Re: What the heck is going on with TI's Burr Brown parts manufacturing? - was Photodiode case

Winfield Hill wrote:
Phil Hobbs wrote:
Winfield Hill wrote:
Jure Newsgroups wrote:
Winfield wrote ...
Basically it's the bf862 with a ferrite in the gate,
a source resistor to Vee for running at 5mA, a drain
resistor plus bypass cap to gnd to reduce the drain
voltage to under 8V (see data***), a coupling cap
to an emitter-follower stage, which is biased at say
-10V, depending on the PD, an LC filter to quiet the
bias-voltage source, a pos/neg bias-voltage header, a
low-noise EF BJT, with resistor to run at 4mA, etc.,
with back-back collector pmbfJ309 JFETs to limit the
current in the event of a PD-sensor short. Finally
there's another ferrite at the output to damp parasitic
RF oscillations (small resistors, commonly used for this
purpose, aren't allowed because of their Johnson noise).
The bootstrap follower has an 80MHz bandwidth, probably
limited by the ferrites, and is capable of driving the
guard shield in a triple-coax cable, etc. Armed with
a pencil, one should be able to draw the schematic.
Do I read correctly, AC coupled ?
Yes. DC to establish the PD's bias and provide a
return path for the low-frequency current. The ac
signal path provides the bootstrap (which is only
important at high frequencies where the opamp has
lost tight control of the summing junction, and
where Cdiode matters) and the return path for the
high-frequency signal current.
Win,

The classical bootstrapped shield approach is very expensive
in SNR unless you're careful (which I assume you are). How
do you avoid a big noise peak due to concealing the RC rolloff?

Actually, the bootstrap is more quiet. The circuit is
going to suffer from the opamp's en-Cin noise without a
bootstrap, but with it, provided the bootstrap follower
is quiet, the opamp's voltage noise is impressed equally
on both sides of the diode capacitance and therefore it
doesn't generate noise current in the capacitance (which
would be seen as noise in the signal). In a sample case
the opamp's e_n is 4.5nV, which is about 5x higher than
the boostrap, so the amplifier enjoys a 5x improvement.
As a bonus, the apparent bandwidth of the main opamp is
greatly increased, or the detector's apparent capacitance
is decreased, however you prefer to look at it, allowing
dramatically more circuit bandwidth than would otherwise
be possible with the opamp.

The bootstrap bandwidth greatly exceeds the transimpedance
bandwidth, so there is no HF rolloff there. As far as the
ac coupling is concerned (mine is 200Hz), the PD's light-
current signal has to go through the feedback resistor/
capacitor, and hence get measured, no matter what the
current path might be on the other side of the photodiode.

BTW, I prefer the bootstrap approach to your common-base
transistor approach because it works well over a wider
current range, all the way to zero DC photodiode current,
where zero can be many many orders of magnitude below Imax.


Win,

I know about bootstrapping the PD...I've used that approach often, and combined it with the common base idea as well. It's bootstrapping the cable shield I meant. The original 'ghost shield' approach from WWII sonar looks like magic but costs beaucoup SNR. The SNR of a bootstrapped RC front end is the same as the SNR of the same amp connected as a follower on the same RC rolloff--only the transfer function changes.

Wide range photometers have different problems than the laser measurements I'm typically interested in, so naturally different tradeoffs are reasonable. In general I'm worried about getting to within 1 dB of the shot noise in some relatively narrow range of photocurrents, maybe a 10:1 range at most. That's not too much of a restriction, though, since to get to 1 dB above shot noise, you have to drop at least 200 mV across Rf, independent of circuit topology. Even with asymmetric supplies or a big DC offset, there's at most a 20 dB (optical) range where a given linear op amp circuit can be quiet enough for my purposes. (Range switching doesn't count.)

My favourite way to get better SNR is coherent detection, which also compresses the dynamic range by half (in dB). With coherent detection, you can pick the local oscillator beam power to be anything convenient, and still get *1 photon RMS* sensitivity on the signal beam. If I'm really stuck in the nanoamps, I'd rather use an avalanche photodiode and save my SNR. APDs have really improved in the last several years, so the noise penalty is much less than it once was.

Some of the reasons I emphasize the common-base approach in my writing are:

1. It's simpler than a bootstrap--one BJT, or one BJT and two resistors for the fancy model, and simpler is prettier. I share Joerg's aesthetics there.

2. Its characteristics are an excellent match for laser noise cancellers (q.v.), which use BJT diff pairs attached to photodiodes.

3. It's much easier to analyze, and thus it occasionally sneaks a bit of device physics into the heads of people who only know SPICE and a few textbook circuits.

<rant>
It never ceases to amaze me how many EEs and applied physicists chicken out of analyzing the noise of a simple BJT circuit. Even new grads do this--people who were doing partial differential equations for class a year or two ago. The physics is dead simple, and it's about 5 lines of algebra, but the great majority of people I talk to simply _will_not_ do it. This is true even though BJTs follow their simple noise models essentially perfectly, and even when their jobs depend on the results. Pathetic.

I've had a few, fewer than 10 out of dozens and dozens I've talked to, who have gone through the math for a common-base stage. Good luck getting them to analyze a bootstrap.

</rant>

Cheers,

Phil Hobbs
.

Quantcast