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

[Eddie]

New adapter in house, so here we go:

View attachment 445537

I wonder why 300 Ohm (green) has better/lower/less crosstalk than 600 Ohm (blue).
I've learned that the lower the load/impedance, the higher the crosstalk (due to higher current).
Is there perhaps a confusion in the legend?

In any case, I find the values quite disappointing. At 300 Ohm to 600 Ohm (Sennheiser HD 650 or HD 800s) I would have preferred at least -100dB.
Or do the values depend on the measurement method?

 

[Eddie]

The small dents in the sine are not audible, they are of very high frequency (see spectrogram, 90 kHz etc).

I loaded the file "g6-7.5.flac" into iZotope RX. This is an excerpt from the spectrogram (window type: Hann, FFT size: 2048, frequency scale: Melody):

It can be seen that the frequency content of these distortions (as I like to call them) already starts at around 5kHz. This is in the audio band. They are very short in duration, maybe 25 samples (at most), which at 192kHz corresponds to about 0.13ms.
Is this audible to any human? I don't know.
But I think the point here is that these distortions just shouldn't be there in the first place, audible or not.

 

[DrZingo]

Because it's performance trickery i.e. not real figures. Not that it matters SQ wise (probably..) but I hate to be deliberately fooled...

//

The device is designed to measure exceptionally well on static, standardized tests (similar to how Volkswagen a few years back made their cars measure exceptionally well on static, standardized emissions tests). Like you I'm suspicious of features like these, we probably need to add some more sophisticated, dynamic tests when evaluating DACs.

 

[MC_RME]

I wonder why 300 Ohm (green) has better/lower/less crosstalk than 600 Ohm (blue).

Good catch. I did not perform the measurements in the 'intuitive' order, and the AP software does not allow any reordering. So the manually given labels in the right legend were in wrong order - my fault. I uploaded the corrected version.

 

[jkim]

I did some additional analysis of the DAC chip's dynamic range adaptation behavior.

Here's a sneak peek:

See my description in this post where I originally posted this topic (see "Additional Analysis" there).

 

[TNT]

The device is designed to measure exceptionally well on static, standardized tests (similar to how Volkswagen a few years back made their cars measure exceptionally well on static, standardized emissions tests). Like you I'm suspicious of features like these, we probably need to add some more sophisticated, dynamic tests when evaluating DACs.

Exactly - obviously, tests can be fooled - @amirm ?

 

[jkim]

Exactly - obviously, tests can be fooled - @amirm ?

@amirm already expressed his opinion on this technique. And I agree with him---it is a valid method. See my post here on this topic.

 

[Billy Budapest]

@amirm already expressed his opinion on this technique. And I agree with him---it is a valid method. See my post here on this topic.

As stated above, I think the technology is not only valid, it’s ingenious.

There is an ESS white paper describing a related phenomenon—when to use digital volume control as implemented in the ESS DACs and at what point to switch over to analog.

Benchmark, too, employs hybrid digital/analog gain control.

It’s really not a controversial method. It’s just creative engineering.

 

[nick_l44.1]

I am aware of that post. I don't think anything is conclusive from there.

EDIT. Here is what I know. The JM20 Max does not seem to produce this clicking noise. It also does not seem to enter power saving mode as other CS431xx devices do. Lastly, it does enter the dynamic range enhancement mode.

Apparently you didn't read my post carefully.

After the signal drops below -11 dB, it takes about 5.5 seconds before the amplifier power down is activated. Until then, no clipping can occur due to the H mode being activated. Therefore, there can be no more than 3 clipping events during the 10 sec example, not dozens.
The DRE/DRC activation delay is much shorter, not more than 1/16 sec. This can be seen from the clipping of the 8 Hz -15 dBFS sine wave. And, unlike H mode, the increase in dynamic range is due to the reduction of gain at low signal levels, not supply voltage. Then at any increase in signal, even as gentle as at 8 Hz, clipping occurs due to insufficient maximum voltage.
By the way, I was not able to repeat the different behavior of noise level at increasing and decreasing signal levels on bare CS43*** that you noticed.

 

[jkim]

Apparently you didn't read my post carefully.
View attachment 445812
After the signal drops below -11 dB, it takes about 5.5 seconds before the amplifier power down is activated. Until then, no clipping can occur due to the H mode being activated. Therefore, there can be no more than 3 clipping events during the 10 sec example, not dozens.
The DRE/DRC activation delay is much shorter, not more than 1/16 sec. This can be seen from the clipping of the 8 Hz -15 dBFS sine wave. And, unlike H mode, the increase in dynamic range is due to the reduction of gain at low signal levels, not supply voltage. Then at any increase in signal, even as gentle as at 8 Hz, clipping occurs due to insufficient maximum voltage.
By the way, I was not able to repeat the different behavior of noise level at increasing and decreasing signal levels on bare CS43*** that you noticed.

So, basically what you say is that the clipping ("clicking" or "crackle' whatever we call it) occurs not due to the H mode (or power-down or power-saving mode) activation, but due to the gain change scheme of the CS chip. Right?

And this gain change should be part of the dynamic range enhancement (DRE). So, essentially it is a side effect of their DRE implementation. And the Russian article attributing the clipping behavior to the H mode activation is not accurate. Right?

By the way, I was able to reproduce your wideband (BW 96 kHz) noise vs. output level. That pattern is mainly due to the level change of ultrasonic noise shaping. The noise vs. output level that I observed reflected just 20 Hz to 20 kHz, which is a separate phenomenon showing the DRE effect. These two phenomena may be related in some way, though.

Lastly, the multitone signal used in the Russian article to show the clipping behavior does not seem to include a tone below 20 Hz. I guess the key is how shortly a signal is presented after the gain is adjusted. OR we may be talking about two different kinds of clipping caused by separate things?

 

[MC_RME]

In any case, I find the values quite disappointing. At 300 Ohm to 600 Ohm (Sennheiser HD 650 or HD 800s) I would have preferred at least -100dB.
Or do the values depend on the measurement method?

I am not aware of any device reaching such values at an unbalanced TRS output. Plus these values need some perspective - you won't be able to hear any difference from 30 dB up. It is more about proper design and build. For me, reaching stable >70 dB at 32 Ohms and up is nothing to complain about.

 

[nick_l44.1]

So, basically what you say is that the clipping ("clicking" or "crackle' whatever we call it) occurs not due to the H mode (or power-down or power-saving mode) activation, but due to the gain change scheme of the CS chip. Right?

And this gain change should be part of the dynamic range enhancement (DRE). So, essentially it is a side effect of their DRE implementation. And the Russian article attributing the clipping behavior to the H mode activation is not accurate. Right?

By the way, I was able to reproduce your wideband (BW 96 kHz) noise vs. output level. That pattern is mainly due to the level change of ultrasonic noise shaping. The noise vs. output level that I observed reflected just 20 Hz to 20 kHz, which is a separate phenomenon showing the DRE effect. These two phenomena may be related in some way, though.

Lastly, the multitone signal used in the Russian article to show the clipping behavior does not seem to include a tone below 20 Hz. I guess the key is how shortly a signal is presented after the gain is adjusted. OR we may be talking about two different kinds of clipping caused by separate things?

Yes and yes. In any case you can point out logical mistakes to me.
I don't understand what a multitone can show either. It is a periodic signal whose rms cannot cross the -11 dB level with a period of more than 5.5 sec.
Varying the gain allows a high DR result, but the THD+N result on this background will be very ugly, approx. -100 dB. In addition, a 24dB noise floor offset is sure to raise a lot of questions. So noise-shaping masks the noise increase. I don't see any other logic.
Peak values are key, not the presence of frequencies below 8 Hz. After lowering the gain, almost any signal causes a short clipping, the gain quickly adapts and the problem disappears. I think the clipping when H mode is activated should be similar, but may not occur as often.

 

[nick_l44.1]

SMPTE IMD vs. output voltage under 300 Ohm load:

Again I see unrealistically low SMPTEs. Intermodulation is one of the products of harmonic distortion. So it is very hard to imagine SMPTE -120 dB, while you get THD+N = -113 dB at 1 kHz, and at 7 kHz my CS43131 shows THD -108 dB. Can you please describe your measurement setup and show a screenshot of the FFT c distortion settings and fft settings?

THD+N vs. frequency is excellent, too

Why do you limit measurement by 10 kHz instead of 20 kHz? And I find measuring THD+N vs Frequency over a wide bandwidth uninteresting, since the same total noise for all signal frequencies hides so many important details.

No issue was found

In fact, you've already found a few. Try measuring THD (without noise) vs frequency from 6 Hz to 20 kHz at -15 dBFS with Fs = 192 kHz. I'm even curious to see what you get with the detected “reverse sweep” effect.

 

[Eddie]

I am not aware of any device reaching such values at an unbalanced TRS output. Plus these values need some perspective - you won't be able to hear any difference from 30 dB up. It is more about proper design and build. For me, reaching stable >70 dB at 32 Ohms and up is nothing to complain about.

OK, I understand, but I'm also a little confused.
For example, the Topping DX7 Pro+ specifies -120dB at 1kHz, but doesn't specify the actual load.
Does it generally mean that such values without a load specification implicitly mean no load? If so, then such values are useless in practice, as they are likely to be significantly worse for real headphones (even with high impedance). And I feel fooled.

 

[Eddie]

Its low-pass digital filter works great:

I see it differently. A great filter should, among other things, be "down" at half the sample rate to effectively suppress aliased images.
In this case, 22.05kHz instead of 24kHz.
Unfortunately, all four CS43131 filters (not counting the NOS "filter") behave the same in this regard.
I think AKM and ESS offer "great" (or rather, correct) filters to choose from.
I don't understand why incorrect filters are offered at all. Perhaps because they are much easier (cheaper) to implement?

 

[MC_RME]

The answer has been mentioned/explained here several times with nice graphics, but I coudn't find it when searching. 24 kHz for 44.1 kHz sample rate was chosen because it relaxes the filter and phase deviation around 20 kHz, while at the same time anything mirroring back (if at all) will never be below 20 kHz. This is an 'industry standard' that in real-world just works.

 

[jkim]

I see it differently. A great filter should, among other things, be "down" at half the sample rate to effectively suppress aliased images.
In this case, 22.05kHz instead of 24kHz.
Unfortunately, all four CS43131 filters (not counting the NOS "filter") behave the same in this regard.
I think AKM and ESS offer "great" (or rather, correct) filters to choose from.
I don't understand why incorrect filters are offered at all. Perhaps because they are much easier (cheaper) to implement?

The answer has been mentioned/explained here several times with nice graphics, but I coudn't find it when searching. 24 kHz for 44.1 kHz sample rate was chosen because it relaxes the filter and phase deviation around 20 kHz, while at the same time anything mirroring back (if at all) will never be below 20 kHz. This is an 'industry standard' that in real-world just works.

I am sure this basic topic must have been discussed before somewhere in the forum. Although other DAC chips offer that earlier-roll-off filter as a choice, most DAC products choose the one that fully suppresses the signal from 24 kHz as a default.

 

[jkim]

Yes and yes. In any case you can point out logical mistakes to me.
I don't understand what a multitone can show either. It is a periodic signal whose rms cannot cross the -11 dB level with a period of more than 5.5 sec.
Varying the gain allows a high DR result, but the THD+N result on this background will be very ugly, approx. -100 dB. In addition, a 24dB noise floor offset is sure to raise a lot of questions. So noise-shaping masks the noise increase. I don't see any other logic.
Peak values are key, not the presence of frequencies below 8 Hz. After lowering the gain, almost any signal causes a short clipping, the gain quickly adapts and the problem disappears. I think the clipping when H mode is activated should be similar, but may not occur as often.

I believe I understand your logic. But based on my tests of the JM20 MAX---the only CS431xx-based DAC I have---, I am beginning to believe the clipping is NOT attributed to its DRE scheme. Here are some points to consider:

1. @MC_RME measured the JM20 MAX's power consumption. He conjectured that unlike other CS431xx-based devices, the CS chip in the JM20 MAX seems to be set not to enter the power-saving mode. The CS431xx datasheet calls this the Standard Class AB Operation mode (ADPT_PWR = 001 or 010), hence not entering the Adapt-to-Output Signal mode (ADPT_PWR = 111 or power-saving state).
2. In my extensive tests, the JM20 MAX does not exhibit any clipping (or crunchy) behavior. @MC_RME also tested it extensively and reported he has not come across any such thing so far.
3. According to my tests, the JM20 MAX clearly shows the DRE effect.

I ran THD sweeps of 8 Hz tones in both the directions of increasing and decreasing levels:

As you can see, there's no anomaly here. Also, see this 8Hz FFT at -19 dBFS:

Nothing wrong again.

Here is my informed guess. The CS431xx, in the Adapt-to-Output Signal mode, must constantly monitor signal levels to determine an optimal set of rail voltages (to maximize power saving and prevent clipping at the same time). See here:

It should use some kind of logic to identify a signal level. My hunch is that some multitone or out-of-band (like a 8 Hz tone) signal may trick this logic and cause the chip to choose the lower set of rail voltages than appropriate. Some rarely occurring music/audio contents (like the sample used in the Russian website) may do the same thing.

 

[jkim]

Again I see unrealistically low SMPTEs. Intermodulation is one of the products of harmonic distortion. So it is very hard to imagine SMPTE -120 dB, while you get THD+N = -113 dB at 1 kHz, and at 7 kHz my CS43131 shows THD -108 dB. Can you please describe your measurement setup and show a screenshot of the FFT c distortion settings and fft settings?

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. Here is an FFT result of SMPTE tones from my Topping D50 III:

The IMD level is -119 dB, but the sweep plot shows it as -132 dB: a 13 dB difference. But for the purpose of comparing devices, the plot works fine.

Why do you limit measurement by 10 kHz instead of 20 kHz? And I find measuring THD+N vs Frequency over a wide bandwidth uninteresting, since the same total noise for all signal frequencies hides so many important details.

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.

In fact, you've already found a few. Try measuring THD (without noise) vs frequency from 6 Hz to 20 kHz at -15 dBFS with Fs = 192 kHz. I'm even curious to see what you get with the detected “reverse sweep” effect.

I have done quite a few tests including what you suggested above. No clipping was found from the JM20 MAX. See my previous post.

 

[DrZingo]

@amirm already expressed his opinion on this technique. And I agree with him---it is a valid method. See my post here on this topic.

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.