I am trying to reduce(flattened) the flutuations in a signal in "realtime applications" using Simulink. For this , I am using a bandstop filter then a lowpass filter. But the delay is very high.(1-2 sec delay is permissible.). However the delay introduced is more than 20 sec and increases as the filter order increases. Is there any other technique to reduce the flutuations without delay?
I tried using filtfilt function but it of no use for real time systems.
Please help in this regard.
"Delay" or phase lag, as it is more commonly known is inherrent to low pass filtering. You can reduce it by increasing the cut off frequency. Otherwise, consider moving average filters. They have good time domain response (probably what you are looking for) but poor frequency domain performance.
Dear Sir,
What type of filter did you use ?.
As Bidal said, when filtering, there is an inherent minimum delay. This is because the filter needs to be a causal system, ie, the impulse response for negative time needs to be zero.
The minimum delay of a filter is given by the Hilbert transform.
There are also some filters that are minimum delay (eg most passive filters, and also the digital IIR Infinite Impulse Response filters), while others (the FIR filters) are not minimum delay.
In order to combat delay, the first thing to do is to use minimum phase filters (IIR or passive), and no FIRs. (They are more difficult to synthesize,)
The second thing to do is tune your amplitude response for the delay you can tolerate, ie not attenuate more than you need. Calculate the power spectral density of what you want to attenuate, and then design an appropriate amplitude response. In closed loop systems, this often leads to second order filters (ie the amplitude response goes down with 40 db/decade from the loop bandwidth onward). I dont know what you will get here.
Cheers,
Henri.
Thank you Sirs Bilal and Henri for your detailed answers.
However since I am using Bandstop filter to get rid of unwanted high power samples, the cut-off frequencies cannot be changed. But Moving average filter seems a good choice if the cut-off frequencies fall in range.
I have tried with almost all possible filters available in FilterDesign&Analysis app in Matlab like IIR(butterworth ,chebyshev,..) and FIR(Windowed, Equiripple and others available) there is a minimum delay of 10sec to suit my application.
Could LMS concept be applied in such case for real time applications?
A recursive least squares (RLS) filter is much like a moving average (MA) filter.
http://gregstanleyandassociates.com/whitepapers/FaultDiagnosis/Filtering/Least-Squares-Filter/least-squares-filter.htm
Dear Pranav,
If your application is not real-time, i.e. you are post processing collected data, you may consider a zero phase digital filter
https://www.mathworks.com/help/signal/ref/filtfilt.html
It is a non-causal filter, so you cannot apply it in real time.
DEAR PRANAV
HOPE BELOW SLIDE WILL HELPS
BEST WISHES
DR MOHAMAD ARINAL
http://dr-mohamad-arinal.simplesite.com/432957824
http://ernie-coffee-printing.simplesite.com/433567200
- Badges
- Science topic