Revel F208?

[RayDunzl]

Re: High passing the mains

Defining, in this case, a digital clip as two or more consecutive full-scale samples...

I've noticed, on some highly engineered recordings, that if some of the low frequency content is removed (say, by a digital crossover), what's left exhibits clipping, where the contribution of the low frequency wave had pulled the high frequencies away from the limit.

Not having seen that mentioned elsewhere, nor any admonitions to attenuate the original data before crossing, I thought I'd throw it in.

Here's an example:

No clips (as defined above) in the original track.

The second track is a copy of the first, high passed in Audacity at 80Hz and 24dB per octave, simulating the signal to be forwarded to the mains DAC.

 

[Blumlein 88]

Re: High passing the mains

Defining, in this case, a digital clip as two or more consecutive full-scale samples...

I've noticed, on some highly engineered recordings, that if some of the low frequency content is removed (say, by a digital crossover), what's left exhibits clipping, where the contribution of the low frequency wave had pulled the high frequencies away from the limit.

Not having seen that mentioned elsewhere, nor any admonitions to attenuate the original data before crossing, I thought I'd throw it in.

Here's an example:

No clips (as defined above) in the original track.

The second track is a copy of the first, high passed in Audacity at 80Hz and 24dB per octave, simulating the signal to be forwarded to the mains DAC.

View attachment 25248

Could you show the same before and after with the spectrogram view. Maybe in Log. Wondering if some other low frequencies are what clips it afterwards.

 

[RayDunzl]

Could you show the same before and after with the spectrogram view.

I looked, couldn't see anything. Try it on some "compressed" track you have. This one was from Random Access Memories (highly engineered).

However, this is what happens.

You can see the tracks are aligned by the markers on the high frequency, and just after, the elimination of the lowest frequency almost leaves the signal driven to the other rail...

Top - original
Bottom - high pass 80Hz 24dB/octave

I'm sure the result would be different depending on the chosen crossover frequency.

Let's see:

Top, original, bottom 50Hz high pass 24dB/octave. Same location in the track.

 

[LTig]

Re: High passing the mains

Defining, in this case, a digital clip as two or more consecutive full-scale samples...

I've noticed, on some highly engineered recordings, that if some of the low frequency content is removed (say, by a digital crossover), what's left exhibits clipping, where the contribution of the low frequency wave had pulled the high frequencies away from the limit.

Not having seen that mentioned elsewhere, nor any admonitions to attenuate the original data before crossing, I thought I'd throw it in.

Let me add something I wondered about recently: when I extended my DIY signal generating program to create the 31 multitone signal as used by AP (thanks to Amir for the info) I realized that the highest peak of the sum of all 31 sinus waves was much lower than I had expected (some 8.x dB if I remember correctly). I had chosen the ampliude of the indiviual sinus waves to be 0dBFS / 31 and wondered that the summing signal didn't come anywhere near the 0dBFS limit. I did not yet investigate deeper into this, but I shall do it.

This means that it is possible to create a broadband signal which is not clipped but will clip when parts of its spectrum are removed by a filter. Strange.

 

[RayDunzl]

Hmm...

Showing original, 50Hz high pass, and 50Hz low pass

Something doesn't make sense.

That little bit of sub-50Hz wouldn't account for the difference.

Is there a big phase shift/delay?

Zooming out, same area in the center. The big hump here is the one above.

Beats me...

Bedtime.

 

[Soniclife]

Audacity is amplifying the filtered result, back to full scale by the looks of it. Look at your red mark, it's louder after than before.

 

[FrantzM]

I love ASR

 

[Laserjock]

Last several posts in wrong thread?

 

[RayDunzl]

Last several posts in wrong thread?

Maybe the bossman can move them. It was an earlier post in this thread that got me all excited. Didn't realize how old it was.

My bad.

 

[RayDunzl]

Audacity is amplifying the filtered result, back to full scale by the looks of it. Look at your red mark, it's louder after than before.

I would not say amplification. The higher frequencies "ride" lower frequencies. The underlying lower frequency wave structure has changed, moving that section to a different level.

If amplification occurred, the marked spikes themselves would be bigger, but they aren't, just relocated.

 

[Soniclife]

I would not say amplification. The higher frequencies "ride" lower frequencies. The underlying lower frequency wave structure has changed, moving that section to a different level.

If amplification occurred, the marked spikes themselves would be bigger, but they aren't, just relocated.

https://forum.audacityteam.org/viewtopic.php?t=95534
More complicated than I guessed, it's that pesky phase stuff again.

 

[RayDunzl]

More complicated than I guessed, it's that pesky phase stuff again.

Experiment:

1000 and 100 Hz waves of .5 amplitude. One sample marked near zero cross in each.

Tracks mixed. Zero crossing marks don't move (expected).

Mixed track high Passed at 500Hz 24dB/octave. The 1000Hz wave appears to have moved earlier in time. The first wave loses some amplitude to "shorten" it in time.

Mixed track low passed at 500Hz 24dB/octave. The 100Hz wave has moved later in time. The marked sample isn't there (filtered out) but you can see the lateness of the zero crossing.

No clipping expected or seen, as the sum of the waves at any point is less than or equal to an amplitude of 1.

 

[Soniclife]

Experiment:

1000 and 100 Hz waves of .5 amplitude. One sample marked near zero cross in each.

Tracks mixed. Zero crossing marks don't move (expected).

Mixed track high Passed at 500Hz 24dB/octave. The 1000Hz wave appears to have moved earlier in time. The first wave loses some amplitude to "shorten" it in time.

Mixed track low passed at 500Hz 24dB/octave. The 100Hz wave has moved later in time. The marked sample isn't there (filtered out) but you can see the lateness of the zero crossing.

View attachment 25270

No clipping expected or seen, as the sum of the waves at any point is less than or equal to an amplitude of 1.

Very nice.

So what happens in something like a mini DSP where causing clipping would be a bad idea? Is it best to attenuate first?

 

[RayDunzl]

Let me add something I wondered about recently: when I extended my DIY signal generating program to create the 31 multitone signal as used by AP (thanks to Amir for the info) I realized that the highest peak of the sum of all 31 sinus waves was much lower than I had expected (some 8.x dB if I remember correctly). I had chosen the ampliude of the indiviual sinus waves to be 0dBFS / 31 and wondered that the summing signal didn't come anywhere near the 0dBFS limit. I did not yet investigate deeper into this, but I shall do it.

I did the same thing a while back with 88 tones (piano keyboard frequencies) with 1/88 amplitude.

Because all of the tones never peak at the same time, the peak amplitude is less than the planned for maximum.

The highest level was at the beginning of the track, where the closest alignment occurred. Even so, the low frequencies have not yet begun to contribute much, so it's still far from full scale.

Later, it looks like band-limited white noise:

 

[RayDunzl]

So what happens in something like a mini DSP where causing clipping would be a bad idea? It's it best to attenuate first?

I think that's what we're looking at, which, in terms of active crossover, I've never seen mentioned or suggested, Not to say it hasn't.

I often re-invent some obvious wheel and somebody finally comes along and says "So what. We know that. Everybody knows that."

 

[DonH56]

If all the tones are random in phase, or frequencies chosen correctly for the record length (a.la. IEEE standards), then at some point the waveforms will all line up and peak at full scale. Otherwise, maybe not... Have to see when all the phases line up and that may be rarely. It is also quite possible to set up a set of tones from a DAC (so deterministic in phase) that never hit full-scale. That is one of the problems the IEEE standards and others strive to avoid by defining test frequencies based upon sample rate and record length to ensure all codes are hit (assuming you use a long enough record length, natch). For ADC/DAC testing we always worked to generate test frequencies ensuring every code was touched by the test but with as small a record length as we could get away with -- which was not always that small. Using the standards also obviated the need for FFT windowing since continuity at end points was guaranteed.

 

[DNM]

If Audacity filters are minimum phase then isn't clipping to be expected without some headroom management?

MiniDSP have a degree of headroom in their DSP's IIRC, as does Roon (User defined) with its PEQ.

 

[LTig]

If all the tones are random in phase, or frequencies chosen correctly for the record length (a.la. IEEE standards), then at some point the waveforms will all line up and peak at full scale. Otherwise, maybe not... Have to see when all the phases line up and that may be rarely. It is also quite possible to set up a set of tones from a DAC (so deterministic in phase) that never hit full-scale. That is one of the problems the IEEE standards and others strive to avoid by defining test frequencies based upon sample rate and record length to ensure all codes are hit (assuming you use a long enough record length, natch). For ADC/DAC testing we always worked to generate test frequencies ensuring every code was touched by the test but with as small a record length as we could get away with -- which was not always that small. Using the standards also obviated the need for FFT windowing since continuity at end points was guaranteed.

Is there a link to this IEEE standard?

 

[amirm]

Is there a link to this IEEE standard?

Not a free one. It cost good bit to get the doc. That is one of the things I don't appreciate about IEEE. People donate their work for free and they put it behind bars wanting lots of money for them.

 

[RayDunzl]

If Audacity filters are minimum phase then isn't clipping to be expected without some headroom management?

Apparently so... though, I don't remember it being discussed (I don't get out much), or recommended. I just see "Throw a crossover in there and relieve your mains of some strain of producing lows." No memory of "But watch out! You might end up with clipping".

MiniDSP have a degree of headroom in their DSP's IIRC, as does Roon (User defined) with its PEQ.

Only if you set some. I have a miniDSP OpenDRC-DI, which could be used as a one in -> two out digital crossover. The control software allows you to set attenuation on the input.

Also a miniDSP 2x4HD, same situation.

On the other hand, you could probably ask a roomful of people that aren't DSP Engineers whether their filters (or any other part of their system or environment is "minimum phase" and not get a knowledgeable answer.

I fit into that sub-species.

"Minimum phase. In control theory and signal processing, a linear, time-invariant system is said to be minimum-phase if the system and its inverse are causal and stable. The most general causal LTI transfer function can be uniquely factored into a series of an all-pass and a minimum phase system."

Uh, let me ask my brother.