🚀 go-pugleaf

In <1992Dec15.181336.3424@iplmail.orl.mmc.com> natd@iplmail.orl.mmc.com (Natalie Day) writes:

>Hi,
>  I am having problems with the following and would like to know if there is a 
>good cookbook approach to the following:

>  Problem:  I am receiving EKG (electrocardiagram) signals as an input to 
>            a system.  The problem is that there is a 60Hz noise riding the
>            signal and I would like to get rid of it.  
>            I am looking for a good 60 Hz notch filter, (digital software 
>            filter).  I have tried a 6 pole IIR filter but I get a lot of
>            ringing.  Does anyone have any suggestions/solutions to my problem?

I've found the best way to reduce this noise (with amazing results)
is to work in the frequency domain. Do an FFT, look at it, and if the
noise causes vertical stripes in the image then multiply the FFT with
a notch filter which blanks out the frequencies from the highest frequency
(ie from the border of the FFT spectum) down to the frequency which
corresponds to the 60Hz signal (you'll have to work this point out - it will
appear on the spectrum as a bright dot indicating strong energy at that
frequency. Anyway - blank out all these frequencies along the vertical 
axis of the spectrum, do an IFFT and there you go.

Apoligies for the non-maths approach, but this technique has been tried
and works bloody well. See pg 247, Digital Image Processing, Gonzalez
and Wintz, 2nd edition. The notch filter (figure 5.13b) is created by
multiplying the FFT (not the mag-log version of it, the real thing)
by a filter (all 0.0 or 1.0) which can be generated easily (write
some code!).

If you refer to figure 5.13a, about 1/3 of the way down (vertically) along
the centre of the spectrum you can see 2 horizontal lines (rather faint,
but they stand out). These correspond to the horizontal 60Hz noise.
Note that the notch filter on 5.13b extends just past this point.
This point for your image must be determined by examining the mag-log
version of the FFT (otherwise you won't see much).

Comments, anyone?

Shckwish!

>   Thanks for you responses!


>                              Nat