## [Audacity-quality] Scientific Filter graph

 [Audacity-quality] Scientific Filter graph From: - 2013-12-26 17:57:05 Attachments: sci_filter_lp_order1_1000hz.png ```Usual disclaimer for any any ignorance, but I am confused when I look at http://wiki.audacityteam.org/wiki/Low_Pass_Filter and see that the graph for Butterworth filter in Scientific Filter does not conform to attenuation of -6 dB at double the corner frequency. Or should I not expect conformance? In the attached image (1000 Hz first order low pass, range set to +6 to -61 dB but the range makes no difference AFAICT), 1000 Hz seems to be at -3 dB as expected, but 2000 Hz is at -7 dB instead of -6 dB. Similarly with first order low pass at 250 Hz, 500 Hz appears to be at -7 dB instead of - 6 dB. The rolloff above double the corner frequency seems to be "correct" (-6 dB per octave). The Manual (most of it verbatim from Norm's -devel post) says for Butterworth that "attenuation increases at approximately 6 dB per octave times the filter order". Why "approximately", or is that another way of describing what I see? Now if I change my Butterworth first order 1000 Hz LP filter to "Chebyshev I", there is less attenuation than Butterworth (at default 1 dB "Passband Ripple") - look at 5000 Hz for example. As greater attenuation is expected with a Chebyshev than a Butterworth filter, is 1 dB the correct default for "Passband Ripple"? At what points in the stopband part of a Chebyshev filter do you expect a first order rolloff of -6 dB per octave? For example with Chebyshev I Passband Ripple of 1 dB (or Chebyshev II Stopband Ripple of 1 dB), 5000 Hz has a gain of -9 dB, but 1000 Hz has a gain of what looks like -16.5 dB. Gale ```

 [Audacity-quality] Scientific Filter graph From: - 2013-12-26 17:57:05 Attachments: sci_filter_lp_order1_1000hz.png ```Usual disclaimer for any any ignorance, but I am confused when I look at http://wiki.audacityteam.org/wiki/Low_Pass_Filter and see that the graph for Butterworth filter in Scientific Filter does not conform to attenuation of -6 dB at double the corner frequency. Or should I not expect conformance? In the attached image (1000 Hz first order low pass, range set to +6 to -61 dB but the range makes no difference AFAICT), 1000 Hz seems to be at -3 dB as expected, but 2000 Hz is at -7 dB instead of -6 dB. Similarly with first order low pass at 250 Hz, 500 Hz appears to be at -7 dB instead of - 6 dB. The rolloff above double the corner frequency seems to be "correct" (-6 dB per octave). The Manual (most of it verbatim from Norm's -devel post) says for Butterworth that "attenuation increases at approximately 6 dB per octave times the filter order". Why "approximately", or is that another way of describing what I see? Now if I change my Butterworth first order 1000 Hz LP filter to "Chebyshev I", there is less attenuation than Butterworth (at default 1 dB "Passband Ripple") - look at 5000 Hz for example. As greater attenuation is expected with a Chebyshev than a Butterworth filter, is 1 dB the correct default for "Passband Ripple"? At what points in the stopband part of a Chebyshev filter do you expect a first order rolloff of -6 dB per octave? For example with Chebyshev I Passband Ripple of 1 dB (or Chebyshev II Stopband Ripple of 1 dB), 5000 Hz has a gain of -9 dB, but 1000 Hz has a gain of what looks like -16.5 dB. Gale ```
 Re: [Audacity-quality] Scientific Filter graph From: Steve the Fiddle - 2013-12-29 11:54:34 ```On 26 December 2013 17:56, wrote: > > Usual disclaimer for any any ignorance, but I am confused when I > look at http://wiki.audacityteam.org/wiki/Low_Pass_Filter > and see that the graph for Butterworth filter in Scientific Filter > does not conform to attenuation of -6 dB at double the corner > frequency. Or should I not expect conformance? > > In the attached image (1000 Hz first order low pass, range set to > +6 to -61 dB but the range makes no difference AFAICT), 1000 Hz > seems to be at -3 dB as expected, but 2000 Hz is at -7 dB instead > of -6 dB. > > Similarly with first order low pass at 250 Hz, 500 Hz appears to > be at -7 dB instead of - 6 dB. > > The rolloff above double the corner frequency seems to be > "correct" (-6 dB per octave). > > The Manual (most of it verbatim from Norm's -devel post) says > for Butterworth that "attenuation increases at approximately > 6 dB per octave times the filter order". Why "approximately", or > is that another way of describing what I see? The graph closely matches the actual frequency response of the filter, so no problem there. I'm unsure about whether the first order filters are doing "the right thing" or not. Certainly the 1st order Butterworth filter does not match the response that is usually shown for a Butterworth filter (for example: http://en.wikipedia.org/wiki/Butterworth_filter ) and it does not match the response of the Nyquist first order Butterworth filter (the "hp" and "lp" functions), which do match the response as typically described. I don't know whether this mismatch is "a matter of implementation" or "a limitation" or "an error", but is so noticeable that users are sure to ask. I think this needs to be answered in the manual. The response of the 1st order Chebyshev filters also looks rather "surprising" when "ripple" is set to zero. Again I don't know if it "should" look like that, but I think it needs to at least be documented. There is little on-iine documentation about first order Butterworth and Chebyshev filters, but the behaviour of these types in the Scientific Filter is not what our current manual describes. If the effect is "correct" then I think it has to be accounted for in the documentation. The description at the top of the manual page says: "Scientific Filter provides 3 different types of filters which together emulate the vast majority of analog filters," If that is the intention of the effect then there should probably also be Bessel, Elliptic and Linkwitz-Riley filters (which are probably more common than Chebyshev type 2). If that is not the intention, then the description needs to be "toned down" by removing the second part of that sentence. Steve > > Now if I change my Butterworth first order 1000 Hz LP filter to > "Chebyshev I", there is less attenuation than Butterworth > (at default 1 dB "Passband Ripple") - look at 5000 Hz for example. > As greater attenuation is expected with a Chebyshev than a > Butterworth filter, is 1 dB the correct default for "Passband Ripple"? > > At what points in the stopband part of a Chebyshev filter do you > expect a first order rolloff of -6 dB per octave? For example with > Chebyshev I Passband Ripple of 1 dB (or Chebyshev II Stopband > Ripple of 1 dB), 5000 Hz has a gain of -9 dB, but 1000 Hz has a > gain of what looks like -16.5 dB. > > > > Gale > > ------------------------------------------------------------------------------ > Rapidly troubleshoot problems before they affect your business. Most IT > organizations don't have a clear picture of how application performance > affects their revenue. With AppDynamics, you get 100% visibility into your > Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro! > http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk > _______________________________________________ > Audacity-quality mailing list > Audacity-quality@... > https://lists.sourceforge.net/lists/listinfo/audacity-quality > ```
 Re: [Audacity-quality] Scientific Filter graph From: Gale Andrews - 2013-12-29 23:14:56 ```| From Steve the Fiddle | Sun, 29 Dec 2013 11:54:26 +0000 | Subject: [Audacity-quality] Scientific Filter graph > On 26 December 2013 17:56, wrote: > > > > Usual disclaimer for any any ignorance, but I am confused when I > > look at http://wiki.audacityteam.org/wiki/Low_Pass_Filter > > and see that the graph for Butterworth filter in Scientific Filter > > does not conform to attenuation of -6 dB at double the corner > > frequency. Or should I not expect conformance? > > > > In the attached image (1000 Hz first order low pass, range set to > > +6 to -61 dB but the range makes no difference AFAICT), 1000 Hz > > seems to be at -3 dB as expected, but 2000 Hz is at -7 dB instead > > of -6 dB. > > > > Similarly with first order low pass at 250 Hz, 500 Hz appears to > > be at -7 dB instead of - 6 dB. > > > > The rolloff above double the corner frequency seems to be > > "correct" (-6 dB per octave). > > > > The Manual (most of it verbatim from Norm's -devel post) says > > for Butterworth that "attenuation increases at approximately > > 6 dB per octave times the filter order". Why "approximately", or > > is that another way of describing what I see? > > The graph closely matches the actual frequency response of the filter, > so no problem there. What explanation do we give for the non-conformance with Nyquist High Pass and Low Pass (strict -6 dB rolloff)? We're sure to be asked what the differences are between Nyquist High Pass and Low Pass and Scientific Filter. Is one filter "better" than the other in any sense? Do we still want to ship High Pass and Low Pass if we require Scientific Filter? > I'm unsure about whether the first order filters are doing "the right > thing" or not. Certainly the 1st order Butterworth filter does not > match the response that is usually shown for a Butterworth filter (for > example: http://en.wikipedia.org/wiki/Butterworth_filter ) and it does > not match the response of the Nyquist first order Butterworth filter > (the "hp" and "lp" functions), which do match the response as > typically described. I don't know whether this mismatch is "a matter > of implementation" or "a limitation" or "an error", but is so > noticeable that users are sure to ask. I think this needs to be > answered in the manual. Can you be more specific as to the mismatch (off list if you think only I need enlightening). Do you mean this difference (white noise 0.8 amplitude): * 1000 Hz 6 dB rolloff Nyquist Low Pass http://gaclrecords.org.uk/bugs/lowpass_1000hz_order1.png * 1000 Hz first order Butterworth Low Pass in ScienFilter: http://gaclrecords.org.uk/bugs/scienfilter_lowpasss_1000hz_order1.png ? > The response of the 1st order Chebyshev filters also looks rather > "surprising" when "ripple" is set to zero. Again I don't know if it > "should" look like that, but I think it needs to at least be > documented. > > There is little on-iine documentation about first order Butterworth > and Chebyshev filters, but the behaviour of these types in the > Scientific Filter is not what our current manual describes. I see no description of first order filters in the Manual. The Chebyshev types in the Manual refer to "magnitude response of the passband" and "ripple value" but these are not defined. I could guess what these are, but why should any one guess in a Manual? > If the effect is "correct" then I think it has to be accounted for in the > documentation. Assuming you (Steve) are not offering to document the non-conformance, could Norm or a developer enlighten us? It is (should be) a requirement that released effects have reasonable documentation. > The description at the top of the manual page says: > "Scientific Filter provides 3 different types of filters which > together emulate the vast majority of analog filters," > If that is the intention of the effect then there should probably also > be Bessel, Elliptic and Linkwitz-Riley filters (which are probably > more common than Chebyshev type 2). If that is not the intention, then > the description needs to be "toned down" by removing the second part > of that sentence. I agree. I was going to say something like: ''Scientific Filter''''' provides three different types of '''''[[Glossary#filter|filter]]''''' with options for '''''[[Glossary#hp|high pass]]''''' or '''''[[Glossary#low|low pass]]''''' operation. The two [http://en.wikipedia.org/wiki/Chebyshev_filter Chebyshev] filters provide faster attenuation of the filtered '''''[[Glossary#frequency|frequencies]]''''' than [[High Pass Filter]]. One unanswered issue is that default Chebyshev I provides less attenuation than default Butterworth, neither do I think I can call the default Chebyshev I "sharper" than the default Butterworth filter (as suggested in the link above). Gale > > Now if I change my Butterworth first order 1000 Hz LP filter to > > "Chebyshev I", there is less attenuation than Butterworth > > (at default 1 dB "Passband Ripple") - look at 5000 Hz for example. > > As greater attenuation is expected with a Chebyshev than a > > Butterworth filter, is 1 dB the correct default for "Passband Ripple"? > > > > At what points in the stopband part of a Chebyshev filter do you > > expect a first order rolloff of -6 dB per octave? For example with > > Chebyshev I Passband Ripple of 1 dB (or Chebyshev II Stopband > > Ripple of 1 dB), 5000 Hz has a gain of -9 dB, but 1000 Hz has a > > gain of what looks like -16.5 dB. > > > > > > > > Gale ```
 Re: [Audacity-quality] Scientific Filter graph From: Bill Wharrie - 2014-01-05 15:30:13 ```Oops - meant to send this to -quality. Sorry for the duplication. -- Bill On 04/01/2014, at 8:04 PM, Martyn Shaw wrote: > Hi there > > This is a complicated thread, but I'll try and respond here, and > enlighten if I can... > > [snip] > You will get smoother, more repeatable results if you use an impulse > for tests like this. Generate a second or so of silence, zoom in near > the start, use the 'draw' tool to set a sample to '1' and then test. Could you be more specific? :) -- Bill ```