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

[nick_l44.1]

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.

I believe that only technologies that are useful are interesting. There is no benefit from DRE, fake dynamic range will not improve sound quality. I agree that clicks are almost impossible to hear, and it is unrealistic to encounter conditions for clipping at 6-8 Hz, but the main question is: for what?
The measurement methodology assumes that thermal noise is constant. What could be the interest for the end listener in the fact that the DAC literally hides this noise from the measurements? At low signal levels this is handled by DRE, at high levels by noise-shaping. The spectrum of music unlike 1 kHz sine is very dense and the noise will not be suppressed, and the dynamic range instead of 130 dBA will be up to 24 dB less (16 times!). However, I must admit that noise-shaping is a very useful technology in other cases.

I can see how the CS engineers did not use simple thresholding for DRE switching

Obviously to hide the use of gain manipulation. And you are right that a high steep step will have too much distortion in the transition.
Well, the DRE level in these chips is exactly 24dB.

You don't have to worry about the sign of the Step value. REW will take care of that.

Oh, thank you! I tried such settings, but the older beta didn't work that way. It worked. Here is the DRE operation in CS43131 when the signal decreases from 0 dBFS. In my case, the dependence is exactly the same as when the signal increases.
Dependence of noise level on signal level, BW96kHz. Measured with Cosmos ADC

So, I have no idea what exactly happens in the jm20 max when the signal goes down. But I doubt very much that it is some algorithm of the DAC chip.

 

[CedarX]

There is no benefit from DRE

I thought the main benefit was to reduce the noise (hiss…) in silence periods or between tracks. Am I wrong here?

 

[jkim]

There is no benefit from DRE, fake dynamic range will not improve sound quality.

I disagree. It is not fake. The main reason why this DRE technique's benefit is unnoticeable in many cases is just because the noise performance of modern devices is already excellent even without DRE.

I agree on what @amirm said when discussing this topic with @staticV3. The following quotes are from here:

AES17 Dynamic Range was designed specifically to counteract mechanisms where DACs would modulate their noise floor depending on the content.

@amirm:
Not so. This was to get around DACs muting with no signal. That is not what is happening here.

Cirrus OTOH just adapted their noise modulation to work with the new measurement stimulus, side-stepping the entire point of it.

@amirm:
Nope. That would be the case if they detected the test and cheated. Them trading digital headroom for analog is a valid method to increase dynamic range. As I explained, HDCD did that also to get to 20 bits of dynamic range in a 16 bit format.

Likewise, as @amirm indicated, the HDCD is not fake. The reason why the effect of HDCD is not appreciated in many cases is because transcoding is not done correctly. Of course, the biggest reason should be that CD's 16-bit resolution is already good enough. There would've been certainly audible benefits if the same technique had been applied to increase the resolution from 12 bits to 16 bits. Similarly, the DRE's audible benefits would be more appreciated if devices without it exhibited an audible noise floor.

In any case---whether its benefit is clearly audible or not---, it is not fake. In fact, I learned that techniques based on essentially the same principle existed even before HDCD, as pointed out by @restorer-john here.

Here is the DRE operation in CS43131 when the signal decreases from 0 dBFS. In my case, the dependence is exactly the same as when the signal increases.
Dependence of noise level on signal level, BW96kHz. Measured with Cosmos ADC
View attachment 446511
So, I have no idea what exactly happens in the jm20 max when the signal goes down. But I doubt very much that it is some algorithm of the DAC chip.

Your results simply show the level change of ultrasonic noise shaping because you used BW 96 kHz and the THD+N calculation span in REW covers the noise shaping region. DRE effects in this case are completely masked by the ultrasonic noise. I reproduced results just like yours when I used your setting. Use BW 48 kHz and THD+N calculation span 20 Hz - 20 kHz. You will have the same results as I did.

 

[jkim]

I thought the main benefit was to reduce the noise (hiss…) in silence periods or between tracks. Am I wrong here?

See my reply to @nick_l44.1 above.

 

[markanini]

Maybe some thought should be given to whether SINAD rankings should weight DRE the same or different. If audible crunchy artifacts happen to be a consequence of DRE it should be a part of the evaluation too.

Other than that, there's no inherent issue with DRE, it's a valid optimization if not implementation sensitive.

 

[jkim]

If audible crunchy artifacts happen to be a consequence of DRE it should be a part of the evaluation too.

Other than that, there's no inherent issue with DRE, it's a valid optimization if not implementation sensitive.

At least the JM20 MAX has no audible or measurable crunchy artifacts. And DRE is present in its measurements, which is part of the reason why I believe the DRE shouldn't be responsible for the crunchy artifacts being reported on CS431xx-based devices.

EDIT. I have found signs of DRE in action from the measurements of CS431xx-based DACs made by @amirm previously.

Because Amir could not find a proper ASIO driver for dongle-type devices for his AP setup, his THD+N sweeps did not show correct results. In fact, the JCALLY JM20 (not the MAX version) was the first device whose THD+N sweep was done correctly:

:

The purple line is what I drew to show the noise level that would've been there if DRE had not been in action.

A byproduct is a tiny linearity error (i.e., error in digital compensation for reduced gain), which is also shown in his measurements:

Amir mistakenly called it a "DC shift" but it's not due to DC bias. The linearity error starts from -12 dB and gradually increases below that level.

Of course, the same thing is observed from his recent measurements of the JM20 MAX:

The SMSL DL100 adopted CS43131:

Actually, the offset starts from -12 dB just like any other devices based on CS431xx.

His measurements of other CS431xx-based devices also showed this tiny linearity error:

In fact, here Amir missed the tiny shift. No perfection, actually The Topping D30 Pro used CS43198.

The Meizu Hifi Audio Pro used CS43131:

Again, a small shift from -12 dB and below.

EDIT 2. By the way, the amount of DRE effects in a THD+N sweep can only be indirectly inferred, because there are other device-specific factors that can affect THD+N results. For example, THD itself, excluding RMS noise, tends to rise as the output level increases.

 

[mc.god]

At least the JM20 MAX has no audible or measurable crunchy artifacts. And DRE is present in its measurements.

EDIT. I have found signs of DRE in action from the measurements of CS431xx-based DACs made previously by @amirm.

Because Amir could not find a proper ASIO driver for dongle-type devices for his AP setup, his THD+N sweeps did not show correct results. In fact, the JCALLY JM20 (not the MAX version) was the first device whose THD+N sweep was done correctly:View attachment 446529:

The purple line is what I drew to show the noise level that would've been there if DRE had not been in action.

A byproduct is a tiny linearity error (i.e., error in digital compensation for reduced gain), which is also shown in his measurements:
View attachment 446530

Amir mistakenly called it a "DC shift" but it's not due to DC bias. The linearity error starts from -12 dB and gradually increases below that level.

Of course, the same thing is observed from his recent measurements of the JM20 MAX:
View attachment 446531

The SMSL DL100 adopted CS43131:
View attachment 446532

Actually, the offset starts from -12 dB just like any other devices based on CS431xx.

His measurements of other CS431xx-based devices also showed this tiny linearity error:

In fact, here Amir missed the tiny shift. No perfection, actually By the way, the Topping D30 Pro used CS43198.

Meizu Hifi Audio Pro used CS43131:

Again, a small shift from -12 dB and below.

That's interesting, so DRE is the origin of that pattern we have seen in many CS43131 devices, that we often called an "early rise in THD+N".

Amir's measurements of Sonata BHD Pro, also not affected by the asio drivers issue, show that same behavior with the 3.5 output but a more linear curve with the 4.4 output

So I'm curious to see what will come out when you will be able to test the TRN Black Pearl

 

[jkim]

That's interesting, so DRE is the origin of that pattern we have seen in many CS43131 devices, that we often called an "early rise in THD+N".

Exactly. The pattern that looked like an early rise in THD+N is mainly due to the DRE being phased out gradually (when the output level increases):

 

[nick_l44.1]

I thought the main benefit was to reduce the noise (hiss…) in silence periods or between tracks. Am I wrong here?

Look here:

The CS43131 reproduces levels very linearly, so there is no audio noise reduction. In pauses, when playback is stopped DRE cannot work and the output signal is muted by other circuits or algorithms.
DRE hides only its own DAC noise.

It is not fake

All right. Let's check what the real dynamic range will be with and without DRE for devices that have DR = 130 dBA according to the standard test. By dynamic range I mean exactly what it is the difference between the loudest and the quietest sound. The loudest would be the test tone, the quietest would be the noise. To keep the reasoning simple, we will consider only thermal noise.

Case 1. No DRE.
At -60 dBFS we have a range of 130-60 = 70 dBA.
As the signal increases, the noise remains the same and by 0 dBFS the range reaches 130 dBA. All is simple.

Case 2. Like CS43131 DRE, but without noise-shaping
At -60 dBFS we have a range of 130-60 = 70 dBA.
By the -36 dBFS level, the range increases to 94 dBA.
That is, up to this point there is no difference from Case 1. Then the most interesting.
Between -36 and -12 dBFS, as you can see on my graph, the noise grows with the signal, their ratio is constant, and the DR also does not change 94 dBA.
Above -12 dBFS the DRE is no longer active and the real range increases to 106 dBA.
So, case2 130 dBA is a fake.

I suppose if a fact is objectively discovered and there are no errors in measurement, no authority can refute it.

However, you don't see such high noise in the THD+N measurement thanks not to DRE (it doesn't work at 0 dBFS at all) but to noise-shaping, which shifts the noise energy above 40 kHz. We get nice measurements at 1kHz, ugly at 10kHz and noise reduction in the region up to 40kHz if the spectrum is sparse. I would prefer to disable shaping and have unremarkable but moderate measurements at different frequencies.

DRE effects in this case are completely masked by the ultrasonic noise. I reproduced results just like yours when I used your setting. Use BW 48 kHz and THD+N calculation span 20 Hz - 20 kHz. You will have the same results as I did.

The DRE by definition has no effects at 0 dBFS. It only works at low levels. In other words, there is one situation when 0 dFBS signal comes to the output and another when signals up to -130 dB can appear above the noise. But this is not possible at the same time. This is an orchestra in which quiet and loud instruments play strictly at different times.
I understand very well where the noise energy is located. This measurement does not determine the noise level in the audible range, it is needed to determine the logic of DRE operation.
You are suggesting that we turn a blind eye to what happens in ultrasound. Will this make the measurements more objective? When measuring THD+N@1 kHz - yes, but I gave above THD+N@10 kHz 24 bit 44100 Hz. Noise and distortion going up the hump like people up a hill is real. By limiting the measurement range to 20 kHz you get THD exactly 50 times better. Would it be true that by measuring distortion you are discarding the highest distortion? I don't think so.

 

[jkim]

Look here:
View attachment 446560
The CS43131 reproduces levels very linearly, so there is no audio noise reduction. In pauses, when playback is stopped DRE cannot work and the output signal is muted by other circuits or algorithms.
DRE hides only its own DAC noise.

All right. Let's check what the real dynamic range will be with and without DRE for devices that have DR = 130 dBA according to the standard test. By dynamic range I mean exactly what it is the difference between the loudest and the quietest sound. The loudest would be the test tone, the quietest would be the noise. To keep the reasoning simple, we will consider only thermal noise.

Case 1. No DRE.
At -60 dBFS we have a range of 130-60 = 70 dBA.
As the signal increases, the noise remains the same and by 0 dBFS the range reaches 130 dBA. All is simple.

Case 2. Like CS43131 DRE, but without noise-shaping
At -60 dBFS we have a range of 130-60 = 70 dBA.
By the -36 dBFS level, the range increases to 94 dBA.
That is, up to this point there is no difference from Case 1. Then the most interesting.
Between -36 and -12 dBFS, as you can see on my graph, the noise grows with the signal, their ratio is constant, and the DR also does not change 94 dBA.
Above -12 dBFS the DRE is no longer active and the real range increases to 106 dBA.
So, case2 130 dBA is a fake.

I suppose if a fact is objectively discovered and there are no errors in measurement, no authority can refute it.

However, you don't see such high noise in the THD+N measurement thanks not to DRE (it doesn't work at 0 dBFS at all) but to noise-shaping, which shifts the noise energy above 40 kHz. We get nice measurements at 1kHz, ugly at 10kHz and noise reduction in the region up to 40kHz if the spectrum is sparse. I would prefer to disable shaping and have unremarkable but moderate measurements at different frequencies.


The DRE by definition has no effects at 0 dBFS. It only works at low levels. In other words, there is one situation when 0 dFBS signal comes to the output and another when signals up to -130 dB can appear above the noise. But this is not possible at the same time. This is an orchestra in which quiet and loud instruments play strictly at different times.
I understand very well where the noise energy is located. This measurement does not determine the noise level in the audible range, it is needed to determine the logic of DRE operation.
You are suggesting that we turn a blind eye to what happens in ultrasound. Will this make the measurements more objective? When measuring THD+N@1 kHz - yes, but I gave above THD+N@10 kHz 24 bit 44100 Hz. Noise and distortion going up the hump like people up a hill is real. By limiting the measurement range to 20 kHz you get THD exactly 50 times better. Would it be true that by measuring distortion you are discarding the highest distortion? I don't think so.

On reading this post, I'm confident that this discussion reached a point where we have expressed our clearly different viewpoints on this topic. How about we let other members express theirs, if anyone is still reading our posts?

 

[multiformous]

I just hope the JM20 Max works in exclusive mode in UAPP on my old Android device. My Tanchjim Space did, but not my Meizu HiFi (non-pro) for some reason. JM12 with JA11 firmware does, as well as Apple and Samsung dongles; but not the CX31993 dongles I've tried. Dunno why.

 

[CedarX]

About the CS431xxx DRE, would adding a constant, 0 dBFS, ultrasonic signal—a little bit like IMD measurements—force DRE, or equivalent techniques, to not impact the DAC measurements?

I think it’s not so much about denying these “trick technologies”, when they have an actual positive impact, but more about understanding what they do, and how much they may distort traditional measurement results.

 

[markanini]

I get can being vary about DRE, it is requires analog circuits to play along, even as solid-state ICs they are susceptible to tolerances and capable of various non-ideal behavior. On the other hand, a well-designed DRE implementation considers such limitations carefully*. While DRE can't be perfect it can be deemed effective and practical for discerning use-until someone present a specific issue tied to DRE.

*Optimizations driven by reducing power consumption also apply here

 

[jkim]

About the CS431xxx DRE, would adding a constant, 0 dBFS, ultrasonic signal—a little bit like IMD measurements—force DRE, or equivalent techniques, to not impact the DAC measurements?

This can be one way to defeat DRE in order to measure the AES17 DR with a -60 dB sine tone. The ultrasonic signal, say a 30 kHz sinusoid, doesn't have to be at 0 dbFS. A -6 dB tone should work.

I think it’s not so much about denying these “trick technologies”, when they have an actual positive impact, but more about understanding what they do, and how much they may distort traditional measurement results.

I get can being vary about DRE, it is requires analog circuits to play along, even as solid-state ICs they are susceptible to tolerances and capable of various non-ideal behavior. On the other hand, a well-designed DRE implementation considers such limitations carefully. While DRE can't be perfect it can be deemed effective and practical for discerning use-until someone discovers a specific issue tied to DRE.

Quite a chunk of debate on this topic comes from the definition of Dynamic Range and its AES17 measurement procedure. Looking at this topic in better perspective now, I can say that the name we have given to this technique, "Dynamic Range Enhancement" is a misnomer, because DR is defined as the ratio of the loudest, clean signal to the noise floor of a device. If you want to be faithful to this definition, the result of DR measurement of CS431xx using the AES17 procedure (implemented in AP as well) is considered not valid because DRE in the chip tricks the method.

But as you guys agreed, the effect of DRE is real, and if implemented correctly with no audible, adverse side effect, should be beneficial.

So, from now on, I suggest calling this technique "Adaptive Signal-to-Noise Ratio Enhancement" or ASNRE.

 

[TNT]

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?

Yes.

//

 

[DrZingo]

Yes.

//

I suspect so too - on a several seconds long reverb tail from a church recording for example. Actually, since sounds typically rise quickly and fade out slowly, you would probably want at least as gentle a fade-in of this noise reduction effect, rather than the sudden "full on" when the signal drops below a certain threshold.
Perhaps the above sweeps should also always be played in reverse, from loud to soft? That would better show the discontinuities.

 

[nick_l44.1]

About the CS431xxx DRE, would adding a constant, 0 dBFS, ultrasonic signal—a little bit like IMD measurements—force DRE, or equivalent techniques, to not impact the DAC measurements?

You're right, it will have an effect. But we still won't get accurate data, because besides DRE there is also noise-shaping, which has different settings for different signal levels. Here is my post on diyaudio, where there is the measurement you suggested.

it is requires analog circuits to play along

The DRE can use a DAC modulator, its gain is digitally controlled.

if implemented correctly with no audible, adverse side effect, should be beneficial

I can only see one relatively useful effect of the DRE. When using very sensitive headphones, you need to lower the digital volume a lot, while reducing the dynamic range. In this case the DRE can partially save the situation by acting as a gain switch. If the DRE turned into volume control, that could be interesting.
But what other profits can this technology offer?

 

[jkim]

Yes.

//

I suspect so too - on a several seconds long reverb tail from a church recording for example. Actually, since sounds typically rise quickly and fade out slowly, you would probably want at least as gentle a fade-in of this noise reduction effect, rather than the sudden "full on" when the signal drops below a certain threshold.
Perhaps the above sweeps should also always be played in reverse, from loud to soft? That would better show the discontinuities.

You guys talk about two different things. The 0.1 dB error in linearity and the 6 dB noise floor reduction are related as they are due to this noise reduction technique, but are two different things. I doubt anyone can hear the effect of a 0.1 dB linearity error especially at such a low level.

But one MAY hear the effect of abrupt 6 dB noise reduction at the -50 dB source level IF it occurs at a level of audible hiss---for example, when extremely (and unrealistically) high amplifier gain is applied.

 

[jkim]

I can only see one relatively useful effect of the DRE. When using very sensitive headphones, you need to lower the digital volume a lot, while reducing the dynamic range. In this case the DRE can partially save the situation by acting as a gain switch. If the DRE turned into volume control, that could be interesting.

This benefit can be huge since quite a few enthusiasts are into sensitive IEMs. Some non-planar headphones are quite sensitive, too. For example, I use Sony MDR-MA900 (12 ohm rated) with the JCally JM20 Max. Even with a -5 dB EQ preamp cut, my Windows volume slider is at 15% for some loud recordings. Also my modded Sennheiser PX100 II is very sensitive (even with a -8 dB preamp cut) and I mostly use 15% to 20% on my Windows volume slider.

Some people are obsessed with a transition hiss from sensitive IEMs. Perhaps, @HissingFree may chime in?

EDIT. I measured the signal-to-noise ratios of the JM20 MAX and E1DA 9039S in stepped sine tone tests:

Note. Signal = 1 kHz sinusoid, Load impedance = 32 Ohms, Fs = 96 kHz, Data span = 20 Hz - 20 kHz.

As expected, the 9039S has overall better SNR. As far as I know, the 9039S is currently the best DAC/amp combo device in regard to noise performance (with respect to THD as well). It even slightly beats my best stack of separate DAC & amp (Topping D50 III & A70 Pro in low gain mode).

The ASNRE ("Adaptive SNR Enhancement") in CS43131 kicks in at about 0.6 Vrms and lowers its noise floor down from there. I believe CS engineers carefully chose these transition levels so it can be effective in practical applications. Below 0.2 Vrms, its noise level is neck and neck with 9039S's.

Adding THD shows essentially the same picture:

 

[JanL]

I received the J20 Max today. Very good indeed for the fairly low price of $30.00 from Crinacle's shop. Driving Audio Technica ATH-R50X with it and the sound is very dynamic and effortless. No hint of any noise pumping at any volume. Whatever the noise reduction technology is, it's very good. Compared to the Realtek headphone DAC that's built in the laptop I'm using, it's a totally different experience.
BTW, it was simply dropped off at my door. No tariffs on this drop shipment from China.