Alt-BEAM Archive
Message #12383
To: beam@sgiblab.sgi.com
From: Bruce Robinson Bruce_Robinson@telus.net
Date: Thu, 23 Mar 2000 07:17:18 -0800
Subject: Re: "Living Machines" question & an experiment
> Pat Wehren wrote:
> >
> > The paper says that if the times of the neurons are roughly the
> > same, they will converge on 100100. I can't for the life of me
> > see why.
>
> Who says there's no feedback/implex? Well, I did once.
>
> OK, I decided to test out this assertion as follows. I set up and ran,
> not one, but two 6Nv loops, completely independent except for a share
> power supply (6 V battery). Decoupling caps all over the place, no
> floating inputs, everything "by the book".
>
> I first selected components that had as nearly identical values as
> possible (helps when you buy resistors and caps by the hundred). Then I
> set up indicator LED's in two lines of 6, each line driven by a separate
> Nv loop, and ran one impulse around each loop just to check that the
> timing was close for all the Nv's. I could watch the two impulses
> marching along their respective rows, and by fiddling with the phase
> shift, I was able to cross-compare the delay times for all the Nv's.
>
> Having satisfied myself that the times were, in fact, "roughly the
> same", I rearranged the LED's so each loop had them set up in two rows
> of three, with the LED's for each loop physically separate (to aid
> observation). Then I introduced two impulses into each loop in a 101000
> configuration, and let them run. The two rows of three made it easy to
> observe the pattern: in a 101000 configuration the impulses would appear
> at opposite ends of a row of three, whereas in a 100100 configuration
> they would march in lock-step down the two rows.
>
> What I found after about 6 hours was that both circuits had evolved to a
> 1-0-3/4-1/4-0-0 pattern. In other words, the leading impulse would move
> ahead just a little before the trailing impulse.
>
> Twentyfour hours later, the pattern had not changed. So for the heck of
> it, without disturbing or interupting the pattern in any way, I plugged
> a low current gearmotor (10 mA) directly into one pair of adjacent Nv's
> on ONE of the loops. This completely disrupted the pattern in the
> affected loop ... it reverted back to the saturated state (101010)
> immediately. As I had not put any kind of capacitor across the motor
> leads, this almost certainly injected a lot of noise into the chip.
>
> But what was fascinating was what happened to the other loop. Mounted on
> a separate breadboard, but sharing the same power supply. In about 4
> forward-reverse cycles of the motor, the second loop shifted to a 100100
> pattern. Just like that, perfect synch, no overlapping. And it is still
> doing that, an hour later.
>
> So a noisy, low-current motor hooked directly to the Nv outputs in one
> loop affected the impulse pattern in an independent loop, with the only
> connection between them being the supply lines. Fascinating.
>
> Bruce
Home