Measurements: "ESS Hump" revisited (Khadas Tone Board V1.3)

[KSTR]

Forgive me if i’m wrong but is not Ben ( a member of this forum ) the designer of SONCOZ also the designer of the KTB ? Maybe he has some insights , this is technical beyond my skill set , so good luck with this curious investigation

I don't know but obviously WesionTek gained knowledge to remove the hump and reduce the noise with the Tone2 Pro (now on par with the D10 B)

However is not that far worse performance than this still is transparent when listening to actual music ?

That's the big question. In a true constructive and collaborational effort I think we could sort that out and I will try to contribute to that effort.

 

[Pdxwayne]

Gut feeling and educated guessing suggests me that if there is an audible difference between a "typical" ESS vs. say, an AKM, then this is it.
I will do loopback music recordings and further analysis and maybe even post ABX'able material.

Thank you for doing all this! I appreciate your efforts!

My first DAC blind test was between ktb and e30. At that time, once level and filter matched, I could not really tell a difference between them. Looking at your charts, I think I selected my test song incorrectly.

If you have found a few good demo songs that have the right kind of output levels, please share.

Thanks again!

 

[DDF]

I don't know but obviously WesionTek gained knowledge to remove the hump and reduce the noise with the Tone2 Pro (now on par with the D10 B)

@KSTR, John from Benchmark posted "All D/A converter chips produce significant common-mode distortion on the balanced outputs. This unwanted common-mode distortion is easily removed with a differential amplifier. If it is not removed, the result is IMD...
...This IMD (and THD) is a direct result of omitting a differential amplifier after the output of the chip...the solution is to place a well trimmed differential amplifier at the balanced outputs of the D/A chip"

 

[restorer-john]

I don't know but obviously WesionTek gained knowledge to remove the hump and reduce the noise with the Tone2 Pro (now on par with the D10 B)

The thing is, CD player manufacturers back in the early 1990s were doing exactly what John Siau wrote in the thread linked above by @DDF. Nothing new in the HiFi world eh?

The differential DAC chip outputs of their earliest (1990 model PWM DAC CXD-2552Q-1) used in parallel pairs on Sony TOTL ($2000+) players and run into a carefully trimmed (1% resistors etc) differential amplifier, then LPF, buffer for SE and then to a separate balanced line driver which was a buffer driving a proper 600:600R transformer for balanced. They were able to equal the high level THD, IMD and noise of the PCM-58P-S model from the year before, but with much better low level THD.

 

[half_dog]

@KSTR, John from Benchmark posted "All D/A converter chips produce significant common-mode distortion on the balanced outputs. This unwanted common-mode distortion is easily removed with a differential amplifier. If it is not removed, the result is IMD...
...This IMD (and THD) is a direct result of omitting a differential amplifier after the output of the chip...the solution is to place a well trimmed differential amplifier at the balanced outputs of the D/A chip"

Khadas Tone Board has a I/V amplifier for each polarity and than a differential amplifier summing the signals. Topping do the same on their D10s but different resistor and capacitors values and some other differences.

https://static.wixstatic.com/media/04a2a6_cebb7cecd25f43d496ca74ab72916ac6~mv2_d_2500_2500_s_4_2.jpg/v1/fill/w_966,h_966,al_c,q_85,usm_0.66_1.00_0.01/04a2a6_cebb7cecd25f43d496ca74ab72916ac6~mv2_d_2500_2500_s_4_2.jpg

Topping D10s USB DAC and Bridge Review

At least we got a common on the stuff we've heard except R2R. My enjoyment in music doesn't involve me doing proper matched test. I just listen to music. So sue me :) No offense, I'm not saying you can't enjoy what you enjoy. Some like to buy overly expensive stuff just to feel good, aesthetics...

www.audiosciencereview.com

@KSTR I have Khadas first gen too and a D10s. Inspired by you I tried to measure them using a Tascam US-366 (not the cleanest ADC but it seems enough to display something).

Test Setup:

• A notebook Vostro 5470 powered by its battery (floating) is connected to an US-366 which serves to records the Distortion Spectra.
• Topping D10s is conneted via USB to another notebook, an G7 7588 (grounded and running REW), which serves as SPDIF source for the KTB and itself.
• KTB is supplied by a 5V Samsung cell-phone charger connected to the USB-C input of the board.
• KTB and D10s outputs go to a Topping L30. This is used to keep the signal around -13dBFS - this is a good level for US 366 inputs to work.
• Test Frequency is 240Hz.
• At Vostro REW uses a 512k-point FFT with 8 averages to display the distortion spectra - the notebook couldn't run 1Meg.

 

[Herbert]

General Question: Why don´t we see the distortions in the multitone tests?

 

[Killingbeans]

Because it's level dependent.

 

[Herbert]

So if the multitone would not be done at full scale,
but let´say -20db, the distortions would be more visible, correct?

 

[KSTR]

Khadas Tone Board has a I/V amplifier for each polarity and than a differential amplifier summing the signals. Topping do the same on their D10s but different resistor and capacitors values and some other differences.

https://static.wixstatic.com/media/04a2a6_cebb7cecd25f43d496ca74ab72916ac6~mv2_d_2500_2500_s_4_2.jpg/v1/fill/w_966,h_966,al_c,q_85,usm_0.66_1.00_0.01/04a2a6_cebb7cecd25f43d496ca74ab72916ac6~mv2_d_2500_2500_s_4_2.jpg

Topping D10s USB DAC and Bridge Review

At least we got a common on the stuff we've heard except R2R. My enjoyment in music doesn't involve me doing proper matched test. I just listen to music. So sue me :) No offense, I'm not saying you can't enjoy what you enjoy. Some like to buy overly expensive stuff just to feel good, aesthetics...

www.audiosciencereview.com

@KSTR I have Khadas first gen too and a D10s. Inspired by you I tried to measure them using a Tascam US-366 (not the cleanest ADC but it seems enough to display something).

Test Setup:

• A notebook Vostro 5470 powered by its battery (floating) is connected to an US-366 which serves to records the Distortion Spectra.
• Topping D10s is conneted via USB to another notebook, an G7 7588 (grounded and running REW), which serves as SPDIF source for the KTB and itself.
• KTB is supplied by a 5V Samsung cell-phone charger connected to the USB-C input of the board.
• KTB and D10s outputs go to a Topping L30. This is used to keep the signal around -13dBFS - this is a good level for US 366 inputs to work.
• Test Frequency is 240Hz.
• At Vostro REW uses a 512k-point FFT with 8 averages to display the distortion spectra - the notebook couldn't run 1Meg.

Thanks for that, and it readily shows the difference even without a notch filter for the ADC. At -13dBFS your ADC the US366 seems to wok well, exactly as you say.

 

[Killingbeans]

So if the multitone would not be done at full scale,
but let´say -20db, the distortions would be more visible, correct?

I would think so. Between -20dBFS and -35dBFS the "grass" would become a bit taller proportionally. Looks like it would be "worst" right around -23dBFS.

 

[KSTR]

@KSTR, John from Benchmark posted "All D/A converter chips produce significant common-mode distortion on the balanced outputs. This unwanted common-mode distortion is easily removed with a differential amplifier. If it is not removed, the result is IMD...
...This IMD (and THD) is a direct result of omitting a differential amplifier after the output of the chip...the solution is to place a well trimmed differential amplifier at the balanced outputs of the D/A chip"

Now the KTB has a standard subtractor stage (with another 1st order filter) after the I/V's so it should do basic CM cancelling fine to about -30dB CMRR.

Anyway, yesterday I closely examined the individual I/V outputs to check for CM content and general behavior and found some interesting things that need more exploration (as time allows)...
With the proper amount of digital DC bias I could manage a stable operating point where for that I/V all distortion is even-order only (still an excessive train, though). This would allow for cancelling with the other I/V... if only it could be made to have the same polarity DC bias.
Alas, the DC in the test signal of course is applied mirrored and then the even-order only condition breaks because it is the wrong bias for the other I/V. But it is a pointer in the right direction that good degree of cancelling is possible when both I/Vs have the same even-order-only profile.

Waiting for the D10B to arrive now, as I'm really curious what the fine-print in the residual is for them.

 

[KSTR]

So if the multitone would not be done at full scale,
but let´say -20db, the distortions would be more visible, correct?

I looked at this and found something very unusual:

The distortion mechanism is so weird that it is simply not appearing with the AP's 32 multitone test signal! No matter what level.
It looks textbook perfect down to the noise floor, what?!

Only when the frequency range is limited so that the upper (600Hz+) frequency space is reduced/empty then all the dirt shows up like it did with single tones, rising the noise floor by 20dB wide-band. And quite independent of level as well, similar to the single sine test.

That's for sure worth an extra thread, quite a surprise. A clean standard 32 tone multitone result does not automatically imply there isn't any significant distortion with other signals or music. That only holds true for classic "analog" type of distortions (and equivalent digital phenomena which we have here as well).

 

[gvl]

The plot thickens. I've been always wondering why the hump is not on any of the multitone tests as afaik they are run at lower level than full scale. Sounds like the question about audibility of the hump just got much harder to answer. I certainly couldn't detect anything of significance comparing a KTB and a Topping D10s.

 

[Pdxwayne]

The plot thickens. I've been always wondering why the hump is not on any of the multitone tests as afaik they are run at lower level than full scale. Sounds like the question about audibility of the hump just got much harder to answer. I certainly couldn't detect anything of significance comparing a KTB and a Topping D10s.

Can you try listening to the following video using your KTB and D10s? Listen to the cymbals. Would KTB sound a little more fatiguing if listening for a long time to song like this?

 

[gvl]

I’m not setup for it and frankly don’t want to spend any more time on it. If anything I thought the D10s was a bit cleaner and holographic, but definitely not something I would be able to detect in an unsighted test, so in the end it didn’t matter and I switched to another DAC anyway.

 

[Pdxwayne]

I’m not setup for it and frankly don’t want to spend any more time on it. If anything I thought the D10s was a bit cleaner and holographic, but definitely not something I would be able to detect in an unsighted test, so in the end it didn’t matter and I switched to another DAC anyway.

No problem. I thought you are still using both ktb and d10s.

I haven't been using the ktb for about a year too. I have moved on to using e30 and x16. I kind of remember that the cymbal was pretty fatiguing when using ktb, thus I thought the YouTube song might be a good candidate for DAC comparison.

 

[Herbert]

Can you try listening to the following video using your KTB and D10s? Listen to the cymbals. Would KTB sound a little more fatiguing if listening for a long time to song like this?

Any "fatiguing" would come from the 128Kbps AAC, besides the lack of dynamics. Even 16 bit PCM from the original recording featured in this video would be more revealing...

 

[Pdxwayne]

Any "fatiguing" would come from the 128Kbps AAC, besides the lack of dynamics. Even 16 bit PCM from the original recording featured in this video would be more revealing...

You can always use similar kind of song in CD quality to compare.

Main question is this: Can song like this reveals the hump issue? Because of mostly low frequency tones, would those below 600hz tones cause issue with higher tones like cymbals.

 

[KSTR]

Could be perhaps some ASRC artefacts?

IMO ASRC has the potential of a periodic patter because of the fractional ratio of input and output frequencies which may slightly change periodically too, the DPLL tuning in, ...

That was my thought as well. The ASRC is also something that sets ESS apart from most (all?) other high performance DAC chips. If the ASRC is to blame, perhaps nudging the input sample rate slightly would change pattern.

I suppose it could also be a quirk of their DWA implementation.

You were right, I stand corrected.

I've now looked at the D10 balanced and found that the specific distortion pattern seen with the KTB is indeed dominated by ASRC interpolation error because in the D10 that strong sine'ish "modulation" of the residual it is basically absent, also in single-ended.
The D10 operates the ESS in slave mode (two local master clocks for 44.1/48 family rates), ASRC off. Additionally confirmed by inspecting output clock drift, none to be seen. The Khadas Tone Board2 Pro, by looking at the PCB, also seems to run synchronous mode.

I/V setup also has an impact, currently under investigation...

 

[phofman]

The D10 operates the ESS in slave mode (two local master clocks for 44.1/48 family rates), ASRC off.

I could find only pictures of D10S which has the 100MHz ASRC clock. Do you have any pictures of the D10 balanced version? Thanks.

Additionally confirmed by inspecting output clock drift, none to be seen.

IIUC ASRC should not produce any output clock drift either, that's the principle and reason for ASRC instead of some dumb FIFO. But maybe I just misunderstood...