High-pass filter in Roon

[Chagall]

I added a high-pass filter in Roon to see if I could hear a difference in layering and overall clarity on certain songs without the bass content. But adding the high-pass filter results in digital clipping. Can anyone explain why?

Scenario:
- set the headroom adjustment to -4dB to accommodate EQ -> no digital clipping
- add one more band to the EQ, set it as a high-pass filter at 30Hz, 36dB/octave slope -> digital clipping

Could this be caused by a phase shift due to the steep filter?

Thanks!

 

[DVDdoug]

I wonder if something "strange" is going on... 4dB is LOT of change!

No filter is perfect, and some filters create ripple in the passband as one of the compromises.

And yes, filters can introduce different phase shifts at different frequencies. When multiple-different frequencies are time/phase shifted differently their peaks align differently and you can get some higher and some lower peaks. Normally all of these frequencies should combine randomly (uncorrelated) so phase shifting wouldn't make much difference, but when the recording is limited/compressed the highest combined peaks are "pushed" down. When they are phase-shifted the peaks align differently making some new higher peaks.

The process of cutting and playing a vinyl record or converting to MP3 also creates some higher (and some lower) peaks. They aren't really "louder" and the average/RMS is the same, but they combine with some higher and lower peaks.

 

[Chagall]

I wonder if something "strange" is going on... 4dB is LOT of change!

No filter is perfect, and some filters create ripple in the passband as one of the compromises.

And yes, filters can introduce different phase shifts at different frequencies. When multiple-different frequencies are time/phase shifted differently their peaks align differently and you can get some higher and some lower peaks. Normally all of these frequencies should combine randomly (uncorrelated) so phase shifting wouldn't make much difference, but when the recording is limited/compressed the highest combined peaks are "pushed" down. When they are phase-shifted the peaks align differently making some new higher peaks.

The process of cutting and playing a vinyl record or converting to MP3 also creates some higher (and some lower) peaks. They aren't really "louder" and the average/RMS is the same, but they combine with some higher and lower peaks.

Thank you for the response!

4dB is a lot? With HD800s, I have 7dB of headroom adjustment and a far more complex EQ.

What I find strange is that, however complex the EQ graph is, I never got the clipping indicator to light up as long as the headroom is equal to or a bit more than the sum of the EQ filter at any given frequency (for HD800s biggest change is in the bass), but as soon as I enable high-pass, I get clipping.

HD 800s -7dB headroom or negative preamp:

DCA E3 -4dB (actually -3dB would be fine also)

But as soon as I do this:

The indicator shows clipping - not all the time, but regularly.

To get rid of it, I need to set the headroom to -8dB?! So that is 4 - 5dB of increased gain around the cutoff frequency? That seems a lot for a phase shift.

 

[Soniclife]

To get rid of it, I need to set the headroom to -8dB?! So that is 4 - 5dB of increased gain around the cutoff frequency? That seems a lot for a phase shift.

How much attenuation do you need if you only use the high pass? I've never needed more than 3db, but I don't use such a steep rolloff.

 

[Chagall]

How much attenuation do you need if you only use the high pass? I've never needed more than 3db, but I don't use such a steep rolloff.

Tried now with just the intro to Get Lucky:

- with only high-pass filter (30Hz, 36dB/octave) -5dB of attenuation to avoid clipping
- with high-pass filter (30Hz, 36dB/octave) and other filters (DCA E3 screenshoot from the previous post) -6dB of attenuation to avoid clipping

Sorry -8dB must have been a different EQ profile, or it was on a different song.

Edit: And of course, if I bypass all filters, 0 dB attenuation is fine.

 

[DVDdoug]

4dB is a lot?

If you are getting 4dB higher peaks (or 4dB of clipping) just from a high pass filter, that surprises me.

...Or maybe it shouldn't have surprised me. I just opened a random file in Audacity, normalized for 0dB peaks, and applied a 30Hz, -36dB/octave filter. The new peak was about +2.5dB.

Then I tried the same thing with a sweep tone. The peak went up a tiny-tiny bit (+0.013dB).

That implies it's mostly phase shift related since it's almost unchanged with one frequency at a time. That fraction of a dB might be ripple, or it might be a small "rounding error" or some other error/imperfection in the filter.

 

[Chagall]

If you are getting 4dB higher peaks (or 4dB of clipping) just from a high pass filter, that surprises me.

...Or maybe it shouldn't have surprised me. I just opened a random file in Audacity, normalized for 0dB peaks, and applied a 30Hz, -36dB/octave filter. The new peak was about +2.5dB.

Then I tried the same thing with a sweep tone. The peak went up a tiny-tiny bit (+0.013dB).

That implies it's mostly phase shift related since it's almost unchanged with one frequency at a time. That fraction of a dB might be ripple, or it might be a small "rounding error" or some other error/imperfection in the filter.

Oh, I thought you meant 4dB of attenuation...

Okay, so basically Roon clipping indicator works as intended, and by adding such a steep high-pass filter, I must attenuate by X dB depending on the song.

That is so counterintuitive, because I'm cutting frequencies, but here it is.

Thanks for the effort!

 

[danadam]

That is so counterintuitive, because I'm cutting frequencies, but here it is.

Here should be easy to see how removing a bass frequency might increase a peak:

 

[Sokel]

I would check electrically for added noise.

 

[Chagall]

Here should be easy to see how removing a bass frequency might increase a peak:

View attachment 466587

Cool, adding HPF to your file increased the amplitude from 0 dBTP to 3,03 dBTP. I'll play around with the slope some more. Thanks!

 

[Chagall]

I would check electrically for added noise.

It seems this is happening purely in digital domain due to the filter. But what did you have in mind?

 

[Sokel]

It seems this is happening purely in digital domain due to the filter. But what did you have in mind?

If the filter is as low as I think it is, depending the filter stacking with the rest of the filters, it could introduce all kinds of effects including quantization noise, as broadband as to add to level (additively across its bandwidth) .

It's a long stretch, but just a consideration.

 

[Chagall]

If the filter is as low as I think it is, depending the filter stacking with the rest of the filters, it could introduce all kinds of effects including quantization noise, as broadband as to add to level (additively across its bandwidth) .

It's a long stretch, but just a consideration.

Oh, okay! I need to read up a bit more about quantization noise, but as I understand, it is a less of an issue for Roon 64-bit float processing.

 

[Sokel]

Oh, okay! I need to read up a bit more about quantization noise, but as I understand, it is a less of an issue for Roon 64-bit float processing.

...or non-issue at all of its own.
Interactions though can be strange, I always check electrically what's going on and I have seen strange things, purely on the digital domain.

A test you could do is to scrap any other filter, leave the subsonic one and see what happens.

 

[Chester]

This is an interesting one @Chagall , thanks for posting about it.

I also have a single high pass filter for my bookshelf speakers and haven’t added any headroom, as I (clearly wrongly) assumed that because I was only cutting, additional headroom would not be required. I haven’t checked it for clipping but I will do.

 

[Chagall]

...or non-issue at all of its own.
Interactions though can be strange, I always check electrically what's going on and I have seen strange things, purely on the digital domain.

A test you could do is to scrap any other filter, leave the subsonic one and see what happens.

Thinking I’ll run a test in Audition with actual music to see what’s happening with true peaks and noise.

Plan is to try the following:
- No filters applied (baseline)
- High-pass filter at 30 Hz with 6, 12, 24, and 36 dB slopes
- Same HPF settings + HD800S EQ, since it’s a more complex
- Maybe one HPF at 80 Hz just to see if that behaves differently

Anything else I should try while I’m at it?

 

[Sokel]

Anything else I should try while I’m at it?

Levels, as to find the sweet spot.
Couple of dB up or down.

 

[Chagall]

Billie Eilish – Bossa Nova
Statistics are for the entire song, including the minimum RMS amplitude.

No filters	Channel 1	Channel 2
Peak Amplitude	-0.2dB	-0.2dB
True Peak Amplitude	0.47dBTP	0.62dBTP
Possibly Clipped Samples	0	0
Minimum RMS Amplitude	-80.64dB	-77.79dB

HPF 30Hz 36dB/octave	Channel 1	Channel 2
Peak Amplitude	4.94dB	5.39dB
True Peak Amplitude	4.96dBTP	5.39dBTP
Possibly Clipped Samples	26371	26162
Minimum RMS Amplitude	-97.21dB	-94.54dB

HPF 30Hz 24dB/octave	Channel 1	Channel 2
Peak Amplitude	5.15dB	5.40dB
True Peak Amplitude	5.16dBTP	5.43dBTP
Possibly Clipped Samples	27995	28408
Minimum RMS Amplitude	-97.19dB	-94.46dB

HPF 30Hz 12dB/octave	Channel 1	Channel 2
Peak Amplitude	3.49dB	3.67dB
True Peak Amplitude	3.49dBTP	3.71dBTP
Possibly Clipped Samples	20282	20382
Minimum RMS Amplitude	-97.16dB	-94.41dB

HPF 30Hz 6dB/octave	Channel 1	Channel 2
Peak Amplitude	2.10dB	1.92dB
True Peak Amplitude	2.10dBTP	1.96dBTP
Possibly Clipped Samples	7280	7474
Minimum RMS Amplitude	-97.17dB	-94.43dB

HD800s EQ	Channel 1	Channel 2
Peak Amplitude	3.72dB	3.59dB
True Peak Amplitude	3.75dBTP	3.66dBTP
Possibly Clipped Samples	90954	89402
Minimum RMS Amplitude	-77.71dB	-74.88dB

HD800s EQ + HPF 30Hz 36dB/octave	Channel 1	Channel 2
Peak Amplitude	6.21dB	6.69dB
True Peak Amplitude	6.23dBTP	6.78dBTP
Possibly Clipped Samples	81086	82042
Minimum RMS Amplitude	-96.47dB	-93.35dB

HPF 80Hz 36dB/octave	Channel 1	Channel 2
Peak Amplitude	3.56dB	3.06dB
True Peak Amplitude	3.56dBTP	3.06dBTP
Possibly Clipped Samples	2474	2039
Minimum RMS Amplitude	-97.26dB	-94.53dB

Takeaways:

• Add digital attenuation always when applying a high-pass filter (HPF).
• The steeper the slope, the more attenuation is needed - but interestingly, the 24 dB/octave filter produced higher peak amplitudes than 36dB/octave.
• At 80 Hz, even though overall levels were lower, peak amplitude still reached +3 dB - so attenuation is still necessary.
• Noise is not being introduced by the filter.