Photodiode readout without transimpedance amplifier

Sticking the output into almost anything will convert current to voltage. I often connect a photodiode to a scope input.

greg

se

s

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.

You might have a look at an article of mine from some years back:

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

do

)

base

hods

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.

do

)

base

hods

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=3DI^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.

back:

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.

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

back:

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* 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.

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

to do

th)

he base

ethods

e l e d

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.

se

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 =3D Iphotodiode * (1+Rf/Rg)

(warning: bad ASCII art follows)

[from photodiode] | +-----------Rf-------------+ | | GND----Rg---+ +-----Iout | | | \ / / +-------+-----|+ \ | | | >---R1--+----+-----| GND--|- / | | | | / R2 V V | | | [V--]----+------+-----+

base

back:

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

--
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

do

ba=

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

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

Hobbs

a

want to do

bandwidth)

into the base

better methods

noise

Linear

be a

back:

the

amp

This

But

really

less

of

electrical

are

because

noise

quoted

on

also

usually

with

do

OK another book for my collection. ISBN please?

e:

a

nt to do

width)

o the base

r methods

oise

ear

e a

tp://electrooptical.net/www/frontends/frontends.pdf.

e

mp

is

t

ly

ss

of

ical

re

ause

e

oted

on

so

ly

h

do

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.

Hobbs

using a

want to do

bandwidth)

into the base

better methods

since noise

Linear

Phil

will be a

back:

the

photodiode amp

the

This

But

really

less

gather

of

operation of

whereas

electrical

noise are

because

the

noise

D*

region

quoted

the

of

limit on

also

=A0Well

summing

usually

win

with

to do

of

near

in

Merci beaucoup.

The day is coming when the best technical libraries are digital and = require=20 membership in a relevant society (Say IEEE, ISME, ASTM).

rote:

ng a

want to do

andwidth)

into the base

tter methods

e noise

Linear

hil

l be a

:

the

e amp

e

This

But

eally

less

r f

ion of

eas

ctrical

e are

because

the

oise

D*

n

quoted

the

of

it on

also

ll

ng

ually

win

with

to do

of

r n

re

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?

Hobbs

using a

that I want to do

and bandwidth)

directly into the base

better methods

low

since noise

detail. Linear

is Phil

will be a

back:

would the

photodiode amp

the

amp. This

resistor. But

there really

is

low

diode

that less

gather

SNR of

operation of

whereas

electrical

noise are

is

hypothesis.)

Johnson

area, because

and the

the noise

merit D*

region

is quoted

basically the

sources of

limit on

and also

=A0Well

summing

usually

;)

you win

problem with

want to do

base

kHz of

basically

of

near

detail in

free

Schwartz,

when

require

Simple. IEEE has become a publisher, not a professional society, nor a = technical=20 society. It is another path to publishing, like Irwin Feerst said, for = the=20 publish or perish segment of our society.