• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

ESS THD ‘Hump’ Investigation

cjfrbw

Senior Member
Joined
Mar 5, 2018
Messages
410
Likes
472
Don't hump investigations belong in the 'Rape Culture' thread?
 
OP
jackenhack

jackenhack

Active Member
Forum Donor
Joined
Oct 25, 2018
Messages
203
Likes
497
Location
Stockholm, Sweden
Ok, update on trying to get Bus Pirate to talk to the DAC. It captures the data without problems but refuses to read or write. I've checked on my oscilloscope, and it's a problem with pulldown of the signal. I'll continue tomorrow to see if I can find a fix. Very irritating. The Bus Pirate is in a weird kind of limbo, and the documentation is horrendous. If you are curious how it works, there is a video on Youtube on how to use it.
It's filmed in Potato-Cam vision, but you will get the gist of it.

 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
I don't know that John wanted to teach people how to fix the problem. :) He may have even misunderstood which issue we are talking about as we thought it was an IMD only issue at the time. Clearly though, John is very good at micro analyzing audio performance and likely enjoys very good support from ESS so has found and fixed this issue. Or avoided it somehow.

I think our effort here to keep analyzing it is good in figure out what is going on. We can get help then from our friendly manufacturers and possibly another push to ESS to listen.

I think I agree with you that he may have been referring to another issue, I have been one eyebrow raised quizzically about his comments since he posted them.

I also started to look at this in detail recently but have been somewhat "time poor" of late. Will delve into it over the Christmas break.
 

hetzer

Member
Joined
Nov 24, 2018
Messages
45
Likes
121
Since the behavior of THD increase is only at specific level, I think the behavior of DEM(Dynamic Element Matching) is cause of the problem.
DEM is a noise shaper that removes the distortion made by multi-bit quantizer inside the delta-sigma modulator. It actually pushes the distortion to ultrasonic frequency so that the dac manufacurers can remove them using analog notch filter.
ESS is the first manufacturer as far as I know to use DEM to remove distortion and noise floor modulation.
This is the patent of ESS about their DEM architecture.
https://patents.google.com/patent/US7116257
For non-engineer, I recommend this link.
https://books.google.ca/books?id=8l...v=onepage&q=cmoset high end audio dac&f=false
The problem is that the DEM which ESS is using is too old. TI, Cirrus Logic, Asahi Kasei also have their own architecture and they develop new algorithms every time they launch a new series of dac. But It seems that ESS didn't make such thing since their first Hyperstream architecture. I am not sure if they really did or not because ESS don't share enough information. But this can be the cause of thd increase at certain range of output level.
For example, this is part of the paper that one of the engineer of Asahi Kasei published at 2000.
Screenshot_20181220-151630.jpg

this is the explanation of 'dip' of SNDR in conventional DWA(Data Weighted Algorithm, most typical method of DEM)
"The sharp dips in the traces are typical artifacts of DWA. Because of the periodic nature of DWA, the shaped mismatch error appears as tones. These out-of-band tones may fold back to inband due to limit cycles that are dependent on the signal level, and degrade the in-band SNDR."
If the problem is caused by DEM algorithm, there is nothing we can do about it.
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
To what extent is the hump a result of increased noise, harmonic distortion, and/or IMD?
 

rajapruk

Senior Member
Forum Donor
Joined
Sep 22, 2018
Messages
472
Likes
489
Since the behavior of THD increase is only at specific level, I think the behavior of DEM(Dynamic Element Matching) is cause of the problem.
DEM is a noise shaper that removes the distortion made by multi-bit quantizer inside the delta-sigma modulator. It actually pushes the distortion to ultrasonic frequency so that the dac manufacurers can remove them using analog notch filter.
ESS is the first manufacturer as far as I know to use DEM to remove distortion and noise floor modulation.
This is the patent of ESS about their DEM architecture.
https://patents.google.com/patent/US7116257
For non-engineer, I recommend this link.
https://books.google.ca/books?id=8lwnzVwBbYcC&pg=PA44&dq=cmoset+high+end+audio+dac&hl=en&sa=X&ei=qAwHUc2NKunQigK1xYCoBA#v=onepage&q=cmoset high end audio dac&f=false
The problem is that the DEM which ESS is using is too old. TI, Cirrus Logic, Asahi Kasei also have their own architecture and they develop new algorithms every time they launch a new series of dac. But It seems that ESS didn't make such thing since their first Hyperstream architecture. I am not sure if they really did or not because ESS don't share enough information. But this can be the cause of thd increase at certain range of output level.
For example, this is part of the paper that one of the engineer of Asahi Kasei published at 2000.
View attachment 19126
this is the explanation of 'dip' of SNDR in conventional DWA(Data Weighted Algorithm, most typical method of DEM)
"The sharp dips in the traces are typical artifacts of DWA. Because of the periodic nature of DWA, the shaped mismatch error appears as tones. These out-of-band tones may fold back to inband due to limit cycles that are dependent on the signal level, and degrade the in-band SNDR."
If the problem is caused by DEM algorithm, there is nothing we can do about it.

DEM is that the same thing as Dither?
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,835
Likes
16,497
Location
Monument, CO
DEM has been around a long time. First time I read about it was in an article (and later book) by Rudy van der Plassche back in the 1990's, I think. (Aside: Every good idea seems to have been invented back in the tube days, or maybe using stone knives and bearskins <cr. Spock>.) It is not the same as dither, and is not part of the noise shaping in the designs I have seen (but I do not claim to have seen them all; it could be incorporated into the noise shaping cells, but that is not its usual goal. It essentially "rotates" among several unit cells and averages their output. The average can be much more precise than using a single cell, and can obviate (very expensive and time-consuming) trimming of internal cells to achieve good matching. The "rotation" can occur at a fixed rate (or be spread; I tried doing the latter to reduce clock noise) but averaging "after" the cells suppresses the noise from the rotation clock. Hopefully.

Noise shaping is what pushes the quantization noise out of the audio band; dynamic element matching enables precision ADC/DACs in multibit delta-sigma designs. One little secret is that most delta-sigma designs these days include multibit quantizers or DACs inside the loop, and those multibit pieces require high precision (or some sort of digital compensation is used).

I am not sure how DEM would cause the problems shown but it is not something I have really though about. It is possible it is only enabled as part of their distortion suppression circuits, and only affects the MSBs (where it is normally used) that are engaged at higher signal levels. The catch is that delta-sigma designs, and most DACs, still toggle the MSBs for low signal levels since zero-crossings mean switching the MSB. There are many architectures that avoid switching large values at zero crossings (virtually every design of mine and most others avoid this like the plague).

IME/IMO - Don
 

Pluto

Addicted to Fun and Learning
Forum Donor
Joined
Sep 2, 2018
Messages
990
Likes
1,631
Location
Harrow, UK
Has anyone had a chance to check out the effect of this thread on the big ugly part of the audio world where tubes, vinyl and subjective emotions reign supreme? It's, obviously, very easy to lose sight of the fact that, despite the mysterious distortion hump in the current iteration, ESS DAC devices pretty much rule the roost right now because of their (otherwise) superlative performance. There are those on audio forums who proudly declare that they find any ESS-based converter “painful” [substitute pejorative of choice].

Are these folks now celebrating that, with Amir's ‘discovery’ of the ESS hump, the reason for the “painful” sound of ESS is now apparent to those objectivist fools?
 

hetzer

Member
Joined
Nov 24, 2018
Messages
45
Likes
121
I apologise for that I used some of the terms incorrectly at my previous post. As DonH56 corrected, DEM is not a noise shaper that handles quantization noise. It converts nonlinear distortion by imperfection of multi-bit dac to noise.

There is a picture that shows what DEM does at US 7,916,058 B1 patent by Texas Instrument. Fig 6a, b, c will be a good example.
Screenshot_20181221-183937.jpg

The 'noise shaper' part that I meant at the previous post is presented at US 6,266,002 B1 patent by Cirrus Logic. The title is "
2nd order noise shaping dynamic element matching for multibit data converters"
Screenshot_20181221-183841.jpg

I just thought that DWA can be the cause of the hump as Asahi Kasei's partial DWA showed removment of hump of signal-to-noise plus distortion on conventional DWA. As DEM algorithm of ESS is well designed as shown on their own presentation, the hump is likely to be by other cause. But I am just telling about possibility since the 'hump' is a very unusual case.
https://www.yumpu.com/document/view/23182504/3-shaping-sigma-delta-dacs-ess-technology-inc
this is a presentation file of ESS and it shows the great dc sweep measurement of their dac. As far as I know, every modern dac chip over 120db SNR shows similar measurement too.
Lastly, this is a white paper from Wolfson(now it is part of Cirrus Logic) with useful information about DEM.
https://statics.cirrus.com/pubs/whitePaper/WP_Design_Eval_AudioDAC.pdf
 

hetzer

Member
Joined
Nov 24, 2018
Messages
45
Likes
121
Has anyone had a chance to check out the effect of this thread on the big ugly part of the audio world where tubes, vinyl and subjective emotions reign supreme? It's, obviously, very easy to lose sight of the fact that, despite the mysterious distortion hump in the current iteration, ESS DAC devices pretty much rule the roost right now because of their (otherwise) superlative performance. There are those on audio forums who proudly declare that they find any ESS-based converter “painful” [substitute pejorative of choice].

Are these folks now celebrating that, with Amir's ‘discovery’ of the ESS hump, the reason for the “painful” sound of ESS is now apparent to those objectivist fools?

The painful sound of ESS dac is only audible for the guys with 'golden ears' who can hear THD under -60db on their reference music and has headphones with lower THD than Toppings DAC.
 

hetzer

Member
Joined
Nov 24, 2018
Messages
45
Likes
121
I forgot to share ess patent of harmonic suppression technique. Sorry for posting too many things but this will be helpful. I'm sure that this is the exact thd compensation function of new ess chips.
This is part of the document
"The digital signal processing provides a method for substantially removing the
even and odd harmonics of the DAC output error."
"The DAC includes a first circuit configured to remove even harmonics from a sigma delta circuit, and a second circuit configured to remove odd harmonics."
Screenshot_20181221-220803.jpg

Digital circuit, different curcuit for odd and even harmonics... This is it!
US 6,897,795 B2
https://patents.google.com/patent/US6987475?oq=US+6,897,795+B2
 
Last edited:

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,835
Likes
16,497
Location
Monument, CO
Downloaded, will read it this weekend. I'd forgotten what fun it is to decipher ideas when engineering verbiage is changed into legalese. Hope the pictures are good. :)
 
Top Bottom