Audio Science: Does it explain everything about how something sounds?

[jkeny]

Sounds reasonable. Do you think the noise floor modulates up and down with the volume of the signal? The complexity of the signal? I imagine the answer is probably "Yes."

Probably. We have an example of this in the ESS DAC video presentation that Mallinson did - he showed the noise floor modulating with volume (which they claim they fixed but subsequent measurements of ESS DAcs show vestiges of this still). All I know is that when I address what I believe are noise issues & noise stability issues in my modifications, I hear audible improvements. I would love to have the time, resources & expertise necessary to do these measurements but I don't.

To me this absence (or reduction) of noise floor modulation is one of the characteristics a good audio reproduction system that makes it mor realistic sounding. Funnily enough - this seems to be more crucial at the front end of the reproduction chain, more than the back end - so DACs more than speakers are the problem areas. In some ways it's not surprising, as whatever signal at the front end gets amplified through the chain, including the noise floor element of the signal. I also just have a gut feeling that we are more sensitive to this noise floor mod than current audibility threshold thinking would have us believe.

Why not? If your theory of the modulating noise floor is correct, these measurement would, I'd think, be critical to quality product development.

Why aren't these measurements done? Probably because it would not be very flattering to equipment & who is going to be the first to make the measurements of their equipment public? I reckon the marketing people have a lot to say about doing or not doing this

 

[Mike Lavigne]

Yes, I like Led Zep or early Black Sabbath in the car, but most classical music simply can't be heard over the road noise, so I basically switch between talk radio and rock while driving my truck.

+1.

sports radio almost 100%, then pop/rock XM when nothing is worth listening to. in the convertible some compilation CD's of driving music.

why even fight the noise. just go home and enjoy.

 

[Mike Lavigne]

Do you have a photograph of the window plugs?
Keith.

I had my contractor make these with Quietrock 545 and some 3/4" maple finish grade ply. there are little attachments at the top that screw into the jam.

heavy buggers; which is what it takes. they establish a room boundary.

works like a champ.

you could use sheetrock, then texture and paint it. or match the wood casework. there is a pleated shade on the window behind the window insert/plug. so from the outside it looks right (would not want to offend the deer or birds behind my barn). the backside of the plug has the same 3/4" ply as the front side. I can remove them or install them quickly.

for a large window some creative thinking would be needed to deal with the weight and moving it around.

 

[Phelonious Ponk]

Probably. We have an example of this in the ESS DAC video presentation that Mallinson did - he showed the noise floor modulating with volume (which they claim they fixed but subsequent measurements of ESS DAcs show vestiges of this still). All I know is that when I address what I believe are noise issues & noise stability issues in my modifications, I hear audible improvements. I would love to have the time, resources & expertise necessary to do these measurements but I don't.

I must have misunderstood you. I thought you said a guy you know on a DIY forum had done these measurements, and I assumed they were not terribly expensive or demanding.

Why aren't these measurements done? Probably because it would not be very flattering to equipment & who is going to be the first to make the measurements of their equipment public? I reckon the marketing people have a lot to say about doing or not doing this

Well, that's a reason not to publish them in your brochures, but I expect that during product development and testing, lots of measurements are done that are not making it to the marketing departments.

Tim

 

[Tony Lauck]

My take on this noise floor issue is that it is actually not being revealed by the steady state signals that are used in testing so we have no idea what the real noise floor is. Further to this I believe it's the modulation of the noise floor that's perceptually important & not the absolute noise floor. And I'm of the opinion that noise floor modulations at a very low level affect our perception of music but not in a way that we are aware that it's noise - more like an imprecision in presentation - imprecision in attack, & decay of instruments - the elements that we mostly use to differentiate instruments. This, I'm sure is how things can snap into focus when some of these noise floor elements are improved

Getting back to steady state Vs more music-like test signals & measurement. Look at IMD measurements - they usually are done with two single tone signals (19 & 20KHz at -3 or -6dB)? & the resulting IMD spurs plotted. Fine but extrapolate this to multitones as the test signal & see what IMD spurs are measured. Also do the measurements at various amplitudes & plot. If you extrapolate this sufficiently with many tones at many different amplitudes analogous to a busy passage in music all these side spurs would form a grass at the base of the FFT measurement i.e a noise floor & when IMD spurs are added on top of other spurs, it may not be at a low level. Now change this pattern of tones & what would we see - a different noise floor i.e a modulating noise floor which shifts with changes in tone pattern & amplitude(music). I've seen the notion of masking being used as the reason why side spurs will not be audible but when all the side spurs are added together is this not perceived as a noise floor?

This is just one example of a measurement that isn't done - correct me if I'm wrong. If it was we may a have a very different picture of our audio devices. This, to me, represents the great unmeasured measurable!! It may turn out that it is not a low level noise floor fluctuation, at all - when added together the IMDs from multitone test signals may be very measurable & way above audibility, I don't know, at the moment.

It could be a significant factor in the tubes Vs SS & TT Vs digital debates/disagreements?

I remember reading Scott Wurcer on DIYaudio a year or so ago posting that he used some multitone test patterns on some manufacturer's DAC evaluation boards & it really separated out the performance/implementation of the DACs. Haven't heard him refer to it since -maybe it wasn't very flattering to his own company's design efforts?

Adding to the confusion, there are two definitions of "noise" floating around. One is noise that is uncorrelated to the signal in any way. This is what you get if you mix in noise downstream of a source (assuming the subsequent electronics is linear). The other is noise that is correlated to the signal. This comes in various forms, as per traditional electronic amplification or mechanical transducers. However, the distortions provided by digital formats and digital converters are more insidious. Here the non-linear effects can be quite complex. You see this with R-2-R DACs if the resistors don't have the exact values. You see this with sigma-delta modulators because their noise varies, even with a change in DC. This is a limitation of the 1 bit format itself as has been proven, however it is possible to shape this noise so there is little variation in the audible band. You see this noise modulation with the PCM format itself without proper dither.

Considering a multi-bit PCM cvhannel, even perfectly implented, with no dither, there is gross distortion that depends on the signal level. With rectangular dither correctly done there is no correlation between the first moment of the noise with the signal. However, the noise power (2nd moment) depends on the signal value. (For example, with rectangular dither if the signal value were 0.0 the digit output would be 0, if 0.5 the digit output would dither from 0 to 1, and if the signal value were 1.0 the digit output would be 1. With TPDF dither then the 2nd moment of the noise is uncorrelated. This is a big improvement (but it does come with more total noise for a given bit depth).

It has been asserted (I don't have a good reference) that it doesn't matter if 3rd and higher order moments are correlated with the signal, as shown by listening tests. I am skeptical, for the same reason I am skeptical of most negative statistical tests. Here is where it becomes possible to do better if the goal is minimizing bits stored and transmitted: use subtractive dither. If done digitally, subtractive dither totally uncorrelates the quantization noise from the signal. In addition, it removes the noise penalty created by adding dither. My experiments showed that even 8 bit PCM can sound musical if subtractive dither is used, although there is a lot of constant background noise that will be spread throughout the sound stage, much like 4 track tape.

It is often possible to read DAC chip spec sheets and reviews and see the effects of noise floor changes with signal. The problem with interpreting these results is that they may include artifacts of the measuring equipment as well as the DAC chip itself. As far as I am concerned, any DAC where the noise floor varies in a measurable fashion is not really first rate. Indeed, one may even get better sound by adding lots of extra noise to completely mask these variations. (This comment applies to people like me who hate "distortion" but don't particularly mind "noise" provided the noise is independent of the music and not grossly excessive. I suspect this includes most subjective listeners and music lovers. The problem is that most equipment is made to measure well on the usual specifications.)

 

[NorthStar]

Yes, I like Led Zep or early Black Sabbath in the car, but most classical music simply can't be heard over the road noise, so I basically switch between talk radio and rock while driving my truck.

Classical music in a convertible sports car, with the roof open, or on a motorcycle, is a no no. :b ...I like Bob Dylan's music and John Mayall and CCR for the countryside roads.
...Headphones...never.

* But I bet there are some deluxe cars built like tanks with very very smooth rides and impervious to road noises (well built and insulated, acoustically treated). ...Mercedez Benz? ...Roll Royce? ...Lincoln Continental? ...Grand Cadillac SuperHighway Seville Transcontinental? ...Acura? ...Tesla? ...

 

[Al M.]

Do you have a photograph of the window plugs?
Keith.

yes, here (post #12):

http://www.whatsbestforum.com/showthread.php?17334-My-minimonitor-subwoofer-system/page2

On the following page there is a technical link to the damping material.

 

[DaveC]

* But I bet there are some deluxe cars built like tanks with very very smooth rides and impervious to road noises (well built and insulated, acoustically treated). ...Mercedez Benz? ...Roll Royce? ...Lincoln Continental? ...Grand Cadillac SuperHighway Seville Transcontinental? ...Acura? ...Tesla? ...

I installed a system in one of my cars years ago using all Focal speakers and an amp with a tube input section. I took the interior apart and added a ton of noise suppression to all the interior panels, floor, firewall, etc and the difference was unbelievable. It was a 1992 AWD Talon/Eclipse with the Mitsu 2L turbo motor, and after the install it was as quiet as a big Mercedes and the doors sounded like a bank vault when they shut. These days I couldn't care less about the system in my car but back then it was fun...

 

[NorthStar]

That was pretty cool Dave. ...Did you listen to Classical music, while driving, ...back then? ...And today?

 

[jkeny]

I must have misunderstood you. I thought you said a guy you know on a DIY forum had done these measurements, and I assumed they were not terribly expensive or demanding.

Ah, Scott Wurcer is not just a DIY guy he is one of the chief engineers at Analog Devices. He is responsible for the design of some of their opamps & has access to a lot of sophisticated equipment.

Well, that's a reason not to publish them in your brochures, but I expect that during product development and testing, lots of measurements are done that are not making it to the marketing departments.

Tim

Indeed, one would think so but he, Scott Wurcer, posted this a couple of years ago & from his post, he didn't describe it as a measuring technique that Analog Devices perform their DACs. I don't know if they do now but I doubt it somehow!

 

[jkeny]

Adding to the confusion, there are two definitions of "noise" floating around. One is noise that is uncorrelated to the signal in any way. This is what you get if you mix in noise downstream of a source (assuming the subsequent electronics is linear). The other is noise that is correlated to the signal. This comes in various forms, as per traditional electronic amplification or mechanical transducers. However, the distortions provided by digital formats and digital converters are more insidious. Here the non-linear effects can be quite complex. You see this with R-2-R DACs if the resistors don't have the exact values. You see this with sigma-delta modulators because their noise varies, even with a change in DC. This is a limitation of the 1 bit format itself as has been proven, however it is possible to shape this noise so there is little variation in the audible band. You see this noise modulation with the PCM format itself without proper dither.

Considering a multi-bit PCM cvhannel, even perfectly implented, with no dither, there is gross distortion that depends on the signal level. With rectangular dither correctly done there is no correlation between the first moment of the noise with the signal. However, the noise power (2nd moment) depends on the signal value. (For example, with rectangular dither if the signal value were 0.0 the digit output would be 0, if 0.5 the digit output would dither from 0 to 1, and if the signal value were 1.0 the digit output would be 1. With TPDF dither then the 2nd moment of the noise is uncorrelated. This is a big improvement (but it does come with more total noise for a given bit depth).

It has been asserted (I don't have a good reference) that it doesn't matter if 3rd and higher order moments are correlated with the signal, as shown by listening tests. I am skeptical, for the same reason I am skeptical of most negative statistical tests. Here is where it becomes possible to do better if the goal is minimizing bits stored and transmitted: use subtractive dither. If done digitally, subtractive dither totally uncorrelates the quantization noise from the signal. In addition, it removes the noise penalty created by adding dither. My experiments showed that even 8 bit PCM can sound musical if subtractive dither is used, although there is a lot of constant background noise that will be spread throughout the sound stage, much like 4 track tape.

It is often possible to read DAC chip spec sheets and reviews and see the effects of noise floor changes with signal. The problem with interpreting these results is that they may include artifacts of the measuring equipment as well as the DAC chip itself. As far as I am concerned, any DAC where the noise floor varies in a measurable fashion is not really first rate. Indeed, one may even get better sound by adding lots of extra noise to completely mask these variations. (This comment applies to people like me who hate "distortion" but don't particularly mind "noise" provided the noise is independent of the music and not grossly excessive. I suspect this includes most subjective listeners and music lovers. The problem is that most equipment is made to measure well on the usual specifications.)

Very good post, Tony

Even though ESS highlighted the modulating noise issue in SD DACS & showed plots of it in a competitor's DAC they seem not to have been able to fully eliminate it as I've seen some measurements done on the DAC by some competent amateurs on DIYaudio which seem to show a variation in noise with different volume settings. Could be an artefact of the measuring device but I don't think so - it was a while ago & I'd need to look back at the posts.

 

[Tony Lauck]

Very good post, Tony

Even though ESS highlighted the modulating noise issue in SD DACS & showed plots of it in a competitor's DAC they seem not to have been able to fully eliminate it as I've seen some measurements done on the DAC by some competent amateurs on DIYaudio which seem to show a variation in noise with different volume settings. Could be an artefact of the measuring device but I don't think so - it was a while ago & I'd need to look back at the posts.

On some of the spec sheets and magazine reviews the only way to see noise floor changes is with different test scenarios, e.g. different test signals that appear at different levels. This makes it certain that the plots are scaled differently and likely that the test instruments were running on different gain settings as well. Sometimes, however, you can get comparisons that are more likely to be consistent.

I did this kind of testing a while back with a very bad 4th order 1 bit modulator that came from a Philips paper. Horrible noise modulation, and that even after putting the filter output through a hard brick wall filter at 20 kHz. It was easy to see what happened when converting a DC signal to DSD. However, the really interesting behavior of modulators isn't exposed with DC signals, it appears with musical transients. These errors can probably be smoked out through null tests, although their correlation with perceived sound quality is going to be contentious. In addition, to measure peak errors (analog in - analog out) one needs to take into account the filter used for removing the strong modulation noise. This exposes one of the main problems with digital audio: there is no standard of accuracy that defines what the "ideal" encoder should do and what the "ideal" decoder should do. It's like the situation with LP records before the RIAA curve, only more subtle (audibly) and more complex (mathematically).

Because of the chaotic behavior of modulators, a tiny change in the input (e.g. a single low order bit in a PCM bit stream) can result in completely different noise. I suspect this effect is magnifying minute differences and creating a lot of audiophile neurvosa. Unfortunately, I don't have a suitable test set up to explore this on my own. (The chaotic state held in the modulators may also account for some cases of CD RIPs sounding different due to offset errors.)

One test that could be done would be to take a PCM file of fairly low amplitude signals (mono, double test for stereo) and then split this into two separate files by adding a random set of PCM samples (large, but not so large as to overflow after the add). The noise would be added in one case and subtracted in the other. Then the two DAC outputs could be mixed. The effect of this process would be to decorrelate any distortion with the music. Just feeding silent music one will probably hear lots of artifacts this way. (Nothing new here, Barry Blesser did this back in the 80's with ladder DACs.) You will not see this approach in commercial products so long as people are spec driven, because DACs built this way are going to be noisy, but clean. They will probably sound better to many audiophiles, but so long as people purchase products based on conventional specifications they will be dead in the water.

 

[Phelonious Ponk]

Ah, Scott Wurcer is not just a DIY guy he is one of the chief engineers at Analog Devices. He is responsible for the design of some of their opamps & has access to a lot of sophisticated equipment.


Indeed, one would think so but he, Scott Wurcer, posted this a couple of years ago & from his post, he didn't describe it as a measuring technique that Analog Devices perform their DACs. I don't know if they do now but I doubt it somehow!

If a modulating noise floor is an issue, it would be critical in a source. If Analog Devices has the tools to do the measurements and don't, even during development, they must not think it reveals anything particularly important.

Tim

 

[jkeny]

If a modulating noise floor is an issue, it would be critical in a source. If Analog Devices has the tools to do the measurements and don't, even during development, they must not think it reveals anything particularly important.

Tim

And there is the most probable explanation - DAC companies using specmanship marketing!!
ESS in that video (now deleted) not only showed the measured differences that revealed noise modulation but Mallinson (the chief engineer) also said it was audible to some (yes, in blind testing) but not to him!!

 

[jkeny]

On some of the spec sheets and magazine reviews the only way to see noise floor changes is with different test scenarios, e.g. different test signals that appear at different levels. This makes it certain that the plots are scaled differently and likely that the test instruments were running on different gain settings as well. Sometimes, however, you can get comparisons that are more likely to be consistent.

I found one of the posts of the Sabre DAC ES9018 which shows noise modulation (on this Head-fi post) - I still have to find the DIYaudio one - found it!


He says this: "Perhaps your dislike towards the 9018 chips is due to the distortion that were tested with AP and confirmed with ESS. Between 1khz and 5khz, with signal between 0db and 60db, there is nonlinear distortion. It looks inaudible, but some people might believe otherwise.
Teal is -35dB input signal, and Blue is -36dB input signal"

These measurements by different individuals shows the same noise "bump" at -35/-136dB

Also an interesting video here of measurement comparisons between ES9018 & AKM4396 DACs, I believe? Commentary is in French, however so if anyone can give technical highlights & timestamp of important measurements, I would be grateful - my French is no longer adequate to comprehend fully.

I did this kind of testing a while back with a very bad 4th order 1 bit modulator that came from a Philips paper. Horrible noise modulation, and that even after putting the filter output through a hard brick wall filter at 20 kHz. It was easy to see what happened when converting a DC signal to DSD. However, the really interesting behavior of modulators isn't exposed with DC signals, it appears with musical transients. These errors can probably be smoked out through null tests, although their correlation with perceived sound quality is going to be contentious. In addition, to measure peak errors (analog in - analog out) one needs to take into account the filter used for removing the strong modulation noise. This exposes one of the main problems with digital audio: there is no standard of accuracy that defines what the "ideal" encoder should do and what the "ideal" decoder should do. It's like the situation with LP records before the RIAA curve, only more subtle (audibly) and more complex (mathematically).

Yes, many different filters with different approaches can be used - as you say, no standard applies - mainly because the resultant issues are considered inaudible.

Because of the chaotic behavior of modulators, a tiny change in the input (e.g. a single low order bit in a PCM bit stream) can result in completely different noise. I suspect this effect is magnifying minute differences and creating a lot of audiophile neurvosa. Unfortunately, I don't have a suitable test set up to explore this on my own. (The chaotic state held in the modulators may also account for some cases of CD RIPs sounding different due to offset errors.)

One test that could be done would be to take a PCM file of fairly low amplitude signals (mono, double test for stereo) and then split this into two separate files by adding a random set of PCM samples (large, but not so large as to overflow after the add). The noise would be added in one case and subtracted in the other. Then the two DAC outputs could be mixed. The effect of this process would be to decorrelate any distortion with the music. Just feeding silent music one will probably hear lots of artifacts this way. (Nothing new here, Barry Blesser did this back in the 80's with ladder DACs.) You will not see this approach in commercial products so long as people are spec driven, because DACs built this way are going to be noisy, but clean. They will probably sound better to many audiophiles, but so long as people purchase products based on conventional specifications they will be dead in the water.

Yes you mentioned this added noise technique of Barry Blesser before - was it ever used/tested in a built design?

 

[Tony Lauck]

Yes you mentioned this added noise technique of Barry Blesser before - was it ever used/tested in a built design?

I recall going to a lecture Barry Blessser gave. (My weak recollection was that it happened at Lincoln Labs in the 1980's.)

I believe there are DACs and ADCs that use these ideas. They are a form of what is now called "dynamic element matching". I suspect that the Pacific Microsonics PM2 uses a variant of these techniques. See Figure 15 of Keith Johnson's patent, US5479168.

 

[Groucho]

The chaotic state held in the modulators may also account for some cases of CD RIPs sounding different due to offset errors

Could someone explain this in more depth..? (thanks)

 

[jkeny]

I recall going to a lecture Barry Blessser gave. (My weak recollection was that it happened at Lincoln Labs in the 1980's.)

I believe there are DACs and ADCs that use these ideas. They are a form of what is now called "dynamic element matching". I suspect that the Pacific Microsonics PM2 uses a variant of these techniques. See Figure 15 of Keith Johnson's patent, US5479168.

Thanks - found this image which explains shows the use of additive noise increasing the slewrate of the signal?

 

[NorthStar]

Also an interesting video here of measurement comparisons between ES9018 & AKM4396 DACs, I believe? Commentary is in French, however so if anyone can give technical highlights & timestamp of important measurements, I would be grateful - my French is no longer adequate to comprehend fully.

John, you don't have a "Translate" feature on your top browser toolbar?
If not, I can certainly give you the technical highlights. ...What do you mean by "timestamp", from the measurements that count?

EDIT: It's a youtube video, I see; we need the text. /// I looked @ some of it, and it is not black on white...the narrator is also exploring several implementations on those DACs, and you cannot simply relate on those graphs from one implementation only. Overall the ESS Sabre DAC appears to be "cleaner", but that's not the full story; the AKM DACs, the Analog Devices DACs, Burr-Brown DACs, ...depending of their Grade (carefully hand-chosen selectively and fine-tuned and matched) and on their implementation, ...dual, quad balanced differential mode...is all the pudding on the cake around the top cherry of the sunday (or yogourt). ...Icing, skating ring. ...Dithering matching each music recording and jitter equaling zero. ...It's like asking for immortality. ;-)

 

[Tony Lauck]

Could someone explain this in more depth..? (thanks)

Some people have noted that a CD track can be ripped with an offset due to hardware and software misconfigurations. The file has all the audible music samples correct and is the right length, but starts at the wrong place, e.g. has a few extra zeros at the start and a few missing zeros at the end. Since these samples are silent, an offset rip and the correct rip ought to sound the same. The only audible difference ought to be a few microseconds difference in the delay between pressing "Play" and the music beginning. Some audiophiles have reported that these rips sounded different.

There are various explanations that have been offered as to how this can possibly be. In the case I mentioned, a sigma-delta DAC has memory of what it has done, which may possibly include the length of time the modulator has been running since playback started. This provides one type of explanation. Other "explanations" have been the ones made by hard-core objectivists, including calling those who heard differences "delusional", etc...