Phase shift and echo trails length illusion

[pkane]

I think @Pdxwayne is experimenting and looking for ways to detect audible differences.

From the number of metrics available in DeltaWave, PKEM is the closest to approaching a perceptually-weighted metric. Others, such as phase differences, RMS of the error signal, correlated-null depth, jitter, linearity, etc., are all engineering metrics. And while known audibility thresholds can be applied to engineering metrics, they are not as easy to interpret for a relative audibility determination.

 

[AnalogSteph]

Had to Google some of the terms and I am not sure I understand FIR vs IIR much.....

Understanding the distinction is kinda important in digital audio though.

The filters you are going to see out on the analog world are generally IIR = minimum phase filters, including RC and LC. For a given filter performance, they take the shortest time for the signal to traverse the passband, but that time will not be constant across frequencies (group delay variation), reflecting in phase response. In the digital domain, IIR filters involve feedback, which is potentially problematic as it allows computational inaccuracy (quantization errors) to accumulate. Nowadays we do have the required accuracy available though.

FIR = linear phase filters always take a constant time (group delay) for signals to get through anywhere, as they are essentially based on time delays. They were a favorite in digital filters early on since quantization errors cannot build up. In the analog world, you may be familiar with comb filtering effects. The effect is also being exploited in SAW (surface acoustic wave) filters, e.g. as RF band filters in mobile phones, or as SAW resonators to build oscillators in the GHz range that is way out of reach for quartz crystals.
In the time domain, FIR filters have a symmetrical impulse response, which gives rise to some peculiarities, including periodic passband ripple being linked to sort of a pre-echo effect (as discussed e.g. by Julian Dunn). Also, a complex filter invariably necessitates a long group delay, which for a fancy anti-alias or reconstruction filter at single speed can be over a millisecond. When you ideally want <3 ms for A/D + processing + D/A for monitoring in a recording application, that can turn into a problem.

FIR and IIR filters can be implemented to give the same exact magnitude response, as seen in modern ADCs and DACs (AK557x come to mind). They will each have their own time domain and phase idiosyncrasies though.

 

[Pdxwayne]

Understanding the distinction is kinda important in digital audio though.

The filters you are going to see out on the analog world are generally IIR = minimum phase filters, including RC and LC. For a given filter performance, they take the shortest time for the signal to traverse the passband, but that time will not be constant across frequencies (group delay variation), reflecting in phase response. In the digital domain, IIR filters involve feedback, which is potentially problematic as it allows computational inaccuracy (quantization errors) to accumulate. Nowadays we do have the required accuracy available though.

FIR = linear phase filters always take a constant time (group delay) for signals to get through anywhere, as they are essentially based on time delays. They were a favorite in digital filters early on since quantization errors cannot build up. In the analog world, you may be familiar with comb filtering effects. The effect is also being exploited in SAW (surface acoustic wave) filters, e.g. as RF band filters in mobile phones, or as SAW resonators to build oscillators in the GHz range that is way out of reach for quartz crystals.
In the time domain, FIR filters have a symmetrical impulse response, which gives rise to some peculiarities, including periodic passband ripple being linked to sort of a pre-echo effect (as discussed e.g. by Julian Dunn). Also, a complex filter invariably necessitates a long group delay, which for a fancy anti-alias or reconstruction filter at single speed can be over a millisecond. When you ideally want <3 ms for A/D + processing + D/A for monitoring in a recording application, that can turn into a problem.

FIR and IIR filters can be implemented to give the same exact magnitude response, as seen in modern ADCs and DACs (AK557x come to mind). They will each have their own time domain and phase idiosyncrasies though.

Is it possible that analog out is using IIR, but when capture with ADC, both the ADC and capture software are using FIR?

In that case, would issues heard with analog chain via stereo system show up in captured audio?

Thanks!

 

[AnalogSteph]

You sound confused.

When recording, your signal path looks roughly like this:
digital --> [DAC] --> [analog stuff] --> [ADC] --> digital

It's a concatenation of 3 "boxes" (which is a fancy way of saying they are in series), each with a specific impulse response in time domain, which in turn has a particular frequency domain equivalent in amplitude and phase response.

The sum effect of all 3 impulse responses is a convolution of their individual impulses responses. In the frequency domain, this means that:
* amplitude responses multiply, or on a logarithmic (dB) scale add up
* phase responses add just like the corresponding group delays do
That makes things quite easy in practice. If you have Bode plots for all 3 systems, you can just add your (dB) responses and phases and get the one for the whole thing.

If you have ever used a convolver plugin in your audio player, that's what that does - it adds a "box" in the signal path. You can do all sorts of things with this, from plain EQ to recreating the sound of an entire concert hall.

So in real life, you are generally going to see a combination of various effects:
* in amplitude response, the sum of FIR + IIR including analog RC / LC filtering
* in phase response, the sum of IIR and analog RC / LC filtering, as any pure time delays are generally zeroed out afterwards (note: you may still see the characteristic exponential signature of a time delay in REW's phase plots if that part wasn't entirely sucessful; well, it looks exponential because the phase plot is half-logarithmic and that's how a linear function comes out if plotted like that)

Your listening chain will have the DAC and some analog part shared with the recording chain, just the things later on are replaced by your amplifier, transducers, air, specific HRTF and auditory system.

This is also why you can detect differences between DACs even on a relatively crummy recording system, at least when it comes to amplitude and phase. (It gets a bit tricky when your ADC noise is 95 dB down and you want to compare DACs with noise levels of -117 and -123 dB, respectively, although it is fundamentally still possible. Detecting miniscule differences in noise floor with sufficient accuracy merely is a bit of a challenge.)

This has turned into a little primer on LTI (linear time invariant) systems, but these are the fundamentals for understanding what happens to signals in a chain of systems.

 

[solderdude]

The problem is with the term pre- and post-echo is that people get the idea that 'echo' can linger on because of these filters or that you can hear the sound as a pre-echo.

In all cases everything well below say... 10kHz does not 'echo' nor 'ring' because of these filters.
Only when using sharp EQ in the audible band that specific band can 'ring' but this is not the case here.

The 'ringing' only takes place when a sudden 'jump' is there in the signal and then it only 'rings' near 1/2 fs and NOT in the actual music signal.
People get the idea when they see the often shown 'post ringing' of filters (caused by an illegal signal only made as an artificial test signal) that such a filter also 'rings' in the mids and can make 'echos' from sounds longer or shorter.

It doesn't work that way. A recorded 'echo' in music is just a recorded waveform and will always be reconstructed in the same way.

 

[AnalogSteph]

Quite correct. The only kind of potentially audible / detrimental pre-echo in practice is the one resulting from periodic passband ripple in FIR filters, as discussed by J. Dunn.

You can actually hear the ringing around fs/2 quite well when lo-fi sample rates <32 kHz are involved. Years ago I needed something low (think 8 or 11.025 kHz) for some sort of demonstration, and the result of resampling CD quality material with good ol' SSRC (extremely steep filter with cutoff extremely close to fs/2) wasn't too pretty-sounding. The SoX VHQ defaults of a 95% fs/2 cutoff are much more well-behaved already, I think.

 

[Pdxwayne]

The problem is with the term pre- and post-echo is that people get the idea that 'echo' can linger on because of these filters or that you can hear the sound as a pre-echo.

In all cases everything well below say... 10kHz does not 'echo' nor 'ring' because of these filters.
Only when using sharp EQ in the audible band that specific band can 'ring' but this is not the case here.

The 'ringing' only takes place when a sudden 'jump' is there in the signal and then it only 'rings' near 1/2 fs and NOT in the actual music signal.
People get the idea when they see the often shown 'post ringing' of filters (caused by an illegal signal only made as an artificial test signal) that such a filter also 'rings' in the mids and can make 'echos' from sounds longer or shorter.

It doesn't work that way. A recorded 'echo' in music is just a recorded waveform and will always be reconstructed in the same way.

Curious, what is the possibility a brand new DAC like X16 might need a little burn-in to get the highs to sounds as good as a well used DAC? When I first got my x16, I kept feeling that E30 have better highs. Now it seems they are pretty similar....Or was it all in my head? Thanks!

 

[AnalogSteph]

You bothered to do some level matching (at least according to this thread), and lo and behold, the differences went away. That's the sort of stuff that happens what you're not comparing apples to oranges for once.

It still is entirely possible to hear difference where there actually are none, of course. That's the crux with sighted listening tests.

 

[Pdxwayne]

You bothered to do some level matching (at least according to this thread), and lo and behold, the differences went away. That's the sort of stuff that happens what you're not comparing apples to oranges for once.

It still is entirely possible to hear difference where there actually are none, of course. That's the crux with sighted listening tests.

Early on, when I said E30 had more "sparkles", I did voltage measurements to do comparisons to make sure there were no voltage mismatched issue.

I am still curious if indeed there is a little burn-in period needed for brand new Gustard x16.

I did not enjoy X16 when I was using it brand new in my stereo setup. I was using my E30 in my stereo setup for two weeks before that. The difference in the highs was obvious to me. No matter if I turned x16 volume up high or low, still something was lacking. It felt like notes stopping too soon.

It got better after about a week of keeping x16 on all the time. Did my brain got burn in or indeed there is a small burn in period for new DAC like X16?

Thanks!

 

[Pdxwayne]

You bothered to do some level matching (at least according to this thread), and lo and behold, the differences went away. That's the sort of stuff that happens what you're not comparing apples to oranges for once.

It still is entirely possible to hear difference where there actually are none, of course. That's the crux with sighted listening tests.

BTW, when I did my KTB vs E30 blind tests (voltage matched and not matched) last year, I have found out that using headphones setup can't really show minor difference in voltage mismatch.

My earlier comments today about E30 and x16 sounded similar is based on headphones setup and not with stereo setup.

Using headphones setup today, using RCA switch to do quick switch, indeed X16 and E30 sounded similar when using my $50 headphones.

I plan to do RCA quick switch with my stereo setup downstairs sometimes this week. Yes, sighted first, before blind.

 

[solderdude]

Warm up time... possible. Would happen every time you switch it on though.
Burn-in... no. There is nothing to 'burn in'.

 

[Pdxwayne]

Warm up time... possible. Would happen every time you switch it on though.
Burn-in... no. There is nothing to 'burn in'.

I didn't really like the sound of x16 even after I let it stayed on for a several days (when in brand new condition). That should be pleanty of time to warm up?

I even bought a new coaxial cable because I thought something was wrong with the optical input of x16. It was that bad in the beginning....

Are there components on the board of x16 that requires a few turned on and turned off to get to optimized operating condition?

 

[solderdude]

When switched on you get settle time first (can be very short) and is caused by capacitors charging up and electronics getting in their operation point.
Usually seconds.
Then you may have 'warm-up' time which can vary depending on components used. Can be minutes to say half an hour.
In that time one may find very tiny measured changes in for instance stability but these are not reaching audible levels unless the design is really crappy.
Nothing in Amir's review of the (Gustard) X16 points towards these kind of issues, but as there was no stability test done we will never know.
On the other hand, when Amir finds such issues and it is audible he would see distortion products 'improve' while testing warranting to investigate.

 

[Pdxwayne]

When switched on you get settle time first (can be very short) and is caused by capacitors charging up and electronics getting in their operation point.
Usually seconds.
Then you may have 'warm-up' time which can vary depending on components used. Can be minutes to say half an hour.
In that time one may find very tiny measured changes in for instance stability but these are not reaching audible levels unless the design is really crappy.
Nothing in Amir's review of the (Gustard) X16 points towards these kind of issues, but as there was no stability test done we will never know.
On the other hand, when Amir finds such issues and it is audible he would see distortion products 'improve' while testing warranting to investigate.

Amir likely got well tested unit and not brand new out of factory unit.
; )

There was one guy with brand new x16 having kind of similar issue, but worse. See https://www.audiosciencereview.com/...x16-balanced-mqa-dac-review.17419/post-663577

He said worse than his Yamaha. Like a blanket over speakers! I wonder what component could cause such issue? Fortunately mine improved over time.

 

[solderdude]

The question is whether or not this was a perception thing or reality.
As there are no measurements and it is purely anecdotal we will never know.

I think we all know what component would be capable of such poor performance that it would reach audible levels AND goes away after a while.
If those kinds of components exist many electronics would be bothered by this and it would be measured often.

On a regular basis I find music to sound 'poor' in quality but a few days later it sounds heavenly. Same music, same gear, same ears.
Willing to bet more people have experienced this.

 

[Pdxwayne]

The question is whether or not this was a perception thing or reality.
As there are no measurements and it is purely anecdotal we will never know.

I think we all know what component would be capable of such poor performance that it would reach audible levels AND goes away after a while.
If those kinds of components exist many electronics would be bothered by this and it would be measured often.

On a regular basis I find music to sound 'poor' in quality but a few days later it sounds heavenly. Same music, same gear, same ears.
Willing to bet more people have experienced this.

Yeah, it happens often to me too. Poor in the morning, but better at night.

But typically the "poor" in the morning was too much harshness. I have yet to experience not enough highs for period of days....But could be my health just coincidentally turned for the worse during the time I first placed the x16 into my stereo setup....

Anyway, I am still glad to do all the measurements. At least now I feel good about both the E30 and x16.

I still plan to do quick DAC comparison test with my stereo setup downstairs. I don't expect to hear much difference now, but will see.

Thanks for all the feedbacks!

 

[pma]

People get the idea when they see the often shown 'post ringing' of filters (caused by an illegal signal only made as an artificial test signal) that such a filter also 'rings' in the mids and can make 'echos' from sounds longer or shorter.

I know what you mean, however your mentioning that post ringing is a result of illegal artificial signal is incorrect and misleading and could confuse the beginners, though it is sometimes a popular pseudo-scientific belief.

As a proof, get a 1kHz analog square and sample it at Fs=48kHz with a proper anti-alias filter (cut everything >=24kHz) and A/D. You will get a digitized wave which reflects input anti-alias (brickwall) filter response, with both pre and post ringing. Now you have a properly digitized, band limited signal and you play it through DAC and catch the analog output by a fast scope. Again you get pre and post ringing previously captured and the signal was not illegal and was not artificially constructed in the digital domain. It was a perfectly valid analog signal band limited by the anti-alias input filter. You may not exclude such signal from considerations.

 

[solderdude]

Yes, correct, the Gibbs effect

I was trying to 'illustrate' the effect what Archimago showed (after my remarks) that tones don't change their decay similarly to that of sharp edges near 1/fs.

Echo trails are not changed or caused by the sharp filters was my point. They are simply in the waveform and cannot possibly be influenced by the filter.

 

[voodooless]

Like a blanket over speakers! I wonder what component could cause such issue? Fortunately mine improved over time.

The brain..

 

[voodooless]

Some more food for thought: https://archimago.blogspot.com/2018/01/audiophile-myth-260-detestable-digital.html