From: Phil Hobbs
Subject: Re: reverse-biased diode noise source?
Date: Tue, 17 Sep 2002 18:18:23 -0400
Organization: IBM T. J. Watson Research Center
NNTP-Posting-Date: 17 Sep 2002 22:18:25 GMT
X-Mailer: Mozilla 4.61 [en] (OS/2; U)
John Larkin wrote:
> > Phil,
> that's neat, except that it's not a lot of signal... something like 18
> pa / rootHz for a 1 mA photodiode current. A simple 50 ohm resistor
> coupled into a TIA would make about that. If you run the pd's 1 mA
> into a good TIA with, say, 1K transimpedance, you're still only up to
> 18 nV/rtHz.
> I do know that 1950-vintage radar jammers used 931 photomultiplier
> tubes as noise sources.
(Longish post warning)
You're right that the noise current isn't too big, but the key is the
impedance level. Photodiodes are very nearly ideal current sources
(with a shunt capacitance). Thus the noise power level can be made as
large as you like by increasing the load resistance, at least until the
RC time constant becomes a problem. It's almost always possible to make
it much higher than the kT per Hz you get from a resistor.
Making an amplifier that's much quieter than the shot noise is as hard
as wiring an LF411 to a 10k resistor--you'll be shot noise limited any
time the photocurrent drops more than 2kT/e volts across the load
resistor (~50 mV at room temperature), provided the resistor noise is
greater than the amplifier noise. (This was the origin of my somewhat
cryptic comment about amplifiers being quiet nowadays.) TIAs are
typically unhappy with 50 ohm source impedances, because the noise gain
becomes so high that the amplifier voltage noise tends to dominate.
Even when it doesn't, the amplifier noise is a far higher proportion of
the total output than with a decent-sized photocurrent, so the
uncertainties and drifts of the amplifier noise will require
Furthermore, the noise power from a resistor is proportional to the
absolute temperature, whereas photocurrent shot noise involves only the
dc current and fundamental constants, so it's easy to make it very
stable with time and conditions, unlike noise diodes and resistors.
Also unlike noise diodes, it's Gaussian to very high accuracy(*) and
shows no measureable 1/f noise.
Photomultipliers are far worse than either, because their noise PSD
depends on their gain, which changes _dramatically_ with stray magnetic
fields, history, age, ambient helium concentration, and bias voltages.
(There are PMTs whose gain changes by a factor of 2 when you turn them
from East-West to West-East in the Earth's magnetic field.) PMT noise
is also very far from Gaussian--it consists of similar-sized pulses
arriving in a Poisson process, with occasional very large blips from
ions hitting the photocathode. Low-pass filtering helps somewhat, but
that won't fix the calibration problem.
(*) I've measured the Gaussian statistics of photocurrent shot noise,
and found that it's Gaussian to within 0.1 dB out to at least 7.1 sigma,
which corresponds to a threshold crossing rate of about 10**-11 times
the bandwidth. I couldn't go any further, because my last data point
gave 2 counts in 69 hours, and the threshold crossing rate goes down as
exp(-0.5(Vthresh/Vrms)**2)--going 1 dB further out would have reduce the
predicted crossing rate to 1 per year in my 1 MHz BW.
For anyone who's interested in this or other photodetection issues and
other low-noise optical measurements, I have some possibly interesting
(The exact URL for the paper containing the Gaussian shot noise plot is