<br><br><div><span class="gmail_quote">2006/3/1, ASSI <<a href="mailto:Stromeko@compuserve.de">Stromeko@compuserve.de</a>>:</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
On Mittwoch, 1. März 2006 17:45, Dave Krooshof - <a href="http://dendriet.nl">dendriet.nl</a> wrote:<br>> The downside however, is that I would not even know how to<br>> do gain properly in a bitstream, let alone filtering.
<br><br>There's a number of PhD thesis' in that question... :-)</blockquote><div><br>I know, and google shows a lot of interesting pdfs that are<br>on the IEEE site, but I can't download those (or I'll have to pay a lot)<br>
</div><br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">> Can anyone point us to information about bistream DSP?<br><br>You may actually have more luck looking up publications on how
<br>binary/trinary pulsed neural networks process information.</blockquote><div><br>Okee. I see. And probably Cellulair Automata too.<br></div><br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Paramount to DSP directly with pulse trains is to chose an efficient<br>(and likely redundant) number system for each operation at hand (for<br>integration and thresholding). That implies frequent number system and<br>sample rate conversions. Multiplication and addition may both be done
<br>by XOR gates in the right representation, while other operation may be<br></blockquote><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">more efficiently implemented by splitting up the signal into several
<br>bitstreams that are processed in parallel before funneled together<br>again.\</blockquote><div><br> </div><br><div>yes, and adding the same signal will not make any gain, unless it's<br>
delayed a bit. But then the question is how much I should delay to<br>
make a nice 6dB gain. Will the number of bits delay be my gain value<br>
from +0 to +6 dB?<br><br> </div><br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">The second key ingredient IMHO is a number of mutually<br>
independent (white) noise sources with adjustable probability of the<br>pulse density (the tricky part here is the seemingly innocent<br>"independent" as you'd need lots of them). And now the best or worst<br>part, depending on how you look at things: you gotta do this in an FPGA
<br>at least (ASIC would be even better), because the most common<br>operations are wildly inefficient to do in software.</blockquote><div><br>This is exactly what we were thinking of!<br><br>thanks<br><br>Dave<br></div><br>
</div>