AW: vocoder dim

[Haible Juergen]

>I'm just designing my 20-Channel Vocoder
>(only Spice-Simulation up to now).

>I'll use 10th order Bessel bandpas-filters (analysis),
>60dB/oct (my tables end at 10th order ;) )
>with 10 op-amps and 20 caps.

Oh well. 20 caps. 20-pole filters. It's easy to do this in 
"Spice-virtual-reality",
isn't it ?? Reality is a bit different, I fear.
Ok, you can surely *build* something like this as well (I just doubt 
it would fit
into a 4 U 19" enclosure ...), but don't expect too much advantage 
over a
smaller design.
First of all, you can get much sharper slopes than 6dB/8ve per pole 
(not in the
infinity, of course, but near the resonant peaks, i.e. where it makes 
a difference).
The art of building a vocoder is a clever *matching* and *staggering* 
of the
different bp filters in one channel *and* in neighboring channels, to 
get steep
slopes for the individual bands without too much phase cancellation 
between
neighboring bands. It's always a tradeoff - and there are no rules for 
a musical
optimum.

Now let's see: With a 2pole design (Paia ...) you can't do much wrong: 
You always
have enormous peaks and spaces, and you can simply choose between 
good
speech intelligibility (high Q) and smooth overall response (low Q). 
Good news
here: You just can use 5% components- can't do much wrong.

With 4pole designs (EMS 2000, Moog, Roland ..) you have the chance to 
make
the bands both broader at the top and steeper at the edges, than with 
a 2pole
design. Here you already have different possible ways to go: resonant 
LP/HP design
(as in the EMS Octave Filter Bank), or staggered BP design. You have 2 
degrees
of freedom: spacing (amount of staggering) and Q of both filters. You 
can either
build such a thing with standard components (1% resistors, 5% caps) 
*and* a decent
number of trimpots, or with fixed components that are selected to 1%.

6pole designs (Sennheiser) are far, far more difficult to calculate. 
Degrees of freedom
are (1) staggering of center bp to the opper and lower bp (speaking of 
one channel),
(2) 	Q of center bp, (3) Q's of upper/lower bp. I have (and friends of 
mine have) spent *weeks*
to try out different variations of this, and the effects of component 
tolerances, with simulation
tools. Don't expect to build such a thing from 1% components without 
*excessive* trimming
orgies ! If you look into Sennheiser service docs, you won't find 
component values for the
filter R's and C's, for good reason: They were hand-adjusted by using 
some E96 resistors,
and soldering an individually calculated 2nd E96 resistor in parallel 
!

What shall I say about 8pole designs? It's hard to imagine that one 
could go on with the same
optimisation process as described so far. The only 8pole which I know 
from its interior (the largest
Synton model), uses 8pole filters that don't really use the 
theoretical possibilities an 8pole
design could give. No more staggering (just 4 low-Q filters with the 
same center frequency), but
still increadible effords to match the components. This compromise 
gives a similar result as
Sennheiser's optimized 6pole technology. So Synton spent a few *more* 
components to be
able to use a much smaller number of *different* components. (I did 
the same, when I started
my - yet unfinished ! - 22-band 8pole vocoder. And I spent many, many 
nights with the multimeter
just measuring and selecting capacitors to 0.5% tolerance.)

Still want to build a 20pole design ?? (;->)


>Ok, the Bessel has not so sharp edges,
>but less "ringing" than Butterworth or Chebychev,
>so I think this will be optimal.
>
>Does this make an audible difference ? Do you agree ?

Yes. Not sure, which amount of ripple is the most "musical" one.


> So it'll be also possible to bypass some bandpass-filter
> stages to get not so steep responses (6bB, 12dB, 24dB & 48dB).
Ø	This will give "cheaper" vocoder sounds.y

As I said, most of the "cheaper" designs that "only" use 4poles, have 
these
filters optimized, so they are very different from a (partial) filter 
in a higher-order
vocoder. Exception: Korg VC-10. This beast actually has 20 bands, but 
they are so
broad, that it isn't considered very well sounding, and many people 
don't believe
that it actually has a large number of bands.


>I'm currently thinking about an bargraph led display of the
>analysis input (one for each channel) with National
>bargraph driver ics.

Very good idea.

>After rectifying the filtered analysis input the signal must be low
>pas filtered to get the synthesis control voltage.
>What kind of lp filter is recommended and what
>is the best cutoff frequency relative to bandpas
>frequency ?

That's difficult to say. Fast is good for speach, but bad for Pad 
sounds
(in terms of distortion). Distortion figure calls for a different lp 
freqency
for each band (most vocoders do that), but I think I have also seen 
designs
which have almost the same lp frequency for each channel's envelope
detector. (Don't remember which one.)
Maybe the best is to make it rather fast (relative to the bp 
frequency), and
add a slew control (see below) to make things smoother for Pad sounds 
etc.
BTW, the Paia only has 1pole lp filtering on the envelopes (very 
clever
use of a NE570 compander IC for rectifier, lp filter and VCA in half 
of a single
chip per channel !). This surely causes side effects, but only the ear 
decides
if they are pleasant or not.

>Some EMS Vocoders have slew regulation for rising and falling
>edge of the analysis detector simultaneously with a single pot
>for all channels. Interesting feeeaaatuuuure.
>How ?
>Voltage controlled RC slew regulation with junction fet as "R" ?
>They also have a freeze feature.
>Could also be done with "RC"-fet as sample & hold.

Haven't seen EMS vocoder schematics, but as the Powertran vocoder has 
the same
feature and was designed by the same person, it's easy to guess that 
it might be done
in the same way.
There's a PWM oscillator which controls an electronic switch in each 
channel, so a variable
R in an RC lowpass is simulated. Freeze is also simple then: Just set 
the PW ratio to
0%, and you have a Sample and Hold in Hold state.


>Any information would be helpfull.

So finally I have written several bits about vocoders together in one 
posting - sorry if it
is too long, and sorry if I discouraged you with your ambitious 
20-pole project.
Hope you found it interesting, nevertheless.

JH.