I was thinking along the lines of my (?) idea of gluing a Am241 source on
a DRAM and using it in a suitably programmed computer as a source of
entropy. I actually got a Am241 source (smoke detector).

While thinking I was trying to picture how I would visualise the faults in
the ram banks and I inevitably came to the ccd or cmos camera idea. Almost
everyone has these, and I have several. So I thought why not put the
source on the ccd front glass and cap the lens. And then I thought - why
use the Am241 at all when the ccd and its preamplifier produce high
bandwidth white noise just like a noise zener and I already have a
digitizer etc in the computer, in the form of a video capture device ? The
radioactive source would likely increase the noise and make it more random
(maybe).

So I think that all that is needed to make good noise on a computer, as in
entropy source for /dev/rnd and key generation and such, is a camera with
the lens capped and some data filtering etc to make sure that most of the
noise is not mains hash harmonics and sampling artifacts.

I did some preliminary (crude) analysis on sampled frame data and it seems
to work right. There are a few device-specific components that must be
removed and the noise is good as is, but it should be used as a seed in a
good quality random number generator imho and analyzed by someone much
better than me I think. I used a fft to look at patterns (need to use a 2d
fft really to look for spatial correlated noise from the pickup grid) and
chi^2 analysis using a public domain program. Looks good so far. The fft
of a white noise data sample is flat 'grass' as you know.  Most of it
should come come from the preamplifier, not from the ccd (capped lens =
maximum agc).

A similar technique could be used to generate high bandwidth noise from a
certified noise diode by connecting it into the input circuits of a video
camera instead of the original ccd/preamp.

It can also work with a non-ccd camera, such as an old clunker vidicon
camera, even if the vidicon is damaged, as it is not needed. Terminating
the target input with a noisy old 1Meg resistor should help.

All that is really needed is a circuit that generates synchro signals such
that the video capture card or digitizer in the computer is fooled into
sampling the high bandwidth noise from the input signal.

I will still try the Am241 source in this context (with the ccd and with
the cmos dram). I will also try other devices put directly in proximity of
the source. The source should make 3x10^4 particles (beta) per second
(0.9uCi activity - says so on the box).

The high bandwidth source could also be used with a PIC with a/d. Cameras
are cheap so it is possible to add a simple bw surveillance camera with
capped lens to your project to obtain noise for a very low cost. The ad
converter would sample the video input at its own speed, or after an
external rapid s&h, running asynchronously or not from the pic (perhaps
with a window at pic crystal period). Readouts that are below a predefined
level will be discarded (to account for black back and front porch and
sync signals). Additional amplification is also possible. I assume that
the video signal would be clamped to the a/d ground level.

Peter

--
http://www.piclist.com hint: The PICList is archived three different
ways.  See http://www.piclist.com/#archives for details.