Allpass for decorrelating L/R?

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

gyraf

Well-known member
Staff member
GDIY Supporter
Moderator
Joined
Jun 4, 2004
Messages
11,857
Location
Aarhus, Denmark
Hi Group,

I'm looking for pointers to existing designs and/or ideas about using allpass filters for de-correlating L/R information while retaining absolute levels.

I think I have seen such functionality once in a phase-meter or correlation meter - though it may have been in something peripherally related to this .

IIRC, the idea was shifting one side 90degrees phasewise, which would allow L/R addition (comparison?) without loosing S-perspective.

The application is that I'd like to be able to do a summed (common) sidechain audio for stereo compression without reverting to simple summing (which would be the same as just using the M-component, ignoring the S-component) - and I'd like to do it already in AC domain, before rectification..

Help!

Jakob Erland
 
makes me think of dolby surround - the analog predecessor  to 5.1. what are you looking for, schematics or a discussion of your concept?
for an other application I have implemented an FIR phase shift filter in a BSS Soundweb. just a bunch of FIR taps. can make it available if you need to do a quick test. analog is a bit more complicated, as far as I am aware you will need a rather large number of allpass filters to do full spectrum phase shift.

- Michael
 
90degrees filters were common for mono summing in the old german broadcast days (W58 etc IIRC)
I do have a few of those cards, but sadly without pinout or schematic :-(

Attached some info and schematic from Studer about their 90° filter.
 

Attachments

  • 1.914.533.pdf
    639.3 KB · Views: 76
Thanks, the Studer approach is exactly what I needed to know..! I really love being part of this group at times like this...!

..what are you looking for, schematics or a discussion of your concept?

I was looking for some sort of hint to where I could have seen this - spent the whole weekend trying to find what it could be..

This would be for the sidechain input(s) of a probably-upcoming G24 passive compressor, that I'd like to not overreact on M-components or underreact on S-components..

Any opponents, or reasons why it would fail in this application? (yes, I'll try it out anyway)

Jakob E.
 
gyraf said:
Hi Group,

I'm looking for pointers to existing designs and/or ideas about using allpass filters for de-correlating L/R information while retaining absolute levels.

I think I have seen such functionality once in a phase-meter or correlation meter - though it may have been in something peripherally related to this .

IIRC, the idea was shifting one side 90degrees phasewise, which would allow L/R addition (comparison?) without loosing S-perspective.

The application is that I'd like to be able to do a summed (common) sidechain audio for stereo compression without reverting to simple summing (which would be the same as just using the M-component, ignoring the S-component) - and I'd like to do it already in AC domain, before rectification..

Help!

Jakob Erland
Hint; look at the Vox AC30 vibrato section.
 
gyraf said:
This would be for the sidechain input(s) of a probably-upcoming G24 passive compressor, that I'd like to not overreact on M-components or underreact on S-components..

Any opponents, or reasons why it would fail in this application? (yes, I'll try it out anyway)

Jakob E.

Yes, you'd to choose between 90º in advance of L or R, which will overreact on one mid point between M and S and under react on the other. If you have something delayed ~90º at left and you do 90º more delay at left you'll loose it, if you add 90º delay at right you'll have it to hard. I do use small delays for putting things to one side or the other, and this would react oddly in that case.

I think doing it in "DC" (or after rectification I'd rather said) would be much more predictable, preferably M and S rectified, then summed as wanted. Is there a particular reason you are trying to avoid this?

JS
 
joaquins said:
I think doing it in "DC" (or after rectification I'd rather said) would be much more predictable, preferably M and S rectified, then summed as wanted.
+1.
There is certainly a good reason why these phase-shift boxes have become obsolete.
Although some have toyed with the idea based on
sin²+cos²=1
By squaring and summing the signal and it's 90°-shifted image, you're supposed to get the amplitude. It may work with a pure sinewave...
Now, supposing it worked, you would get a different flavour of signal describing the amplitude, but in a compressor, that is not generally what you want; most often you want an analog-or of the highest.


 
I'm trying to use this for sort of a continuous variable LR/MS compression scheme that shouldn't be too obviously ignoring the S-component.

This scheme in stead of having to rectify and time on L+R+M+S individually

But as always, the acid test will be extensive listening and analysis of user interface functionality

Thanks.

Jakob E.
 
Interesting subject. Help me understand though,  Are both channels shifted 90 deg? or just one.
If you shift both L&R 90 deg, there is no difference between L&R.  Both L&R would still be in phase.
I understand 90 phase summing equals -6dB. The Studer article seems to indicate both channels are shifted.
Did I read this wrong?
 
rlaury said:
Interesting subject. Help me understand though,  Are both channels shifted 90 deg? or just one.
Both channels are shifted, but differently, and actually the phase difference can be pretty close to a perfect 90°, even with a somewhat simple circuit.
BTW, it's as close as can be to impossible shifting a signal 90°, even with the most sophisticated digital process; the signal has to be delayed by a 1/4 wave of the lowest frequency. Only differential shift works.
I understand 90 phase summing equals -6dB.
No, -3dB.
The Studer article seems to indicate both channels are shifted.
Did I read this wrong?
No.
 
If you shift both channels 90 deg, then they are  in phase still. What gives? Both channels or just one?


abbey road d enfer said:
rlaury said:
Interesting subject. Help me understand though,  Are both channels shifted 90 deg? or just one.
Both channels are shifted, but differently, and actually the phase difference can be pretty close to a perfect 90°, even with a somewhat simple circuit.
BTW, it's as close as can be to impossible shifting a signal 90°, even with the most sophisticated digital process; the signal has to be delayed by a 1/4 wave of the lowest frequency. Only differential shift works.
I understand 90 phase summing equals -6dB.
No, -3dB.
The Studer article seems to indicate both channels are shifted.
Did I read this wrong?
No.
 
Hi,

Casual reading of this thread makes me think of an old  Studer 'shuffling' circuit (as implemented on a plug-in card-system), but I might be missing the point. Who knows it provides further search-terms though ?
 
clintrubber said:
Hi,

Casual reading of this thread makes me think of an old  Studer 'shuffling' circuit (as implemented on a plug-in card-system), but I might be missing the point. Who knows it provides further search-terms though ?


More detailed reading of this thread learned that exactly that Studer-circuit had already been posted! ;-)
 
rlaury said:
If you shift both channels 90 deg, then they are  in phase still. What gives? Both channels or just one?
As I wrote earlier, it is impossible to shift one signal by 90°. So Both channels are shifted, by a continuously variable degree, but the shift difference is contant at 90°.
See in the attachment how the phase response of the two signals are parallel, with a constant difference, while the amplitude is perfectly constant, to a fraction of dB.
 

Attachments

  • phase shifter.jpg
    phase shifter.jpg
    43.4 KB · Views: 31

Latest posts

Back
Top