From: Phil Hobbs
Subject: Re: photodetector circuit, high speed with good ambient light rejection
Date: Sat, 30 Nov 2002 21:53:02 -0500
Organization: Posted via Supernews, http://www.supernews.com
X-Mailer: Mozilla 4.61 [en] (OS/2; U)
References: <3DE69C97.email@example.com> <3DE76E53.firstname.lastname@example.org> <3DE90887.A854256A@rica.net>
Kevin Aylward wrote:
> John Popelish wrote:
> > Winfield Hill wrote:
> > (snip)
> >> Hobbs points out that with moderate signal currents that create a
> >> high r_e, an external bias current can be added to the BJT's
> >> emitter and a like current taken from the collector. This approach
> >> relies on the absence of shot noise in resistor-generated currents.
> I think he is a bit of a smart arse for stating "because of the Pauli
> exclusion principle...metal resistors..no noise.." etc. There is no shot
> noise in resistors because there is no basic mechanism to generate shot
> noise in resistors. Its not even worth mentioning, imo.
The Pauli principle is the physics that prevents shot noise in metal
resistors. Shot noise occurs in a great many conduction processes, in
fact any time that there is nothing forcing the electrons to be highly
correlated--metals are the exceptions, but are of course common in
circuits. People often try replacing metal resistors with, for example,
current mirror sources, and are surprised at the poor results.
> I'll have a think on what you are showing here. In the meantime, its
> worth pointing out that this two transistor circuit is just an obtuse
> way of achieving the more common npn/pnp compound pair that looks like
> one npn transistor. Topologically, they are *identical*. That is, the re
> of the top emitter pnp transistor is reduced by the hfe of the lower
> npn, so that the npn can be ran at a lower current, thereby reducing its
> base current shot noise.
Um, no. That is, not unless you're going to call all circuits where one
transistor drives another identical. Both transistors are NPN, for one
thing. I don't think you've understood how the circuit works.
> The simplest improvement/simplification is simply to replace the 2
> bipolar with a low noise jet. Your then left with basically 1nv/sqrthz
> (e.g U309) voltage noise and essentially zero current noise. This
> results in a much simpler circuit with equivalent or better performance,
> and the analysis is also much simpler to boot!
Jerry Graeme's book on transimpedance amps has a lot of circuits using
discrete front ends on ordinary op amps. However, getting 1 nV/sqrt(Hz)
from a JFET requires a *big* JFET, for which read "high-capacitance".
U309s are nowhere near 1 nV/sqrt(Hz)--the data sheet I'm looking at says
10 nV, typical, and that's at 10 mA drain current. At 1 MHz, that will
contribute 6 pA/sqrt(Hz), which is 17 dB above the shot noise of 2
> In addition, this technique only works for the large diodes considered
> here, e.g. 100p, in the sense of achieving the *best* S/N at a high BW.
> The idea is that one is converting the 100p diode capacitance to the
> input capacitance of the op-amp via the low re of the transistor. If the
> diode was a high speed one in the 1pf range, and the circuit was
> optimised as such, the capacitance at the opamp input node would kill
> you. You would be back to square one, i.e. a current source with
> capacitance, driving a virtual earth. There is no such thing as a free
Quite so--small diodes and high photocurrents make life very much
easier, and should be used where possible. However, in the context of
optical measurements, we commonly have to use large photodiodes to get a
reasonable photocurrent--there's an invariant called the etendue, which
is the product of the image size and the solid angle of illumination.
For a given level of coherence, you can't focus an image below a certain
size, which can easily be tens of thousands of wavelengths' diameter.
Laser measurements or optical communications have a different set of
tradeoffs, and I don't know a great deal about building 10 Gb/s
photoreceivers. I picked a difficult but reasonably common sort of
problem encountered in building optical instruments.
You might be surprised at the number of people who build beautiful
optical instruments and then go and lose 20 dB of SNR in a crummy front
end amplifier because they don't know any better.
IBM T. J. Watson Research Center