=== Important Findings ===
Out of curiosity, I measured the TRN Black Pearl again in its NOS (Non-Oversampling) mode. A 32-tone signal @ -20 dBFS---at which the Cirrus hump is at its largest---was used. In fact, I did this test and reported results a few posts earlier
here and also
here. But at THOSE times, the signal chain was:
- A 32-tone signal was generated at the sample rate of either 96 kHz or 44.1 kHz.
- AND the Black Pearl played it at the same sample rate as the signal was created.
This assumes a true NOS playback situation, i.e., the DAC plays originally recorded (i.e., sampled) digital content with
no resampling or filtering.
However, I asked myself, "what if CD audio (sampled at 44.1 kHz Fs) is played by this DAC w/ NOS selected
but at a much higher sample rate?" In fact, Windows should upsample the signal, and its upsampling quality must be good (see
here).
It turns out that such playback achieves what we want: neither a Cirrus hump and nor NOS artifacts!
Below are FFT captures of a
32-tone signal @-20 dBFS generated at 16 bit / 44.1 kHz Fs (i.e., CD audio), played by the Black Pearl (
NOS) in
various sample rate settings in Windows 11.
First up is
44.1 kHz playback sample rate:
No Cirrus hump distortion, but as expected, huge NOS distortion from broken waves. Also notice some high frequency roll-off b/w 10 k - 20 kHz which is typically observed from NOS D/A conversion.
Next,
96 kHz playback sample rate:
Still quite large NOS artifacts but this clearly indicates that the signal is being resampled.
192 kHz playback sample rate:
Upsampled to 192 kHz, NOS distortion is much lower. No high-frequency roll-off, either. At this resampling rate, I won't worry about the NOS distortion---it is no longer NOS, anyway.
How about
384 kHz playback sample rate:
Upsampled to 384 kHz, there is just the CS431xx's regular noise shaping effect alone. There's a hint of distortion at 384 kHz, but it is lower than the noise shaping peak.
Now that the absence of a Cirrus hump has been confirmed, how about the Part II tests described in the review?
The C Major clip first. Played with NOS at Fs = 192 kHz:
No clicks.
The Dune soundtrack clip:
Clean.
Then, we want to ask an important question,
why do we not have a Cirrus hump nor the distortion observed in Part II tests
when NOS is selected? In the review we conjectured the distortion could be due to improperly implemented dynamic range enhancement (DRE) in the CS431xx chips. Actually, in the NOS mode DRE is turned off! See below:
In the NOS mode, noise stays the same regardless of the signal level (red solid line). The full noise reduction from DRE, when it is turned on, is 24 dB (green solid line) just as observed
here.
These data provide convincing evidence that the distortion we're dealing with is caused by DRE, because the distortion is gone when DRE is not in action.
One may ask if the DAC's noise performance becomes horrible with DRE not engaged, since DRE contributes whopping 24 dBs to noise reduction according to the above measurements. In fact, the DRE effect in reality is
not 24 dBs. The reason is that the noise level measured above covers wide bandwidth (20 Hz - 90 kHz) and hence is dominated by the noise shaping peak in the ultrasonic region (> 50 kHz). This setting was intended to accurately measure the 'DRE gain' which is solely digital gain and reflected well in the noise shaping. But the noise level in the audio band (20 Hz to 20 kHz) is dominated by analog (i.e., thermal) noise. With DRE, analog noise in this passband is very low but still there's a limit. The dynamic range (DR) of the TRN Black Pearl's balanced output
with DRE turned on is
Left:
, Right:
whopping 132.7 dB! Actually this performance challenges the noise limit of my measurement setup (E1DA Cosmos ADCiso and Scaler). I have never seen this high DR in Amir's AP measurements of any DAC. No doubt the Black Pearl's power supply circuits must have a very nice design. But the real reason how it achieves this astonishingly high DR is simply because of DRE.
The question is, do we really need this high DR? Absolutely not. We cannot hear the difference between two DACs with this high DR and much lower DR. Let's see the DR of the same Black Pearl's balanced output
with DRE turned off (i.e.,
with NOS):
Left:
, Right:
Its DR with no DRE is
decent at 120.5 dB. Sure, the difference b/w DRE being on and off is somewhat large (12 dB), but a DR of 120.5 dB is still very nice.
What about
50 mV SNR with DRE turned on?
(left ch. shown; right ch. is 0.1 dB lower)
With
DRE turned on, the Black Pearl's SNR at 50 mV into its unbalanced output is
amazingly high 96.5 dB! This number would actually be at the very top of Amir's headphone amp 50 mV SNR chart:
Now,
50 mV SNR with DRE turned off:
(left ch. shown; right ch. is 0.1 dB lower)
With no DRE (i.e.,
with NOS)
, the SNR decreases to
85.3 dB. Is this bad noise performance? Not at all. I cannot hear any hiss from my 7Hz x Crinacle Zero:2 when hearing -120 dBFS white noise (not supposed to be heard) with the
device volume set to max. In fact, an SNR of 85.3 dB at 50 mV is very similar to (just 2 dB worse than) what I measured from the Qudelix 5K.
Would I choose DRE in order to have unnecessary 96.5 dB SNR @ 50 mV and swallow the Cirrus hump distortion? Definitely not.
I was curious of other CS431xx-based DACs that support selection of filters. Tried the Tanchjim Stargate II and
JCally JM20-ProFiiO KA15. Observed exactly the same behavior as above.
Key Takeaways:
- When a CS431xx-based DAC can be set to 'NOS' mode the DAC no longer applies DRE (dynamic range enhancement) and does not exhibit Cirrus hump distortion---to make sure, all the DACs supporting filter selection need to be checked but this seems to be the CS431xx chips' default behavior when NOS is selected.
- Playback at a sample rate of 192 kHz or higher, if a good upsampling algorithm is supported by the operating system, will mitigate NOS artifacts. Whether upsampling is supported, and how good it is, will depend on the operating system, though.
- Some may wonder if the DAC's noise performance is still good with DRE not engaged. I can assure that it is still decent. You won't hear hiss even from sensitive IEMs.
What may seem like a hair splitting exercise to some can be far from it. It's a bit like the butterfly effect - tiny errors and flaws get compounded and propagated, and it all adds up to an undesirable end result. And that's why the tiniest details matter.
