Review of TRN Black Pearl: Portable USB DAC & Headphone Amp with 10-band PEQ

[staticV3]

So I'm guessing that's the main point of comparison for that graph?

Yes, to show how frequently different CS43131 devices trigger DRE, which results in those clicks (->lines in the spectrogram)

 

[Matias]

This review is insanely well done. Huge respect, @jkim.

 

[jkim]

So I'm guessing that's the main point of comparison for that graph?

Right, @staticV3 said it well. The reason why the spectrogram shows the ultrasonic range is simply because otherwise we cannot see the distortion as it is masked by audio content in the human hearing range. In fact, most of the time, the distortion is masked not just in the graph but also in actual hearing. It is only noticeable when there's not much high-frequency content in the signal, like in the Dune soundtrack clip.

 

[LDvo]

The comment made by @HissingFree is well said. In my view, the kind of test that needs to be added to a conventional suite of tests is an analysis of dynamic, not steady-state, signals over time. A spectrogram analysis of a level-changing multitone signal, shown in my review, is such an example. Although this analysis was used to show the DRE artifacts, I believe it makes sense to include such a test in a standard 'battery of tests' for the purpose of checking any potential issue of a similar kind.

For example, I created a test signal made of 11 tones concentrated in the bass region, whose FFT looks like:
View attachment 465900

Then added fade-in and out effects (-50 dB through -3 dB) over time:
View attachment 465901

Here's a spectrogram of this signal played by the Black Pearl w/ its Fast-LL filter (i.e., DRE enabled):
View attachment 465904
DRE-induced distortion is nasty.

Of course, distortion is gone in NOS mode of the Black Pearl:
View attachment 465905

But do you know what is surprising? According to this spreadsheet, some CS431xx-based devices exhibit limited DRE artifacts, which I suspect to be attributed to an adjustment made by those DAC manufactures on some DRE-related parameter in the chip (likely 'DRE decay rate'). For example, DRE artifacts of the following devices

• Fosi Audio DS2
• JCally JM20
• JCally JM20-Max
• A few more devices

cannot be detected by steady-state multitone tests. In fact, the C Major test signal is an ideal test signal. For instance, note that the above test signal simply keeps increasing its level and then keeps decreasing it; that is, it's just one cycle. In response to this signal, these devices behave surprisingly nicely. For instance, the spectrogram of the signal played by the JCally JM20:
View attachment 465906
shows nearly no distortion (yes, nearly; we see a faint sign of it by zooming the plot).

Lastly, see below spectrograms of some music played by the Black Pearl with the Fast-LL filter (DRE enabled) in the upper panel and JCally JM20 in the lower panel:View attachment 465907
Note that I just picked a random song (i.e., did not choose this particular song to show large distortion effects). I did not choose this song to tell something about the distortion's audibility, either. Just to compare the two devices' different behavior.

As you can see, the distortion is much less frequent for the JM20. I bet this would not be audible since in most cases it should be masked whenever it overlaps with a wide frequency spectrum in audio content.

In fact, as suggested by the measurements in the "Other Characteristics" section of my review, the reason why the JM20, or another device showing the same behavior, does not distort is simply because DRE is not fully active much of the time! Whenever the signal level reaches a strong level, DRE becom es inactive, or does not increase its effect. It simply waits, or just wants to decrease DRE gain if the signal gets stronger. Only after the signal goes below -50 dBFS, it turns the DRE back on. This is very passive DRE.

I would call this a good, albeit minimal, DRE implementation. Then, the question is, did the designers of these devices deliberately adjust this DRE parameter for this effect? I absolutely have no idea.

see below spectrograms of some music played by the Black Pearl with the Fast-LL filter (DRE enabled) in the upper panel and JCally JM20 in the lower panel

I could be wrong here - but don't they look the same on both devices?

 

[staticV3]

I could be wrong here - but don't they look the same on both devices?

Look at the number of vertical lines extending above 20kHz - those are audible clicks resulting from DRE.

 

[LDvo]

Look at the number of vertical lines extending above 20kHz - those are audible clicks resulting from DRE.

Thanks! I didn't look above 20K as it is beyond audible spectrum
- I will try to reproduce this on my device if i can hear this distortion.

 

[staticV3]

Thanks! I didn't look above 20K as it is beyond audible spectrum
- I will try to reproduce this on my device if i can hear this distortion.

Please read from #58 onward.

 

[LDvo]

Please read from #58 onward.

Got it now. Thanks again.

 

[MaxwellsEq]

Thanks! I didn't look above 20K as it is beyond audible spectrum

Clicks (impulses) have all frequencies from 0Hz to above 20kHz in them. Where you see the line going above 20kHz, it's also going down below 20kHz towards 0Hz. Hard to pick out in the music. This is because it's now become part of the music.

But the key thing to grasp is it should not be there and is messing up the sound in the audio frequencies.

 

[LDvo]

Clicks (impulses) have all frequencies from 0Hz to above 20kHz in them. Where you see the line going above 20kHz, it's also going down below 20kHz towards 0Hz. Hard to pick out in the music. This is because it's now become part of the music.

But the key thing to grasp is it should not be there and is messing up the sound in the audio frequencies.

Aha, i see - that is why i could not pick those distortion impulses before. Thanks for explanation!

 

[Robbo99999]

Right, @staticV3 said it well. The reason why the spectrogram shows the ultrasonic range is simply because otherwise we cannot see the distortion as it is masked by audio content in the human hearing range. In fact, most of the time, the distortion is masked not just in the graph but also in actual hearing. It is only noticeable when there's not much high-frequency content in the signal, like in the Dune soundtrack clip.

Is there any indication how loud the distortion is during those clicks? I guess in dBFS.

 

[lewdish]

What is the purpose of these companies telling us about their USB Bridges?

 

[jkim]

Is there any indication how loud the distortion is during those clicks? I guess in dBFS.

Good question. Will perform some analysis when I get a chance.

What is the purpose of these companies telling us about their USB Bridges?

I guess in most cases they want to say USB connection and data transmittance is handled by a reliable brand SoC (e.g., XMOS, Savitech, Comtrue)?

 

[pangcukaipang]

It's weird. Last week, I finally able to listen to the Dune soundtrack normally, but today the distortion happened again. Haven't changed anything. Setting on my Windows 11 is 32 bit/384 kHz and on my Walkplay is Class AB, High gain, NOS.

Do you have a step by step workaround?

 

[Jean-Paul Petrosyan]

The ADPT_PWR or Class H mode is a different feature from DRE. I initially thought the chip's Class H operation would be the culprit for the distortion, just like Roman at RAA did. But it is clear now that DRE is the reason behind the distortion.



32-tone PCM signal (-24 dBFS) created at 24-bit/96kHz, realtime converted to DSD and played by Neutron MP through the Black Pearl's DSD mode:
View attachment 465620

Notice DRE-induced distortion in the lower frequency region as well as higher-frequency distortion/noise seemingly from PCM-to-DSD conversion.

Same 32-tone PCM signal @ 24-bit/96kHz played by Neutron MP with 4x oversampling through the Black Pearl with Fast-LL filter:
View attachment 465625
As expected, with DRE enabled, there's Cirrus-hump distortion.

Same 32-tone PCM signal @ 24-bit/96kHz played by Neutron MP with 4x oversampling through the Black Pearl in NOS mode:
View attachment 465624
Free of distortion, clean and normal.

Now, I will show you very interesting data.

Periodic white noise at Fs 384kHz (BW 192kHz) into Black Pearl in NOS mode:
View attachment 465622

Same periodic white noise at Fs 384kHz (BW 192kHz) into Black Pearl with Fast-LL filter:
View attachment 465623

What do you see? Yes, absolutely zero difference! Also look at the roll-off. It is NOT a fast roll-off, meaning that the CS43131 does not apply reconstruction filtering when the input signal is at Fs 384 kHz. I checked the other filter options, too, and found they have no effect at all. So, feeding PCM audio oversampled to 384 kHz into the Black Pearl in NOS mode essentially disables just DRE and makes no other difference in reconstruction filtering, because the DAC's internal filter is not active in this case.

Does this answer your question?

Thank you very much for such a detailed response!

>CS43131 does not apply reconstruction filtering when the input signal is at Fs 384 kHz. I checked the other filter options, too, and found they have no effect at all. So, feeding PCM audio oversampled to 384 kHz into the Black Pearl in NOS mode essentially disables just DRE and makes no other difference in reconstruction filtering, because the DAC's internal filter is not active in this case.

Is this a good thing? Are there any caveats to this?

Also, have you done anything measurement with Class H vs Class AB engaged to see what difference does that make, by any chance? ADPT_PWR being in should produce some added distortion by Cirrus Logic's own admission, after all.

 

[Matias]

I am also curious of in depth measurements between class H and AB of the mighty CS43131.

 

[Jean-Paul Petrosyan]

The comment made by @HissingFree is well said. In my view, the kind of test that needs to be added to a conventional suite of tests is an analysis of dynamic, not steady-state, signals over time. A spectrogram analysis of a level-changing multitone signal, shown in my review, is such an example. Although this analysis was used to show the DRE artifacts, I believe it makes sense to include such a test in a standard 'battery of tests' for the purpose of checking any potential issue of a similar kind.

For example, I created a test signal made of 11 tones concentrated in the bass region, whose FFT looks like:
View attachment 465900

Then added fade-in and out effects (-50 dB through -3 dB) over time:
View attachment 465901

Here's a spectrogram of this signal played by the Black Pearl w/ its Fast-LL filter (i.e., DRE enabled):
View attachment 465904
DRE-induced distortion is nasty.

Of course, distortion is gone in NOS mode of the Black Pearl:
View attachment 465905

But do you know what is surprising? According to this spreadsheet, some CS431xx-based devices exhibit limited DRE artifacts, which I suspect to be attributed to an adjustment made by those DAC manufactures on some DRE-related parameter in the chip (likely 'DRE decay rate'). For example, DRE artifacts of the following devices

• Fosi Audio DS2
• JCally JM20
• JCally JM20-Max
• A few more devices

cannot be detected by steady-state multitone tests. In fact, the C Major test signal is an ideal test signal. For instance, note that the above test signal simply keeps increasing its level and then keeps decreasing it; that is, it's just one cycle. In response to this signal, these devices behave surprisingly nicely. For instance, the spectrogram of the signal played by the JCally JM20:
View attachment 465906
shows nearly no distortion (yes, nearly; we see a faint sign of it by zooming the plot).

Lastly, see below spectrograms of some music played by the Black Pearl with the Fast-LL filter (DRE enabled) in the upper panel and JCally JM20 in the lower panel:View attachment 465907
Note that I just picked a random song (i.e., did not choose this particular song to show large distortion effects). I did not choose this song to tell something about the distortion's audibility, either. Just to compare the two devices' different behavior.

As you can see, the distortion is much less frequent for the JM20. I bet this would not be audible since in most cases it should be masked whenever it overlaps with a wide frequency spectrum in audio content.

In fact, as suggested by the measurements in the "Other Characteristics" section of my review, the reason why the JM20, or another device showing the same behavior, does not distort is simply because DRE is not fully active much of the time! Whenever the signal level reaches a strong level, DRE becomes inactive, or does not increase its effect. It simply waits, or just wants to decrease DRE gain if the signal gets stronger. Only after the signal goes below -50 dBFS, it turns the DRE back on. This is very passive DRE.

I would call this a good, albeit minimal, DRE implementation. Then, the question is, did the designers of these devices deliberately adjust this DRE parameter for this effect? I absolutely have no idea.

Oh, so according to that spreadsheet, iBasso DX170 doesn't have an additional amplification stage? I wonder how it manages to achieve such output power then...

 

[jkim]

Thank you very much for such a detailed response!

>CS43131 does not apply reconstruction filtering when the input signal is at Fs 384 kHz. I checked the other filter options, too, and found they have no effect at all. So, feeding PCM audio oversampled to 384 kHz into the Black Pearl in NOS mode essentially disables just DRE and makes no other difference in reconstruction filtering, because the DAC's internal filter is not active in this case.

Is this a good thing? Are there any caveats to this?

There's no caveat I can think of. 352.8 k or 384 kHz is the maximum sample rate possible in the DAC. So, with 352.8 k or 384 kHz input, no oversampling-based digital filtering is possible, anyway.

Also, have you done anything measurement with Class H vs Class AB engaged to see what difference does that make, by any chance? ADPT_PWR being in should produce some added distortion by Cirrus Logic's own admission, after all.

I am also curious of in depth measurements between class H and AB of the mighty CS43131.

I actually performed a test a while ago and found no difference using test signals I could think of at the time. Will do some more testing when I get a chance.

Oh, so according to that spreadsheet, iBasso DX170 doesn't have an additional amplification stage? I wonder how it manages to achieve such output power then...

That's certainly a mistake. Some unknown op-amps must've been used in the iBasso DX170.

 

[amanieux]

what is trn black pearl output impedance on 3.5 and 4.4 ?

 

[amanieux]

Windows upsampling (or any upsampling) includes interpolation.

if all my source music is 48khz/16 bit, will it make an audible difference if i set my usb dac on windows10 to 384khz/16 bit instead of 48khz/16 bit ? is 8x upsampling audible in practice ( in theory yes with lowered jitter i guess but with an untrained listener such as 99% of us) ?