EH 16 second dly

[Haible Juergen]

	>Some DDLs used Delta Modulation, based on assumption you have high
	>sampling
	>rate and "slow" signal (audio) so next sample can be the same as
	>previous one
	>or one "step" up or down, so you need only 1.5 bit resolution. 

1.5 bit ??

Seriously, the encoding was done with 1 bit, but, as you said, with much
higher
sampling rate. Similar to the "Bitstream" converters that are everywhere
nowadays,
but the method (Delta Modulation) was a predecessor of today's
Sigma-Delta-Modulation.
Main difference is that the information that was encoded was the difference
between
two adjacent samples instead of their absolute values.

	> based on assumption you have high sampling
	>rate and "slow" signal (audio) so next sample can be the same as
	>previous one or one "step" up or down

The art of building a good Delta Modulation system was choosing the right
step size.
And changing it continuously, which lead to the name "Adaptive Delta
Modulation".

You could also use a fixed step size if you made the sample frequency high
enough.
But then you'd get a bit stream not smaller than a serialized 16 bit (or
whatever)
encoding - or you'd get audible errors. These errors are of the nasty kind,
as they basically
affect the slew rate of fast transients - a 741 opamp would sound good in
comparison.
Adapting the momentary step size to the current program material would
reduce this
error. But ADM is not ADM - there were algorithms that were way different.
The tempting thing about even the cheapest implementation of ADM was that it
had
a very high "static" SNR at full (small signal) bandwidth ("no brickwall
filter"). You
almost had a studio quality SNR and bandwidth as long as you ran signals
without 
fast transients thru the device. I once built a ADM delay according to an
article in ELRAD
(which had a decent algorithm, though not as sophisticated as the best of
Deltalab's),
and it was wonderful for pad and strings sounds. No need to set an input
level ... virtually
"infinite" dynamics. But run a DX-7 piano thru the same device, and you get
nasty
aliasing. It was better with reduced input level (not sure why that is !),
but the signal
would never be clean.

The Deltalab Effectron I which I own today sounds much better. I did a
patent search
on "deltalab" recently, and up came a multitude of patents, with slight
improovements
of the algorithms for each one - all quite impressive.

I think this is only of historical interest today. Memory is cheap, so why
bother to
use complicated modulators at all. ADM is more or less a DM with intrinsic
compander.
using a higher sampling rate would get rid of the compander.
Higher sampling rate would also mean higher memory clock rate - possible
today,
but not the most economic solution. So use an improoved  1-bit encoder and a
digital filter 
to calculate 16 bit words - and you have a sigma delta ADC and you're back
in the standard
PCM format. The wheel re-invented (;->).

JH.