[sdiy] Non maximal-length LFSR

[Tom Wiltshire]

This seems unlikely to me.

It was my understanding that maximal length sequences pass through all the possible shift register values once and only once (after all, how could it produce a value twice and not repeat the next part of the sequence). The zero or max value is the only missing value.

Since all maximal length sequences produce the same values, and are similarly distributed, I don't see how one could sound significantly different to another. The distribution is key, since a rising binary count also produces all the output values, but it doesn't sound remotely random.


On 3 Mar 2016, at 09:10, mskala at ansuz.sooke.bc.ca wrote:

> Not all maximal-length sequences are created equal.  Using a polynomial
> with very few taps may mean compromising the quality of the sequence in a
> way other than shortening its length.  It'll create a simple correlation
> structure between the sequence and shifted versions of itself.  When
> people apply LFSRs to semiconductor testing and error detection, they seem
> more pleased with the sequences that come from LFSRs with more taps, and
> I've read some interesting research suggsting cellular automata may
> produce nicer sequences for semiconductor testing than LFSRs even though
> they are in some sense mathematically equivalent (same sequence length,
> same polynomials under the covers).  Whether this will make a difference
> for audio I don't know, but it wouldn't surprise me if you built a two-tap
> register and a four-tap register, listened to them both, and found that
> the four-tap register sounded better.