Master Thread: Are Measurements Everything or Nothing?

[Dan Lavry]

Nice work. Now we need to expand the argument. So far, we talked about a full scale 1KHz sine wave, and what happens when you effectively truncate it. You decrease the signal to noise and generate some distortions. It still "looks like" a sine wave with little alterations.

But take a real instrument, where the character of the sound is defined by harmonics. Low level harmonics (with small amplitude) require lower order bits. If digital “push down” some lower bits signals under the noise floor, it may eliminate the fine details that define the instrument sound.

For volume control, I chose analog attenuation. The main reason (if I recall) was the noise issue. Say feeding an 18dBu signal directly to a 4dBu line level requires 14dB attenuation. With digital attenuation, the noise of the DA (even with no digital signal) is fed directly to the power amplifier input. With analog attenuation the noise is reduced by 14dB.

Can anyone hear 14dB more noise, some added distortions and some loss of details (harmonics)? It depends on the music, the setup and of course the listener. Driving consumer level is 26dB more noise. My view as a designer is to avoid degradation if it can be avoided (and be ready for worse case scenario). It calls for providing some MARGIN for the best ears under the most critical conditions (say a great mastering room).

Analog attenuation is nearly perfect. Digital attenuation causes some alteration of the data. Doing it in digital would be much cheaper and easier, and it is “not noticeable” to almost all the customers. I still want to avoid all possible issues with large margins, beyond human hearing capabilities.

But isin't it the case that in a 24 bit or higher dac, the 16 bit data is padded out to 24bits before passing to the actual DAC function. As long as digital volume control is done after padding then we are dealing with a 24 bit noise floor. In fact the 16 bit noise floor is part of the padded signal, and being 48dB above the 24bit digital noise floor, is itself turned down without SNR loss for that first 48dB.

Yes - if digital volume is carried out at the 16 bit domain perhaps there is an issue - but I don't think it is in typical DAC with preamp function, even when the source is 16 bits.


Edited to add:



I fully agree that if one is only looking at the numbers/measurements, and ignoring audibility - then analogue attenuation is going to beat digital for most metrics - possibly excluding channel matching depending on the analogue implementation.

Leaving sound and audibility aside for a moment, there are other factors in choosing digital or analog attenuation.

A DA for professional use output level is often 24dBu/18dBu balanced, or 18dBu/12dBu unbalanced. Some studios required a front panel 0.1dB fine adjustment (+/-0.7dB). High signal level (and balanced) offers some advantages.

But most power amplifiers and other gear with analog input requires less signal. Listening in real time to professional signal can be done with two DAs. One DA for professional level, the other DA receives different data. That approach is "complicated".

I used the analog attenuator to ELIMINATE the need for a second DA.

There is a MAIN output (L/R XLR’s) for the DA. But there is a secondary MONITOR output (L/R XLR’s). The monitor input simply taps the MAIN output signal through analog resistor dividers. If set to 0dB attenuation the MAIN and MONITOR are the same, it is the same DA signal.

I could have done it digitally inside the Quintessence DA. Intercept the data, digital attenuation, second DA and associated circuitry for the MONITOR output. Add that to all we talked about earlier…

I am not saying that doing a clean analog is easy. I think it was worth it to remove the need for a second DA in many situations. Also, the MONITOR output sound is as close to activity on the MAIN.

 

[Sal1950]

You guys are making my brain hurt trying to follow all this deep digital tech.
But thanks for the education anywho, where do I send the tuition check? .

 

[Dan Lavry]

You guys are making my brain hurt trying to follow all this deep digital tech.
But thanks for the education anywho, where do I send the tuition check? .

Here is the simple part. Say there is some DA running 24dBu balanced cable to the “other room” for some mixer or what not. The signal level is high and the all the DA bits are active.

But you may need less voltage to drive a lot of analog gear’s inputs. For REALTIME monitoring, if you use digital attenuation it will have to use a second DA to receive and convert the attenuated data. Accurate monitoring calls for DA matching, there is wiring involved, setup…

The analog attenuation is real-time. Instead of two DA’s you have one MAIN DA outputs and a second MONITOR analog outputs. The monitor is the same signal as the main, but through an attenuator. There is no issue of DA matching, degradation of specifications (and possible sonic impact?)

I lot of it is about work flow. With analog, you switch between DA inputs, and a new attenuation comes up instantly. The display shows the last setting for each DA input. To change it you rotate the knob. Push the knob to alter between 3dB steps and 0.25dB steps.

The analog MONITOR output is also useful in many cases that don’t need the MAIN (professional level signals). It can provide up to 24dBu and attenuate by as much as 80dB. It is an internal attenuator for any purpose.

You don’t need to be an EE or have a good ear to accept that resistor dividers are “cleaner” solution. Resistor dividers are simple. Compare to adding a second DA made with a lot more resistors, caps, critical tolerances and active semiconductors and more…). And you don't loose bits...

 

[danadam]

But in the second pass, the -60dB distortion tone from the 1kHz tone will be added to that created in the first pass. So double voltage to -54dB. After two more passes an additional 6db. Four passes later, another 6dB. So after 8 passes the distortion tone is at -42dB

Here's 1 kHz and power-of-2 number of loops, up to 32x. So in this case (pure tone) it seems to work that way, at least for the lower harmonics and up to that many repeats

 

[antcollinet]

some lower bits signals under the noise floor, it may eliminate the fine details that define the instrument sound.

But then we *have* to start talking about audibility. If the noise floor is inaudible then so are sounds down at that level. It does't matter if they dissappear into the noise floor, since our ears can't detect them anyway.

And that is ignoring the fact that for audible noise and low level tones we can hear tones below the level of Wideband noise in any case.

 

[antcollinet]

You guys are making my brain hurt trying to follow all this deep digital tech.
But thanks for the education anywho, where do I send the tuition check? .

Puts up hand.

Not claiming my spouting off is worth it - but if money is on offer

 

[Newman]

Analog attenuation is nearly perfect. Digital attenuation causes some alteration of the data. Doing it in digital would be much cheaper and easier, and it is “not noticeable” to almost all the customers. I still want to avoid all possible issues with large margins, beyond human hearing capabilities.

Hi Dan, we actually have a thread dedicated to that topic.

I would be interested in your thoughts on my primary input to that topic, link.

cheers

 

[Dan Lavry]

Here's 1 kHz and power-of-2 number of loops, up to 32x. So in this case (pure tone) it seems to work that way, at least for the lower harmonics and up to that many repeats

View attachment 536128

Noise does accumulate with loops. The mechanism is different. Incoming analog noise is treated as signal. The converter electronic noise causes errors which ends up as more noise.

But loops adjusted to gain of 1 don’t increase the signal. Lets start with a signal of 1KHz 0dBFS tone with 2KHz distortion at -70dB.

The AD does not accumulate “previous distortion” with the distortion of the current sample. There is no “previous distortion” we start from scratch for each sample. The converter will of course convert both the 1KHz tone and the 2KHz harmonic. The contribution of the -70dB harmonic to distortion is negligible. The 0dB tone is responsible to the -70dB distortion. The harmonic in the input itself has negligible effect. So you end up with the same distortion…

Small signal (the harmonic) operate on a very small part of the transfer curve. If you want to put numbers on it, if the converter provides -70dB distortion at full scale, it will do so much better for tiny signal.

So you can run music loop 32 times and when it sounds the same as first time loop does not indicate quality.

 

[antcollinet]

Noise does accumulate with loops. The mechanism is different. Incoming analog noise is treated as signal. The converter electronic noise causes errors which ends up as more noise.

But loops adjusted to gain of 1 don’t increase the signal. Lets start with a signal of 1KHz 0dBFS tone with 2KHz distortion at -70dB.

The AD does not accumulate “previous distortion” with the distortion of the current sample. There is no “previous distortion” we start from scratch for each sample. The converter will of course convert both the 1KHz tone and the 2KHz harmonic. The contribution of the -70dB harmonic to distortion is negligible. The 0dB tone is responsible to the -70dB distortion. The harmonic in the input itself has negligible effect. So you end up with the same distortion…

Small signal (the harmonic) operate on a very small part of the transfer curve. If you want to put numbers on it, if the converter provides -70dB distortion at full scale, it will do so much better for tiny signal.

So you can run music loop 32 times and when it sounds the same as first time loop does not indicate quality.

The distortion does accumulate. Afrer the first pass the original 1kHz tone and it's second (and other) harmonic(s) both exist in the signal input to the second pass. In the sceond pass, distortion applies to both. Of course we can discount the distortion of the already tiny harmonic as below audibility. But the distortion tone created in the second pass from the fundamental - is just as large as that already existing from the first pass - they add together.

You can see it clearly happening in the measurement provided by @danadam.

But the issue is not about determining quality of the DAC/ADC - it is a demonstration of just how inaudible typical ADC/DAC distortion really is - even after 10 passes it is still (with typically well performing converters) inaudible. It puts the lie to the statement "The sound difference is night and day" Or people worrying about an additional ADC/DAC stage when they feed active/DSP speakers, or an AVR, with an analogue signal.

It is a demonstration for the self declared Golden Eared.

 

[rdenney]

So here is a question: Say you loop AD in series with DA. The unit’s flatness response is perfectly flat bout it has a 0.1dB increase at and around 5KHz. The converter act like a very mild EQ boost.

So if you loop it 10 times, the 5KHz range gets amplified by 1dB. For +/-0.2dB boost you get “free” +/-2dB boost.

Listening to the outcome can give a good converter a bad name. I would think this flatness response issue would override the increase in noise and distortions. The flatness response deviation increases linearly with number of loops.

Distortions, noise, flatness all count. The question requires some way to quantify it. I would think that looping the converters to “uncontrolled” 1-2dB error is a way to force the process to magnify what can be very small error for real life use.

Comments are welcome.

Dan, the hypothesis we see posited here repeatedly is that a good-measuring DAC can still sound bad if the listener is sufficiently skilled and the surrounding system sufficiently “revealing”. The ASR-(generally)accepted rebuttal is that an e.g. AP 5500 has better listening skills than any human, and is far more revealing than any surrounding system. (I would also even make that claim for an HP 8903, which is ~30 or 40 dB less sensitive than the AP.)

I think you are arguing a different point, hence stating it plainly so you know the perspective many are responding from.

If I can’t hear any effect in the tenth generation feedback loop, and the initial conversion measures well (an important qualifier), then clearly the conversion process itself is not coloring the sound audibly (either due to response non-linearity, noise, or distortion effects—which add non-linearities, of course), even if the test underestimates some effects due to canceling. The burden of evidence is not on those who can’t hear a difference, it’s on those who claim they can.

Rick “thinking pro gear needs a lot of other features not measured by SINAD and frequency response, such as accurate gain staging, precisely correct voltage levels, etc.” Denney

 

[Dan Lavry]

Dan, the hypothesis we see posited here repeatedly is that a good-measuring DAC can still sound bad if the listener is sufficiently skilled and the surrounding system sufficiently “revealing”. The ASR-(generally)accepted rebuttal is that an e.g. AP 5500 has better listening skills than any human, and is far more revealing than any surrounding system. (I would also even make that claim for an HP 8903, which is ~30 or 40 dB less sensitive than the AP.)

I think you are arguing a different point, hence stating it plainly so you know the perspective many are responding from.

If I can’t hear any effect in the tenth generation feedback loop, and the initial conversion measures well (an important qualifier), then clearly the conversion process itself is not coloring the sound audibly (either due to response non-linearity, noise, or distortion effects—which add non-linearities, of course), even if the test underestimates some effects due to canceling. The burden of evidence is not on those who can’t hear a difference, it’s on those who claim they can.

Rick “thinking pro gear needs a lot of other features not measured by SINAD and frequency response, such as accurate gain staging, precisely correct voltage levels, etc.” Denney

Yes, there is a slow accumulation in most cases. The distortions don't seem to contribute more but the fundamental does in most cases. But the distortions and noise grow slowly.

I am not involved in listening. I do have a question:
Say You listen or measure a AD followed by DA and hear distortions or sonic alteration. How do you figure out which unit is responsible for it?
Is it the AD or the AD?

A poor unit in series unit in series with a ggod one does not let you pin point good from bad?

I am not including speakers or room acoustic in the question. The question is the same for many combinations such as power amp and speaker. Mic pre and AD and more.

Is it not best to measure each unit separately to avoide impact from another unit in series?

You may not be able to separate speaker from power amplifier, and the performance may depend on matching impedance and more.

I am talking about cases of units that can be measured separately from each other.

 

[Dan Lavry]

Rick “thinking pro gear needs a lot of other features not measured by SINAD and frequency response, such as accurate gain staging, precisely correct voltage levels, etc.” Denney

I tend to agree with Rick. DA for home listening is somewhat different from music production, where ease of use and easy “work flow” may make a big difference to people that spend hours, days and years with the gear.

For a the Quintessence DA I included easy to operate fine level calibration, Precise levels, MONO mode, INVERT, multiple inputs, additional output with an attenuator (big knob and large display) and many other features for studio use.

The AD is of course not meant for “home use”. The extra features make life easier in the studio, such as automatic sample rate detection (in external sync mode). A large clear “sample accurate” audio display with peak hold feature, a second independent output or built in multiple word clock outputs and more.

Of course, the main thing is the sound quality. I don’t produce music, just doing design. But I know that many users utilize various features to make their lives easier. It took me many years to learn that electronic circuit design for excellent performance is different than product design where feature play a role in real life.

My goal is to provide the easiest user interface to learn and operate. Sound comes first, and there is not much talk about features, but they should not be ignored. Features may not matter to all to some, but very desirable to many others.

 

[danadam]

Say You listen or measure a AD followed by DA and hear distortions or sonic alteration. How do you figure out which unit is responsible for it?

Hear? With current electronics I treat it as: "I'll cross that bridge when I come to it"

But presumably you find better measuring devices, replace one of the two and check if the difference is still audible.

Is it not best to measure each unit separately to avoide impact from another unit in series?

Isn't the measuring device itself another unit in series? It just has to have significantly better specs than the measured device.

 

[Dan Lavry]

You said: “Isn't the measuring device itself another unit in series? It just has to have significantly better specs than the measured device”.

I use Audio Precision test analyzer. Most audio equipment manufacturing companies do. Their strength has always been in the analog side, signal generation and filtering, but they evolved over 40 years much further. The other factor is breaking down the performance into separate entities – flatness response, phase, noise. It enables to focus on one aspect at a time.

Manufacturing: This is pro audio gear, not mass production consumer stuff. Each unit is tested, out in “burn in” for days and retested.

Design and performance: As I mentioned, the AP system provides separate information about various parameters. Let’s take a simple one, amplitude vs. frequency response. The system can yield a very precise plot in a minute. The accuracy can be 0.01dB, no human can do that.

You said: "Hear? With current electronics I treat it as: "I'll cross that bridge when I come to it"

Good music production requires musical ear. The person can be musical but at 30 years old 16Hz bandwidth is not bad (14KHz at 40 years). But much of the audience is young, hearing “more stuff”.

So it is my responsibility to make sure that there is no “funny stuff” going on. It will be too late to have some youngster say: daddy, there is some real screech hurting my ears.

The big question: is measurements an answer to that a human hears? I can’t answer that, it involves electro-mechanical devices (mics speakers, headphones) and room acoustics... But I have an answer for converters, amplifiers and more, all dealing with signals (digital and analog). The common thread is linearity (keeping the waveform intact).

The conversion basic idea is to express the SAME information in different formats. This is not about the ear; it is about doing better than the ear. It is about measuring voltage and viewing binary digits, while keeping the waveform unchanged. That is strictly technical stuff about achieving accuracy and reducing errors.

So yes, I sleep well at night knowing that I covered everything to the best of my ability. I rely on testing and measurements. I don’t want to hire some young person to listen to 250 tone and figure out the relative volume accurately.

Now, back to my original point: when connecting independent 2 units in series, AD and DA, DA and power amp, Mic pre and AD and more, looking at the outcome does not tell which unit is the problem. Most of the world operates on the principle that good system is based on good individual components. Certainly in cases of units in series. It makes sense to evaluate each unit separately.

 

[antcollinet]

AD and DA, DA and power amp, Mic pre and AD and more, looking at the outcome does not tell which unit is the problem.

Absolutely agreed. But again - the purpose of the multiple loop ADC DAC demos is not to evaluate the equipment.

It is a demonstration of just how inaudible the errors introduced by those two processes really are.

 

[solderdude]

Here's 1 kHz and power-of-2 number of loops, up to 32x. So in this case (pure tone) it seems to work that way, at least for the lower harmonics and up to that many repeats

View attachment 536128

Has there been done a similar test but with multitones ?

 

[danadam]

Has there been done a similar test but with multitones ?

No, but I can do it. What kind of multitone would you like to see? (preferably something I can generate with REW)

 

[solderdude]

preferably something with say 20 to 32 tones and maybe pink-noise shaped to mimic a music signal (in spectrum).

I would expect to see the noise to come up mostly ?

 

[Dan Lavry]

Absolutely agreed. But again - the purpose of the multiple loop ADC DAC demos is not to evaluate the equipment.

It is a demonstration of just how inaudible the errors introduced by those two processes really are.

Big difference between listening to tones and music:

The noise does accumulate running an AD DA loop. What about the harmonics? I thought we were talking about listening, not a 1KHz tone or similar. In the 1KHz tone case Say the result with one pass (not great) is around %0.005 (-86dB) distortion at the 2KHz. After 10 times, it becomes %0.05, the distortion grew to -66dB.

But what if that -66dB 2KHz distortion “accumulate” with already existing tone -20dB tone at 2KHz. Add -66dB to a -20dB tone is a different picture.

Typically, the fundamental is stronger, often around 10-20 dB higher than the first few harmonics. Each instrument’s actual harmonic energy will “cover” the its own distortions. You do it enough time (a lot of noise) and you may make a dent in the ratio of the harmonic to the steady fundamental.

The question about multitone frequency is related to my point above. The absolute wrong thing to do is a frequency sequence such as 1KHz, 2KHz, 3KHz….20KHz. The harmonic distortions fall on existing tone, just like the case of looping music around.

Audio Precision sets the frequency spacing on a logarithmic scale to make sure that the distortions are not masked by the signal. I would not want to listen to it...

Multitone testing huge advantage is saving test time. One “snapshot” instead of slowly “accumulating point by point”. It is used for mass production.

 

[Dan Lavry]

Big difference between listening to tones and music:

The noise does accumulate running an AD DA loop. What about the harmonics? I thought we were talking about listening, not a 1KHz tone or similar. In the 1KHz tone case Say the result with one pass (not great) is around %0.005 (-86dB) distortion at the 2KHz. After 10 times, it becomes %0.05, the distortion grew to -66dB.

But what if that -66dB 2KHz distortion “accumulate” with already existing tone -20dB tone at 2KHz. Add -66dB to a -20dB tone is a different picture.

Typically, the fundamental is stronger, often around 10-20 dB higher than the first few harmonics. Each instrument’s actual harmonic energy will “cover” the its own distortions. You do it enough time (a lot of noise) and you may make a dent in the ratio of the harmonic to the steady fundamental.

The question about multitone frequency is related to my point above. The absolute wrong thing to do is a frequency sequence such as 1KHz, 2KHz, 3KHz….20KHz. The harmonic distortions fall on existing tone, just like the case of looping music around.

Audio Precision sets the frequency spacing on a logarithmic scale to make sure that the distortions are not masked by the signal. I would not want to listen to it...

Multitone testing huge advantage is saving test time. One “snapshot” instead of slowly “accumulating point by point”. It is used for mass production.

Let me simplify the above:

One of the things that make running AD – DA loop different than testing how a converter works is THE SAME MUSIC EVERY TIME.

This is a DIFFERENCE. It should be easy to understand and at least acknowledge. But what is the difference?

Somewhat simplified:

The instrument generates harmonics at its fundamental and harmonics at integer multiples of the fundamental. Say the 0dB instrument generates harmonics amplitude at -10dB or -20dB. The musical instrument character is defined by the harmonics.

The electronics, say amplifier or converter adds distortions. The harmonic distortions are at the same multiple of the fundamental. But if the added distortion is far enough bellow the fundamental, the harmonic distortion has little impact. A distortion 30dB below the instrument tone cause 0.27dB. The -40dB or -50dB FFT spike is off by a quarter dB. A distortion 40dB below the fundamental changes the harmonic level by 0.0684dB.

The ear is very sensitive to harmonic ratio, but the loop test does not alter it much. Please check my calculations, I am an old guy trying to recollect some stuff. But I think a pretty poor amplifier can sound “not that bad” for some solo instrument, say clarinet. But add more instruments and it can be is a mess, especially for loud music. They call it inter-modulation distortions…

I realize that I am over simplifying, trying to make a point about AD -DA loop test. I have a lot more to say about it, maybe later.