[sdiy] LFSR digital noise source

Gordonjcp gordonjcp at gjcp.net
Tue Nov 12 09:24:53 CET 2019

On Mon, Nov 11, 2019 at 03:01:20PM -0800, rsdio at audiobanshee.com wrote:
> The real problem here is all of the overhead required to get it working. A microcontroller doesn’t do anything when you first solder it to a board, or plug it into a breadboard. So, you have to buy a programmer and attach it to your prototype. Even then, you’re not done. You have to write the code, test the code, and iterate. Finally, even after you’ve finished getting all of the software bugs out, the non-volatile memory can forget your program, and you might be back to square one.

Programmer?  Buy something with a bootloader flashed onto it, like the
ubiquitous Arduino.  A couple of quid for the part and all you need to
program it is a USB cable.  There's not much to the code, and I don't
see why you'd be chasing bugs - just don't write it with bugs, the
code's almost impossible to get wrong.

If your non-volatile memory is forgetting your program you're using
defective chips, because that's kind of the whole point of the
"non-volatile" bit.

> There are certainly problems that can’t be solved without a microcontroller. That’s why we have them. But they’re not always a win compared to logic circuits that perform the correct function as soon as power is first applied, without the need for additional tools, circuit adaptors, and software development. The most expensive part of any product design is the software. If you can design something without software, you’ve saved yourself the biggest cost. Whether you’re considering a one-off project or a product that you want to manufacture, there are many functions that are best done in hardware, without software.

That's the problem though.  Logic circuits rarely perform the correct
function when power is first applied, unlike microcontrollers.  You've
got all sorts of crazy-ass latchup conditions, you've got to make sure
everything starts up and resets cleanly, and that's all assuming the
40-year-old chips you have to buy aren't defective to begin with.

Yes, you could redesign something like the Transcendent 2000's noise
source to use modern shift registers and XNOR gates to replace the
40-year-old unobtainium but now you've got a whole new set of problems
to contend with.

> > It's incredibly tweakable, without having to find long-obsolete parts
> > that were profoundly shitty when new and are unlikely to have been
> > improved by lying around for 30 years.
> There’s no reason to design with obsolete parts that have been sitting around. There are plenty of modern, in-production logic chips. They are also available in a variety of performance levels, from the slowest TTL to much faster. Many of these chips operate at 40 V peak-to-peak, which an MCU can’t possibly handle without external help.

The "obsolete parts that have been sitting around" are all dead.  They
were shitty parts when they were new (because CMOS was inherently shitty
in the 1970s) and now they're shitty parts that have suffered from 40
years of being zapped with static.

Even if you do all that, redesign it to use chips that don't die if you
look at them sideways, solve all the weird glitches and latchups, and
all the rest, you'll end up with a crappy too-short LFSR that gives a
distinct ChufffChufffChufff repeating sequence.  So now you've got to
redesign it with four times as many shift registers - way more
opportunity for it to misbehave and way more board real estate - and
you've solved one of its problems.

Just use a microcontroller.  It's the price of a coffee, it definitely
works, it needs no other supporting components (okay maybe a capacitor
on the output to centre it around 0V) and it doesn't do that Prophet 5
steam train chuff.


More information about the Synth-diy mailing list