noise lecture

TIA noise calcs. e_n-Cin noise.

And AoE!

an interesting tid bit is theoretically you cannot measure noise with an oscope because if you wait long enough the peak approaches infinity.

The p-p value of random noise is undefined.

might want to talk about crest factor too.

I only have an hour! I do specify that measurements are RMS.

Most scopes measure RMS these days.

Eyeballing p-p/RMS = 5 is not too bad.

That looks to be a long lecture. If B is bandwidth you are missing a B in the shot noise term.

I tend to separate technical noise, (most of your list) and fundamental noise, shot noise and Johnson noise. Where there is some hope, with better technique, you can reduce the technical noise.

As Win said, in opamps, noise gain vs signal gain.

The big electrostatic interferences I see (freq < ~1MHz) are the florescent lights, my DSO display and brick-on-rope SM power supplies.

George H.

Grounding an shielding seems like a whole 'nother lecture.

Oh I think mako means Gaussian noise (amplitude ) and white noise (spectrum)

I get white noise from a reversed biased zener.. but it's far from Gaussian.

George H.

However, waiting some multiple of the lifetime of the universe is not generally an experimentally viable procedure.

It is a pretty good heuristic that less than 1 in 200 values is more than 3 sigma from the mean. (assuming gaussian statistics)

Cauchy (Lorentz) distributed noise is an entirely different problem.

It occurs when someone inadvisedly computes a ratio of two small noisy Gaussian random variables such that the result is greater than one. Geologists are inclined to do this with monotonous regularity.

It occurs in spectral line collisional broadening as well.

Depends on the distribution that your random noise is drawn from.

We always used p-p ~ 6sigma as a rough rule of thumb. Assuming Gaussian random noise.

John Larkin wrote in news: snipped-for-privacy@4ax.com:

You lie. How do I know this? Because the most you could do in an afternoon lecture with that many topics is read off each type, mention one sentence about each and move on. There is no way to effectively address each of those in an afternoon 'lecture'.

The best you'll do is a mere mention. How droll.

Fox News Noise: a TV set tuned to Fox News makes an excellent source of high entropy wideband uncorrelated random phonemes

John Larkin wrote in news: snipped-for-privacy@4ax.com:

The reason RMS and p-p does not work with random noise is because random noise is not sinusoidal. And because each pulse rarely lasts longer than a single spike event, so does not have a 'frequency' either.

The things (hardware/processing) that resolve what the RMS value is need more than a single wave pulse event to make the determination.

You might mention something about:

- Noise Figure

- Noise Factor

- Noise Temperature

For RF, perhaps atmospheric and galactic noise: Also mention that most low frequency RF noise comes from about 8 million lightning hits per day: (Check "Detectors" box on the left)

"Any sufficiently advanced communications technology is indistinguishable from noise". (apologies to Arthur C. Clarke)

We wish.

There is a high periodicity of those phonemes, period 1hr (and presumably 0.5hrs, I wouldn't know)

Matched filters (e.g. the susceptible audience) easily lock onto the phonemes and dig /a/ signal out of the noise.

There may even be anti-correlation properties with news sources.

It tends to be lopsided, and gets better at higher current. One trick is to use two, at opposite bias polarity, and sum them. Or use many!

What voltage zener are you using?

People sell official noise diodes, for lots of money.

You can measure the RMS of a random noise voltage very accurately. The RF guys do it all the time.

P-P is not well defined.

We have Sutro Tower.

The RF get rectified by everything, especially opamp front-ends.

I'll hit the basics and hand out the list. Individuals can meet me later if something especially interests them.

I can cover each topic briefly.

Do you watch TV? I don't. The bandwidth is way too low.

This should be three or four lectures, 30-45-min each, targeted to one aspect. Otherwise, boring, overload.

You can space them out by a week or two.

That's a great idea for an article in the Journal of Irreproducible Results. "Statistical Properties of Fox Noise"

Could be an Ig Nobel in it for you!

-- john, KE5FX