Ian, et al, 2010/8/10 Ian Parker <ianpark...@gmail.com>
> > A μ sec is nothing even when we are considering time critical functions > like balance. > Not true!!! What most people miss are the stability requirements for closed loop control systems. These are crystal clear in an analog world, but some (many?) people think that they don't apply in a digital world. They do. While the thing you are controlling may live in a millisecond world, you can't have a sharp frequency cutoff point, or the system absolutely WILL oscillate at that frequency for complex reasons. If you don't understand those reasons, then you shouldn't be working in this area. So, how sharp is "sharp". The rate of roll-off must be <12db/octave, as 12db/octave represents a 180 degree phase shift, and any more makes it positive feedback. This must continue to high frequencies, at least past the point of "unity gain", meaning that any positive feedback dampens itself out because there is less than unity gain. Lets take an example. Suppose your feedback system senses things and administers corrections, the corrections being, say, at a gain of 1000 times the error, and you want the feedback to work >1kHz. Rolling off at 6db/octave to stay away from the 12db/octave unstable point, means that the permissible unity gain point is 1MHz. Of course, for really precise positioning you might want higher gain and frequency stability, which pushes you above 1MHz. Note that delays look just like phase-linear low-pass filters, and indeed some electronic designs use phase-linear low-pass filters to achieve short delays. Hopefully you noticed "low pass" here, namely, delays introduce their own VARIABLE phase shifts so that at higher frequencies, they become positive feedback and fundamentally unstable. IMHO the "logic" that does the feedback control MUST NOT incorporate a network, unless that network is VERY short, e.g. in the same room. Even then, network protocols will probably delay things too long to run stably. Hence, while networks may be OK for higher functions, forget them as part of any real-world feedback loops. Note in passing that this SAME discussion applies to neurons in complex feedback configurations (like brains). People now think that neurons are *S L O W* when they may be just "compensated" (using analog terminology, meaning that their high frequency response rolls off at ~6 db/octave) to operate in a feedback world. Note that they ARE able to do some things REALLY FAST (e.g. fast edges, doubled pulses, etc.) so perhaps they are really running in a world that is ~2 orders of magnitude faster than anyone (else) has yet thought possible. This would RADICALLY change projections of how much computer it would take to emulate a brain, perhaps by ~2 orders of magnitude. Steve ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
