DAC filter passband ripple

[pjug]

In another forum, I became aware of how DAC filter passband ripple causes echo of the signal in the time domain. This is not Gibbs effect, but a real reproduction of the signal at very low levels, and typically within a millisecond before and after the proper signal (in the case of a sharp linear phase filter). So a true echo. I am starting to do some listening to see if I can hear this and so far I don't think so (at ~-67dB echo level, ~0.7 msec time spacing). What is the feeling around here on this. I haven't read much on ASR about this, although I have seen some comments by @KSTR and @AnalogSteph mentioning it.

Does it make sense to resample externally to avoid this issue? Even if inaudible, it seems to me this is a much bigger deal than the difference between, say, 105dB and 120dB SINAD.

 

[DVDdoug]

So a true echo. I am starting to do some listening to see if I can hear this and so far I don't think so (at ~-67dB echo level, ~0.7 msec time spacing).

I'm not an expert on what's audible, but probably not. Our ears don't have much resolution in the time-domain. i.e. You probably can't hear 1-cycle at 10kHz and if you hear something you certainly wouldn't perceive a particular frequency or pitch.

Does it make sense to resample externally to avoid this issue? Even if inaudible

No. If your purpose is audio and the enjoyment of music, etc., it doesn't make sense to make "improvements" that are not audible. Put your time, effort, and money to what you can actually hear!

 

[pjug]

I'm not an expert on what's audible, but probably not. Our ears don't have much resolution in the time-domain. i.e. You probably can't hear 1-cycle at 10kHz and if you hear something you certainly wouldn't perceive a particular frequency or pitch.

In my case It isn't really much effort since I am using Roon anyway. Just make sure everything is upsampled to a high sample rate before going to the DAC, then apply a slow filter setting with the DAC. And no harm so why not?

 

[MaxwellsEq]

When doing two-way telephone conversation a rule of thumb is to minimise echos to under 75ms. Below that number people generally don't perceive an echo.

 

[pjug]

When doing two-way telephone conversation a rule of thumb is to minimise echos to under 75ms. Below that number people generally don't perceive an echo.

The <1msec echoes would not be perceived as echoes. On the other site some said it could affect sound stage or the sound of attacks.

 

[bennetng]

although I have seen some comments by @KSTR and @AnalogSteph mentioning it.

Does it make sense to resample externally to avoid this issue? Even if inaudible, it seems to me this is a much bigger deal than the difference between, say, 105dB and 120dB SINAD.

...and me.

SMSL SU-9 Balanced DAC Review

Can someone address this Reddit post on this subject.. is this guy right? I’m not an EE and honestly don’t know. This person claims anything besides brick wall is wrong wrong wrong. He conveniently forgets minimum phase filters are used in minimum latency applications like for stage or other...

www.audiosciencereview.com

DacMagicPlus Filter Mirror Images - Short Study

Several times stronger passband ripple and about 20dB poorer stopband attenuation. BTW, the screenshot you showed looks like the Hybrid one (intermediate phase) rather than Brickwall (linear phase)...

www.audiosciencereview.com

Digital filter "game"

Some of you probably have seen a lot of filter plots like these on DAC datasheets: How about making these things without using specialized software like Matlab, Octave and such? They are not particularly easy to use. So many people just "see" them rather than try to figure out the meaning...

audiosciencereview.com

The point I want to make is that regardless of audibility, a digital filter with a fixed length (e.g. in typical DAC chips) can be tuned to optimize for accurate cutoff, lowest ripple or deepest attenuation, but overly biased to one of of them will worsen the other two parameters, it is all about trade off.

Why I said "regardless of audibility"? Because I did some simulations to mimic different filter characteristics with limited filter length similar to typical DAC chips but I can't hear a difference, but it doesn't mean there are no objective, measurable differences, just like 105dB vs 120dB SINAD.

 

[RandomEar]

Looks like for large differences (audio playing vs. total silence) in clean test signals, the time resolution of the human ear can be as good as 3 ms [1]. But what you describe isn't just about time resolution and it's not a clean test signal. There is a huge difference in volume (> 60 dB) and the added signal is extremely similar to the base signal. Therefore, I would assume that you will have to look into masking. The problem is similar to the question of audio compression, where this phenomenon plays an important role.

Knowing that most people can't identify even rather dissimilar signals like distortion past -40 dB with very few people reaching -50 dB [2], I wouldn't expect this to be audible in any way. But it's certainly not wrong to investigate and/or test that.

EDIT: It would also be cool if you could show or link an example, where such an "echo signal" was captured.

 

[MaxwellsEq]

Minimum-Phase Filters | Introduction to Digital Filters

www.dsprelated.com

https://www.analog.com/media/en/training-seminars/design-handbooks/Basic-Linear-Design/Chapter8.pdf

Why Not IIR?

Why Not IIR?

repforums.prosoundweb.com

 

[pjug]

Looks like for large differences (audio playing vs. total silence) in clean test signals, the time resolution of the human ear can be as good as 3 ms [1]. But what you describe isn't just about time resolution and it's not a clean test signal. There is a huge difference in volume (> 60 dB) and the added signal is extremely similar to the base signal. Therefore, I would assume that you will have to look into masking. The problem is similar to the question of audio compression, where this phenomenon plays an important role.

Knowing that most people can't identify even rather dissimilar signals like distortion past -40 dB with very few people reaching -50 dB [2], I wouldn't expect this to be audible in any way. But it's certainly not wrong to investigate and/or test that.

EDIT: It would also be cool if you could show or link an example, where such an "echo signal" was captured.

To make a test track to ABX with an original, I just mixed in two -67dB copies of a music track, offset by + 0.67msec and - 0.67msec, back into the original. To verify that this looked like passband ripple, I ran a comparison in DeltaWave, and the delta of the spectra had the ripple exactly as expected. Right now I am traveling, but I can show this when I return at the end of the week if anyone is interested.

 

[AnalogSteph]

Although the specifics in terms of audibility would depend on the frequency of our periodic ripple (and hence echo displacement), my impression is that the levels found in typical DACs by major manufacturers (±0.00x dB) seem to be generally unconcerning, even if the late Julian Dunn would probably not be impressed by all of them. With some manufacturers you won't need to worry even if numbers are substantially higher, e.g. Cirrus who (a) have an affinity for IIR filters to begin with and (b) don't generally show more than 2 ripple periods per passband, making for very short echo displacement.

The cases I've found to be actually problematic have generally been in "get the job done audio", like AC97 and HD Audio codecs, with specified / measured periodic ripple exceeding ±0.1 dB (i.e. -44 dB pre-echo). Even in this area, I don't think any new Realtek chips introduced since 2009 have remotely concerning levels of DAC passband ripple, although the ADC filters on their lesser chips are a bit "eh" (if still better than a bunch of classic consumer-level ADCs, and any AK55xx below 192 kHz for that matter...). Unknown also-rans chosen by OEMs for their attractive pricing can be hit and miss though. I should really take some measurements of the 92HD93 in my E6330, that thing clearly benefits from upsampling (= reducing echo displacement)...