[bigshot]

And reserving 10dB as a headroom would more than cover any reduced volume caused by EQ.

 

[ADUHF]

Right, but I'm saying that you need a 10dB headroom on your amp over maximum normal listening volume. The added 10dB headroom would give you room to double the weaker signal.

Here is the question I've been answering...



The question was how much headroom does an amp need properly render transient peaks? I pointed out that 10dB more than covered that. Now we're talking about EQ. 10dB is enough to cover that too.

For playing back commercially recorded music, a 10dB buffer over the amp's maximum volume is plenty enough.

On my AVR, the volume is measured in terms of -dB below zero, which is the theoretical limit of the amp. Generally, I play modern movies (which are more dynamic than most music) at -9 to -15dB, depending on the level of the soundtrack on the disc. I've never run into an instance where my amp was insufficient.

You can go ahead and buy an amp as big as Hoover Dam, but it's not going to make any difference. When it comes to headroom, enough is all you need. That's true of a lot of things in home audio... frequency extension, transparent noise floor and distortion, lossy audio artifacting, etc.

Audiophools are always hedging their bets, arguing that enough is never enough. "If a response that goes to 20kHz is good, then one that goes to 40kHz must sound better, right?" "If I gain ride the fadeouts on songs by a massive amount, I can hear the digital noise floor- therefore a noise floor below -120dB is necessary." "I used to have an amp with .01% distortion. Now I have one with .005% and it sounds much better." You can see bologna like this in a lot of the threads on Head-Fi. One person mentions a threshold of transparency or recommended spec and then all the armchair experts try to think of a rare exception that might mean that figure is a little too low. Then another armchair expert tries to one up the first armchair expert by mentioning an even more unlikely exception. Again and again until we are running down a rabbit hole into La-La Land. It has nothing to do with how an amp or DAC actually performs in the real world. It's all inside of people's heads. It's ego gratification. They aren't actually helping anyone put together a good sounding stereo system. "I'm not a doctor, I just play one on TV."

Correct me if I'm wrong (as that seems to be a distinct possibility here ), but if your amplifier and speakers were set up in the usual way, then wouldn't the 0 dB position on the amp's volume control be calibrated to produce an 85 decibel level at the listening position from your speakers with a -20 dbFS signal?

 

[bigshot]

No, zero is the theoretical clipping point of the amp- full rated power output. The actual volume produced depends on the efficiency of the speakers. I’m able to produce 80dB at a -9 to -15dB setting depending on how dynamic the movie is.

 

[ADUHF]

I think one of the problems here is that we are all using different benchmarks for the decibel level. I'm trying to use the decibels at either 0 or -20 dBFS, based on the signal. You are using the dB for sustained peaks. And VNandor is using the average sustained dB level, or RMS.

I think it probably all comes out similarly in the end. But it makes it very confusing, because we're all on different pages in terms of what the reference point for headroom should be.

 

[ADUHF]

Fwiw, this was my final though re the 85 dB at -20 dBFS reference listening level...

The 85 dB reference level is supposedly fairly loud though. So most folks actually listen at about -10 dB (or even lower) below that. Which would put any 0 dBFS transients more in the 95 to 105 dB range.

The 95 dB figure above is pretty close to the level that you have suggested for sufficient headroom. So it seems to me that we are pretty much on the same page on that.

I believe that 85 dB (with a -20 dBFS signal) is the target dB level for unity on the master volume control though on alot of home theater setups. So that is what I would personally try to aim for in my setup. The chances are remote that you'd ever need that much volume in normal listening though.

As stated above, most people will set the master volume on their HT systems more in the -10 (or lower) range when listening normally. Just as you are doing on your amp. Which means the effective headroom will be closer to the 95 dB range in the treble and midrange. And up to about 105 dB in the bass. And the level for a -20 dBFS signal will be around 75 dB.

Maybe this is all wrong though.

 

[bigshot]

It’s easier to just calculate the maximum output in dB and say if it’s over 90 to 95dB, you’re probably fine. Most people who are asking this question just want reassurance that they haven’t bought the wrong amp for the job. Or they’re shopping and they want to know what to look for on the spec sheet.

 

[bigshot]

The reason I choose sustained peaks is because that is what people use to adjust the volume when they are listening to music in their living room. “Too loud” usually means the overall sustained volume is too loud. Other ways of measuring would be different depending on how compressed or dynamic the recording is. For instance, a median level with a highly compressed pop song would sound a lot louder to human ears than the exact same median level with dynamic orchestral music like Tchaikovsky’s Pathetique. The perceived volume of what “loud” sounds like to ears is what matters most. 80dB is just over the border into uncomfortable for most, if not all people.

 

[ADUHF]

Everything you're saying above makes sense. Fwiw, I made a couple edits to my last post above to try to clarify what I'm talking about a bit more.

 

[VNandor]

Fwiw, your comments about EQ make some sense, if you're going by the volume of an amp based on listening. Perhaps less so, if you are using measurements, because the goal of EQ is generally to make the loudness more uniform across the frequency range at your target dB level.

I think bigshot's point on this is simply that the change in overall levels between an EQ-ed and un-EQ-ed headphone would generally be too small to make a noticeable difference in the sound performance, even if you have to raise the amp a few dB's to compensate for the equalization.

The difference between equalized and unequalized levels are often so big there are several EQ plugins that offer loudness compensation for a better A/B-ing so you don't get distracted by the difference in levels and pick the louder one by default. Obviously it depends on how you use the EQ but it's used in this way often enough to warrant this function.

How would EQ increase the dynamics of transient peaks?

I'm sure you are not actually interested in my explanation so I'll just bring an example and you can check if it holds true for your usual cases as well.

I wanted to grab some typical EQ curve from a database like oratory's or something from AutoEQ but these use aggressive EQing so I just picked the EQ curve that I use instead. It's a gentle low shelf with 4.5dB gain combined with a cut around the midbass to keep the overall bass level around the same. The point is that I'm using a gentler than usual EQ curve. The only reason this example is a bit suggestive is because I not only picked some club music (notorious for being loud) I picked something that's louder than usual even for that type of music. The basic intuition is that if changes are made to music that's been highly optimized for loudness, there's a good chance the changes are going to ruin the loudness the producers worked so hard for.

So applying this EQ results in the following:

Both signals are peaking at 0dB. You can clearly see the EQ'd signal's average level is reduced. The global gain that had to be added is -4.5dB so obviously the reduction can't be much more than that. Since the music was loud (or rather, well-balanced) to begin with, the reduction ended up not far off from that, at around at 3.7dB.
Again, if the amp wasn't already pushed to its limit, this wouldn't matter, but if it were, some loudness would have to be sacrificed in order to EQ the signal. (because you aren't allowed to add more gain, that's the definition of pushed to the limit, right???)

If it's still not obvious that even reasonable EQ can meaningfully increase the difference between "sustained" peak levels and the peak level I could calculate the sustained peaks instead of average but I would need an actual time window for that because there's no standard or definiton of what counts as "sustained" peak. I'm actually interested what counts as sustained peak to you. This could turn 5 pages of back and forth discussion into 2 posts as @ADUHF noted.

By the way this (obviously) strays away from the original topic at hand. The question was simple and I think it was already answered at the first page in a clear and practical manner. This is mostly about how using EQ influences what should be picked as the target peak level.

 

[ADUHF]

https://en.wikipedia.org/wiki/DBFSFwiw, this was my final though re the 85 dB at -20 dBFS reference listening level...



The 95 dB figure above is pretty close to the level that you have suggested for sufficient headroom. So it seems to me that we are pretty much on the same page on that.

I believe that 85 dB (with a -20 dBFS signal) is the target dB level for unity on the master volume control though on alot of home theater setups. So that is what I would personally try to aim for in my setup. The chances are remote that you'd ever need that much volume in normal listening though.

As stated above, most people will set the master volume on their HT systems more in the -10 (or lower) range when listening normally. Just as you are doing on your amp. Which means the effective headroom will be closer to the 95 dB range in the treble and midrange. And up to about 105 dB in the bass. And the level for a -20 dBFS signal will be around 75 dB.

Maybe this is all wrong though.

Fwiw, I did a bit more reading on this. And the above appears to be essentially correct. (Though I probably haven't done a totally perfect job of explaining it.)

The reference level for unity on the master volume control is apparently always the same, in both professional and home theaters. And in movie mastering suites. And is actually based on the maximum amplitude level possible in the digital audio content, represented by 0 dBFS. The target level for each channel at 0 dBFS is 105 decibels SPL at the listening position. This is for everything except the low frequency effects, where a 10 dB greater target of 115 dB SPL is used.

The target levels for 0 dBFS can be set in a couple different ways. Since it is often impractical to use the above decibel levels at full scale, a pink noise signal of -20 or -30 dBFS is usually used instead. And (as you might guess) the target for each channel is 85 dB SPL* for the -20 dBFS pink noise, and 75 dB SPL* for the -30 dBFS pink noise. Calibrating the unity volume setting to either one of these levels ensures that 0 dBFS is at the desired 105 dB SPL reference level for each channel. There is often an automatic boost for the LFE of +10 in these test signals, to ensure that it also hits the 0 dBFS target of 115 dB.

According to this article, the -20 dBFS level of 85 dB has historically been the target for the average level of dialogue. Which leaves an additional 20 dB of dynamic range (or headroom) in the soundtrack for the effects.

This is what is used for both mastering and final display in a professional setting. And it has also been adopted for the reference level in home theater equipment. In practice though, the unity reference level of 105 dB at 0 dBFS is generally too loud in the smaller space in a home. So a lower level around -10 dB below the unity reference level is often used for actual listening instead. Which reduces the peak 0 dBFS level from 105 to 95 dB. And the 0 dBFS level for LFE from 115 to only 105 dB. And the average level for dialogue from 85 to 75 dB.

https://www.thx.com/questions/what-is-the-reference-level/

(*C-weighted SPL meters are the industry standard for measuring this. But according to one of the links above, A-weighting can be more accurate for setting the individual channels to the same level at the listening position in a home theater, which is what's most important.)

 

[bigshot]

I can't puzzle all that out. Let me ask an easier question... I know that EQing subtractively lowers the volume level. But couldn't you just normalize the equalized track back up to zero and it would have the same basic compression and overall perceived volume that it had before? I suppose you could do a weird EQ around a specific mix element and make that jump way up compared to the rest, but that isn't generally how EQ is used by consumers.

I don't see how high levels of compression can be undone by EQ. If you could just explain how that can happen in laymen's terms it would help me understand. Also, remember that we're talking about playback of music that has been mixed and mastered. We aren't talking about mixing from a multitrack recording. Some of the stuff you were citing above were referring to a mix, not playback of a commercial CD.

 

[ADUHF]

I can give you some quick answers to the above while yer waitin for VNandor to get back.

There are hard limits on how high you can push the volume of audio content in the digital domain. The loudest anything can get is 0 dBFS. So if you want to boost the level in just the bass, for example, with a digital EQ, what you effectively have to do is lower the volume in the rest of the frequency range instead, to make enough headroom for the bass boost. The bass levels stays at around 0 dBFS (the max digital limit), but everything else has to come down to somewhere around -3 or -4 dBFS.

If you don't do it that way, and try to boost the bass above 0 dBFS, then the transient peaks in the lower frequencies in your music could potentially be clipped. So if you want to raise the volume in any part of the frequency range, the rest of the frequency range basically has to come down to make some room for that.

I'm less familiar with dynamic range expansion functions. But I believe there are some apps which are smart enough to analyze your audio content, and make an educated guess at what type of compression was used. And then to reverse it, on the fly in some cases. Which will also result in a reduction in the overall volume. Because the only way to increase the dynamic range without clipping off the louder parts of the music is to do most of the expansion in a downward direction, deeper into the negative dBFS range. The loudest parts of the music, which were close to the digital limit of 0 dBFS before the expansion, will still remain fairly loud. But the quieter parts will get even quieter.

As long as you have ample headroom in your amplifier to accommodate these reductions in volume in the digital domain, there's no problem. Because you can boost the volume on the amp by a few more dBs to compensate for the reduced overall levels in the digital content.

If you are already pushing the limits of your player or amp's capabilities though (which can be a distinct possibility if you're trying to drive a higher impedance headphone without a dedicated amp), then you'll just have to make due with some reduction in volume to gain the benefits of the equalization. (Which would also sort of defeat the purpose of re-expanding the dynamic range.)

 

[VNandor]

I know that EQing subtractively lowers the volume level. But couldn't you just normalize the equalized track back up to zero and it would have the same basic compression and overall perceived volume that it had before?

It's unlikely a track would sound more dynamic after the EQ + normalization back to 0dBFS, however it still would sound quieter.
If you look at the waveforms I posted, you can see that I normalized to EQd signal back to 0dBFS but still, the overall level decreased. I can assure you the perception of dynamics didn't change, the track still starts off loud and then goes to very loud. The perception of overall volume still decreased though.

I don't see how high levels of compression can be undone by EQ. If you could just explain how that can happen in laymen's terms it would help me understand.

It doesn't undo compression. Compression reduces dynamic range in a very specific and easy to understand way. EQ just changes the dynamic range and although the changes in dynamics can be predicted from the input and the actual curve it is not intuitive at all.
The change isn't actually guaranteed to increase the dynamic range (I'm sure you can craft an example if you think in terms of pure sine waves) but with properly mixed and mastered modern music that makes use of compression and limiting, it is practically guaranteed to increase the dynamic range although it would be most likely noticed by the diminished loudness, not by a subjective perception of increased dynamics.

Here's an other example that might help you to understand what i'm talking about. I clipped a full scale sine wave at -6dB so no samples would go over 0.5 (or -0.5).
I then added some random EQ to a couple of cycles. The bottom track is the full scale sine wave I started with.

Technically, the EQ'd part isn't clipped anymore. I think you know that every signal can be looked at as pressure/voltage change over frequencies and phase (just a sum of sine waves) instead of pressure/voltage change over time. They represent the exact same signal.
There's only one specific balance of frequencies that creates this particular clipped sine wave. If you change that balance with an EQ for example the signal isn't clipped anymore. It's not useful in the sense of "unclipping" the signal because once it got clipped, there's no way to tell what were originally above the clipping. The best someone could do is to make an educated guess based on the context around the clipped parts.

This same concept applies to compression as well. The EQ will change the the difference between the peak and average level but not in a way it would undo the effect of compression. The changed signal wouldn't reflect the original uncompressed signal just like how the "not clipped" signal doesn't reflect the original signal.

Some of the stuff you were citing above were referring to a mix, not playback of a commercial CD.

As a consumer, you will get a properly mastered track. For a lot of genres, this means whatever you get got passed through a limiter as a last step. Even though the limiter operates in the time domain, the end result can be viewed at as a specific balance of frequencies. If you change that balance with an EQ during playback, it will reintroduce some "dynamics" but not the kind that was there before the limiting. This effect will be most likely perceived as reduced volume not as bigger dynamics along with the increased/reduced bass/mid/treble depending on what the EQ was intended to do in the first place.

You can see in my first example that even renormalizing the track to 0dBFS doesn't bring back all the volume. How much you can get back depends on the track and on the EQ curve. I just want to point out that people don't equalize their track, export the result and normalize them. They let the DSP work during playback so the near 0dBFS peaks aren't even guaranteed.

If you already have a program that lets you quickly EQ, export and normalize tracks you could look into all of that yourself. The drop in levels will be most prominent with loud tracks and aggressive use of EQ.

 

[bigshot]

If you look at the waveforms I posted, you can see that I normalized to EQd signal back to 0dBFS but still, the overall level decreased.

That is what I don't understand. If that second waveform is normalized up to zero, there has to be a spike in there somewhere that pushes up to the top. That would be a pretty big spike compared with the overall level of the music. My question is, where is it and what caused it? Are you sure that second waveform was normalized up? I don't see anything in there up near zero.

People may not EQ and then normalize, but they will EQ and then turn up the volume. To the ears, that can sound louder or softer, however you set the volume. An amp with 10% headroom would easily compensate if it is quieter. Is this something that only really shows up in waveforms and not actual sound in your living room? When I EQ and I play back music and adjust the level, I don't hear any difference except for the equalization curve. I don't hear any change in dynamics.

I know from experience that waveform depictions aren't at all a fair representation of how music sounds. Maybe this is one of those anomalies that makes me distrust them for anything but the basics.

Here's a for instance... Take a bass heavy rap track. The sub bass is normalized up to zero in the original mix to give it more energy and a big fat sound. The musical content around 1-3kHz only comes up to about 60% of peak. Now you do an EQ cut to take all the frequencies around 2-3kHz down, and normalize it up. The 1-3kHz content gets boosted in volume from 60% to 100% as part of the normalization. The track would now sound thinner, but also louder, not quieter. Right? At least that is what I would think. Whether it sounds louder or softer to human ears depends on the EQ curve you're using and the amplitude of the frequency you are subtractively boosting.

 

[ADUHF]

I can't really speak for VNandor on this. But when I refer to EQ in this context, what I'm mostly talking about is using equalization to correct the tonal and timbral imbalances in my headphones, as opposed to the music. So I will use the same EQ curve for listening to everything. Consequently, no part of the frequency response can be normalized back up in the way you describe, because it might lead to clipping of the near 0 dBFS transients in that range.

If you are EQ-ing individual tracks, or trying to (re-)expand their dynamic range (which is a different, but somewhat related process), then you can potentially play around a bit more with the headroom in the track, to maybe even out the volume a bit more. Especially if you have a way to visualize its waveform.

When you're EQ-ing a pair of headphones though, then you really have no way of knowing where the transients are likely to show up. So the correction curve has to be designed in such a way that it does not clip any part of the frequency range. And that means all of the adjustments in the frequency response have to be below the 0 dBFS clipping level.

If your EQ curve has both positive (+) and negative (-) gain adjustments, then a preamp with a negative gain also has to be added somewhere in the correction chain to normalize the levels in the EQ curve so all the adjustments are below, or do not exceed 0 dBFS.

Analog EQ's can work exactly the same way, btw, if you are using them to correct the response in a stereo system, or a pair of headphones. (Versus using them to adjust the tonal balance in a specific music track.) The first thing I tried for EQ-ing my headphones was a 31-band DBX graphic EQ, which allows you to both raise or lower the amplitude at different frequencies. If you raise any of the frequency bands by more than a dB or two above the unity setting, then a corresponding correction in the opposite or negative direction may also be necessary on the EQ's overall Gain setting. Or you could potentially end up clipping or distorting some of the transients in that frequency range.

Most analog EQ's will have a little bit of extra headroom in their 0 Gain setting. If the levels are raised by more than a couple dBs though, on any of the frequency bands, then the overall levels have to be reduced by approximately the same amount using negative Gain to prevent the audio from being clipped or distorted.

Since the effects of these headphone/stereo EQ correction curves are "global", and applied to all music tracks in the same way, they have to be robust enough to avoid clipping at any frequency range. Which means all the adjustments effectively have to be made in a downward direction, so they're below the clipping points of 0 dBFS, if you're using a digital EQ. Or 0 (aka unity) Gain if you're using an analog EQ.