Why is an amplifiers distortion lowest on high volume?

[ninetylol]

Sorry if this has been asked before, I couldnt find an answer so far.

When looking at headphone amp measurements the distortion always goes down the more power you are putting out of it, except for when its clipping at the end. Is that correct? What is the physical explanation for this?

Would it be wise to turn the amp up to like 80% (so its not clipping) and change volume with the DAC?

Also is there any guide to read measurements which answers questions like this?

 

[pma]

This is a never ending question, the reason is that at low and very low volume you read only noise which is then wrongly interpreted as distortion.

 

[Matias]

The distortion (or noise) may be somewhat constant while you increase the signal level, so that dividing one by the other, the distortion ratio decreases with power increase. Until at one point where distortion increases with power, then the graph goes parallel or increases.

 

[majingotan]

SINAD (the noise to signal factor (NOT noise floor) decreases here) obviously increases as you turn up the volume level to top of full scale. BTW, power amps do not have a volume control so if there's a volume control, it's an integrated amp or an active speaker like the HS7 that I have that has an attenuator that is configured to accept low level signal (-10 dB) or line level aka 2Vrms (unbalanced) / 4Vrms (balanced) signal (+4 dB)

If I put the level to -10 dB and I'm using a full scale 2V input, I will drive the power amp to clipping in this case. If you have an integrated amp, you never need to worry about that BTW.

 

[MRC01]

A useful way to think of this: distortion is correlated with the signal, noise is not. If the noise is more or less evenly spread across a wide range of frequencies (tape hiss, dither), it tends to sound less "bad" than distortion.

Typical amp designs have fixed gain with variable attenuation for the signal. So the noise is roughly at a constant level. Here, the volume control changes the signal level but not the noise, so the SNR is best at full output for obvious reasons. We see this here in Amir's measurements comparing SNR at full-scale to 50 mV: as you turn down the volume, you also reduce SNR. Since SINAD includes both noise and distortion, it is similarly affected depending on the relative mix of distortion & noise.

All else equal, gain increases both noise & distortion, and reduces bandwidth ("the more gain, the more pain"). So if you want to listen at lower volumes, it's better to pass a full-scale signal through less gain, than to attenuate the signal going through a high gain circuit. With source signals at 2 Vrms (or 4 Vrms for balanced), typical listening levels are much less than unity gain. Some (less common) amp designs do exactly this: change the gain with volume, so the noise is not at a constant level. These at full output could be better, or worse, or the same, as at lower volumes.

Here's a practical theoretical example: suppose the signal coming off the DAC chip peaks at 2 Vrms and you want to listen with peaks at 50 mV. That's a 32 dB reduction in level.

Design 1: analog stage (headphone amp) having +6 dB of gain (so max volume is 4 Vrms), with a potentiometer in front, attenuating by -38 dB.
Design 2: analog stage (headphone amp) having -32 dB of gain, accepting the full scale signal from the DAC.

Design 1 is more common, though in most cases, or with equal attention to engineering, design 2 should have higher SINAD, lower noise & distortion.

 

[solderdude]

Would it be wise to turn the amp up to like 80% (so its not clipping) and change volume with the DAC?

That actually depends on the way the amplifier is constructed.

Most amplifiers have the volume control in front of the gain stage.
The result is that it is not possible to clip the input signal at low volume levels but as a downside has the highest self noise which is not (or marginal) volpot position dependent.
The reason it could be wise to turn up the volume of an amp in this case and adjust volume with the DAC in this case is the better channel balance of the digital volume control.
The best way here is set the volume of the DAC to max. Set the amp vol. control so it goes as loud as you want it and then adjust the volume digitally. There is no benefit in S/N ratio.

Something like the O2 or Atom have the volume control after the gain stage. This has the disadvantage that one can clip the input signal even when the volume control is all the way down. Selfnoise is the same as the amplifier described above when the volpot is fully open and the same op-amps and gain is used. BUT the selfnoise of the amplifier becomes less when the volume pot is set lower. Of course there is a limit so selfnoise will still be there.
The reason it could be wise to turn up the volume of an amp in this case and adjust volume with the DAC in this case is the better channel balance of the digital volume control.
The best way here is set the volume of the DAC to max. Set the amp vol. control (lowest gain) so it goes as loud as you want it and then adjust the volume digitally. There is no benefit in S/N ratio.

Then there is a rare breed of amplifiers (think Meier) who control the gain of the amplifier.
When properly designed it should not clip the input signal and has a lower self noise, distortion and wider bandwidth when the volpot is turned down.
The reason it could be wise to turn up the volume of an amp in this case and adjust volume with the DAC in this case is the better channel balance of the digital volume control.
The best way here is set the volume of the DAC to max. and adjust the volume on the amp.

As has been mentioned above.. the seemingly higher distortion is basically a measurement artifact. It is caused by the signal to noise ratio.

Note that the choice of opamps and volpots (linearity/tracking not SQ) can be poor to excellent with all 3 designs.
The Meier tested on ASR does not perform better than some cheap headphone amps. Should the designer have chosen a better design and better parts he could possibly have been at the top of the charts.

Also note that all of this is really only of importance when one uses really sensitive headphones. With normal sensitivity headphones the noise floor is below the audible limit and not of any consequence except in the measurement plots.

 

[MRC01]

...Then there is a rare breed of amplifiers (think Meier) who control the gain of the amplifier. When properly designed it should not clip the input signal and has a lower self noise, distortion and wider bandwidth when the volpot is turned down. The reason it could be wise to turn up the volume of an amp in this case and adjust volume with the DAC in this case is the better channel balance of the digital volume control.

A volume control that changes the gain of the circuit should have perfect channel balance at all settings, just like the DAC output. There is no pot; instead it's changing metal film resistors in the gain-feedback loop. This is seen in practice with measurements here. With this kind of amp there is no reason to adjust volume with the DAC. You're best off using full output from the DAC and attenuating in the amp.
Meier isn't the only example - the RME ADI-2 also has this feature. It changes the analog gain in large steps, with digital attenuation between the steps.

 

[solderdude]

A volume control that changes the gain of the circuit should have perfect channel balance at all settings, just like the DAC output. There is no pot;

It still uses a pot and still has channel imbalance. The only difference is the pot is in the feedback path. This has other challenges though.
In case of the RME it just has an output stage where the gain can be varied in certain steps. It isn't anywhere the same as the volume control of the Meier amp. The problem with this type of volume control is that it has a limited volpot range.

 

[MRC01]

It still uses a pot and still has channel imbalance. The only difference is the pot is in the feedback path. This has other challenges though.

That sounds like it could work, but I've never seen it used. The Meier amp doesn't use a pot in the feedback loop. Its volume knob switches different fixed resistors in the feedback loop. That's why the volume changes in discrete steps and it has perfect channel balance at all settings.

In case of the RME it just has an output stage where the gain can be varied in certain steps. It isn't anywhere the same as the volume control of the Meier amp. The problem with this type of volume control is that it has a limited volpot range.

The RME has 4 steps of analog gain spaced 6 dB apart, and because they're so far apart it uses digital attenuation between these levels. If the RME instead had 32 (or more!) levels spaced closer together, then it wouldn't need digital attenuation at all, and its volume control would be similar to the Meier amp.

 

[solderdude]

That sounds like it could work, but I've never seen it used. The Meier amp doesn't use a pot in the feedback loop. Its volume knob switches different fixed resistors in the feedback loop. That's why the volume changes in discrete steps and it has perfect channel balance at all settings.

In case of that specific amp yes and those FET switches are also the reason for its technical limitations in distortion levels.
The reason why one does not see it in practice is the limited volume control range. Also the circuit would have a lot of gain.
It is outclassed by regular circuit designs.

Using a potmeter in the feedback loop is possible but an intermitted wiper will cause huge output spikes where with normal volume control the signal would simply be 'off'.

The gain control is not part of the volume control. The volume control is purely digital and the gain control is so one can match the FSD output level to the gear that it must drive.

 

[bennetng]

RME ADI-2 DAC is No.1 in this regard and the volume control is partially digital. JDS Labs Atom uses a traditional pot also got a high score so there is no indicator that which design is the best. When putting other factors like channel balance and price into account it is even more complex.

The red one in the graph is only $120.
https://www.audiosciencereview.com/.../fiio-a5-portable-headphone-amp-review.10468/

BTW, JDS Labs actually recommends turning down the volume in the computer slightly before sending data to the DAC, which is in fact a wise suggestion. Deliberately leaving digital headroom in DAC design and mastering are only for stubborn people who are unconditionally refuse to use digital volume control.
https://www.audiosciencereview.com/...review-of-jds-labs-usb-ol-dac.2244/post-61171
https://blog.jdslabs.com/2017/04/odac-vs-ol-dac/#comment-757

 

[ninetylol]

JDS Labs actually recommends turning down the volume in the computer slightly before sending data to the DAC, which is in fact a wise suggestion. Deliberately leaving digital headroom in DAC design and mastering are only for stubborn people who are unconditionally refuse to use digital volume control.
https://www.audiosciencereview.com/...review-of-jds-labs-usb-ol-dac.2244/post-61171
https://blog.jdslabs.com/2017/04/odac-vs-ol-dac/#comment-757

If im using optical from the motherboard to the DAC should i turn Windows volume to 98 or something? How is a optical signal changed in volume and wouldnt Windows somehow reduce the audio Signal in quality by adjusting volume?

 

[bennetng]

If im using optical from the motherboard to the DAC should i turn Windows volume to 98 or something?

The amount of ideal digital attenuation depends on audio source, and that's exactly why JDS Labs don't implement a fixed amount of headroom into their products. That means "98 or something" also depends on what you are playing.

How is a optical signal changed in volume and wouldnt Windows somehow reduce the audio Signal in quality by adjusting volume?

Only if you reduce a lot, like 20dB or more. If your playback software support volume management (e.g. ReplayGain) you can scan your music collection, an ideal reduction value will be automatically applied in per-track or per album basis. After this you can still use a secondary volume control of your choice to further adjust the volume.

Read the following for more information:
https://archimago.blogspot.com/2019/06/guest-post-why-we-should-use-software.html
https://www.audiosciencereview.com/...-music-players-foobar-jriver.7412/post-188072

 

[MRC01]

In case of that specific amp yes and those FET switches are also the reason for its technical limitations in distortion levels. The reason why one does not see it in practice is the limited volume control range. Also the circuit would have a lot of gain. It is outclassed by regular circuit designs. ...

That limitation was a cost-cutting decision made for the Corda Jazz. It disappears if one uses metal film resistors instead of FET switches.

... [with the RME] The gain control is not part of the volume control. The volume control is purely digital and the gain control is so one can match the FSD output level to the gear that it must drive.

RME disagrees with you. They implemented a featured called "Auto Ref Level" that automatically switches the gain control as the volume knob is turned. It always uses the lowest analog gain with the highest (least attenuated) digital signal. For example if you turn it down by -18 dB, it uses the lowest gain (+1 dB) with no digital attenuation. This is cleaner than using the highest gain (+19 dB) with 18 dB of digital attenuation.

So if the cleanest way to attenuate the signal is a low gain analog stage, why only have 4 steps spaced 6 dB apart? They could instead have, say, 64 steps spaced 1 dB apart. This would be even better, especially at low output levels. The only reason is to reduce cost & complexity.

RME ADI-2 DAC is No.1 in this regard and the volume control is partially digital.

That 93 dB 50 mV SNR was taken from the RME's dedicated IEM output jack. The 50 mV SNR on the JDS Atom and Corda Jazz were taken from their standard 1/4" output jacks. The RME's 50 mV SNR from its 1/4" jack was 85 dB.

 

[bennetng]

That 93 dB 50 mV SNR was taken from the RME's dedicated IEM output jack. The 50 mV SNR on the JDS Atom and Corda Jazz were taken from their standard 1/4" output jacks. The RME's 50 mV SNR from its 1/4" jack was 85 dB.

You also need to take into account that the RME is a DAC+amp combo and it is not completely fair to compare with an amp-only product.

 

[MRC01]

If im using optical from the motherboard to the DAC should i turn Windows volume to 98 or something? How is a optical signal changed in volume and wouldnt Windows somehow reduce the audio Signal in quality by adjusting volume?

If the engineers who made the recording did it properly, there's no reason to turn down the volume because the digital peaks will be at 0 dB or less - not clipped, with no intersample overs. Benchmark describes an example of a Steely Dan recording that has digital clipping. Simply put, it was recorded (or digitally encoded) too hot which is a mistake. In that example, if the engineers lowered the level by 1 dB it would not be clipped. Sadly, this is a common mistake due to the loudness wars. Recording this hot doesn't increase the dynamic range because they aren't even near to using the full available dynamic range; even the lowest levels are still well above the noise floor. The sole purpose is to make it sound louder. IMO, intersample vs. true clipping is a distinction without a difference. Intersample still means the level was too hot, and either way the real solution is to lower the level.

If recordings are made properly (no clipping), then there's no reason to turn down the volume. The PC should send the bits directly to the DAC without any modification (volume or otherwise).

Incidentally, why does everything recorded have to sound so damn loud all the time? All this dynamic compression only makes it sound unnatural and fatiguing. Dynamics constitute an important dimension of the musical experience; compression squashes the life out of the music.

 

[bennetng]

There is a misconception that inter or non-intersample over must be related to loudness war. In fact it is not necessarily true.
https://www.audiosciencereview.com/...nd-dacs-with-volume-controls.5432/post-121344

 

[MRC01]

That depends on how broadly or narrowly one defines the loudness war. Recording at lower levels eliminates the problem entirely. And lowering the level is loss-less, since the 16-bit dynamic range is virtually never used.
So the only reason to use these high levels that cause clipping or intersample overs, is to make the music sound as loud as possible.

 

[bennetng]

Recording at lower levels eliminates the problem entirely

Obviously it has nothing to do with recording level. Only novice users introduce clipping during recording.
The subsequent process (e.g. using a sampler/rompler/synthesizer, EQ, resampling, lossy encoding etc) often cause changes in phase which increases the chance of clipping in the playback chain, even if the music is not particularly loud.

 

[MRC01]

Right. That's a strong argument for always recording in 24-bit (or more, but 24 bit should be sufficient), so you can keep everything 6-12 dB below peak without any loss of resolution, giving headroom for whatever processing you want to do. Then when all processing is done, you have the final waveform, shift the peak levels just below 0 dB and then convert it to 16-bit or whatever your distribution format is.
Where is the need to use levels so hot you risk clipping or intersample overs?