From: Mike Monett
X-Mailer: Mozilla 2.02 (Win16; I)
Subject: Re: Binary Sampler
Date: Thu, 16 Jan 2003 11:54:58 -0500
NNTP-Posting-Date: Thu, 16 Jan 2003 11:54:23 EST
Organization: Bell Sympatico
> Actually, even if it did servo on the mean and there was no noise,
> there would still be errors.
Yes, I think it servos on the mean, and yes, there will be an error.
It samples above and below the signal, which creates a deadband
where it does not respond to changes in the signal. The deadband can
be made quite small.
Paradoxically, a small amount of noise helps signal recovery by
eliminating the deadband, the same was dithering in an ADC.
> Mike isn't sampling the recovered signal synchronously, so the
> noise averaging has to happen over time.
I don't know what you mean by "synchronous recovery", but yes, the
signal is swept with another offset in frequency.
> The averaging function can be modeled as a window.
I don't know why you want to do this, since you lose the history of
the sampler that gives it the noise-rejection property.
> Errors result from the convolution of the window and the sine
This is an artifact of the modeling, not related to the operation of
> The errors can be reduced by either narrowing the window or
> increasing the sample rate.
> Narrowing the window increases the bandwidth, reducing the noise
> cancellation and the usefulness of the circuit. The only way to
> reduce the noise and the error is to increase the sample rate.
> The error can be made to approach zero, but only at the cost of
> infinite sample rate.
It doesn't have to go to infinity. All it has to do is be better
than current technology, which is easy to do.
> Given a signal frequency and a finite sample rate, noise
> cancellation performance trades off with error magnitude.
This is not quite true. The noise performance is shown in a series
of simulations I did, but I took the web page down since it took too
long to load. I plan to condense it and put it back on my web site
The performance is improved by increasing the ripple amplitude. In
the simulation, usable signals are recovered down to 60dB below the
I plan to do more simulations using band-limited Gaussian noise, but
it will take a while to develop the software.
Thank you for your comments!