ESS claim DAC pushes the boundaries of audio science, They claim results are supported by blind testing Options

[wakibaki]

Apologies if this topic has been posted previously, I did a search but found nothing.

http://www.youtube.com/watch?feature=playe...p;v=1CkyrDIGzOE

The above link shows a youtube video of an ESS presentation at RMAF.

Granted it is a sales presentation, it details advances ESS claim in the design of their sigma-delta DACs.

Foremost in the claims is the fact that some people can evidently hear features well below the noise floor which were previously held by the majority of engineers to be inaudible. This is said to be confirmed in blind testing.

Large changes in the values of variables in the state space are said to contribute to non-periodic steady-state noise normally invisible using conventional measurement techniques. This is visible when instrumentation is used to examine the Noise-vs-DC Offset.

This is said to be minimized in ESS products by techniques designed to control the state-space related noise by rapidly quenching state variable excursions.

ESS also claim that a further artifact of sigma-delta modulators, variable (non-linear) excess phase, which causes oscillation in S-D modulators, has been eliminated by making their devices unconditionally stable, and that this too contributes to greater acceptance of the devices by some listeners, although the reasons for this are apparently not clearly understood.

w

[Arnold B. Kruege...]

Apologies if this topic has been posted previously, I did a search but found nothing.

http://www.youtube.com/watch?feature=playe...p;v=1CkyrDIGzOE

The above link shows a youtube video of an ESS presentation at RMAF.

About 10 minutes into the presentation it misidentifies dither as noise shaping. How many more serious simple errors do I have to tally up before I justify not wasting any more time with it?

[2Bdecided]

About 10 minutes into the presentation it misidentifies dither as noise shaping.

No, he says the signal is dithering around, and puts the word dither in quotes. I think he knows full well it's not conventional dither he's talking about.


I can well believe the graphs he's showing. I would question audibility. I would question how the DAC responds to real-world signals. It's not just that it's hard to measure and show the (real) performance issues he's describing once you use real signals - it's that the noise in real signals randomises those problems so well that both measurement equipment and ears struggle to pick it up any more. There are/were SDM problems that did crop up in an apparently unpredictable ways with some real signals, but most companies believe those problems are solved.

It's a fascinating talk. There are some great, insightful examples in there (not least the very last one). However, he accepts the audibility of high (ultrasonic) frequencies as read, so...

I looked at the YouTube channel it's in - there are many other fascinating talks in the same channel (many from familiar names) which I wish I had time to listen to - but you will note very quickly that many of the people talking do not take the same approach to audio as HA - so while you may learn something, expect to find things that you totally disagree with...

http://www.youtube.com/user/RMAudiofest?feature=watch

Cheers,
David.

[Arnold B. Kruege...]

About 10 minutes into the presentation it misidentifies dither as noise shaping.

No, he says the signal is dithering around, and puts the word dither in quotes. I think he knows full well it's not conventional dither he's talking about.

Please provide the time in the video where that takes place.

Maybe we're not watching the same video, because the one I'm watching starts mentioning noise shaping at 8:02, but the slide' graphic shows dither. The slide specifically mentions noise shaping, and AFAIK dither has not been mentioned at all at this point. Actually, the voice comments at that point seems to be describing the operation of a successive-approximation converter. But the slide says quite clearly "Noise shaping".

[2Bdecided]

Maybe we're not watching the same video, because the one I'm watching starts mentioning noise shaping at 8:02, but the slide' graphic shows dither. The slide specifically mentions noise shaping, and AFAIK dither has not been mentioned at all at this point. Actually, the voice comments at that point seems to be describing the operation of a successive-approximation converter. But the slide says quite clearly "Noise shaping".

If you use low order noise shaping without dither, it'll look a bit like those graphs. The graphs are clearly drawn in a graphics package, not real (he's got the step size wrong on the green "noise shaped" graph for one thing), but he is showing noise shaping.

Of course, he could equally well be showing lousy dither. If you think you can tell the difference between an artistic representation of non-optimal dither, and an artistic representation of low order noise shaping, then can I respectfully suggest that you don't know what at least one of them really looks like.

Cheers,
David.

[Arnold B. Kruege...]

Maybe we're not watching the same video, because the one I'm watching starts mentioning noise shaping at 8:02, but the slide' graphic shows dither. The slide specifically mentions noise shaping, and AFAIK dither has not been mentioned at all at this point. Actually, the voice comments at that point seems to be describing the operation of a successive-approximation converter. But the slide says quite clearly "Noise shaping".

If you use low order noise shaping without dither, it'll look a bit like those graphs. The graphs are clearly drawn in a graphics package, not real (he's got the step size wrong on the green "noise shaped" graph for one thing), but he is showing noise shaping.

Can't be.

Just to remind you, here's a really pretty good definition of noise shaping from Wikipedia:

"
Noise shaping is a technique typically used in digital audio, image, and video processing, usually in combination with dithering, as part of the process of quantization or bit-depth reduction of a digital signal. Its purpose is to increase the apparent signal to noise ratio of the resultant signal. It does this by altering the spectral shape of the error that is introduced by dithering and quantization; such that the noise power is at a lower level in frequency bands at which noise is perceived to be more undesirable and at a correspondingly higher level in bands where it is perceived to be less undesirable. A popular noise shaping algorithm used in image processing is known as ‘Floyd Steinberg dithering’; and many noise shaping algorithms used in audio processing are based on an ‘Absolute threshold of hearing’ mode
"

The key part of this definition is "Its (noise shaping) purpose is to increase the apparent signal to noise ratio of the resultant signal. It does this by altering the spectral shape of the error that is introduced by dithering and quantization; such that the noise power is at a lower level in frequency bands at which noise is perceived to be more undesirable and at a correspondingly higher level in bands where it is perceived to be less undesirable."

The long and the short of it is that noise shaping takes place in the frequency domain. Until one shows something going on in the time or frequency domain, they aren't sensibly talking about noise shaping.

Since the diagram he shows is clearly in the amplitude domain, he is not sensibly talking about noise shaping. My most charitable judgement is that he was talking about dithering even though his discussion of pump gas prices sounded more like a discussion of successive approximation.

Bottom line, I think this guy was giving a mish-mash of every lecture and book he ever read or heard about DACs, about all 3 of them (!!!), jammed aimlessly into something that made sense to him, but can't actually make sense to anybody who has a proper understanding of what he was conflating and muddling through.

This post has been edited by Arnold B. Krueger: Jan 25 2013, 16:38

[benski]

The long and the short of it is that noise shaping takes place in the frequency domain. Until one shows something going on in the time or frequency domain, they aren't sensibly talking about noise shaping.

What are you talking about?

You read the wikipedia article but you clearly did not look at the math. Error-shaping clearly happens in the time domain. The gas station example is a near-perfect example of first-order error-shaping. Noise-shaping happens by adding dither and a filter. The fact that we can color via the filter does not make any of the math or practical operations happens in the frequency domain.

This post has been edited by benski: Jan 25 2013, 17:28

[2Bdecided]

What are you talking about?

You read the wikipedia article...

ouch! FWIW, while I wouldn't even claim to be up to speed on noise shaping these days, never mind a world expert, I have at least designed them in the recent-ish past.

...but you clearly did not look at the math. Error-shaping clearly happens in the time domain. The gas station example is a near-perfect example of first-order error-shaping. Noise-shaping happens by adding dither and a filter.

You don't need dither for noise shaping. You just "shape" the quantisation noise.

This is good, but sadly some of the images are rather fuzzy...
http://www.essex.ac.uk/csee/research/audio...aping%20IOA.pdf

The HA knowledge base is fine too, though note in this picture...
http://wiki.hydrogenaudio.org/index.php?ti...oiseshaper3.png
...you don't need the dither, and strictly speaking you don't need H(z) (i.e. the filter) either, as the noise still gets shaped just by delaying it by one sample and subtracting it from the input. That just gives a very simple rising-with-frequency characteristic.

Cheers,
David.

This post has been edited by 2Bdecided: Jan 25 2013, 17:55

[benski]

and strictly speaking you don't need H(z) (i.e. the filter) either, as the noise still gets shaped just by delaying it by one sample and subtracting it from the input. That just gives a very simple rising-with-frequency characteristic.

Ahh, but it's still technically a filter

[2Bdecided]

and strictly speaking you don't need H(z) (i.e. the filter) either, as the noise still gets shaped just by delaying it by one sample and subtracting it from the input. That just gives a very simple rising-with-frequency characteristic.

Ahh, but it's still technically a filter

I know delay one sample and subtract is a filter. H(z) is "the filter" in the article that pictures comes from...
http://wiki.hydrogenaudio.org/index.php?title=Noise_shaping
The point is you are still "noise shaping" without putting an explicit filter (e.g. some psychoacoustically optimised thing) in the feedback loop.

And it is just such a simple topology that really could give rise to graphs rather like those in the presentation. Due to noise shaping. Without dither.

Cheers,
David.

This post has been edited by 2Bdecided: Jan 25 2013, 19:50