[sdiy] Syncable digital noise source.

[Tom Wiltshire]

On 1 Jul 2008, at 18:30, ASSI wrote:

> On Dienstag 01 Juli 2008, Tom Wiltshire wrote:
>> I mentioned the noise source on the Nord Lead 2 a couple of weeks
>> ago . This allows you to reset the LFSR to a known start state. This
>> means you can sync the noise to an oscillator. According to Clavia, a
>> given LFSR sequence gives fixed formants in the frequency response.
>> They've got a nice diagram of it on page 40 of the manual (http://
>> www.clavia.se/products/manuals.htm).
>
> In case anybody is wondering, these plots are in the Nord Lead 2X
> manual, not the Nord Lead 2....

Ah, sorry. My mistake.

>> Anyway, since then, I've tried putting a reset input on my PIC noise
>> source. Then I fed a squarewave from an SH101 into the reset input.
>> The result was not worth the limited effort, in my opinion. Whilst
>> the formant-shaping produces noise with certain "flavours", it didn't
>> produce much that sounded good to my ears. Obviously you can start
>> the LFSR in many different places and get many different flavours, so
>> perhaps I didn't explore enough. I wasn't encouraged to by the
>> results I got. The sound isn't pitched enough to be really useful,
>> and as unpitched noise goes, pure white or pink have a smoother
>> quality whilst the synced noise often sounded "spattery" and uneven.
>
> That's because you implemented a real pseudo-random generator and  
> Clavia
> most likely did not - there are no strong formants even on
> sub-sequences of a good pRNG unless you make the sub-sequence very
> short, on the order of the length of the LFSR.  The reason why it  
> could
> perhaps work like described in the Clavia manual is when they use
> the "state" of the pRNG as their noise wave and not the single bit
> produced each time it is stepped.  This produces strong
> auto-correlation in the noise signal as each random bit will  
> produce an
> exponential pulse and such sequences can have strong formants that are
> related to the length of the shift register and the number of bits you
> take out for each step; the order and spacing of bits also plays into
> this.  The result is largely unpredictable, but once you find a good
> combination of parameters, they can be reproduced with ease.  It is
> also possible that in this mode they use non-maximum length sequences
> which can be coloured even more strongly.

That's really helpful to know.

  I've seen the auto-correlation effect that you mention by plotting  
outputs from an LFSR. The easiest way to see this is to use a whole  
byte of the LFSR as the output, but only shift the LFSR a single bit.  
This gives you the exponential pulses, and they are quite visible in  
the output.

So the tricks for a good synced noise LFSR seem to be the exact  
opposite of what you need for good white noise; use a really short  
register, take too many bits out of it, and don't shift it enough!  
The result will have strong formants and appalling auto-correlation,  
but might sound quite decent when synced!

Thanks, you've encouraged me to have another play with this.

T.