two cases.
If your noise is temporally correlated (i.e. colored) and it’s power-density spectrum overlaps with the signal then try a Wiener filter. If your signal and noise are well separated in the frequency domain then a band-pass filter tuned to the signal may be sufficient.
I don’t understand what you are trying to do from the fragments of information you have provided. What is the frequency range of the sampled waveform that you would like to enhance (I.e.the signal) and what is the frequency range that you would like to suppress (I.e. the noise)? And what is the nature of the signal (heartbeat?) and noise (breathing?). If the signal and or noise are periodic but impulsive (I.e. short duration) then you may be better off identifying them and separating them in the time domain.
if you can specify the properties of the signal and noise then there is no need to make it adaptive.
Hugh Lachlan Kennedy
Sorry for my poor explanation of the problem!
First, my purpose is to find breathing rate from Photoplethysmography, and the information is in the spectrum from 0.05 to 0.75Hz.
Secondly, I will use a technique (like EWT, EMD, EEMD...) to get the sub-band, and use the sub-band to approximate breathing signal.
Thirdly, the highest peak in the spectrum of the sub-band is the information I want.
However, noise peak sometimes is higher than the true peak, so the frequency range that I want to enhance and suppress overlapped.
You can see the example which I took before .
I agree with separating them from time-domain, but is there any feature or technique?
By the way, I think the problem is similar to using PPG signal to estimate heart rate, but meeting some peaks which caused by moving artifacts.
However, for breathing, it cannot use acceleration signal to fix wrong estimation like estimating heart rate because the noise is unknown.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python