Photodiode readout without transimpedance amplifier
renning
I'm trying to develop a photodiode readout pre-amp without using a
current-to-voltage converter; the post-processing steps that I want to do
require a current as a signal. The best performance (SNR and bandwidth)
that I have come up with is to feed the photocurrent directly into the base
of an emitter follower NPN, but I'm thinking that there are better methods
than this in terms of SNR. Does anyone have any ideas?
Regards,
Raoul
GregS
Sticking the output into almost anything will convert current to voltage.
I often connect a photodiode to a scope input.
greg
mi...@sushi.com
For bandwidth, you probably want to feed the diode into a low
impedance. Generally you only want the bandwidth you need, since noise
is proportional to the square root of bandwidth.
Graeme's book goes into photo diode preamps in painful detail. Linear
Tech has some bootstrap circuits in their app notes. There is Phil
Hobbs website. I don't think the answer to your question will be a
simple usenet post.
Phil Hobbs
You might have a look at an article of mine from some years back:
http://electrooptical.net/www/frontends/frontends.pdf.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net
mi...@sushi.com
wrote:
I've read the majority of your website. I recall there is an amp with
cascode (low impedance) input. If the photo current is low, that means
the gm of the transistor is low, so I think either you or Graeme has a
circuit with feedback around the cascode element. Lots of good tricks
in these front end designs. It really is one of the areas of
electronics where you roll your own and do better than the chip
designs.
Anyway, I think the subject material is complicated enough that
cracking a book is warranted, especially regarding stability.
There are off the shelf amps for lab use. I got one from UDT for
peanuts ($5 at the ham swap meet). I also got one from an auction. The
market must be small, so these preamps are virtually free. However,
they are very low bandwidth, which I presume is due to
overcompensation since the manufacturer doesn't know the specs of the
photodiode in use.
Thee was a vendor about two years ago on ebay selling what looked like
new old stock from UDT. I got a few of the PIN10D with the integrated
BNC. Very sexy. Hard to believe a diode that big is only leaking 2nA
with 10V bias. I'm never ceased to be amazed at what these solid state
physic gurus can come up with.
There is a vendor on ebay at the moment selling API "pulls" for around
$1 each in lots of ten. Not the greatest specs, but good for hacking.
I suspect the supply is endless since there are multiple auctions.
mi...@sushi.com
While I've got your ear, do you have a write up on NEP? Why would the
noise be specd in power instead of current.
Here is my non-expert interpretation of NEP. If I had a photodiode amp
with a low impedance input (known value), then I could convert the
power to currrent with P=I^2R, with R the impedance of the amp. This
should also work for a TIA with the R being that of the resistor. But
say you hooked the diode up to a high impedance amp. Then there really
is no power transfer in the ideal case.
I've noticed the NEP is larger as the device gets larger. This is
counterintuitive to chip design, where large generally means low
noise, at least for active elements. However, I could see a diode
putting out more noise if it has more area. So it would seem that less
is more here, but if you consider the larger diode also can gather
more light, it might be a break even situation.
Phil Hobbs
I bought 75 pieces of a 1.3 micron InGaAs avalanche photodiode module
for about 50 cents apiece. Probably $200 each new. I'll figure out
some killer application for them, but I don't know what it is yet.
Re front ends:
The bootstrapped biased cascode is pretty good above 500 nA or so,
unless your photodiode is really chubby.
For lower level stuff, it's often better to do a few parallelled BF862s
run single ended, with another few BF862s and something like a 2N3904 as
a bootstrapped bootstrap. With the right component values you can get
within a few dB of the shot noise in a 1 MHz bandwidth with ~30 nA of
photocurrent and 30 pF of photodiode capacitance. I did one of those
for a Far Eastern consumer products maker earlier this year. There are
a couple more turns of the crank available there as well.
Interestingly, the reason they cared was that CF lamps produce optical
crap out to beyond 1 MHz. Even at 1 MHz, you have to find a spot in
between the emission lines and sit there, looking through a narrow
interference filter.
Phil Hobbs
Noise equivalent power is the optical power required to get a SNR of
unity (0 dB). There's a lot of confusion generated by the operation of
square law photodetectors (i.e. all of them).
The energy deposited by N photons per second is always h*nu*N, whereas
once they're detected, it's I**2*R, i.e. (eN)**2*R. Thus the electrical
power goes like the square of the optical power. Signal and noise are
always uncorrelated, so when you square a noisy current, the SNR is
squared as well. (The cross term 2*<S*N> is zero by hypothesis.)
Most photodetectors have capacitance, leakage, and a little Johnson
noise. Electrically the noise power generally goes as the area, because
each square millimetre contributes the same as all the others, and the
contributions are uncorrelated. Optically, this means that the noise
equivalent power goes as the diameter (square root of area).
Thus detector families are often compared using the figure of merit D*
(D-star), which is defined as
D* = sqrt(Area)/NEP(1 Hz BW),
at some convenient modulation frequency that's out of the 1/f region
(often 1 kHz). Because of the optical vs electrical units, D* is quoted
in cm*sqrt(Hz)/W--which is a pretty weird unit.
D* isn't much use in the visible and near IR, where it's basically the
capacitance that sets the noise floor--the intrinsic noise sources of
silicon PDs are generally negligible. The capacitance sets a limit on
how high a feedback resistor you can use for a given bandwidth, and also
causes horrible noise gain due to its differentiating action. Well
within the feedback bandwidth, hanging a capacitance C on the summing
junction of an op amp produces a noise current
iN_Vamp = 2*pi*C*e_NAmp
where e_NAmp is the 1-Hz noise voltage of the amplifier. This usually
dominates for low light, large PDs, and unimaginative designs. ;)
Since the NEP goes as the diameter, for a constant light level you win
by going to larger areas, but again that isn't usually the problem with
silicon PDs. Big PDs are expensive enough that you usually want to do
optical things to avoid them where possible.
For audio-range jobs you can do fun things by putting a common base
stage on a solar cell--3 square inches of sensitive area and 20 kHz of
bandwidth.
mi...@sushi.com
wrote:
I was doing good up to this point:
"D* isn't much use in the visible and near IR, where it's basically
the
capacitance that sets the noise floor--the intrinsic noise sources of
silicon PDs are generally negligible. "
How does the spectrum effect D*? If anything, the sensor peaks at near
IR, so that would be optimal SNR.
The rest of the stuff I got since it is explained in painful detail in
Graeme's book.
whit3rd
> I'm trying to develop a photodiode readout pre-amp without using a
> current-to-voltage converter; the post-processing steps that I want to do
> require a current as a signal. The best performance (SNR and bandwidth)
> that I have come up with is to feed the photocurrent directly into the base
> of an emitter follower
Don't you mean feed the photocurrent into the emitter? A grounded-
base
transistor has the low-input-impedance quality that you seek, and the
stray capacitance of the photodiode usually is higher than (limits the
bandwidth more) than the Miller capacitance of a well chosen
transistor.
One can also use a transimpedance circuit (the op amp current/voltage
converter) with a two transistor current mirror connected to the op
amp output,
so the photodiode load is the pseudoground input pin, and the output
is the programmed current of the current mirror. That current mirror
can
get you much better compliance than the raw photodiode, and provides a
current gain (if you split the current from the feedback-connected
collector).
Iout = Iphotodiode * (1+Rf/Rg)
(warning: bad ASCII art follows)
[from photodiode]
| +-----------Rf-------------+
| | GND----Rg---+ +-----Iout
| | | \ / /
+-------+-----|+ \ | |
| >---R1--+----+-----|
GND--|- / | | |
| / R2 V V
| | |
[V--]----+------+-----+
Phil Hobbs
> I was doing good up to this point:
> "D* isn't much use in the visible and near IR, where it's basically
> the
> capacitance that sets the noise floor--the intrinsic noise sources of
> silicon PDs are generally negligible. "
>
> How does the spectrum effect D*? If anything, the sensor peaks at near
> IR, so that would be optimal SNR.
>
> The rest of the stuff I got since it is explained in painful detail in
> Graeme's book.
I could have said that D* is no use with silicon and short-wavelength
InGaAs detectors, and not much use with germanium. The reason is that
those ones are dominated by the effects of capacitance, whose effect is
completely dependent on the amplifier's voltage noise.
It isn't the spectrum of the light that matters, it's mostly the choice
of detector material. D* becomes useful with long-wavelength InGaAs,
InAs, InSb, HgCdTe, triglycine sulphate, PbLaZrTiO3, PVDF, LiTa03, and
so on. ;)
Of course it doesn't really apply to thermal detectors either, because
their noise goes like the thermal conductivity rather than the geometric
area.
Cheers
Phil Hobs
renning
>
>> I'm trying to develop a photodiode readout pre-amp without using a
>> current-to-voltage converter; the post-processing steps that I want to
do
>> require a current as a signal. The best performance (SNR and bandwidth)
>> that I have come up with is to feed the photocurrent directly into the
>se
>> of an emitter follower
>
>Don't you mean feed the photocurrent into the emitter? A grounded-
>base
>transistor has the low-input-impedance quality that you seek, and the
>stray capacitance of the photodiode usually is higher than (limits the
>bandwidth more) than the Miller capacitance of a well chosen
>transistor.
>
>One can also use a transimpedance circuit (the op amp current/voltage
>converter) with a two transistor current mirror connected to the op
>amp output,
>so the photodiode load is the pseudoground input pin, and the output
>is the programmed current of the current mirror. That current mirror
>can
>get you much better compliance than the raw photodiode, and provides a
>current gain (if you split the current from the feedback-connected
>collector).
>
>
>(warning: bad ASCII art follows)
>
>[from photodiode]
> | +-----------Rf-------------+
> | | GND----Rg---+ +-----Iout
> | | | \ / /
> +-------+-----|+ \ | |
> | >---R1--+----+-----|
> GND--|- / | | |
> | / R2 V V
> | | |
> [V--]----+------+-----+
>
Thanks to everyone for the tips!
What I was trying to do is feed the annode current into the base of an NPN,
with the collector at rail. (not the cathode current, which I guess is
commonly drawn from the emitter of a grounded base transistor). Then I
would have a gain of Beta, still have a current as signal, and not have to
worry about any transimpedance stages. Is there any point to this, or am I
just going to drown myself in noise?
---------------------------------------
This message was sent using the sci.electronics.design web interface on
http://www.Electronics-Related.com
Phil Hobbs
You're trying to reinvent the phototransistor. Phototransistors are
bad--their gain goes like beta, which is not a well controlled
parameter. You'd be far better off either doing your signal processing
with the photocurrent directly, maybe with a common base stage or a
bootstrap to help get rid of the capacitance problem, or else using a
TIA and a resistor.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
email: hobbs at electrooptical dot net
http://electrooptical.net
JosephKK
OK another book for my collection. ISBN please?
mi...@sushi.com
ISBN-10: 007024247X
You can get the details from amazon. I got mine for $30 used with free
shipping from half.com.
When you click on it on Amazon, it suggests the book "Building
Scientific Apparatus" or something like that. I came across that in a
used book store, It would have been a cool book if it wasn't so out
dated.
Berkeley and a few other bay cities have a chain called Half Price
Books. The often have cool stuff in their engineering section. I get
the old "classic" electronics books there from time to time. Schwartz,
Terman, etc. I got Papolius on Stochastics today. These are books when
engineers were engineers, not bit jockeys. Palo Alto used to have
great geek used book shops.
I'm dreading the day when all the books are ebooks.
JosephKK
Merci beaucoup.
>
>When you click on it on Amazon, it suggests the book "Building
>Scientific Apparatus" or something like that. I came across that in a
>used book store, It would have been a cool book if it wasn't so out
>dated.
>
>Berkeley and a few other bay cities have a chain called Half Price
>Books. The often have cool stuff in their engineering section. I get
>the old "classic" electronics books there from time to time. Schwartz,
>Terman, etc. I got Papolius on Stochastics today. These are books when
>engineers were engineers, not bit jockeys. Palo Alto used to have
>great geek used book shops.
>
>I'm dreading the day when all the books are ebooks.
The day is coming when the best technical libraries are digital and require
membership in a relevant society (Say IEEE, ISME, ASTM).
mi...@sushi.com
Yep. I would go to Terman on the weekends and read journals. Now they
are all on line and you need an account. I don't know if you can get
one without being a student.
Since the reproduction cost is nil, you would think journals would
just be free. WRONG! Look at the IEEE. You get paid nothing for peer
reviewing articles. There is a fee to get the paper published. So just
why does it cost anything to get these papers?
JosephKK
Simple. IEEE has become a publisher, not a professional society, nor a technical
society. It is another path to publishing, like Irwin Feerst said, for the
publish or perish segment of our society.