Audibility thresholds of amp and DAC measurements

[Serge Smirnoff]

Some changes are not sound quality degrading such as slow phase changes and gentle sloping frequency response.
This will show up in nulling but may have no influence on perceieved sound quality.

To remove even non-audible changes. Yes, it will be some overkill but it doesn't matter, we already can manufacture devices of required accuracy. As a mass product they will be cheap anyway regardless of their target accuracy.

 

[solderdude]

That's not the point.
How do you deal with quite visible 'distortions' that are not of any audible consequence yet will give measurable differences with nulling ?

 

[Serge Smirnoff]

That's not the point.
How do you deal with quite visible 'distortions' that are not of any audible consequence yet will give measurable differences with nulling ?

The point is to design the chip that will not produce any visible 'distortions'. Manufacture it in mass and install to "all" audio devices. The problem of audio quality will be solved. I would start with the market of personal digital audio. Required s-level can be defined relatively easy for these devices.

We need to stop thinking in the paradigm of manufacturer-driven audio market. The market can also be consumer-driven.

 

[j_j]

And my point is to make it small enough in order not to deal with all that complicated psychoacoustic stuff, which we can't properly sort out and which results to marketing speculations only. Why you insist on keeping the distortion, no matter what kind of, if it can be eliminated completely?

I've advocated no such thing. Why are you making things up? Now I"m the one "keeping the distortion" after you told me that I can't back out linear modifications, which is quite possible! I'm the one saying "back them out if you must have them", so now I'm "keeping" them? You are simply having an argument. Stop it.

And, no, you can never COMPLETELY eliminate error in any analog system (or digital system with any substantial processing). Never. Physics prevents it, and the level of noise due to physics is, you will find, surprisingly close to limits you might want to reach. Ask anyone who's built a mic preamp or an LP preamp. (p.s. that RIAA EQ in the LP preamp is a classic example of backing out a linear modification, you know)

There is no such problem anymore, both DeltaWave and Matlab code removes time inaccuracy with ease. This is deterministic operation and can be done with any predefined accuracy.

Yes, yes, including partial sample alignment. Been there, done that. Over and over. You have bothered to go to the site I pointed out long ago that's the "fft workshop". It has octave code there that, among other things, will measure delay to an annoying accuracy (especially annoying when some folks latency claims are examined, not your problem).

3d plots of error signals are not what you need. You're still not plotting error (as a function of time and frequency) compared to SIGNAL as a function of time and frequency. Everything you talk about is signal dependent. You did notice that I said for a normal listening room with normal equipment, 110dB total SNR is probably sufficient. That assumes room noise and a peak limited system equal to most "hi fi" systems. A PA system for a large space will require more.

As to software that's available, I've probably written some of what you're using. Certainly I've written many linear-system identification routines, detection of distortion product routines, etc. Goodness. Now you're not only condescending and putting false words in my mouth, you're telling me to go refer to my own work? SERIOUSLY, dude?

 

[solderdude]

The point is to design the chip that will not produce any visible 'distortions'. Manufacture it in mass and install to "all" audio devices.

The high-end audio market IS consumer driven and really does not care about the lowest possible distortion numbers at all.
Only a very small minority of the human population wants this.
Also it is not needed to reach vanishingly low numbers.

 

[j_j]

I'm pretty sure Df levels for transducers will be within 0 -20dB range.

What are you even comparing? Pressure, Calrec output, what?

 

[Serge Smirnoff]

And, no, you can never COMPLETELY eliminate error in any analog system (or digital system with any substantial processing). Never. Physics prevents it, and the level of noise due to physics is, you will find, surprisingly close to limits you might want to reach. Ask anyone who's built a mic preamp or an LP preamp. (p.s. that RIAA EQ in the LP preamp is a classic example of backing out a linear modification, you know)

"Completely" in various listening environments means different df levels. For example a DAP showing Df ~ -60dB with m-signal will be perfectly transparent for most listeners, no matter what other measurements are showing. For comparison, Hugo 2 has Df = -34dB, MDAC4 from #254 has Df = -70dB.

You're still not plotting error (as a function of time and frequency) compared to SIGNAL as a function of time and frequency.

Should I? I created df metric exactly for the purpose to avoid extensive analysis of distortion.

As to software that's available, I've probably written some of what you're using. Certainly I've written many linear-system identification routines, detection of distortion product routines, etc. Goodness. Now you're not only condescending and putting false words in my mouth, you're telling me to go refer to my own work? SERIOUSLY, dude?

I'm not sure about DeltaWave but for sure you have no relation to Matlab code referred.

 

[Serge Smirnoff]

What are you even comparing? Pressure, Calrec output, what?

I didn't yet. And it might turned out that such measurements are not required as long as both production and consuming sides use similar transducers.

 

[Serge Smirnoff]

The high-end audio market IS consumer driven and really does not care about the lowest possible distortion numbers at all.
Only a very small minority of the human population wants this.

Yes! And this is result of marketing efforts of manufacturers. Mix of irrelevant measurements and misconceptions is very comfortable environment for them. They will keep it as long as possible. My point is to develop measurable criterion of audio quality on consumer side and to correct asymmetry of information on the audio market. The asymmetry is the real cause of marketing lower quality audio products as higher quality ones. This is a known bug of market economy and can be corrected only from consuming side. We should clearly define what we need and communicate this well grounded demand to manufacturers.

Also it is not needed to reach vanishingly low numbers.

Real-world s-levels are not 'vanishingly low numbers', best audio devices on the market already show them in df measurements. To be honest I can listen music through poor quality equipment and still get pleasure. But games and VR will require very accurate audio path to preserve spatial information. And possibility to listen music close to what author hear in studio (in headphones as an example) is also very attractive. So, it's better to have those numbers low enough. Manufacturers know how to achieve them.

 

[pkane]

I'm not sure about DeltaWave but for sure you have no relation to Matlab code referred.

Nothing in DeltaWave computation algorithms is borrowed, except for the base FFT routine. Everything else was developed by me, from scratch.

 

[Serge Smirnoff]

Everything else was developed by me, from scratch.

Take my respect for that, I can easily imagine the trove of work as I also developed my Matlab code from scratch.

 

[j_j]

"Completely" in various listening environments means different df levels.

So, for "one number" what do you assume?

Should I? I created df metric exactly for the purpose to avoid extensive analysis of distortion.

1% second order distortion is almost harmless. 1% 10th order distortion is unbelievably (**(&awful. Therefore ...

I'm not sure about DeltaWave but for sure you have no relation to Matlab code referred.

Fair enough. I've written the same thing, and given it away, but I don't doubt you've written it, too.

 

[j_j]

Nothing in DeltaWave computation algorithms is borrowed, except for the base FFT routine. Everything else was developed by me, from scratch.

Well, now we know at least 3 people who have written the same thing. Note, I don't doubt it.

Note, it was the ineffectiveness of pure SNR with linear processing backed out that led to much more complex measurements, too. I thought once upon a time your idea was the way to go, too. Sorry.

 

[j_j]

I didn't yet. And it might turned out that such measurements are not required as long as both production and consuming sides use similar transducers.

Highly unlikely. And you still need to decide what to measure in "real" space.

 

[j_j]

But games and VR will require very accurate audio path to preserve spatial information.

That's complicated. Some distortions will screw things up horridly, other huge distortions won't matter at all as far as spatial information. That is a disturbingly complicated space to examine. I do not speak without experience in that space.

 

[j_j]

The high-end audio market IS consumer driven and really does not care about the lowest possible distortion numbers at all.
Only a very small minority of the human population wants this.
Also it is not needed to reach vanishingly low numbers.

The high-end market is coocoo for cocoapuffs. Agreed. My apologies to the breakfast cereal.

 

[pkane]

Well, now we know at least 3 people who have written the same thing. Note, I don't doubt it.

Note, it was the ineffectiveness of pure SNR with linear processing backed out that led to much more complex measurements, too. I thought once upon a time your idea was the way to go, too. Sorry.

My idea isn’t quite the same as Serge’s. DeltaWave is a collection of tools that’s used to visualize and make audible differences between two recorded digital files. In the process it corrects for both, linear and nonlinear phase and amplitude differences, measures jitter and linearity and provides tools for determining audibility. DF metric was added recently, as an experiment after Serge posted his idea here, on ASR. It appears very similar to the rms null metric DeltaWave computes, so was trivial to add. And yes, DeltaWave is free and not commercial in any way.

 

[j_j]

... In the process it corrects for both, linear and nonlinear phase and amplitude differences, measures jitter and linearity and provides tools for determining audibility.

Note for readers. "nonlinear phase" does not necessarily mean a process that is not linear. A standard RC first order filter is nonlinear in terms of phase, but it a completely linear process, until you get to an actual zero in the system. "Linear phase" is better described as "constant delay", but the term "linear phase" permeates the lexicon. "nonlinear phase" simply means that the system being measured has different time delays at different frequencies, for instance like the allpass filter mentioned a few pages ago.

 

[solderdude]

We should clearly define what we need and communicate this well grounded demand to manufacturers.

Here is the problem... there is no 'we'. There are many we. All of them ask (not demand) different things from manufacturers and some of them are happy to oblige. Some manufacturers cater for audiophools, others for audiophiles, others like to play the measurement game and yet others follow their own path, often supported by enough sales.

Real-world s-levels are not 'vanishingly low numbers', best audio devices on the market already show them in df measurements.

Where are the df measurements for the best audio devices on the market ?
Have you measured ADI-2 DAC (or similar), Tom's HPA1 and devices like the Heresy and Magni 3+ on the low end ?
How is their df ? What more can we demand from them ? Everyone keeps coming out with new stuff every few years, usually with improvements or other pricepoints.

So, it's better to have those numbers low enough. Manufacturers know how to achieve them.

Yes, for those desiring the lowest possible 'distortion' (not everyone wants this). And yes manufacturers know how to achieve this and some actually build them as well. Consumers have a choice already. This is the upside of a free market. Downside is one can also buy crap as one cannot demand from manufacturers not to make it. Once people know something is crap sales will drop and a successor is made. Maybe even crappier or better.

 

[Serge Smirnoff]

So, for "one number" what do you assume?

At the moment my expectation of s-level for DAPs is -50dB -- -60dB. Df measurements of state of the art DAPs can help to define it almost precisely.

Where are the df measurements for the best audio devices on the market ?
Have you measured ADI-2 DAC (or similar), Tom's HPA1 and devices like the Heresy and Magni 3+ on the low end ?
How is their df ? What more can we demand from them ? Everyone keeps coming out with new stuff every few years, usually with improvements or other pricepoints.

You can choose on this page the devices of your interest, I will compute df levels for them from supplied recordingd. Also you can do this yourself but I'm ready to help.