??? Look at any thread containing measurements of tube amplifiers or e.g. Kusunoki-style TDA1543 dacs. The point is that measurements reveal imperfections which are there no matter how it sounds. Or are you claiming that e.g. FFT showing THD of 1.0% is not correct?
1% THD may be correct, but what's the crest factor, what's the time-domain residual? What is the time-variance with frequency, with volume level? If only one thing is measured (i.e. time-averaged, steady-state HD) then that thing will dominate WYSIATI, which tends to result in cognitive error.
Also, what is the alternative? ESS dac sound? IIRC @KSTR said if he turned up the type of hump distortion effects he found, they did sound like glare and they did have a fatiguing effect. He went so far as to say the effects might be audible if listening to certain types of music. What wasn't investigated was how his sensitivity to such distortion effects compares to the sensitively of other people who say they can't stand ESS dac sound. Maybe its so bad for them they will happily tolerate 1% distortion of a more benign type. Doesn't necessarily mean they like distortion in an absolute sense, nor that they like steady-state distortions.
Also, what is the alternative? ESS dac sound? IIRC @KSTR said if he turned up the type of hump distortion effects he found, they did sound like glare and they did have a fatiguing effect. He went so far as to say the effects might be audible if listening to certain types of music. What wasn't investigated was how his sensitivity to such distortion effects compares to the sensitively of other people who say they can't stand ESS dac sound. Maybe its so bad for them they will happily tolerate 1% distortion of a more benign type. Doesn't necessarily mean they like distortion in an absolute sense, nor that they like steady-state distortions.
Last edited:
You seem to have a fixation that measurements are only time averaged steady-state FFTs. There are a variety of other measurements. E.g. what about the measurements made by @KSTR when successfully investigating ESS hump.
Try learning by making measurements yourself instead of always referring to measurements made by others.
Try learning by making measurements yourself instead of always referring to measurements made by others.
I saw the distortion waveform, but did he find out the actual cause? (mismatch in the multibit dac, HF noise, etc?)the measurements made by @KSTR when successfully investigating ESS hump.
IIRC those measurements were extremely time averaged distortion residuals (i.e. block-averaged). Apparently the only way to get the waveforms out of the noise. Some discussion:
https://www.audiosciencereview.com/...ted-khadas-tone-board-v1-3.30136/post-1057416
https://www.audiosciencereview.com/...loopback-analyzer-software.27844/post-1158150
Of course humans can to some extent hear signals buried in noise. IMHO particularly so when the 'distortion' signal is highly correlated with the music signal being listened to.
Regarding the underlying cause, IIRC that was left to future work. Demodulation and or slew limiting in the opamp were two suggested possible mechanisms.
https://www.audiosciencereview.com/...ted-khadas-tone-board-v1-3.30136/post-1057416
https://www.audiosciencereview.com/...loopback-analyzer-software.27844/post-1158150
Of course humans can to some extent hear signals buried in noise. IMHO particularly so when the 'distortion' signal is highly correlated with the music signal being listened to.
Regarding the underlying cause, IIRC that was left to future work. Demodulation and or slew limiting in the opamp were two suggested possible mechanisms.
Instead claiming that a device failing in basic measurements sounds good is an indication that you prefer non-linearities and colorations.
If you think it is more than just a very weak indication, then this is the logical fallacy that Mark has been warning about.
To give you an example, my valve DAC has a couple of peculiarities compared to most audio DACs. Incomplete list:
1. Far higher noise floor
2. Headroom to prevent clipping on intersample overshoots
3. Anti-imaging filter that suppresses images from the Nyquist frequency onward (as opposed to 0.55 fs)
4. Building it costs a substantial amount of money and involves potentially lethal voltages (-300 V) as well as fine-pitch soldering (two 80-pin 0.5 mm pitch connectors and two TQFP48 packages, among other things)
So far everyone who has made one likes it very much. That could be because they like a bit of background noise, but it could just as well be that the noise just doesn't bother them and that it is due to some of the other things I listed (or forgot to put in the list, or am not even aware of myself). It could even be number 4: maybe it's all psychological and related to all the trouble they had to go through to get it to work.
I'm sure the Y scale on the plots mean something to him, but I sure have no idea what amplitude I'm looking at lolIIRC those measurements were extremely time averaged distortion residuals (i.e. block-averaged). Apparently the only way to get the waveforms out of the noise. Some discussion:
https://www.audiosciencereview.com/...ted-khadas-tone-board-v1-3.30136/post-1057416
https://www.audiosciencereview.com/...loopback-analyzer-software.27844/post-1158150
The indication may be a very weak or stronger. In SET amplifiers with several % distortions this indication is probably quite strong.If you think it is more than just a very weak indication, then this is the logical fallacy that Mark has been warning about.
To give you an example, my valve DAC has a couple of peculiarities compared to most audio DACs. Incomplete list:
1. Far higher noise floor
2. Headroom to prevent clipping on intersample overshoots
3. Anti-imaging filter that suppresses images from the Nyquist frequency onward (as opposed to 0.55 fs)
4. Building it costs a substantial amount of money and involves potentially lethal voltages (-300 V) as well as fine-pitch soldering (two 80-pin 0.5 mm pitch connectors and two TQFP48 packages, among other things)
So far everyone who has made one likes it very much. That could be because they like a bit of background noise, but it could just as well be that the noise just doesn't bother them and that it is due to some of the other things I listed (or forgot to put in the list, or am not even aware of myself). It could even be number 4: maybe it's all psychological and related to all the trouble they had to go through to get it to work.
Regarding your valve DAC I would guess that the reasons for liking the sound are 4 and something you failed to mention: people usually like something they have built.
In my book it is a success if he managed to lower the distortions significantly in Khadas Tone Board. That is more than many others have managed.I saw the distortion waveform, but did he find out the actual cause? (mismatch in the multibit dac, HF noise, etc?)
Did he?In my book it is a success if he managed to lower the distortions significantly in Khadas Tone Board. That is more than many others have managed.
Regarding your valve DAC I would guess that the reasons for liking the sound are 4 and something you failed to mention: people usually like something they have built.
Thanks for the addition. I thought it was covered by 4, but maybe it isn't; people may also like something they have built when building it was not expensive, dangerous and/or difficult.
In any case, the point was that it doesn't necessarily have to be 1, I'm glad we agree on that.
Last edited:
I did, but it sort of ended on a cliffhanger. The cap between the inputs of the diff amp looks legit, but they keep mentioning "distorted common mode signals" without ever measuring the actual common mode signalMaybe you should read the ASR thread.
To me it looks more that others chimed in and confused the matter.I did, but it sort of ended on a cliffhanger. The cap between the inputs of the diff amp looks legit, but they keep mentioning "distorted common mode signals" without ever measuring the actual common mode signal
I would think high price is an "expectation bias driver" when buying things. Many DIYers are frugal and actually are drawn to DIY due to falsely expecting to save money.people may also like something they have built when building it was not expensive, dangerous and/or difficult.
It's still under investation and it's clear that there are a bunch of different factors which have so far not been clearly plucked apart.I saw the distortion waveform, but did he find out the actual cause? (mismatch in the multibit dac, HF noise, etc?)
To recap what I did and what I think I do know by now:
- The KTB has visible humps in IMD+N / THD+N vs level plots, and I could confirm the THD+N with my AP.
- To dig in deeper, I used the AP only as signal conditioner, notch and make-up gain and captured the residual output which I could peel out from the noise by sample-synced time-domain averaging (btw, the red lines in the waveforms are the zero-signal residual noise RMS level, for ease of comparision of signal-correlated distortion vs. noise floor, that is, whether it would stick out in the THD+N or not). The corresponding spectra, again with a zero-signal RMS noise marker line, started to make much more sense once I could look at the time signal. What I found particularily interesting was the smoothness and extremely systematic nature of the residual over most of the period. The D10B residual was also very systematic but way rougher (and of course a bit lower in level, generally).
- Looking at the KTB's I/V schematic we find it has little voltage gain (low-valued feedback resistor) and more importanty, a 100Ohms (inherited and AD797-related) "angst" resistor in series with the integrator cap, degenerating it.
- These two choices alone tend to produce more dirt vs audio signal, even with a perfect OpAmp, compared to the solution found in the Topping D10B which has only 10R series and twice the value for the feedback resistor. However, in a quick try, porting the I/V-values to the KTB did not fully work out.
- OTOH, temporarily hacking the D10B and swapping in a MC33078 OpAmp for the LM4562, I found to my surprise that the distortion/artifacts at my low test frequency were basically not impacted. Only at higher frequencies the H2/H3/.. "normal" distortion rises and when putting out very high audio frequencies at full level the 33078's output would slew and have a nasty recovery from this. Thus, a lowly OpAmp does not appear to be the all-dominating factor in that specific environment, and probably in general.
- BTW, I indeed looked at the CM signal, passively summing the KTB's integrator outputs as well as the D10B outputs. Not the exact same picture but even in the D10B there is additional even order distortion in the CM signal.
- Going back to the KTB, the PCB is very unfriendly wrt modding, QFN IC packages, 0402 R/C's, dense as hell... meh. So was looking for a minimal hack that improves things, and the main idea is a shunt cap across the I/V inputs and largely reduced series resistors. Stability is not affected much (checked via SPICE'ing). Still it's a hack and not very extensively tested with a lot of component variations.
- As of today I don't know whether the original OpAmps in the KTB do or don't contribute much to the error. Original OpAmps have been unsoldered by now and OPA1612 are already waiting to be soldered in... together with some R/C changes I hope to get close to D10B levels wrt purity but it's not at all clear that this strategy alone will work out perfectly (and no more than two tries). There are other differences to D10B, the Vref scheme for example and the unknown but likely to be different register settings of the ESS chip.
- Further down the road is a replacement evaluation I/V-board for the D10B in form of a piggy-back where I can freely experiment with the I/V-setup in all parameters and maybe could succeed with robustly lower residual distortion than stock. This will also implement a fully usable single-ended output which is the main motive for the project anyway.
Allo reported getting ES9038Q2M HD down below datasheet spec. IIRC it was around -127.5 dBFS according to their AP. Problem was they found it didn't sound very good that way, so they backed off a little and left it closer to datasheet spec. IIRC they were using a small amount of filtering before the I/V inputs, and supercaps on the power rails including differentially across the rails.
Seems to me @IVX has has also reported very good numbers with selected ES9038Q2M chips, and done so while using minimal supporting circuitry.
OTOH IME supporting circuitry has a significant effect on SQ. For example using opamp buffers for AVCC (as Topping did with D90SE) often sounds 'better' than using an LDO regulator. Does it show up well in HD numbers? Doubt it. Maybe it has something to do with opamp buffers not needing Class II ceramic output caps for stability, don't know.
Seems to me @IVX has has also reported very good numbers with selected ES9038Q2M chips, and done so while using minimal supporting circuitry.
OTOH IME supporting circuitry has a significant effect on SQ. For example using opamp buffers for AVCC (as Topping did with D90SE) often sounds 'better' than using an LDO regulator. Does it show up well in HD numbers? Doubt it. Maybe it has something to do with opamp buffers not needing Class II ceramic output caps for stability, don't know.
Last edited:
It is possible to get really good THD numbers with ES9038Q2M as I have shown here. Usually this requires THD compensation which I do not particularly like as I have said before. But getting THD+N to really low levels is made difficult by the odd noise shaping of the chip (i.e. lowering MCK increases noise).
As the THD compensation is level/FS/channel dependent it is quite possible that it has adverse effect on sound. That is why I prefer AKM chips which can have low THD that is consistent regardless of FS or level.
As the THD compensation is level/FS/channel dependent it is quite possible that it has adverse effect on sound. That is why I prefer AKM chips which can have low THD that is consistent regardless of FS or level.
Thanks for all the details!To recap
If you put the IV on a daughterboard, it's hard to ground it with low inductance to the main board, so all the HF noise and common mode at the output of the dac may be a problem! Recommend filter on main board to avoid turning the mezzanine board into an antenna...
Yep, I think I'm aware of the pitfalls wrt EMC and signal integrity. I do have some background and will try to implement proven best practices as per Ott's book etc. Starting with a 4-layer PCB...Thanks for all the details!
If you put the IV on a daughterboard, it's hard to ground it with low inductance to the main board, so all the HF noise and common mode at the output of the dac may be a problem! Recommend filter on main board to avoid turning the mezzanine board into an antenna...
The main idea is to isolate the I/V's input for RF as much as possible so that the I/V proper doesn't have to handle the nasty part of any glitch energy. That is, ideally I would place shunt caps across the DAC pin outputs and to the very local (Vref) ground. This part I cannot easily place of the piggy-back board for the reasons you've mentioned but there are via holes close to the chip where I might be able to place these 3 caps per channel, I also could claim the now unused PCB area and pads of the original I/V which still would be quite good from a RF standpoint.
On the piggy back the I/V-current inputs will have small inductors and resistors in series (plus I might try larger CM ferrite bead as well), and similar arrangement of cap shunts again, referenced locally. The shunts are basically the input caps of a second order subtracting (differential) lowpass filter with OpAmp. GND plane reference is at the back edge near the outputs, as is filtered +-/rails entry. All low-Z connection crowded at one end of the board, and massive ground planing all around should avoid any unintentional antennas.
I haven't started the PCB design process as I'm still figuring out which detailed circuit options I really need to have some degrees of freedom. For example, allowing the output voltage at the DAC chip pins to be adjusted from (close to) 0V to 3.3V Vref requires some thought...
- Home
- General Interest
- Everything Else
- How we perceive non-linear distortions