The impact of high-pass filter on the step response

[Kvalsvoll]

Would be really great to understand why this is happening

It is the hp filter causing ringing in the time domain. Any changes in the frequency domain will always be followed by a change in the time domain, and this is the result of a high-pass filter limiting the frequency response, when fr is limited -> time gets longer. Laws of physics, that can not be eliminated.

So, is it audible? Yes, indeed. A very low hp at 10hz is audible, on a full-range system, with the right content. Bass will sound sort of slower, sluggish, and there seems to be more bass. But it is not audible, if the content does not have energy in the affected frequency range, and any negative effects of the filter is insignificant compared to distortion caused by over-excursion.

And it is not audible, if volume is kept lower, so the low frequency part disappear below audible limit. For normal low volume, a f range of 30hz is sufficient.

The solution is obviously to have full-range reproduction down to well below 20hz.

 

[ppataki]

Hi, what is the actual implementation of the minimum phase and linear phase high passes? (what processor or software?)

And particularly, how many taps at what sample rate, are you using for the linear phase?

As mentioned in the first post, I was using CraveEQ
But I have done comparative analysis with Ozone EQ, FabFilter Pro Q3, Jriver's built-in PEQ and the results are the same or similar; point being that minimum phase HP filter will increase post-ringing while linear phase HP filter won't

 

[ppataki]

It is the hp filter causing ringing in the time domain. Any changes in the frequency domain will always be followed by a change in the time domain, and this is the result of a high-pass filter limiting the frequency response, when fr is limited -> time gets longer. Laws of physics, that can not be eliminated.

Thank you, that is helpful
Now what I still don't get is how come that the same filter in linear phase mode won't do that

 

[ernestcarl]

Now what I still don't get is how come that the same filter in linear phase mode won't do that

But it likely isn't the same filter. Open the convolution files and really take a close look at the filters in REW, does it actually look the same?

The latter three are supposed to be the "same" or identical LR4 filters while the first one serves as a control sample.



Even the one generated by REW (from first IR and created in REW's EQ module) doesn't look the same -- exporting and re-importing said filter with different settings will change it a bit as well.

 

[Kvalsvoll]

Thank you, that is helpful
Now what I still don't get is how come that the same filter in linear phase mode won't do that

But it does, it is just that the ringing is moved in time so that it appears symmetrical, so you get some preringing and less tail after. It may sound different too, I have not verified that, so who knows. The only way to reduce ringing is to reduce the sharpness of the filter.

 

[Kvalsvoll]

And if someone wants to check for audibility, it is easy to rig a controlled experiment for this. Just apply the filter on a sound sample, and compare to original, using foobar abx. You can use music, or a step signal. Step signal is good for this, as it is transient and has a frequency distribution with increasing low frequency energy.

 

[ernestcarl]

Connected to the example I gave in this post:

The impact of high-pass filter on the step response

Recently I have been playing around with some high-pass filters on my system (these are my fronts and these are my rears) For music I don't really need a high-pass filter but when I run some Dolby or DTS tests or watch a movie at high volume levels I need to apply a high-pass filter to prevent...

www.audiosciencereview.com

The subwoofer is convolved with a mixed minimum phase EQ purely generated from REW's EQ in contrast to mixed phase FIR filter entirely from rePhase. Check out the wavelet spectrograms as well.

*Correction is only partially linear to economize. Normally, I would split the EQ work between JRiver and the convolver -- in this example you will notice that I used more taps.

 

[gnarly]

As mentioned in the first post, I was using CraveEQ
But I have done comparative analysis with Ozone EQ, FabFilter Pro Q3, Jriver's built-in PEQ and the results are the same or similar; point being that minimum phase HP filter will increase post-ringing while linear phase HP filter won't

Hi , my reason for asking was mainly to find out about your linear phase implementation.
Because, depending on the number of taps and sampling rate used. the linear phase high pass might not be doing anything,
and give you an impulse that looks like no high pass is in place.

So if i may ask again, how many taps at what sample rate?


And for sure, it's well known min phase has post ring.
And complementary linear phase xovers don't pre or post ring. Emphasis on xovers.

The question for me is, with a linear phase system high pass , why aren't your measurements showing pre-ring?
I think they should be, and suspect that low freq resolution to be ineffective... (due to tap count/sample rate)

My suspicion might be bogus, but something appears wrong with your linear phase high-pass measurement.

 

[ernestcarl]

@gnarly and anyone else interested in DSP filter design, can you tell me which of the three filter versions in the attachment below is less objectionable or somewhat better?

I tried replicating my original mixed phase filter in REW as a new min phase filter design by overlaying the curves in the prediction tab. It seems to me that version 2 which is simplified even further may actually be "objectively better" -- perhaps? But I'm not really sure... Now no pre-ringing and less post-ringing?

 

[ppataki]

Hi , my reason for asking was mainly to find out about your linear phase implementation.
Because, depending on the number of taps and sampling rate used. the linear phase high pass might not be doing anything,
and give you an impulse that looks like no high pass is in place.

So if i may ask again, how many taps at what sample rate?


And for sure, it's well known min phase has post ring.
And complementary linear phase xovers don't pre or post ring. Emphasis on xovers.

The question for me is, with a linear phase system high pass , why aren't your measurements showing pre-ring?
I think they should be, and suspect that low freq resolution to be ineffective... (due to tap count/sample rate)

My suspicion might be bogus, but something appears wrong with your linear phase high-pass measurement.

I have actually attached the measurements in one of my earlier posts in this thread - if you open them you will see that the LP filter provides almost the same frequency response (there is fraction of dB difference) as the MP filter - so actually it does what it has to do
I suspect that what we see happening is what @Kvalsvoll was saying above, that 'simply' the ringing is shifted to the 'left' instead of shifting it to the 'right' - which is clearly visible on the wavelet diagrams; it is just that as you can see on the diagrams the amount of shift is far less vs with the MP filter shifting it the to 'right'

Re. taps and frequency response: frequency response is 48kHz and I cannot tell you the number of taps since I don't know how many taps CraveEQ is using under the hood

 

[gnarly]

I have actually attached the measurements in one of my earlier posts in this thread - if you open them you will see that the LP filter provides almost the same frequency response (there is fraction of dB difference) as the MP filter - so actually it does what it has to do
I suspect that what we see happening is what @Kvalsvoll was saying above, that 'simply' the ringing is shifted to the 'left' instead of shifting it to the 'right' - which is clearly visible on the wavelet diagrams; it is just that as you can see on the diagrams the amount of shift is far less vs with the MP filter shifting it the to 'right'

Re. taps and frequency response: frequency response is 48kHz and I cannot tell you the number of taps since I don't know how many taps CraveEQ is using under the hood

Hi ppataki,

Please forgive me if I'm barking up a tree, but I think the CraveEQ is maybe doing some funky things that don't fit normal ways of thinking about minimum phase and linear phase.
I'm also getting strange results in REW from the mdat you gave in #6, and they also don't match the wavelets in the thread starter.

Strange results first:
I removed the 1/12th smoothing on the measurements, and applied FDW9 instead which has the same 1/12th octave smoothing.
All I was trying to accomplish was to see the phase better.
And was very surprised to see how SPL changed.
You can see in the small measurement panes on the left, how all the SPL traces developed large notches.
Here's the Flat L Flat FIRS after FDW9. I've no clue what's going on vs the flat trace that appears without FDW, and regular 1/12th smoothing.

And here is the spectro I get for that same flat FIR measurement, either with or without FDW9.
And it is so different than the wavelet in the opening post. What spectro settings were you using, maybe that's it ?

Here's the spectro for low shelf + HP LinPhase.
It looks about the same as Flat FIRS (which it should I think other than low boost)
But again, it is way different than the opening post. And clearly shows pre-ring, (and post).

Here's the step response i get of that low shelf + HP LInPhase
Appears to match the pre-ring in the spectro pretty well.

Anyway, bottom line is dunno what's going on....
I can't help but think there is some funky stuff happening in the Crave EQs

 

[ernestcarl]

And it is so different than the wavelet in the opening post. What spectro settings were you using, maybe that's it ?

I have not rechecked the difference in the mdat on this one, but the wavelet settings from previous graphs do not use the ones you have enabled. The step response, though, is similar to what I’ve seen from before with crave EQ engaged *(or was it curve eq?). If the EQ filter applied in Crave could be exported and examined (even an analogous one recreated in rePhase) maybe that would help explain the behaviour of the filter better. My suspicion for now is the entirety of the filter design is linear phase causing the ringing. Normally lin phase EQs is used sparingly.

*smoothing might be 1/1 or 1/2 and linear (% peak).

 

[ernestcarl]

I removed the 1/12th smoothing on the measurements, and applied FDW9 instead which has the same 1/12th octave smoothing.
All I was trying to accomplish was to see the phase better.
And was very surprised to see how SPL changed.
You can see in the small measurement panes on the left, how all the SPL traces developed large notches.
Here's the Flat L Flat FIRS after FDW9. I've no clue what's going on vs the flat trace that appears without FDW, and regular 1/12th smoothing.

BTW, frequency dependent windowing discards energy outside the set window, hence the notches you saw -- that's very much normal and expected behavior in many small rooms. A regular smoothing filter is not the same as it uses the full IR.

I've installed the crave eq 2 plugin out of curiosity, and I don't understand how the time value setting in "linear" mode works. I'm only getting an additional ~30-40 ms of delay from their DSP in the chain after selecting "1000ms".

*presuming this is the correct plugin as I've previously got confused "Curve EQ" (which pptaki also have used) and this one


-------


The ff. may clear things a bit:

Both plugins are embedded in the chain. In post #6 the third measurement was the only one that appears to have used entirely min phase EQ. I am guessing in the fourth measurement Curve EQ remains in min phase mode with Crave EQ applying the linear phase HPF. But @ppataki should be able to clarify more of what's going on.

 

[gnarly]

@gnarly and anyone else interested in DSP filter design, can you tell me which of the three filter versions in the attachment below is less objectionable or somewhat better?

I tried replicating my original mixed phase filter in REW as a new min phase filter design by overlaying the curves in the prediction tab. It seems to me that version 2 which is simplified even further may actually be "objectively better" -- perhaps? But I'm not really sure... Now no pre-ringing and less post-ringing?

Hi, I've looked at the measurement set and really don't know what to say about them.
I'm not a fan of using REW for looking at phase, because i think it has become so full of options, that it's become complicated to the point of nearly unusable.
Like wise, I don't look at spectrographs for timing or ringing etc.

I tune using Smaart, where all i pay attention to is magnitude and phase. Then after tuning for flat mag and phase, I'll measure with REW to see how all the ancillary graphs look.
Impulse and spectrograph/waterfalls will always look good. Step response will show pre-ring. I never concern with post-ring.


So when I look at the phase traces in the 4 measurements in your set, I'm bewildered about their phase traces. They aren't anything I know how to work with.
Maybe because they are taken in-room? I dunno.
I tune a sub outdoors to tie to the main speaker with flat mag and phase. And then forget about phase indoors.
For example, here's a sub i tuned on the driveway. it's flat phase trace is of course easy to tie to any main speaker after deciding on xover freq.

This has become so easy, it's all i do anymore. I'm really not very good with minimum phase work, other than driver EQs before lin phase xovers, and system hp and lp.
Sorry I'm not more help...

 

[ernestcarl]

So when I look at the phase traces in the 4 measurements in your set, I'm bewildered about their phase traces. They aren't anything I know how to work with.
Maybe because they are taken in-room? I dunno.
I tune a sub outdoors to tie to the main speaker with flat mag and phase. And then forget about phase indoors.

Well, in-room measurements at the MLP has a high probability of mangling the phase response almost making it unrecognizable from the default (quasi)anechoic response. I've seen worse looking in-room transfer functions from very expensive speaker systems.

Sorry I'm not more help...

No, problem. The differences are rather small in the grand scheme of things -- I could A/B the three sets and still likely won't be able to "hear" which filter sounds superior -- though, notice a difference? Absolutely, but, not as to which is truly perceptually the better one.

 

[ppataki]

BTW, frequency dependent windowing discards energy outside the set window, hence the notches you saw -- that's very much normal and expected behavior in many small rooms. A regular smoothing filter is not the same as it uses the full IR.

That is correct; if I turn on FDW, the FR will change
Personally I don't use FDW but smoothing instead

And it is so different than the wavelet in the opening post. What spectro settings were you using, maybe that's it ?

Currently I am not at home but in the evening I will post my wavelet settings here

Both plugins are embedded in the chain. In post #6 the third measurement was the only one that appears to have used entirely min phase EQ. I am guessing in the fourth measurement Curve EQ remains in min phase mode with Crave EQ applying the linear phase HPF. But @ppataki should be able to clarify more of what's going on

Correct, the only difference in the measurements is that CraveEQ was in Analog mode (min phase) or in Linear Phase 75ms mode - I mean only the last CraveEQ instance in the chain, where I apply the HP filter (the other CraveEQ instances were all in Analog mode all the time)

Later today or in the coming days I can run similar tests replacing the last CraveEQ instance with Fabfilter Q3, Ozone EQ, etc.
Last time (a few years ago) when I did that as far as I remember I came to the same conclusion (=linear phase HP causing some pre-ringing and min phase HP causing a lot of post-ringing)
I don't think that CraveEQ is to blame here; I think that is how generally these filters behave but again I will post more measurements once I have time to do so

 

[Frgirard]

That is correct; if I turn on FDW, the FR will change
Personally I don't use FDW but smoothing instead



Currently I am not at home but in the evening I will post my wavelet settings here



Correct, the only difference in the measurements is that CraveEQ was in Analog mode (min phase) or in Linear Phase 75ms mode - I mean only the last CraveEQ instance in the chain, where I apply the HP filter (the other CraveEQ instances were all in Analog mode all the time)

Later today or in the coming days I can run similar tests replacing the last CraveEQ instance with Fabfilter Q3, Ozone EQ, etc.
Last time (a few years ago) when I did that as far as I remember I came to the same conclusion (=linear phase HP causing some pre-ringing and min phase HP causing a lot of post-ringing)
I don't think that CraveEQ is to blame here; I think that is how generally these filters behave but again I will post more measurements once I have time to do so

I came to the same conclusions as you with Fabfilter Pro Q and the HPF of my RME Babyface

 

[fluid]

My approach to avoiding over excursing a line array is to equalize the low end roll off to a specific target using PEQ rather than applying steep filters.

In this image you can see the difference between a target in green that has a 1dB/oct fall above 40Hz with the low end being representative of a 22Hz 24dB/oct filter in REW and a 30Hz 24dB/oct filter. There is much more protection in the blue curve from dinosaur footprints and the like without really losing low end.

I do not find it sounds right to just drop off a cliff at some point to try and extend the lowest point before roll off by a few Hz. What I prefer is to have the magnitude and phase at the listening position be as close to a minimum phase representation as it can be. This ends up with a mixed phase filter where there is some time correction but it does not result in a totally symmetric impulse with a lot of pre ringing. This only works with the right windowing to remove large non minimum phase variations, smoothing isn't going to work.

 

[Chromatischism]

It looks like this:

View attachment 226374

Here you are:

View attachment 226375

This would definitely be audible in a movie with a heavy LFE scene. With the filter on, the impact would be much less.

 

[ppataki]

Currently I am not at home but in the evening I will post my wavelet settings here

here is the wavelet settings I use:

I have a hard time doing any measurements this week (whole family being home) but will post them as promised whenever I have some time......