adaptive sampling frequency

"mustafa umut sarac" schreef in bericht news: snipped-for-privacy@h48g2000cwc.googlegroups.com...

I think you're nuts.

That's an idiotic statement, at the very least.

Last week Cornell sent me an email offering to license an invention that digitizes a signal only when the level crosses bit boundaries, so it outputs low rate samples when the signal isn't changing much.

Seems sort of silly to me, given real-world issues like noise and signal processing.

John

Sounds like a compression algorithm to me. Some of them Do work, but this is no breakthru idea.

Luhan

No, it's actually in the adc itself. It's like a classic flash with, say, 1023 comparators, each comparing the input against taps on a resistive divider. They just don't do anything until one of the comparators transitions, or something like that.

Looks like I deleted the email.

John

But that's merely a form of adaptive sampling by adapting the sampler rather than compressing afterwards (and yes, a form of compression) which has been around for decades. I remember talking to a few friends about sampling and not bothering to sample unless there was a change in the input at least 20 years ago. (I still have the notes from those meetings)

Cheers

PeteS

Then your notes may invalidate the Cornell patent.

John

The patent isn't worth anything - I used that technique when I programmed the PDP-8 to handle the experimental data for my Ph.D. work. I knew I was looking at a monotonic decay, so I had the computer sample the data until there was a significant change in voltage, then work out the average value of all the samples stored since the last value had been stored, then store that average (and the time) if there had been a siginficant change in voltage.

The initial samples were summed over 2^n intervals (double precision) and divided by 2^n by shifting., which dealt with the noise. I adjusted n by hand, but a high pass filter would have extracted the noise to allow "n"to be set automatically.

Much the same idea was employed in the Cambridge Instruments Electron Beam Tester 1989-91, but it wouldn't be useful in a patent case, since it wasn't publicly documented. The digital signal processing was all done in 100k ECL, and all the sample intervals were restricted powers of two to avoid the necessity for a proper multiplier.

Worked fine in both application - and in 1968 it meant that I could get my data into the 3k of 12-bit memory left available after I'd loaded my

Well, Bill's stuff is certainly more impressive, but it would be amusing if you mentioned it to Cornell the next time they ask you to license this 'astounding' invention ;)

The details:

First meeting: May 7 1983 Discussion on reducing hardware and software requirements for the experimental voice encoder [back then, of course, memory was at a real premium as you remember]. We used a power supply for basic encoding simply because we could read it with a scope to correlate (at least roughly) with the data for proof of concept.

Sampled data : Power supply under varying loads to prove the principle of the project

Issue: fixed sampling takes too much memory

Possible resolution:

"As the supply only changes significantly when the load changes, set the circuitry to trigger a comparator when the input changes significantly enough to require a sample. So the output may be reconstructed, tag a new sample (where sampling has been suspended) with the sample frame ID."

Those are my raw notes. There are also schematics (included a sampler counter so the sampler count could be tagged at the start of a new sample sequence). Takes me back.I'll need to dig out at least three notebooks to reconstruct the whole thing, but I have no doubt I could :)

Cheers

PeteS