[sdiy] digital delay, was ...

[harrybissell]

Hi Glen....

wow.  too much for inline :^P

I think the compander time constants are a problem not for each individual
band... but for the difference between (lets say) the 40Hz band and the
10KHz band.  The 40Hz band will need a time constant in the hundreds of
milliseconds range... the 10K might only need milliseconds.

Also... the compander around a BBD is a bastard (ie trade off)... the CV does not
travel though the time delay along with the audio signal... the CV are tied
direct. MOST
folk don't care... at least the BBD is quieter.  To do it right, you need to delay
the
CV as well.

Pitch shifting : holy grail from hell ???   I wished to use a pitch shifter to
bring the
frequency of guitar up to where I could P/V convert it... but alas... the dome
filters
for the 80Hz low end signal added up to about 16ms of group delay... far to much
to make the
tracking any quicker.  I suspect this system would suffer from the same problems.

I agree... this is 'pie in the sky' conjecture... probably none of us would commit
to trying
it in real life (much less with tubes :^) because the complexity would be
prohibitive... by
that time, the best A/D and D/A in the world and a clock in the GHz and no
anti-alias
filters start to look not only good, but a price bargain  :^P

I agree... it would have a character all its own.  At dinner tonight, my family
was discussing
Phil Spector... who is credited with trying to fatten a vocal by duplicating a
tape, then trying
to record a copy in sync with the original master. The song was called "The Big
Hurt" and
was probably the godfather of all the BBDs in the world (flanging)... So you might
accidently
come up with a new effect and then people will talk about YOU at dinner instead
:^P

H^) harry

Glen wrote:

> At 10:27 PM 9/12/03 , harrybissell wrote:
>
> >The trade off is always ripple vs response time.  The high frequency signals
> >could
> >have very acceptable ripple / response time... but the low frequency signals
> >would
> >have to use MUCH longer times, or have undesirable artifacts in the sound (IM
> >
> >distortion)
>
> I'm still not sure how the companders would be such a problem. Let me see
> if we are thinking of the same sort of system here:
>
> First, we filter the incoming signal into different frequency bands,
> creating separate channels for each frequency band. I suppose this would
> probably be similar to the filters used in active crossover networks,
> perhaps with steeper slopes.
>
> Then, we place pitch shifting devices after each set of filters. Each
> frequency band will be both spectrally relocated and spectrally compressed.
> The low frequency bands will become much higher in pitch. The high
> frequency bands will become lowered in frequency. Everything will be moved
> somewhere toward the middle of the audio spectrum. At this point, there
> will no longer be any high frequencies or low frequencies in the signal flow.
>
> Then, we compress each channel with the first half of our compander. When I
> said that the companders were tuned, I meant that each compander was tuned
> for the spectrally-altered narrow-bandwidth signal it was meant to process
> in this application. Since there are no longer any low frequencies or even
> any very high frequencies, much less of a compromise will need to be made
> in the response time of the compressors. Each of them might even use the
> same circuit. The main thing is that we wouldn't have to have a more
> sluggish response typical of a compander that had to deal with low
> frequencies on a regular basis.
>
> There should also be other advantages to the narrow-band mid-range signals
> we are processing at this point. Since we aren't handling any high
> frequencies, we can also use somewhat lower clock rates for longer delay
> times--without worrying about getting too close to the Nyquist frequency.
> We can also use somewhat larger capacitors in the feedback paths of any
> op-amps we are using in this area, without worry about their effects on
> high frequency audio performance. This should help a little to cut down on
> some higher frequency noise. I'm sure there are lots of little things we
> could enhance, with the knowledge that we no longer have to process a
> wideband audio signal.
>
> Next, we delay each channel with BBD devices, which are all synchronously
> driven from the same master clock. There is also the possibility here to
> use multiple BBDs, in parallel, for each separate channel. This would also
> help improve S/N. I've seen some application notes referring to running two
> of them in a balanced-audio  fashion, for example. I think this was largely
> an attempt to attenuate clock signal bleed through. Simply having two or
> more of them in parallel will help reduce the effects of thermal noise,
> even if a balanced-audio design isn't selected. All sorts of parallel and
> series-parallel combinations are possible, including the extreme of
> multiple balanced-audio circuits in series-parallel.
>
> After being delayed, each channel is then re-filtered to remove residual
> clock noise, out-of-band thermal noise, and whatever other out-of-band
> noises that might have accumulated by this point.
>
> Then all the channels are expanded with the second half of our compander
> circuits. Once again, the narrow bandwidth of the signals they will process
> should make their task easier. This restores the dynamic range of the
> original signal. As a byproduct, the expansion also makes our preceding
> filter stages seem to have steeper slopes. This should also be beneficial.
>
> After expansion, each channel is then pitch-shifted back to its original
> spectral location, restoring the spectral range of the signal.
>
> Then, we merge the channels together into a single, full-bandwidth signal.
>
> It's all rather simple, actually. (Did I really say that?)   :)
>
> If anyone ever built this thing, it would certainly have a "character" all
> its own, and possibly a shot at a Guinness world record. Handcraft all the
> BBDs with legions of vacuum tubes, and I think that some sort of Guinness
> world record becomes a sure thing. I'm just not sure what you would be
> setting the world record for.   :)
>
> later,
> Glen