(Unofficial) Review of JCALLY JM20 MAX: USB-C Headphone Dongle with High Unbalanced Output Power

[Lambda]

I must say i enjoy this deep dives in quirks and edge cases of a device way more then just comparing SINAD numbers that are (if they are good) basically meaningless.
Its no longer impressive or interesting to me to see a dac with 110dB sinad in "ideal" conditions.

for me its way more interesting to see how, when and if the device under test behaves unexpected.

 

[CedarX]

I must say i enjoy this deep dives in quirks and edge cases of a device way more then just comparing SINAD numbers that are (if they are good) basically meaningless.
Its no longer impressive or interesting to me to see a dac with 110dB sinad in "ideal" conditions.

for me its way more interesting to see how, when and if the device under test behaves unexpected.

What we are still missing is, perhaps, the positive side of these quirks and edge cases: do they provide audible benefits, even if, strictly speaking, it’s no longer transparent?

 

[jkim]

But in post number 8 you yourself showed that there are linearity problems, i.e. the digital and analog gain aren't exatcly matched. Surely it's only a valid method if they are.

That has nothing to do with the validity of the method itself which was the point of the discussion. The 0.1 dB shift is just an inconsequential implementation error. Would you hear the effect of this 0.1 dB shift in linearity?

 

[Palladium]

Yeah keep putting that electron microscope over a $25 gadget while $10000+ stuff gets a free pass because "they must know what they doing".

 

[Lambda]

Yeah keep putting that electron microscope over a $25 gadget while $10000+ stuff gets a free pass because "they must know what they doing".

Why even look at 1000€ stuff if 25€ stuff used right is more then good enough?

Why not look deeper at extensive stuff if a view meaningless number are meaningless high...
My point was how good something works when it works good is relevant (we can get more then good enough for 9€)
How good it performs if it strudels and performs bad is interesting.

 

[nick_l44.1]

I am beginning to believe the clipping is NOT attributed to its DRE scheme

I see errors in your reasoning:
1. “JM20 Max H mode is turned off” - ok. I don't see any reason to doubt it, but it doesn't show a direct correlation. Here is the proof. I have two dongles with bare CS43131 and CS43198. Both exhibit clipping at frequencies below 8Hz and inaudible clicks. The first one shows signs of a working power-saving mode: when the signal drops to -12 dBFS, the current consumption drops immediately, and then after about 5-6 seconds it drops some more. The second dongle shows nothing of the kind.
2, 3. Judging by the strange behavior of your jm20 max, DRE is present in it, but not working. At least it doesn't work correctly. In stepped sine measurements the signal rose and fell linearly, so the response to more complex signal behavior is unknown. If you look at the “backward” curve, up to -50 dBFS the DRE does not work. If you measure the DR under these conditions, you get no more than 121 dBA instead of 127(?) dBA.

8Hz FFT at -19 dBFS

I think this is confirmation that the DRE here is not working correctly, it's likely that something is interfering with it.
That the CS431** are monitoring the signal is beyond doubt.

It turns out that REW (or at least the beta version I used) does not correctly plot the IMD sweep results

Indeed. Release V5.31.3, July 25th, 2024 works correctly. imho it is better to redo graphs with such errors (I saw your warning).

Because I used BW 48 kHz for this measurement to exclude the ultrasonic noise shaping's effect. This is just like what Amir does when a DAC exhibits this issue. Increasing the BW will make the results dominated just by the ultrasonic noise level.

That's the point, these noises affect the sound quality, and ignoring them reduces the objectivity of the measurements. A dense noise hump degrades HF interference even worse than normal harmonic distortion. Look at the effect of ADC (and DAC) Fs when measuring cs43131 SMPTE:
Fs=44.1 kHz IMD SMPTE = -107.3 dB

Fs=96 kHz IMD SMPTE = -101.6 dB

Fs=192 kHz IMD SMPTE = -99.6 dB

Fs=384 kHz IMD SMPTE = -96.9 dB

The filter settings of 20...20000 Hz are the same everywhere, but the smpte degrades by about 10 dB due to the influence of noise entering the digitization bandwidth.
Also look at CS43198 THD+N 10 kHz. The test signal is 24/44100, digitized at 768 kHz. My ADC also has a noise hump at 364 kHz, but it only goes up to -120 dB in this mode at 512k FFT.

I understood that the jm20 max suppresses the DRE for some unknown reason, the behavior below 8Hz is different and getting clipping/clicks to appear is problematic or impossible. Ok. Unlike ultrasonic noise, I don't consider neither clicking nor DRE working or not working as a big problem. Just showing what I suggested you measure obtained for the CS43198. The right side should be similar to jm20 max. The THD calculation was done using the 9 lowest harmonics, signal level -20 dBFS. Notice that the THD results are worse in the audible range because of the low signal level.

 

[Miley_B]

Look at the effect of ADC Fs when measuring SMPTE:

Does it produce the same results when the FFT length is changed in line with the Fs change?

 

[nick_l44.1]

Does it produce the same results when the FFT length is changed in line with the Fs change?

The length of the FFT does not affect the result, except when it is too small.
However, I have now noticed that this effect is not present on my cs43198 dongle. The smpte measurements above were made on CS43131 (meizu hifi), repeated tests of it show similar results with SMPTE dependence on Fs.

 

[jkim]

I see errors in your reasoning:
1. “JM20 Max H mode is turned off” - ok. I don't see any reason to doubt it, but it doesn't show a direct correlation.

What we are doing here is inductive inference (or reverse engineering) based on observations of a few samples. Any such inference can be wrong. Only the designer of the chip should know exactly what elements of it causes the clipping behavior. What I'm saying is that the 8 Hz tone or multitone distortion should not be attributed to the DRE alone. It could be due to H mode transition error or something else (or multiple things combined).

Here is the proof. I have two dongles with bare CS43131 and CS43198. Both exhibit clipping at frequencies below 8Hz and inaudible clicks. The first one shows signs of a working power-saving mode: when the signal drops to -12 dBFS, the current consumption drops immediately, and then after about 5-6 seconds it drops some more. The second dongle shows nothing of the kind.

I believe you as I do not have those dongles. If this is true, it means that the H mode transition alone may not be the reason for the clipping.

2, 3. Judging by the strange behavior of your jm20 max, DRE is present in it, but not working. At least it doesn't work correctly. In stepped sine measurements the signal rose and fell linearly, so the response to more complex signal behavior is unknown. If you look at the “backward” curve, up to -50 dBFS the DRE does not work. If you measure the DR under these conditions, you get no more than 121 dBA instead of 127(?) dBA.

I don't see what point you wanted to get across. What do you mean by "doesn't work correctly"? What is the "correct" operation? The DRE behavior measured on my JM20 MAX is from the designed logic, which also makes sense. DNR measurement with -60 dB signal will always work as intended. Sure, one may say it is cheating as CS set the threshold to -50 dB, though.

I think this is confirmation that the DRE here is not working correctly, it's likely that something is interfering with it.

I'm not sure of your point here, either. How does that FFT show the DRE is not working correctly?

That's the point, these noises affect the sound quality, and ignoring them reduces the objectivity of the measurements. A dense noise hump degrades HF interference even worse than normal harmonic distortion. Look at the effect of ADC (and DAC) Fs when measuring cs43131 SMPTE:
Fs=44.1 kHz IMD SMPTE = -107.3 dB
View attachment 446099
Fs=96 kHz IMD SMPTE = -101.6 dB
View attachment 446100
Fs=192 kHz IMD SMPTE = -99.6 dB
View attachment 446101
Fs=384 kHz IMD SMPTE = -96.9 dB
View attachment 446102
The filter settings of 20...20000 Hz are the same everywhere, but the smpte degrades by about 10 dB due to the influence of noise entering the digitization bandwidth.

This is good to know. But not surprising. And even the worst case does not look terrible.

I understood that the jm20 max suppresses the DRE for some unknown reason, the behavior below 8Hz is different and getting clipping/clicks to appear is problematic or impossible. Ok. Unlike ultrasonic noise, I don't consider neither clicking nor DRE working or not working as a big problem. Just showing what I suggested you measure obtained for the CS43198. The right side should be similar to jm20 max. The THD calculation was done using the 9 lowest harmonics, signal level -20 dBFS. Notice that the THD results are worse in the audible range because of the low signal level.
View attachment 446106

Of course, the increase of THD on the right side should be present on any implementation of a 431xx-based DAC. This rising THD is mainly due to its nonlinear response in the ultrasonic range, as seen in your 10 kHz FFT. THDs of tones above 10 khz are all ultrasonic. The level of this ultrasonic garbage is still well controlled in my view---it won't cause heat or damage to amps or speakers. Will it be audible? Surely not.

 

[nick_l44.1]

What we are doing here is inductive inference (or reverse engineering) based on observations of a few samples. Any such inference can be wrong. Only the designer of the chip should know exactly what elements of it causes the clipping behavior. What I'm saying is that the 8 Hz tone or multitone distortion should not be attributed to the DRE alone. It could be due to H mode transition error or something else (or multiple things combined).

Alas, that is what we are doing. I can't imagine any other possible cause for the 8Hz phenomenon. DRE will not have such consequences only if the chip has the ability to raise gain in advance, buffering a significant fragment of the digital signal.

If this is true, it means that the H mode transition alone may not be the reason for the clipping.

Clipping was observed only at signal levels below -12 dBFS, so the H mode with a threshold of -11 dBFS cannot switch at all under these conditions. And that clipping from H mode can occur only briefly and once in 5.5 seconds I have already written.

I'm not sure of your point here, either. How does that FFT show the DRE is not working correctly?

It's not about FFT, it's about the dependence of the noise level on the signal level. DRE assumes noise reduction, but up to -50 dB this does not happen, i.e. DR does not increase. It is very difficult to call correct operation of the functionality, in which the output characteristics differ greatly when the signal rises and falls.

Will it be audible? Surely not.

This noise itself is of course not audible, but its effect on audible sounds may well be. Amplifiers with high distortion on the HF never sound bright and clear. But I do not propose to go into the preferences of listeners. According to the statistics of CS*** chip distribution, most people are clearly happy with their sound. I just think that measurements should be as objective and exhaustive as possible.

 

[jkim]

Clipping was observed only at signal levels below -12 dBFS, so the H mode with a threshold of -11 dBFS cannot switch at all under these conditions. And that clipping from H mode can occur only briefly and once in 5.5 seconds I have already written.

It may not be about the "transition" to the H mode. It may be due to implementation specifics of the H mode itself. We cannat rule that out.

It's not about FFT, it's about the dependence of the noise level on the signal level. DRE assumes noise reduction, but up to -50 dB this does not happen, i.e. DR does not increase. It is very difficult to call correct operation of the functionality, in which the output characteristics differ greatly when the signal rises and falls.

Have you seen my "Additional Analysis" here? It maintains the "off" state up to -50 dB only if the signal starts above -11 db AND subsequent signal levels are continuously decreasing with no long enough pause. Otherwise it actively adapts the DRE to a new signal level. Basically the algorithm is "aggressive" at reducing the DRE effect when the signal tends to increase in its level, but is very "cautious" in increasing the DRE when the signal level tends to decrease---because the subsequent signal could ramp up abruptly. This makes good sense and is smart.

Anyway, I think all this conjecture-based discussion has reached a point beyond which no productive findings can be made further. Interestingly, some WalkPlay-enabled device built on CS43131 allow access to firmware where the Class H/AB mode can be switched. I plan to test one in the near future. We may have a clearer picture at that time.

 

[viazniki]

the new TRN Black Pearl is killing my JM20 Max get him so cheap from Linsoul for 27£ total 7day delivery took is crazy good on pc and android

 

[jkim]

the new TRN Black Pearl is killing my JM20 Max get him so cheap from Linsoul for 27£ total 7day delivery took is crazy good on pc and android

What a coincidence. I have ordered it, too. Can you quicky check if the WalkPlay app supports change of some DAC parameters like Class AB/H operation?

 

[viazniki]

What a coincidence. I have ordered it, too. Can you quicky check if the WalkPlay app supports change of some DAC parameters like Class AB/H operation?

not have just 8 eq and new firmware just since get him on my pixel 8a similar to the guy who review him in headfi just mine somehow look better show the dac only all same is great dac love him so great sound quality make and cheap

 

[nick_l44.1]

Have you seen my "Additional Analysis" here?

Uh, sure. This all makes the behavior of the device very unpredictable. By the way, can you describe how you perform measurements with decreasing signal strength?

 

[jkim]

Uh, sure. This all makes the behavior of the device very unpredictable.

Why do you want to "predict" its behavior while listening to music or other contents? Sure, it will definitely confuse people like us who want to characterize a device through measurements.

By the way, can you describe how you perform measurements with decreasing signal strength?

Nothing special. Just used REW's stepped RTA.

 

[nick_l44.1]

Why do you want to "predict" its behavior while listening to music or other contents? Sure, it will definitely confuse people like us who want to characterize a device through measurements.

Measurements are exactly what is needed to know the distortion added by the device. In this case, no one can say exactly what the THD+N will be. A difference of up to 6 dB is very large.

Nothing special. Just used REW's stepped RTA.

I can't figure out how you have the stepped sine set to a negative step to generate signals down from 0 dBFS.

 

[JohnPM]

It turns out that REW (or at least the beta version I used) does not correctly plot the IMD sweep results: a constant dB subtraction.

That has been fixed for the next build. Total input rms was being used as the reference for plots instead of the relevant tone level for the chosen IMD measure. If you encounter an error please take a moment to report it e.g. using [email protected].

 

[jkim]

Measurements are exactly what is needed to know the distortion added by the device. In this case, no one can say exactly what the THD+N will be. A difference of up to 6 dB is very large.

Are you suggesting audio industry should not develop technologies that are not amenable for conventional measurement methods and standards? I for one have no problem with that. For example, if each and every DAC starts adopting DRE of some kind at some point in the future, measurement methodology must also evolve to take that into account.

As a methodologist and algorithm developer (not in audio), I can see how the CS engineers did not use simple thresholding for DRE switching. They wanted to make DRE kick in a wide range of signal levels but at the same time prevent clipping. Actually, I would have been thrilled if they had been able to make the DRE effect as large as 12 dB---even by a more complex algorithm---with no adverse side effects!

I can't figure out how you have the stepped sine set to a negative step to generate signals down from 0 dBFS.

Simply enter the Start, End, & Step values. You don't have to worry about the sign of the Step value. REW will take care of that.

 

[jkim]

That has been fixed for the next build. Total input rms was being used as the reference for plots instead of the relevant tone level for the chosen IMD measure. If you encounter an error please take a moment to report it e.g. using [email protected].

Thanks, John. I will surely report any error later on.