On the subject of "ringing"

[John_Atkinson]

33 minutes ago, adamdea said:

Why not just use a real thing like a castanet?

 

Because you have to capture the sound of the castanets to digital using an A/D converter whose antialiasing filter's behavior is unknown. You can't solve an single equation that has two variables.

 

John Atkinson

Editor, Stereophile

[Fokus]

12 minutes ago, John_Atkinson said:

 using an A/D converter whose antialiasing filter's behavior is unknown. You can't solve an single equation that has two variables.

 

Look at the datasheet of the ADC chip in the convertor.

 

The vast majority of them have the same filter strategy anyway ...

 

[Fokus]

11 hours ago, buonassi said:

 

Does preringing only affect HFs near Nyquist?  The entire bandpass is affected I thought?  

 

Preringing has less to do with the frequency domain, rather the time domain as I understand it,

 

 

1) How could that be?

 

Take a filter that rings at F. Take a signal without energy at F. Filter it. There is no ringing. How could the filter then affect the entire passband?

 

2)  Time = frequency in a linear system. The audiophile industry want you to believe otherwise, but that does not make it true.

 

 

[mansr]

13 hours ago, John_Atkinson said:

The high-frequency rolloff of the Ayre QA-9's Listen filter seems pretty benign. You can find my measurements at https://www.stereophile.com/content/ayre-acoustics-qa-9-usb-ad-converter-measurements

 

Below is fig.2  on that page, showing the frequency response at –1dBFS with the Listen filter,  analyzed in the digital domain, and data sampled at 192kHz (left channel blue, right red), 96kHz (left green, right gray), 48kHz (left cyan, right magenta)  The vertical scale is 1dB/div. The response at 20kHz is down by just a fraction of a dB at all 3 sample rates, though the rolloff is slow, which will potentially allow some aliasing at single Fs rates with some music.

 

John Atkinson

Editor, Stereophile

 

Something is off here. A 48 kHz sampled signal by definition has no content above 24 kHz. What is that graph actually showing?

[adamdea]

Just now, John_Atkinson said:

 

Because you have to capture the sound of the castanets to digital using an A/D converter whose antialiasing filter's behavior is unknown. You can't solve an single equation that has two variables.

 

John Atkinson

Editor, Stereophile

I assumed that what you were testing was the adcs' digital decimation filter not the analogue filter before the modulator. 

If you can sample a castanet at crazy hf  then there wouldn't be a problem knowing that the anti aliasing decimation  filter is at a higher rate than the 96 khz used in your test example for the A/Ds. Clearly one would expect the 96 khz filter in your example to dominate the response. 

 

Alternatively if we are testing the very limits of the A/d process then I appreciate you could model something digital matching the envelope of the castanet. Perhaps this is what you meant by your diagnostic signal. What is confusing me is the specification of the diagnostic signal. To be fair IMHO it should match what a real world signal might be and should be diagnostic of the effect (if any) of minimum phase on the rising edge of a transient 

[Shadders]

10 minutes ago, Fokus said:

 

1) How could that be?

 

Take a filter that rings at F. Take a signal without energy at F. Filter it. There is no ringing. How could the filter then affect the entire passband?

 

2)  Time = frequency in a linear system. The audiophile industry want you to believe otherwise, but that does not make it true.

 

 

Hi Fokus,

Even if you energise a filter with ringing at F, with a frequency F, the filter does not ring.

 

Example, a 192kHz sample rate signal, filter set to cut off at 50kHz, any signal in the frequency 40kHz to 60kHz has NO ringing.

 

All the filter does is attenuate the signal according to the transition band and stop band characteristics. No ringing at 50kHz.

 

Regards,

Shadders.

[adamdea]

33 minutes ago, adamdea said:

I assumed that what you were testing was the adcs' digital decimation filter not the analogue filter before the modulator. 

If you can sample a castanet at crazy hf  then there wouldn't be a problem knowing that the anti aliasing decimation  filter is at a higher rate than the 96 khz used in your test example for the A/Ds. Clearly one would expect the 96 khz filter in your example to dominate the response. 

by way of example. AKM adcs apparently will output fs= 768 kHz  https://www.akm.com/akm/en/product/datasheet1/?partno=AK5397EQ

presumably we aren't concerned about ringing at that rate?

[Spacehound]

16 hours ago, Ralf11 said:

 

when coupled with a lack of methodology it is no surprise that people reject this

 

if you want to be believed you will need good strong evidence

 

unusual results require unusual proof -- and anything appearing to contravene known physical laws is indeed unusual

His methodology is superb:

 

"Martin Colloms was able to....."  (one occurrence, used many times)

or

"Many members agree"  ("many" appears to approximate 5, used many times, it is not known if it is the same 5 every time but you won't get on his 'approved testers'  list if you fail the initiation ceremony)

or

Insults

 

Beat that.

[John_Atkinson]

35 minutes ago, mansr said:

Something is off here. A 48 kHz sampled signal by definition has no content above 24 kHz. What is that graph actually showing?

 

Look more closely at the graph using the color coding for the traces I supplied in my earlier posting.

 

John Atkinson

Editor, Stereophile

[Shadders]

19 hours ago, John_Atkinson said:

which captures a band-limited impulse without any ringing before or after:

 

John Atkinson

Editor, Stereophile

Hi,

It was stated by Archimago et al in another thread that accuracy when making statements is a key aspect. In the digital domain, an impulse is a single sample, usually of value 1, where preceding and post the impulse, all values are zero. In the analogue world, then an impulse is technically a infinitely narrow pulse whose area under the "curve" is 1 - also known as a delta dirac function :

https://en.wikipedia.org/wiki/Dirac_delta_function

 

Maybe call it a band limited pulse. As per mansr request - what is the shape of this pulse - if bandlimited, then smooth corners will be present ?. Are all frequencies contained in this pulse up to the band limit ?

Or have you just taken an impulse, passed it through a filter, and used the output as the bandlimited pulse ?

 

Regards,

Shadders.

[Spacehound]

21 minutes ago, John_Atkinson said:

 

Look more closely at the graph using the color coding for the traces I supplied in my earlier posting.

 

John Atkinson

Editor, Stereophile

Can we  assume that after Jim's  "illegal"  impulse this is the 'approved' one?

 

My impulse is to think yours is  obfuscation.

[mansr]

3 minutes ago, John_Atkinson said:

Look more closely at the graph using the color coding for the traces I supplied in my earlier posting.

Oh, I see now the traces stop at different places. That's far too many overlapping traces in one graph if you ask me.

 

Now that's settled, I'm anything but impressed. An anti-aliasing filter that is down only a few dB at Nyquist will result in severe aliasing if the input has any content above this frequency. For the higher sample rates, this might not be a problem, but for 44.1/48 kHz, a lot of music has content extending far enough above Nyquist that aliases could easily fold back well into the audible range.

[Spacehound]

15 hours ago, mansr said:

New physics is pretty much confined to high-energy accelerators these days.

Shame there's not a downtick.

One day there might be.

[Spacehound]

2 hours ago, Shadders said:

Hi,

I am surprised that pre/post ringing has not been resolved or defined as an issue in the audio world.

 

How can such an issue be continually discussed as either it exists, or does not exist, and where and how it occurs ?.

 

If the AES was a professional organisation, then surely this aspect would have been well understood by the participants/members ? and published papers on the effect presented for the edification of all ?

 

Regards,

Shadders.

The AES isn't a professional organisation in the  generally accepted sense.

[Shadders]

7 minutes ago, Spacehound said:

The AES isn't a professional organisation in the  generally accepted sense.

Hi,

Yes - i am beginning to realise that now. I think they are doing themselves a disservice by publishing papers that contain a fair amount of pseudo science.

Regards,

Shadders.

[Spacehound]

2 hours ago, mansr said:

First of all, pre-ringing isn't a well-defined term. In fact, it is never used outside of audio. Let us thus define it for the sake of this discussion as an impulse response having at least one negative excursion prior to the (positive) peak.

 

Symmetrical impulse response plots are often centred around time zero. On the face of it, this amounts to (part of) the response preceding the input. Such a system is termed non-causal and cannot be physically realised. We nevertheless use this representation because it is mathematically convenient. Now remember, we are dealing with a time-invariant system, which means a time-shift of the input results in a time-shift of the output. Thus, provided the impulse response is finite, we can make it causal simply by shifting it such that it becomes zero for all negative time values. The only effect of this is the addition of a constant delay to the output. Importantly, the frequency response is not affected.

 

In practice, a digital filter incorporating such a time-delay is trivially constructed. An all-analogue realisation is trickier, although it can be accomplished through the use of various delay elements.

 

Have I convinced you yet?

Yes you have,, it is causal but doesn't appear so how it's usually measured and calling it 'pre' is silly.

 

But you presume too much. I was expecting the answer to come from s....k 

[Hifi Bob]

Here’s what ringing can look like with a real world signal.

The same castanet sample has been decimated from 96k to 44100 (using SoX) with successively steeper filters. I can’t see any ringing with the default filter, can just about see it with the ‘steep' filter; it’s easily seen with insanely steep filter.  In any case, ringing falls outside of the range of human hearing, so unless it causes distortion in the playback equipment, should not have any audible effect.

[Shadders]

36 minutes ago, Hifi Bob said:

Here’s what ringing can look like with a real world signal.

The same castanet sample has been decimated from 96k to 44100 (using SoX) with successively steeper filters. I can’t see any ringing with the default filter, can just about see it with the ‘steep' filter; it’s easily seen with insanely steep filter.  In any case, ringing falls outside of the range of human hearing, so unless it causes distortion in the playback equipment, should not have any audible effect.

Hi,

I cannot read the text at the top defining which response we are seeing.

 

A few questions - was the signal that energised the filter the only signal being tested, or was the castanet a sample with other sounds in play at the time ?.

 

What was the length of the filter ?

 

Let us assume that n was the number of taps, then did you

1. Remove the initial n samples from the filter output response which will have been caused by the initial conditions of the filter being zero.
2. Remove the last n samples which will be the start of the injection of zero's into the filter, since there are no sound samples left.

With the steepening of the filter order, which is an increase in the number of taps, all you are seeing is the transient response of the filter. The greater the order of the filter, then the longer the ringing, where the number of ringing samples is equal to the number of taps.

 

So, small number of taps, short transient response, large number of taps, long transient response.

 

Once the transient response has decayed, then all you see is the steady state response of the filter - that is, it will function as it is supposed to - attenuate or pass frequencies according to the filters response.

 

Regards,

Shadders.

[Hifi Bob]

46 minutes ago, Shadders said:

Hi,

I cannot read the text at the top defining which response we are seeing.

 

Perhaps download the file to make the text (and also the ringing) easier to see?

 

In any case the commands were:

 

sox castanets-2496.wav ringing-1.flac rate 44100
sox castanets-2496.wav ringing-2.flac rate -s 44100
sox castanets-2496.wav ringing-3.flac rate -b 99.7 44100
 

 

46 minutes ago, Shadders said:

 

A few questions - was the signal that energised the filter the only signal being tested, or was the castanet a sample with other sounds in play at the time ?.

 

Just castanets.

 

46 minutes ago, Shadders said:

 

What was the length of the filter ?

 

The filter lengths (if I am interpreting the debug output [with -V -V] correctly) are, for increasing steepness: 409, 2045, 6805.

 

46 minutes ago, Shadders said:

Let us assume that n was the number of taps, then did you

1. Remove the initial n samples from the filter response which will have been caused by the initial conditions of the filter being zero.
2. Remove the last n samples which will be the start of the injection of zero's into the filter, since there are no sound samples left.

I just ran the commands above.

 

46 minutes ago, Shadders said:

With the steepening of the filter order, which is an increase in the number of taps, all you are seeing is the transient response of the filter. The greater the order of the filter, then the longer the ringing, where the number of ringing samples is equal to the number of taps.

 

Well yes, but it is not the transient response per se, but the effect on a real world signal (since some have said that testing with impulses is not representative of real music/sounds).

 

46 minutes ago, Shadders said:

 

So, small number of taps, small transient response, large number of taps, large transient response.

 

Once the transient response has decayed, then all you see is the steady state of the filter - that is, it will function as it is supposed to - attenuate or pass frequencies according to the filters response.

 

Regards,

Shadders.

 

[Shadders]

16 minutes ago, Hifi Bob said:

 

Perhaps download the file to make the text (and also the ringing) easier to see?

 

In any case the commands were:

 

sox castanets-2496.wav ringing-1.flac rate 44100
sox castanets-2496.wav ringing-2.flac rate -s 44100
sox castanets-2496.wav ringing-3.flac rate -b 99.7 44100
 

 

 

Just castanets.

 

 

The filter lengths (if I am interpreting the debug output [with -V -V] correctly) are, for increasing steepness: 409, 2045, 6805.

 

I just ran the commands above.

 

 

Well yes, but it is not the transient response per se, but the effect on a real world signal (since some have said that testing with impulses is not representative of real music/sounds).

 

 

Hi,

I only have Linux OS at the moment - i checked the SoX page - seems to be Windows or Mac only ?UPDATE : Just checked is in my repo's.

 

I have Octave - so how can i download the files for the castanets ?

 

The text i wrote is stating that the transient response is a real world response. That is, a filter when energised will have an output response which is [transient response + steady state response], regardless of the signal input.

 

Your SoX example is just that, the filter output is : [transient response + steady state response]. The ringing is the transient response.

 

Regards,

Shadders.

[mansr]

3 minutes ago, Shadders said:

I only have Linux OS at the moment - i checked the SoX page - seems to be Windows or Mac only ?

As de facto co-maintainer of Sox, I can state with certainty that it runs on Linux.

[Shadders]

Just now, mansr said:

As de facto co-maintainer of Sox, I can state with certainty that it runs on Linux.

Hi mansr,

Thanks - just did a quick check, exists in repo - and have installed.

Regards,

Shadders,

[audiventory]

On 25.01.2018 at 1:32 AM, fas42 said:

Does pre ringing matter? No.

 

And there are no know me serious researches of ringing.

 

Filters with post ringing have double energy in post ringing area, comparing with pre+post-ringing filters. Thus it is not clear, that "post-ringing only" filter is better than "traditional" one. And phase response non-linearity of "post-ringing" filters is no good, theoretically.

[Fokus]

3 hours ago, Shadders said:

Even if you energise a filter with ringing at F, with a frequency F, the filter does not ring.

 

Example, a 192kHz sample rate signal, filter set to cut off at 50kHz, any signal in the frequency 40kHz to 60kHz has NO ringing.

 

Wrong.

 

Take this input signal, sampled at 96kHz. It occupies a band from DC to 47 kHz, i.e. it is properly band-limited and thus a legal signal in the 96k space. Observe its clean, monotonic leading edge.

 

 

If we downsample this signal to 48kHz with a steep linear phase filter cutting at 24 kHz we get this

 

 

Our linear phase filter has imprinted its pre-ringing at its transition frequency.

 

 

[Hifi Bob]

20 minutes ago, Shadders said:

 

 

I have Octave - so how can i download the files for the castanets ?

 

Shall dropbox the files soon, when I have a moment.