Filter in vs out -peak level-

[zamproject]

Hi all

Ok this is digital so it's maybe better at -truth table- sub forum, but also EQ relative...

I was recently hired for -mastering- job only, well I'm not a ME, but I have ear and do the job as good as possible for what is requested...

The point is I usually -master- most of the producing/mixing going out of the room and invest in the basic tool few years ago to export DDP
As 98% of the sound (for me) is done at rec/mix side, the master is usually a very little limiter, basic safety filter, authoring and DDP export or whatever transfert format is requested for the final medium.

For the previously mentioned master, I use the Filter/EQ plugin of my tool (I won't name it for now...) more than before and realise strange behaviour with the meter (peak sample ?) at input vs outpout

With only a simple low cut at 30Hz (whatever order) and Q at 12 (which seem non resonant by visual graph and ear) my output meter show peak louder than input meter ?!?

I forward this observation to the dev, thinking of a bug in the peak in/out detection and coding, and the answer is:

With each filtering, individual samples and thus also the peak value can be higher than before. This is completely normal.

Is there something I'm missing ?!?

Non resonant high or low pass should result in lower output level (full range) !?! at least in analog world ?

Cheers
Zam

 

[abbey road d enfer]

In the digital world, there is something called intersample peaks. Just imagine you have two samples close to the max; the actual waveform can pass by these two points, but in between, it can exceed 100%FS.
There are some specialized meters that capture intersample peaks.
Now a HPF with a Q of 12 seems to be highly resonant to me...just sayin'

 

[zamproject]

Hey Abbey

I check with ISP before and after the EQ in the chain.
I don't know what is the setting/algo behind the EQ peak meter regarding this, inter-sample detection or not, there is no user option for this one. But I suspect it also read ISP as the result show the same peak difference.

As note about Q, it's the same observation if I set the HPF at 6dB/octave, first order, where no Q setting is available (obviously ?)
I don't know what Q references they used, I just see that below 12 it's soft knee on the graph EQ curve and over that a bell/resonant curve occur.

I don't check precisely at which frequency the output is softer than input, but with the usual 30 or 40Hz HPF output peak are louder

I'm more concerned about the reply I get, which is counter-intuitive to me, are they right or just fooling me ?
Is it possible that in digital domain something -normal- happen at frequency corner (event with first order filter) that can be interpreted louder ? like shifting frequency can produce louder ISP ?

I should test a frequency corner at 44.1Hz which is a round division of 44.1kHz

I just wanted to help the devs who helped me on a previous bug/corrupted setting that occur some times ago.
And now I'm puzzled

Cheers
Zam

 

[john12ax7]

Your intuition is only partially correct. Loudness is really about power not instantaneous voltage. Take a square wave and apply a LPF. The peak voltage will actually increase due to the ringing you now have (look up gibbs phenomenon). The power, however, will indeed be less, which is essentially an integration of the area of the signal.

So it's certainly possible for peaks to be higher. But I also question the Q of 12 setting, that is extremely high if it is indeed actually Q. You could build an oscillator with much less.

 

[zamproject]

Yes I should not said loud, but higher for peak.
For sure my loudness meters show less LUFS at output (I usually use short term)

I agree the Q setting number seem not related too usual, i don't remember... Q=1 is for 1 octave BW@3dB ?
for this EQ:
setting 24dB/oct @100
Q@12 = -3@100
Q@6 = -6@100
Q@3 = -9@100
and it's resonant over 12
Q@14 give unity @100
and for +3 its about 17...

Your intuition is only partially correct

Yes I admit I have issue to get it.
How same sample sequence but with lower -value- at each sample can produce higher ISP, I mean the slope of (interpreted) waves should be flatter in the attenuated zone.

look up gibbs phenomenon

Dose this apply to digital ?

Sorry I'm not enough educate to digital audio computing

Cheers
Zam

 

[abbey road d enfer]

Yes I should not said loud, but higher for peak.
For sure my loudness meters show less LUFS at output (I usually use short term)

I agree the Q setting number seem not related too usual, i don't remember... Q=1 is for 1 octave BW@3dB ?

For asymptotic filters (i.e. that tend asymptotically to - infinity), the BW notion is debatable.
A Butterworth (maximally flat) LPF or HPF has a Q of 0.707

for this EQ:
setting 24dB/oct @100
Q@12 = -3@100
Q@6 = -6@100
Q@3 = -9@100

That is weird. Seems their "Q" figure is the actual amount of attenuation at cut-off.

and it's resonant over 12

It's coherent with the fact that a non-resonant 24dB/oct HPF has a max attenuation at cut-off of 12dB.
But still weird to use a standardized notion for describing something that different, although related.

Dose this apply to digital ?

Yes it does. First discovered by mathematicians exploring teh Fourier series (19th century!).

 

[zamproject]

Tks Abbey

Yes Q setting is strange with this one

A Butterworth (maximally flat) LPF or HPF has a Q of 0.707

45° at -3dB I remember that

Yes it does. First discovered by mathematicians exploring teh Fourier series (19th century!).

So the conclusion is I'm wrong and the dev is right, I won't complain and argue with the guy then

I have a misconception of what happen in a filter , where non resonant filter can produce resonance.

 

[john12ax7]

Forget about ISP and digital for the moment. Just consider that cutting in the frequency domain doesn't necessarily mean lower amplitude at every point in the time domain. The attached shows a square wave with different colors for effectively different amounts of LPF. The peak amplitudes are higher even as frequencies are cut.

There might actually be other things going on with the particular plugin. Try some others and compare the results, this would help confirm if there is an implementation issue.

 

[abbey road d enfer]

So the conclusion is I'm wrong and the dev is right, I won't complain and argue with the guy then

No; The dev is wrong, using "Q" for a different parameter.

I have a misconception of what happen in a filter , where non resonant filter can produce resonance.

No. Non-resonant filter do not resonate.
The Gibbs phenomenon is not a resonance at all. It changes the peak value, though.

 

[zamproject]

Tks john for the graph

I think I get it now

The Gibbs phenomenon is not a resonance at all. It changes the peak value, though.

Yes, now I can see without math that the half wave square area is lager than the red one (HPF at higher setting right?)
So less RMS level but it peak over at 1/4 and 3/4 of time/wavelength.

Thanks to both of you for refreshing and loading my brain cells

Cheers
Zam

 

[ccaudle]

I check with ISP before and after the EQ in the chain.

What is the difference in level between in and out? You can check with an oversampling meter for the "true peak" value (i.e. including inter-sample peaks) of the input. A filter could move a peak which was previously between sample points onto a sample point, so that a meter which does not read inter-sample peaks could show different (higher) value, but if you checked both with a true peak meter they would not be different.

(look up gibbs phenomenon)

Should be less relevant in this case since the filter was high-pass, not low-pass.

There is a separate effect of filtering which can affect any high, low, or band pass, which is where the phase shift associated with the filter can cause parts of the signal which did not have aligning peaks to align after the filter, at least for short periods of time. If you were listening you probably would not notice, and if you were using analog level meters you probably would not notice, but it shows up if you are using digital peak meters which can catch even the shortest time of signal peaks aligning.

 

[zamproject]

What is the difference in level between in and out?

Don't remember for the songs/album I mastered, I had most important things to deal with

Now with a white noise file

filter set at 100Hz 24dB/oct and Q for -flat- response.
dif between meters with ISP ON before and after filter I get +0.2dB
dif between In/Out meters of the filter +0.4dB

for a 6dB/oct no diff at ISP meters and +0.1 at EQ meters

For sure I catch nothing by ear, and don't care that much if there is inconsistency in numbers that showed on the computer screen, I just notice this and try to understand afterwards.
As the reply from dev was somhow expeditious I'm glad you're all here to give some explanation

 

[kags]

Many digital EQs act this same way. I always demonstrate to my students that EQs often overload when using a HPF. As a rule, I leave at least 3dB of sample peak headroom before applying EQ while mastering program material.

Have you tried using linear phase EQ? When I do have a need for HPF in mastering (which is rare), I always use linear phase EQ. Otherwise, just use a gentle bell cut in the subs to reduce bass but don’t remove all the low info. There is music down there bellow 30Hz.

Your particular EQ may have other problems but this phenomenon is common.

 

[MicUlli]

A 24dB HPF means two 2nd order biquads cascaded. Tuned to Butterworth response (maximally flat) the two biquads have an inherent Q-factor of 0,54 and 1,31. You can see that it is possible to overload the stage with Q = 1,31 for the case that it is the first in the cascade. As already mentioned before the EQ implementations differ. So headroom always matters and should be at least 20dB*LOG10(Qmax)...

 

[abbey road d enfer]

Have you tried using linear phase EQ? When I do have a need for HPF in mastering (which is rare), I always use linear phase EQ.

Don't you have issues with pre-echo? It happens one period of the dialled frequency before. A 40Hz linear-phase HPF has pre-cho 25ms before the attack. 25ms is enough to be perceived as a parasitic signal.

Otherwise, just use a gentle bell cut in the subs to reduce bass but don’t remove all the low info. There is music down there bellow 30Hz.

In most productions, except some classical pieces, I'm not so sure it qualifies as "music".
If you listen to current production, below 30Hz there is mainly noise.
In live sound, the only purpose of subwoofers extending down to 30Hz is shaking trousers and providing erotic suffocation.

 

[zamproject]

There is music down there bellow 30Hz.

Like Abbey, as I don't record/mix symphonic with 5 stings double bass or 62' pipe (8Hz ) , I always put a HPF, usually 30Hz.
To me it's a technical move, avoiding possible noise and unnecessary energy that most of playback systems can't reproduce correctly anyway.

 

[kags]

Don't you have issues with pre-echo? It happens one period of the dialled frequency before. A 40Hz linear-phase HPF has pre-cho 25ms before the attack. 25ms is enough to be perceived as a parasitic signal.
I
In most productions, except some classical pieces, I'm not so sure it qualifies as "music".
If you listen to current production, below 30Hz there is mainly noise.
In live sound, the only purpose of subwoofers extending down to 30Hz is shaking trousers and providing erotic suffocation.

Hey Abbey - I can't think of a single instance where pre-echo or pre-ringing has been audible in a real-world situation. For delicate music styles, I would always audition more than one way to process tracks/mixes and choose the best combination of musical results and the least artifacts. As always, gentle slopes help minimize distortions in any kind of filter. Have you had instances of audible pre-echoes caused by linear phase filters?

I often compare HPF vs low-shelf cut or bax-shaped roll-offs on lead vocals and other instruments and I and my clients almost always prefer shelves and bells over HPF. If a HPF is necessary, then linear phase filters always sound better to me. I have performed this exercise hundreds of times with hundreds of clients and the results are consistent. I've always attributed the coloration caused by (minimum phase) HPF filters to phase response, phase delay, and group delay that occurs for decades or even octaves above the EQ point. I use many different plugin EQs and have found some better than others, but the overall impression of minimum vs. linear for HPF filter duties remains constant for me. For other EQ applications, minimum phase EQ is almost always fine, perhaps with the exception of extremely steep notch filters used to remove resonances.

The audible effect of high pass filters was first brought to my attention by the head of a label who I've produced and engineered many records for—jazz, folk, Latin, and other high-quality acoustic projects. The first time I mixed a project for him he mentioned that the lead vocal sounded not as natural as he remembered it sounding during the tracking session. Our goal was to create a very natural feeling and un-processed sounding master and I had been extremely gentle in my processing of the instruments and vocals, especially compared to the top-40 stuff I often work on(!). This made me really pay attention to how EQ and especially HPF affect the timbre of vocals. I would urge everyone to try some listening tests for themselves and see what works for them. Obviously for highly-processed vocals, this won't matter, but it is always worth considering.

As far as what audio lives below 30 Hz, there are plenty of songs where the fundamental bass note reaches low C (32.7 Hz) and even below — and these are pop songs, not organ music or log drums. "Royals" by Lorde is a good example of a clear C bass fundamental that should be clearly heard. A HPF filter with a corner at 30 Hz or even 20 Hz will lower that note by at least a few dB, so I prefer to be as gentle as possible when cleaning out the basement. While mixing individual tracks I might HPF many instruments, especially miked instruments, reverbs, and other sources with musically unnecessary lows/subs, and the artifacts of filters on those instruments tend to be unnoticeable.

A good mix engineer will clean up low-frequency problems that eat headroom and muddy mixes so that during mastering there should be very little need for a "safety" HPF. I have this conversation often with many other professional MEs who feel the same. My clients also prefer masters where the low end is not treated with a HPF. I've done many taste tests with artists and producers whose ears I trust and almost always they (and I) prefer leaving the HPF out, but perhaps still shaping the bottom with a bell or shelf. Try it for yourself. As always, if it sounds good, it is good!

On a related note, I spec, install, and tune a lot of monitor systems for studios and I try to make sure that professional systems are flat to at least 30 Hz, and often below 25 Hz. I have had several studio owners and producers tell me that they have experienced for the first time being able to easily tune sub-basses and 808 kicks without resorting to headphones. This is another indication that the area around 30 Hz is critical to understanding the fundamental frequencies in pop, EDM, and urban genres. Other genres, like rock, punk, jazz, and orchestral probably have their fundamentals (kick drum and bass) in the 45 Hz to 100 Hz range and may tolerate a more aggressive HPF.

I'm in no way claiming that my opinion is what everyone experiences or believes, but it has served my career well and I can reliably hear and demonstrate what I hear and make my choices based on what my clients and I prefer our music to sound like.

 

[abbey road d enfer]

Have you had instances of audible pre-echoes caused by linear phase filters?

Yes, with large PA systems. Subwoofers sound weird when using linear-phase High Pass and x-over.

I've always attributed the coloration caused by (minimum phase) HPF filters to phase response, phase delay, and group delay that occurs for decades or even octaves above the EQ point.

AFAIK, phase/delay distortion happens below cut-off, not above.
And I'm not too convinced about the audibility of phase.
In particular, linear-phase filters have a constant group delay, which in turns means that phase is proportional to frequency.
In other words, a filter with a constant delay of 1ms results in 360° @1kHz, 720° @2kHz, and so on.
Linear-phase does not mean constant-phase.

This made me really pay attention to how EQ and especially HPF affect the timbre of vocals.

EQ and HPF are part of the same process of modifying the spectrum. They should not be considered as separate entities. Interaction between the two must always be taken into account.

"Royals" by Lorde is a good example of a clear C bass fundamental that should be clearly heard.

So, it's music that is destined to be fully appreciated only by people who have a system that goes down to 20Hz?

This is another indication that the area around 30 Hz is critical to understanding the fundamental frequencies in pop, EDM, and urban genres.

What systems do listeners use?

 

[zamproject]

For the particular master (I don't mix/record it) which make me start this discussion, analyser show me some tracks with high rumble in the 10-15Hz (can't hear it), to me it make sense to reduce them with HPF.

A good mix engineer will clean up low-frequency problems that eat headroom and muddy mixes so that during mastering there should be very little need for a "safety" HPF

I mix analog, I clean my multitrack playback, DAW EQ or console EQ, (I am good ? )
But things and noise can happen in between when recording back my 2 track, especially in the 1-20Hz range, like fader sliding noise or inherent low noise of the console or treatment at insert.
So in my use case I feel the need for a safety and cleaning HPF when the track enter the mastering chain (digital).

Other genres, like rock, punk, jazz, and orchestral probably have their fundamentals (kick drum and bass) in the 45 Hz to 100 Hz range and may tolerate a more aggressive HPF.

Yes, obviously the song and arrangement have to be taken in account.
Still for very low kick, which usually have more moving rumble (going lower note few ms after attack) due to low tension, at mix it happen I use HPF close to initial resonance for better definition and reduction of the portamento/glissando effect that I usually don't find that pleasing in the low/sub range. But again it depend of the song.

Cheers
Zam

 

[kags]

Yes, with large PA systems. Subwoofers sound weird when using linear-phase High Pass and x-over.

AFAIK, phase/delay distortion happens below cut-off, not above.
And I'm not too convinced about the audibility of phase.
In particular, linear-phase filters have a constant group delay, which in turns means that phase is proportional to frequency.
In other words, a filter with a constant delay of 1ms results in 360° @1kHz, 720° @2kHz, and so on.
Linear-phase does not mean constant-phase.

EQ and HPF are part of the same process of modifying the spectrum. They should not be considered as separate entities. Interaction between the two must always be taken into account.

So, it's music that is destined to be fully appreciated only by people who have a system that goes down to 20Hz?

What systems do listeners use?

Abbey - I'm sure you are correct about the effects of linear phase EQ in PA systems with subwoofers - especially the way subs are employed at EDM festivals and in clubs. My biggest annoyance in those systems is protective limiting to keep the DJs from overloading the system, but with slow attack times. That creates some really strange musical effects! Not to mention subs that cross-over — no, not even crossover— they simply overlap the mains at ridiculously high frequencies.

I've attached an image that purports to show the phase and group distortions from a minimum phase EQ (from Izotope's Ozone 10). I'm sure they have simplified the graph, but it clearly demonstrates distortions that occur above the knee of a HPF. Regardless of the image, I can clearly hear the difference between a vocal that has been high-passed (even well below 100 Hz) as compared to one that has not been high-passed, but has some tonal shaping on the bottom.

I think the largest audience for pop music listens on headphones that are generally flat or even exaggerated down to 20 or 30 Hz. Even Apple Air Pods produce a lot of output down to 20 Hz. Besides that, as a creator I need to hear if there are problems in the low subs to even decide if I need to worry about remove said frequencies. Simply adding a "protective" HPF is not an appropriate solution to me. People with limited response sound systems will enjoy music without artifacts in the midrange thanks to the mixer judiciously applying filters.

As for @amproject and the initial issue - yes, feel free to clean up any subsonics that you see (or even better hear). I would urge you to try a few different techniques to remove any low frequencies and see if there is a method that sounds best to you, on your system. I usually go for a bell-shaped dip centered as low as the EQ will allow and then widen the Q as needed to remove the appropriate frequencies. Only a few dB of cut should be necessary to push the offending lows down to an acceptable (inaudible) level. The highest bass energy of most pop music will be between 40 Hz and 200 Hz, but 20 Hz content might only be 15 dB lower than 40 Hz content.

Always, if it sounds good it is good, but do yourself a favor and audition some alternatives to simply hitting everything with a HPF.

Attached are screenshots that show Izotope Ozone 10's phase and group distortions and another graph showing Apple Airpod's frequency response (as measured by Sonarworks).