Skip to main content
Topic: I have a crazy question about anyone seeing 90deg or hilbert in olden days. (Read 207 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

I have a crazy question about anyone seeing 90deg or hilbert in olden days.

Some of you probably know that I am working on audio processors -- and I have run into an odd situation.  I am finding that some processing is working better (with simple stereo -- not quad or explicitly matrixed stuff that I  definitely know of) if I apply hilbert transforms in some of the signal streams -- similar to this:

IN -- process -- hilbert/90deg -- process differently -- hilbert/90deg -- OUTPUT

Mathematically & technically - I know exactly what is going on, but it seems to be odd using the hilbert/90deg filter like this back in the 1970's.   I understand using 90deg for quad encoding -- maybe I am seeing some stealth use of a SQ/QS quad encoder or decoder for sound effect of some kind.

The difference in quality using the two hilbert stages instead of simply passing the signal through is singificant (like 20-30%, not light and day.)   I am suspecting that it might be related to the peak detector and waveform shape differences -- (the processing is using the signal level information.)   Hilbert maintains same RMS signal, but with different phase and waveform.  Somehow this is beneficial.

Just wonder if anyone who saw/involved 'back in the day' heard of this kind of thing.   Otherwise, I would guess that it has something to do with DSP vs. analog differences that I haven't seen before.  (The stuff that I am doing is NOT in textbooks or in most research articles.)

Just wondering -- and interested in any input.


Re: I have a crazy question about anyone seeing 90deg or hilbert in olden days.

Reply #1
Since I haven't gotten any responses so far -- I am going to publically hazard a guess as to what is going on -- and it is pure guesswork since I didn't know how the music was originally encoded.  Will also give some background:

I noticed that the amount of compression in some music (obviously now DolbyA) might have been done twice in a row.   It seems like the decoding is generally better if done L+R then M+S, but again, I don't know how it was originally recorded/encoded.  L+R then L+R also works, but has a few more disruptions in the sound.

Specifically, the song is 'SuperTrouper' and other songs from ABBA Gold.   They seem to have done some kind of dual encoding -- maybe for effect.  Almost every commerical copy of SuperTrouper has these disruptions in extreme at the beginning (if you listen for the disruptions then you can do nothing but hear them everytime the song plays in the future), and I am seeking to remove them from a pristine copy coming from a South American release of the music (which was apparently not DolbyA decoded.)

My GUESS as to the improvement using the hilbert/90deg filter technique is related to the problems of undoing two compressions in a row is that there can be artifacts that compound.   These artifacts tend to be concentrated mostly during the gain attacks (and some during the decays), and can partially be represented as a kind of intermodulation that produce sidebands.  The time domain nature of these artifacts are relatively sharp transients of one kind or another.   I have definitely improved the situation by slowing down the attack/decays to a reasonable bandwidth,  but the very worst problems still persist until using the 90deg filters.

Details of my guess:
When serially decoding the signal twice, these transients that happen from the first decoding will cause misbehavior on the second decoding.   When doing the hybrid thing, it is spreading the transient waveform and changing the sharp edge into a more controlled square wave type shape (akin to clipping before transmitting with a single sideband transmitter causing sharp edges, even though the modulation was square wave.)  So, there was not likely a use of a hilbert/90degree filter during encoding, but the filter tends to mitigate some of the damage done by the serial decoding.  Of course, after processing the 90deg phase shift signal, then I convert it back by doing a 90degree shift again (thereby causing a total 180deg shift or inversion.)

The above is the wild-a** guess that I have done so far.   I don't know what is really going on, but I am seeing a very worthwhile improvement.

Anyone have any ideas?


SimplePortal 1.0.0 RC1 © 2008-2018