On the Distortion of Cirrus Logic CS431xx-Based Devices: A Comparative Review

[CedarX]

It seems that PEQ are often (always?) implemented based as biquads—or maybe this is what the integrated DSP provide.
The biquads formulas are Fs-dependent. It seems that most PEQ implementations use Fs=48kHz to calculate the biquads, but don’t re-calculate them based upon the actual Fs of the signal. Not necessarily a big issue as it affects mostly high frequencies where you shouldn’t EQ anyway…
Is the above correct? And if yes, is this an acceptable way to implement PEQ?

 

[torgeirs]

Normally biqads can have the same coefficients regardless sampling frequency.
They will hit the same frequencies regardless.
No problem if you are not very close to the frequency extremes.
May be some trouble with phase at the highest highs and truncation in the lowest lows

 

[torgeirs]

If the software saves on word length, there may be a some distortion in bass

 

[Jean-Paul Petrosyan]

It seems that PEQ are often (always?) implemented based as biquads—or maybe this is what the integrated DSP provide.
The biquads formulas are Fs-dependent. It seems that most PEQ implementations use Fs=48kHz to calculate the biquads, but don’t re-calculate them based upon the actual Fs of the signal. Not necessarily a big issue as it affects mostly high frequencies where you shouldn’t EQ anyway…
Is the above correct? And if yes, is this an acceptable way to implement PEQ?

Why one shouldn't EQ high frequencies?

Also, my EQ settings typically include peak filters centered around 30 Hz with a Q value of ~0.5-07 and/or Low Shelf Filters at ~70-110 Hz with same Q values. Does that mean I'm introducing extra distortion?

 

[jkim]

Since we are into the details of artifacts and low level distortion.
Low frequency PEQ with high frequency samplerate often intoduce distortion. If not 64 bit prosessor. How is this implementation?

At least the TRN Black Pearl's onboard PEQ does not add any distortion:

EDIT. I think we want to move this discussion somewhere else if we want to continue it. It is not on the topic of this thread.

 

[torgeirs]

Why not show normal THD with and without PEQ? Not shure the distortion will show with multitone.
Because most of the energy in the measurement signal will be to close to sampling frequency

 

[torgeirs]

That is same samplerate and PEQ, but only 20 hz sine. Long FFT

 

[torgeirs]

Heres a thorough article on filter precision https://firmware-developments.com/WEB/DOC/REF/ZauchaIIR_AES.pdf

 

[torgeirs]

At least the TRN Black Pearl's onboard PEQ does not add any distortion

So that sentence is not possible if its not infinite number of bits precision

 

[A Surfer]

Is the distortion going to be audible or simply measureable?

 

[torgeirs]

That’s what I want to know. But as all the other Cirrus artifacts has shown. The measurement signal must reveal the distortion or artifact. And not hide it as the multitone will regardig low frequency truncation distortion

 

[Jean-Paul Petrosyan]

Should we start a dedicated thread for this matter? I would be very interested in learning about that more, though my understanding of the relevant topics is very limited.

I thought that DSP precision is supposed to increase with higher sample rates (at least that's what I read before) and that oversampling may be beneficial when using EQ or other DSP. Turns out it's the opposite?

I'm a heavy EQ user, so this concerns me a lot.

 

[jkim]

So that sentence is not possible if its not infinite number of bits precision

Of course, I did not mean that literally, but rather 'at least the TRN Black Pearl's onboard PEQ does not add any (measurably meaningful) distortion.' And in my experience, if a test with a multitone signal shows a clean response, a single tone test always also showed no problem, but not the other way around. Sure, in theory there must be an exception, but I haven't seen such a case empirically. In the above, another reason why I used a multitone test was to show that the EQ filter is actually in effect.

Single-tone FFTs with a 20 Hz sinusoid look like:

There's virtually no difference.

And the topic of PEQ implementation is not directly related to this thread, i.e., DRE-induced distortion of CS431xx-based devices. So, if anyone thinks this particular topic has not been adequately discussed at ASR, please create a thread in an appropriate forum. Let's stop discussing it here.

 

[torgeirs]

Wow, that is really impressive measurement at 384 fs, 20 hz PEQ and 20 hz signal!
Just 1 dB added distortion at those low levels.
Thanks

 

[torgeirs]

I thought that DSP precision is supposed to increase with higher sample rates (at least that's what I read before) and that oversampling may be beneficial when using EQ or other DSP. Turns out it's the opposite?

Yes for the low end frequencies. But at 64 bit prosesing and more its negligible like on this device measured here.

 

[torgeirs]

Sorry @jkim
Somehow I drifted over on this thread from the https://www.audiosciencereview.com/...usb-dac-headphone-amp-with-10-band-peq.64331/
thread without noticing
So my posts from #276 should have been in that thread. Bummer

 

[torgeirs]

And in my experience, if a test with a multitone signal shows a clean response, a single tone test always also showed no problem, but not the other way around. Sure, in theory there must be an exception, but I haven't seen such a case empirically

Just a comment on this from a measurement theory perspective.
The multitone will mask the real proplem. Try 32 bit PEQ on some software. Single tone will show the distortion but multitone will mask it.
At multitone the lowest frequency is attunated maybe 20 dB compared to a single tone, else the signal will clip

Multitone expose gain bandwidth induced distortions. This distortion orginates from the opposite end of the spectrum

 

[jkim]

Just a comment on this from a measurement theory perspective.
The multitone will mask the real proplem. Try 32 bit PEQ on some software. Single tone will show the distortion but multitone will mask it.
At multitone the lowest frequency is attunated maybe 20 dB compared to a single tone, else the signal will clip

Multitone expose gain bandwidth induced distortions. This distortion orginates from the opposite end of the spectrum

Of course, no single form of measurement can cover everything. However, if I have to choose just one test for the purpose of checking potential problems (not a certain targeted problem), I will select the AP-style 32-tone test, as I agree to (and also experienced) what @Rja4000 mentioned in one of his reviews:.

Multitone measurement is always very informative, as it allows to assess a lot of different things in one single measurement:
You see the noise, all kind of distortions (shown at the frequency where you'll hear their effect), the frequency response, and even the crosstalk.

 

[torgeirs]

Well guess bass distortion not caused by gain/bandwith limitations will pass unnoticed then. And low octave PEQ troubles is one of them.

 

[torgeirs]

But then the Cirrus chip don’t have IIR filters so not so relevant for that chip.
But guess there will be a floodwave of IIR enabled DAC chips or USB audiochips in near future. Actually it is already here.
Again, sorry to clutter your thread,