[sdiy] Random design, was Re: When vintage stuff comes as an ideal....

cheater00 . cheater00 at gmail.com
Sat Jun 14 16:16:30 CEST 2014

On Thu, Jun 12, 2014 at 3:40 AM, Dan Snazelle <subjectivity at hotmail.com> wrote:
> I know more than one person who believes that THE ELUSIVE VINTAGE SOUND comes from IS BECAUSE OF: wide tolerance caps and carbon comp 5% resistors, And also from (which seemed weird to me without data to back up its effects) the widespread use of HAND DRAWN, round,  WIDE pcb traces on single layer or dual layer pcbs...quite often without solder-mask.
> but then, my internal reply to this is always, "variation among units is not the same as making ONE prototype sound amazing"
> whether or not they could get the repeatability (and longevity of sound as caps age over decades)
> of modern manufacturing methods seems to me more an answer as to why vintage analogs always differ
> i assume anything built in ( relatively small numbers ( compared to a dx7 or an ipad) for niche markets....weird musicians, djs, etc will have a built in mystique or even "energy"/magic/etc
> i dont know if  its crazy to think super fans 35 yrs from now will be arguing over how todays products " got that sound "
> of course todays companies are battling OVER-hype
> over-exposure
> other than HARDCOPY patents, magazine articles, or obscure books....or maybe clubs or college music departments....i dont think a lot of todays Common knowledge about filters/ vcos/etc was so common back in the days when the legends were built
> so thats strike one against modern companies !!
> when people DONT KNOW, their minds fill in
> but when even non sdiy types ask me if a circuit of mine is based on xxxxxx OR based on such and such part, no matter how i answer if I choose to answer -- it will already have influences the way they will perceive/hear it !! before they even turn it on for the first time!!
> imagine the killing youd make if you went back in a time machine to 1961 armed with your 2014 synth designs, and a tractor-trailer truck full of OTA's, LEDS, and oh yeah, tl072s!!!!!

I think you have some points right and some points wrong.

In my opinion, "vintage sound" is attributed 100% to survivor bias:


Here's the process of an item "acquiring" the "killer tone" as I see it:

1. 10 000 units of the model have been produced.
2. the first owner of each buys it. If he's a good musician, and the
thing doesn't sound so good, he either has it hotrodded or sells it
3. After the thing becomes "old", the owner has one of the following
options. For example, think what you'd be doing with a fairly
low-level Akai sampler that was taking up space in your garage today:
a) keep it and keep using it
b) keep it and store it
c) sell it on
d) throw it away

If you own a thing, it'll usually go from a to b to c or directly from
b to d. It's unlikely it'll climb back from d/c to b or even a. (Here
we're assuming the steps c and d actually take time during which they
might be aborted, e.g. you decide not to sell the thing after all).

4. If the thing was sold on and not thrown away, the next owner keeps
on using it. He might have it serviced, hotrodded, etc, so basically
we're looping around to point 2 again. After which point 3 and point 4
again come.

5. After several iterations of points 2-4, this model of equipment has
gone through an amazing amount of sieves and improvements. The units
that are not in the landfill have had to defend themselves many, many
times. They haven't acquired "killer sound" - they have always had it
- but the ones that didn't stopped existing. Eventually, all that's
left is the ones with "killer sound", and the ones that didn't have it
turned to oblivion.

6. At this point, in 2014, someone who hasn't used this unit back in
the day picks up one of those carefully selected and hotrodded units,
and goes: "wow all of them must have had the same sound!" However,
that's a false generalization. All the units you can find might have
the "killer sound" but this does not mean that all units produced had
"killer sound". Large variance meant that some things were great (and
survived) whereas others weren't (and were destroyed).

I haven't touched on the subject of items acquiring cult status, which
can skew survivability. There is no rule as to when in an unit's life
cycle such "cult status" is achieved, therefore it escapes logical
reasoning and vast generalizations such as the above.

The steps above hold for any antique items, be it synths, furniture,
books, or uranium glass bowls.

Now you might ask if the same will happen with gear nowadays. The
answer is a resounding "NO". Nowadays gear falls into one of three

a) it has "killer sound" built in
b) it doesn't have "killer sound", but it can be modded to have "killer sound"
c) it doesn't have "killer sound" and never will.

The class of products in (b) is very small. It's limited to fully analog items.

The class of products in (c) is very large.

The steps I described above with selection through subsequent owners
only happens if you have large variance in production. On the other
hand, all production methodologies try to minimize variance between
units, usually achieving virtually zero differences. Current
semiconductors and passive devices are very stable and well
controlled. The "antiquification" of items can only happen if they
have large variance. You may hit a pretty good sweet spot of you do a
production yield that is tightly controlled, but you will only hit
"out of this world killer sound" unknowingly, by allowing a random
error in your production. This random error just might nudge the
circuit into an even better "sweet spot".

Quite often the random error which creates "killer sound" will be very
large: maybe an envelope will sound amazing because the sound ends up
being low-pass filtered or something like that due to an off-spec
capacitor. Or maybe a filter will oscillate at a very low frequency
creating a beating DC offset which makes the sound acquire a slowly
changing non-linearity. Random processes in design are an important
design consideration. In one of my topics of interest, which is
computer programming, design using random processes yields amazing
results which have not been available by any other means. For example,
random code deletion or branch removal. Quantum annealing in
rendering. Randomized service breakage (Amazon Chaos Monkey). Random
testing (QuickCheck). Evolutionary automatic code generation. Car

Most of those will not be possible in analog electronics because it
would take much too long to create randomized units; a computer can go
through thousands of randomized versions of a test subject in one
second; it takes thousands of seconds to randomize one subject in
electronics. I would however be interested in how people can imagine
allowing random design in their products. Using large tolerance
devices is not the right way to do it anymore - a 20% capacitor will
in fact be between +19.5% and +20.5% of the value (or -19.5% and
-20.5%), this is for reasons of maximizing binning yield at the
manufacturers houses.

Perhaps we could start spraying/painting our boards with conductive or
capacitive lacquer or paint which can be easily washed off. Spray a
random pattern on your filter and listen to how it changes. Wash,
rinse, repeat - quite literally. Once you hit something interesting
try to incorporate that in your design. Kinda similar to the
(reportedly made up) design story behind the 808 cymbal. Maybe we
could use something alcohol-based, which would evaporate on its own.

The second way I see: Current devices still don't have perfect
temperature stability. Maybe applying a hot air gun, at around 150C,
would be a good idea. Heat it up, let the components drift, see where
that leads you.

There's also the option of making your own devices. Artisan FETs made
from photo diodes. Artisan memristors. Artisan cat whisker diodes.
Artisan carbon composition capacitors. But that is a very slow process
and might not always yield good results; the great thing about it is
that it'll never yield the results you want.

Sometimes designers are smarter than they should be, expecting that
specific, tightly controlled values will yield best performance,
whereas in reality hitting just off to the side of the bulls eye.

One way of randomizing design is opensourcing it. Look at the x0x b0x.
A thousand people have built it, with slight variations. Some have
learnt from it. Each time, they experimented. Soon, discussion was
born, and a knowledge base was built up on what minute enhancements
can be made to the vanilla project.

I don't know of any other ways of randomizing design. Do you?

More information about the Synth-diy mailing list