Transimpedance amplifier (TIA) oscillating at above 1 GHz. Need Help

[julian....@gmail.com]

Hey Folks,

I need to design a broadband TIA for photo current amplification.
I am using a LMH6629.
I simulated the circuit in Tina-TI an it is working pretty well for a feedback resistor of 330Ohm and a feedback cap of 4pf and a signal of 10MHz.
However the simulation is showing oscillation if I increase the cap to something above 20pf. In my opinion it should be stable for higher caps and only start oscillating if the cap is to small.

However, I developed the board and now it is oscillating at 1.2 GHz with a perfect sine, even without a signal applied!
This oscillation is nearly independent of the feedback cap. I tried form 1pf up to 10pf.
Only if the cap is missing totally (still around 2pf from resistor) the oscillation is getting worth.
So, I think it has nothing to do with the feedback cap but some other issue.

I really have no ideas left..

Schematic: http://imageshack.us/a/img211/617/schematic.png
Board-top: http://imageshack.us/a/img585/4955/boardtop.png
Board-bottom: http://imageshack.us/a/img822/2739/boardbottom.png

Cheers,
Julian Arnold

[Phil Hobbs]

Wow, that's one ugly schematic. Conventionally you use a triangle for
an op amp, and put the input on the left and the output on the right.

With no photodiode and no feedback cap, you're trying to run the LMH6629
at a noise gain of 2, whereas it's only barely stable at a noise gain of
10. It's also a really hot amplifier, with a GBW of nearly 10 GHz, so
the layout is going to be very critical.

You overestimate the capacitance of the feedback resistor itself--it'll
generally be between 0.05 and 0.15 pF, depending on the type.

The photodiode capacitance is key to making this thing stable, because
at high frequency the noise gain is set by the ratio of the input and
feedback capacitances--you need a PD whose capacitance is at least 10
times the feedback cap to make it even marginally stable.

What sort of photodiode are you planning to use?

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510 USA
+1 845 480 2058

hobbs at electrooptical dot net
http://electrooptical.net

[julian....@gmail.com]

Hey,

Actually it is so ugly because it was my first approach to add a custom part to my layout editor ;)

How do you calculate the noise gain of 2?
Without the feedback cap and the diode it should be 1, shouldn`t it?

I am using a pin diode with 1.2pf of capacitance at 3.3V reverse bias.
I thought the more input capacitance the more unstable the circuit is so I tried to keep it small..

Cheers,
Julian

[Phil Hobbs]

> Wow, that's one ugly schematic. Conventionally you use a triangle
> for an op amp, and put the input on the left and the output on the
> right.
>
> With no photodiode and no feedback cap, you're trying to run the
> LMH6629 at a noise gain of 2, whereas it's only barely stable at a
> noise gain of 10. It's also a really hot amplifier, with a GBW of
> nearly 10 GHz, so the layout is going to be very critical.
>
> You overestimate the capacitance of the feedback resistor
> itself--it'll generally be between 0.05 and 0.15 pF, depending on the
> type.
>
> The photodiode capacitance is key to making this thing stable,
> because at high frequency the noise gain is set by the ratio of the
> input and feedback capacitances--you need a PD whose capacitance is
> at least 10 times the feedback cap to make it even marginally
> stable.
>
> What sort of photodiode are you planning to use?
>
> Cheers
>
> Phil Hobbs


>

> Hey,
>
> Actually it is so ugly because it was my first approach to add a
> custom part to my layout editor ;)
>
> How do you calculate the noise gain of 2? Without the feedback cap
> and the diode it should be 1, shouldn`t it?
>
> I am using a pin diode with 1.2pf of capacitance at 3.3V reverse
> bias. I thought the more input capacitance the more unstable the
> circuit is so I tried to keep it small..
>
> Cheers, Julian
>

(Context restored)

The noise gain is the closed-loop noninverting gain of the stage. With
no PD and no cap, you've got 1.2k input and 1.2k feedback resistors, for
a noninverting gain of 2, which is guaranteed to oscillate with this part.

If your PD is only 1.2 pF (which is a bit hard to believe at zero bias),
then if the pads are small and the traces short, you should have ~ 1 pF
from the board, 1.2 pF from the PD, plus the differential input
capacitance of the amp. This number is suspiciously absent from the
datasheet, but I'm guessing that it's about 2 pF. Thus your total C_in
is about 4 pF.

To maintain stability, your feedback capacitance needs to be less than
0.4 pF. I'd try replacing your PD with a 1-pF capacitor and running
with no feedback cap, first. If it still oscillates, try pressing your
finger down so that R_f squashes right into the pad of your finger, and
see if you can make it stable. If so, choose the right C_f and you're
probably done. If not, try reducing R_f.

With too high an R_f, the phase shift of R_f*C_in can destabilize the
loop even if the amp were unity-gain stable. The corner frequency of
1.2k and 2 pF is only 66 MHz, so you could also have an oscillation in
the low hundreds of megahertz if the 1.2 GHz one didn't take over.

Do you really need such a hot amplifier?

Cheers

Phil Hobbs


--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

[John Larkin]

Is the pd connected to pad1-pad2?

Why is there a resistor across the photodiode? To Vcc? That will make troubles.

What is Vcc?




--

John Larkin Highland Technology Inc
www.highlandtechnology.com jlarkin at highlandtechnology dot com

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators

[Julian Arnold]

Thanks a lot for the responses!

Unfortunately I need to build up the circuit for a university project where I need to transmit an OOK modulated signal at around 100 MHz...

The resistor across the diode is not soldered to the board.
I think I should remove it from the schematic.
It was placed there to remove biasing voltage if the diode would not be connected to VCC, which is at 5V.

So if the stability of the circuit depends on the noise gain, which needs to be above 10, than wouldn´t it be impossible to get it stable without a signal applied to the diode? Because without the resistor across the diode the noise gain should always be 1 without applying a signal.

Cheers,
Julian

[Julian Arnold]

Am Dienstag, 14. Mai 2013 10:25:29 UTC+2 schrieb Julian Arnold:

Just made a few measurements with different caps and no optical signal applied:
R_f = 1,2k
C_d = 3pf => V_o: Sine wave with 440mVp-p at 418MHz

R_f = 1,2k
C_d = 3,9pf => 448mVp-p at 477MHz

R_f = 1,2k
C_d = 1pf => 190mVp-p at 1,2GHz

[John Larkin]

Phil is the expert here, but I'd go for a parallel, 100 MHz tuned LC network
driving the gate of a phemt.

If s/n is not critical, just dump the photodiode current into a cheap MMIC.

[George Herold]

On May 14, 10:54 am, Julian Arnold <julian.arnol...@gmail.com> wrote:
> Am Dienstag, 14. Mai 2013 10:25:29 UTC+2 schrieb Julian Arnold:
>
>
>
>
>
> > Hey Folks,
>
> > I need to design a broadband TIA for photo current amplification.
>
> > I am using a LMH6629.
>
> > I simulated the circuit in Tina-TI an it is working pretty well for a feedback resistor of 330Ohm and a feedback cap of 4pf and a signal of 10MHz.
>
> > However the simulation is showing oscillation if I increase the cap to something above 20pf. In my opinion it should be stable for higher caps and only start oscillating if the cap is to small.
>
> > However, I developed the board and now it is oscillating at 1.2 GHz with a perfect sine, even without a signal applied!
>
> > This oscillation is nearly independent of the feedback cap. I tried form 1pf up to 10pf.
>
> > Only if the cap is missing totally (still around 2pf from resistor) the oscillation is getting worth.
>
> > So, I think it has nothing to do with the feedback cap but some other issue.
>
> > I really have no ideas left..
>
> > Schematic:http://imageshack.us/a/img211/617/schematic.png
>
> > Board-top:http://imageshack.us/a/img585/4955/boardtop.png
>
> > Board-bottom:http://imageshack.us/a/img822/2739/boardbottom.png
>
> > Cheers,
>
> > Julian Arnold
>
> Thanks a lot for the responses!
>
> Unfortunately I need to build up the circuit for a university project where I need to transmit an OOK modulated signal at around 100 MHz...
>
> The resistor across the diode is not soldered to the board.
> I think I should remove it from the schematic.

So the PD is across pad 1 and pad 2? And R5 1.2k is not in the
circuit?
(I was wondering what R5 was doing.)

I must admit I'm also a bit confused by Phils advise that you need
more C on the input. But I've never used a opamp in a TIA that wasn't
unity gain stable.

Your circuit looks very similar to one in the LMH6629 data sheet...
but there they do have a 10 pF PD on the input.

Finally I assume that you've got some nice bypass caps on the power
supplies.

Please let us know when (and how) you get this working.
I might have learned something today :^)

George H.

> It was placed there to remove biasing voltage if the diode would not be connected to VCC, which is at 5V.
>
> So if the stability of the circuit depends on the noise gain, which needs to be above 10, than wouldn´t it be impossible to get it stable without a signal applied to the diode? Because without the resistor across the diode the noise gain should always be 1 without applying a signal.
>
> Cheers,

> Julian- Hide quoted text -
>
> - Show quoted text -

[Julian Arnold]

Yes, the Resistor across the diode is not uses, sorry for that.
The schematic actually is:

http://imageshack.us/a/img211/2842/schematicx.png


What I am confused about is, that the datasheet tells, that at least a gain of 10 is required for a stable operation. What gain does this refer to? If it is the noise gain, than a stable operation could never be possible without a signal applied because than the noise gain is 0.
Furthermore, if I use no feedback cap, than the TIA is nearly oscillating from the upper to the lower rail. With the feedback cap it is only oscillating with an amplitude of around 70mV. These 70mV even stay the same if I reduce the feedback resistor form 1,2kOhm to 330Ohm.
When I press my finger onto the feedback network, the amplitude is only increasing slightly.

Maybe Phil can take away some of our confusion tomorrow;)

Cheers,
Julian

Am Dienstag, 14. Mai 2013 10:25:29 UTC+2 schrieb Julian Arnold:

[Tim Wescott]

What is your major? Because you need to go back and renew your knowledge
of (if you're an EE) or learn (if you're not) the basics.

Specifically, the concept of "gain" is one that comes from circuits that
have been linearized assuming their response to infinitesimal
perturbations. So the noise gain of the circuit remains the same as long
as the linearized models of the circuit elements remain the same.

You have the diode zero-biased, which is an error (it should be reverse-
biased for speed's sake). Even so, with a small enough signal applied to
the diode it's characteristics remain essentially the same -- it's going
to act like a current source in parallel with a capacitance. So your
gain does not change.

The noise gain, as Phil stated, comes from the capacitance of the diode
(which is higher than it ought to be because of the zero bias) and C5.
Unless you shine so much light on the diode that its voltage gets over a
couple of hundred millivolts, its capacitance and conductance aren't
going to change much.

You should have that diode reverse-biased, though. A diode's capacitance
varies with applied voltage, decreasing as the diode gets less reverse
biased (or, to an extent, as it gets forward biased, until it is
conducting so much that "capacitance" ceases to be a meaningful property
to assign to it). Reverse-bias the diode and you can get more bandwidth
out of the assembly.

--
My liberal friends think I'm a conservative kook.
My conservative friends think I'm a liberal kook.
Why am I not happy that they have found common ground?

Tim Wescott, Communications, Control, Circuits & Software
http://www.wescottdesign.com

[Tim Wescott]

On Tue, 14 May 2013 13:09:33 -0700, Julian Arnold wrote:

(Top posting fixed)

> Am Dienstag, 14. Mai 2013 10:25:29 UTC+2 schrieb Julian Arnold:
>> Hey Folks,
>>
>>
>>
>> I need to design a broadband TIA for photo current amplification.
>>
>> I am using a LMH6629.
>>
>> I simulated the circuit in Tina-TI an it is working pretty well for a
>> feedback resistor of 330Ohm and a feedback cap of 4pf and a signal of
>> 10MHz.
>>
>> However the simulation is showing oscillation if I increase the cap to
>> something above 20pf. In my opinion it should be stable for higher caps
>> and only start oscillating if the cap is to small.
>>
>>
>>
>> However, I developed the board and now it is oscillating at 1.2 GHz
>> with a perfect sine, even without a signal applied!
>>
>> This oscillation is nearly independent of the feedback cap. I tried
>> form 1pf up to 10pf.
>>
>> Only if the cap is missing totally (still around 2pf from resistor) the
>> oscillation is getting worth.
>>
>> So, I think it has nothing to do with the feedback cap but some other
>> issue.
>>
>>
>>
>> I really have no ideas left..
>>
>>
>>
>> Schematic: http://imageshack.us/a/img211/617/schematic.png
>>
>> Board-top: http://imageshack.us/a/img585/4955/boardtop.png
>>
>> Board-bottom: http://imageshack.us/a/img822/2739/boardbottom.png
>>

> Yes, the Resistor across the diode is not uses, sorry for that. The
> schematic actually is:
>
> http://imageshack.us/a/img211/2842/schematicx.png
>
>
> What I am confused about is, that the datasheet tells, that at least a
> gain of 10 is required for a stable operation. What gain does this refer
> to? If it is the noise gain, than a stable operation could never be
> possible without a signal applied because than the noise gain is 0.
> Furthermore, if I use no feedback cap, than the TIA is nearly
> oscillating from the upper to the lower rail. With the feedback cap it
> is only oscillating with an amplitude of around 70mV. These 70mV even
> stay the same if I reduce the feedback resistor form 1,2kOhm to 330Ohm.
> When I press my finger onto the feedback network, the amplitude is only
> increasing slightly.
>
> Maybe Phil can take away some of our confusion tomorrow;)

First: feel free to make your posts conform to normal USENET usage,
instead of whatever bizarre format Google Groups is imposing today.

Second: What in heck do believe the definition of "gain" is, and why do
you think that the noise gain of the amplifier changes with changing
signal?

[tm]

"Tim Wescott" <t...@seemywebsite.com> wrote in message
news:uPGdnQEpUYuJMA_M...@giganews.com...

Don't you have that backwards? The capacitance decreases with increasing
reverse bias.

[Tim Wescott]

Yes I do -- thanks for catching that.

I think I started to say it one way, and changed in mid-sentence to the
confusion of all.

It's easy to remember if you know your semiconductor device physics: the
capacitance is formed by two "plates" consisting of the N- and P-regions,
with a "dielectric" formed by the depletion region around the junction.
The more reverse bias, the wider the depletion region and the LESS (see,
I got it straight that time!) capacitance.

[George Herold]

Geesh, don't be too hard on him Tim. I was wondering myself about the
circuit operation.


To the OP, So I also don't see why it's enough to keep peak noise gain
(~Cin/Cfb) at ten. But that is the place the circuit wants to
oscillate.
So try a 10pF input C.

The other way to make it happen, is to reverese bias the PD (with a 9V
battery perhaps) into a resistor (maybe 50 or 100 ohms) and then use
the 'screaming' opamp (non-inverting) at a gain of 10-20 to 'gain up'
the resistor voltage.

I've done that with little PD's and mini-circuits amps.

(Tim's certainly correct about reverse biasing the PD to whatever it
can take.. if you want speed.)

George H.

>
> Specifically, the concept of "gain" is one that comes from circuits that
> have been linearized assuming their response to infinitesimal
> perturbations.  So the noise gain of the circuit remains the same as long
> as the linearized models of the circuit elements remain the same.
>
> You have the diode zero-biased, which is an error (it should be reverse-
> biased for speed's sake).  Even so, with a small enough signal applied to
> the diode it's characteristics remain essentially the same -- it's going
> to act like a current source in parallel with a capacitance.  So your
> gain does not change.
>
> The noise gain, as Phil stated, comes from the capacitance of the diode
> (which is higher than it ought to be because of the zero bias) and C5.
> Unless you shine so much light on the diode that its voltage gets over a
> couple of hundred millivolts, its capacitance and conductance aren't
> going to change much.
>
> You should have that diode reverse-biased, though.  A diode's capacitance
> varies with applied voltage, decreasing as the diode gets less reverse
> biased (or, to an extent, as it gets forward biased, until it is
> conducting so much that "capacitance" ceases to be a meaningful property
> to assign to it).  Reverse-bias the diode and you can get more bandwidth
> out of the assembly.
>
> --
> My liberal friends think I'm a conservative kook.
> My conservative friends think I'm a liberal kook.
> Why am I not happy that they have found common ground?
>

> Tim Wescott, Communications, Control, Circuits & Softwarehttp://www.wescottdesign.com- Hide quoted text -

[George Herold]

> Tim Wescott, Communications, Control, Circuits & Softwarehttp://www.wescottdesign.com- Hide quoted text -
>
> - Show quoted text -

Hmm, Hi Tim, I was reading his post differently.
(Your reading is more accurate.)
But reading between the lines, I thought he was asking about the noise
gain as a function of frequency. Which is only one at low
frequencies.

George H.

[Tim Wescott]

Well, he should know this stuff!

> To the OP, So I also don't see why it's enough to keep peak noise gain
> (~Cin/Cfb) at ten. But that is the place the circuit wants to
> oscillate.

I don't like the term "noise gain" in this usage. What's _important_ is
to keep the open-loop gain of the amplifier in its stable region. The
open-loop gain is determined by the amplifier characteristics and the
feedback network you put onto it.

It _happens_, by coincidence, that the noise gain (or non-inverting gain,
your pick) is also wholly determined by the feedback network you put onto
the amplifier.

So you can use that coincidence to equate a certain noise gain to
stability. But I still don't like it.

> So try a 10pF input C.

Slapping a cap on the input will reduce the signal available from the
diode at high frequencies, though. Your best bet is to decrease the
value of C5.

(This is so freaking counter-intuitive for me, by the way, or I have my
head up my ass -- I'm used to voltage amplifiers where you _increase_ the
feedback capacitance for more stability, because too much capacitance on
the input destabilizes the thing).

> The other way to make it happen, is to reverese bias the PD (with a 9V
> battery perhaps) into a resistor (maybe 50 or 100 ohms) and then use the
> 'screaming' opamp (non-inverting) at a gain of 10-20 to 'gain up' the
> resistor voltage.
>
> I've done that with little PD's and mini-circuits amps.

I believe that doing that will get you a bunch of resistor noise in your
circuit. But if the signal's strong enough it's certainly a safe thing
to do.

> (Tim's certainly correct about reverse biasing the PD to whatever it can
> take.. if you want speed.)

I hope so! It's what I do in these circumstances!

[Tim Williams]

What?

Tim

--
Deep Friar: a very philosophical monk.
Website: http://seventransistorlabs.com

"Tim Wescott" <t...@seemywebsite.com> wrote in message

news:uPGdnQApUYtvMA_M...@giganews.com...

[Jamie]

Tim Wescott wrote:

> On Tue, 14 May 2013 17:58:05 -0400, tm wrote:
>
>
>>"Tim Wescott" <t...@seemywebsite.com> wrote in message
>>news:uPGdnQEpUYuJMA_M...@giganews.com...
>>
>>>On Tue, 14 May 2013 07:54:18 -0700, Julian Arnold wrote:
>>>
>>>
>>
>>>You should have that diode reverse-biased, though. A diode's
>>>capacitance varies with applied voltage, decreasing as the diode gets
>>>less reverse biased (or, to an extent, as it gets forward biased, until
>>>it is conducting so much that "capacitance" ceases to be a meaningful
>>>property to assign to it). Reverse-bias the diode and you can get more
>>>bandwidth out of the assembly.
>>>
>>>--
>>
>>Don't you have that backwards? The capacitance decreases with increasing
>>reverse bias.
>
>
> Yes I do -- thanks for catching that.
>
> I think I started to say it one way, and changed in mid-sentence to the
> confusion of all.
>
> It's easy to remember if you know your semiconductor device physics: the
> capacitance is formed by two "plates" consisting of the N- and P-regions,
> with a "dielectric" formed by the depletion region around the junction.
> The more reverse bias, the wider the depletion region and the LESS (see,
> I got it straight that time!) capacitance.
>

You're just lucky! :)

Jamie

[Julian Arnold]

Am Dienstag, 14. Mai 2013 10:25:29 UTC+2 schrieb Julian Arnold:

Sorry for the questions about basic stuff.
I am still studying and have not heard so much about OpAmps so far,
but should not the gain always be a function of frequency?

Furthermore Tim said:
"The open-loop gain is determined by the amplifier characteristics and the
feedback network you put onto it"

but is not the open-loop gain only determined by the amplifier and can be approximated by a single pole response?

[George Herold]

Sure I was just suggesting this as a means to test the stability...
not a permanent solution.

>
> (This is so freaking counter-intuitive for me, by the way, or I have my
> head up my ass -- I'm used to voltage amplifiers where you _increase_ the
> feedback capacitance for more stability, because too much capacitance on
> the input destabilizes the thing).

I agree! But I guess there is some price to pay for using an opamp
that is not stable at unity gain.

>
> > The other way to make it happen, is to reverese bias the PD (with a 9V
> > battery perhaps) into a resistor (maybe 50 or 100 ohms) and then use the
> > 'screaming' opamp (non-inverting) at a gain of 10-20 to 'gain up' the
> > resistor voltage.
>
> > I've done that with little PD's and mini-circuits amps.
>
> I believe that doing that will get you a bunch of resistor noise in your
> circuit.  But if the signal's strong enough it's certainly a safe thing
> to do.

Grin, sure more noise... The TIA with a 1k feedback R will have the
johnson noise of the 1k R plus the opamp voltage noise (at gain 10).
If he was to use 100 ohm to ground and a gain of ten after (for the
same output signal level) that will have the equivalent noise of a 10k
ohm resistor... but lots of times the noise doesn’t matter that much.
Especially if he’s got a big hairy laser with lotsa photons. Then
(ignoring any laser noise) the shot noise can swamp the johnson
noise. For the TIA case I know the ‘break even’ point is when there
is 50mV of “photocurrent” (I = 50mV / R_feedback). So at a guess
maybe 500mV of signal for the R to ground with x10 gain.

George H.

> > (Tim's certainly correct about reverse biasing the PD to whatever it can
> > take.. if you want speed.)
>
> I hope so!  It's what I do in these circumstances!
>
> --
> My liberal friends think I'm a conservative kook.
> My conservative friends think I'm a liberal kook.
> Why am I not happy that they have found common ground?
>

[George Herold]

Hi Tim, just re-reading your post.
Noise gain is pretty much entrenched in the PD literature.
I don't really like the term either. I know it caused me hours of
head banging when I first started to study PD's and TIA's. But let's
just stick with it for consistencies sake.

I’m not sure why you are concerned that we equate stability with
gain.
(But you probably just have a deeper understanding than moi.)
So here’s a question. If I take the same opamp out of the TIA and
just put resistors around it to set the gain. Then it will oscillate
unless the gain is greater than 10. Now if I was to try and roll off
the gain at high frequency, by putting a small cap across the feedback
R. Would that cause it to oscillate?
(I’ve only used minimum gain of 10 opamps a few times... and rolling
off the gain was *not* what I wanted.)

George H.

>
> > So try a 10pF input C.
>
> Slapping a cap on the input will reduce the signal available from the
> diode at high frequencies, though.  Your best bet is to decrease the
> value of C5.
>
> (This is so freaking counter-intuitive for me, by the way, or I have my
> head up my ass -- I'm used to voltage amplifiers where you _increase_ the
> feedback capacitance for more stability, because too much capacitance on
> the input destabilizes the thing).
>
> > The other way to make it happen, is to reverese bias the PD (with a 9V
> > battery perhaps) into a resistor (maybe 50 or 100 ohms) and then use the
> > 'screaming' opamp (non-inverting) at a gain of 10-20 to 'gain up' the
> > resistor voltage.
>
> > I've done that with little PD's and mini-circuits amps.
>
> I believe that doing that will get you a bunch of resistor noise in your
> circuit.  But if the signal's strong enough it's certainly a safe thing
> to do.
>
> > (Tim's certainly correct about reverse biasing the PD to whatever it can
> > take.. if you want speed.)
>
> I hope so!  It's what I do in these circumstances!
>
> --
> My liberal friends think I'm a conservative kook.
> My conservative friends think I'm a liberal kook.
> Why am I not happy that they have found common ground?
>

[Julian Arnold]

Am Dienstag, 14. Mai 2013 10:25:29 UTC+2 schrieb Julian Arnold:

Good question, George!

[Frank Miles]

On Wed, 15 May 2013 09:05:11 -0700, George Herold wrote:

>So here’s a question. If I take the same opamp out of the TIA and
>just put resistors around it to set the gain. Then it will oscillate
>unless the gain is greater than 10. Now if I was to try and roll off
>the gain at high frequency, by putting a small cap across the feedback
>R. Would that cause it to oscillate?

Yes - if that the capacitive reactance is low enough (Zf/Zi < Amin)*.
By inserting a zero** - a resistor in series with this feedback capacitor
to keep Zf/Zi > Amin you should be stable, though of course you won't be
killing the HF gain as much as you might have wanted. The other classic
approach is to add capacitance from the inverting node to ground to
increase the noise gain even with the added feedback capacitance. Of course
there may be other sources of instability, e.g. capacitive loading.

* -evaluated in the vicinity of f=GBW/|Zf/Zi| to GBW.
**- a frequency domain zero, of course.

[George Herold]

Thanks Frank, I'll see if I can dig up a linear chip with a minimum
gain requirement and play in LT spice. So I should be thinking about
(at least) another pole in the open loop transfer function. (That
looks to really complicate the math.) Hey maybe Julian can run 'his'
screaming opamp in the TI spice simulator and post some pics. (You'll
have to calculate 'right' C to put accross the feedback R.)

George H.

[George Herold]

> George H.- Hide quoted text -

>
> - Show quoted text -

For anyone following along at home, here's an LTspice file (with
LT1127 opamp)
If you change C1 from 10 pF to 1 pF it will stop with the
oscillations. I don't know what to make of the 10k ohm * 10 pF = 1e-7
second time constant.

George H.

Version 4
SHEET 1 896 680
WIRE 192 80 176 80
WIRE 304 80 272 80
WIRE 176 144 176 80
WIRE -256 208 -256 176
WIRE -128 208 -128 176
WIRE 304 208 304 80
WIRE 304 208 112 208
WIRE 400 208 304 208
WIRE 304 224 304 208
WIRE 400 256 400 208
WIRE 176 288 176 272
WIRE 112 304 112 208
WIRE 144 304 112 304
WIRE 304 320 304 304
WIRE 304 320 208 320
WIRE 400 320 304 320
WIRE 480 320 400 320
WIRE 576 320 560 320
WIRE -256 336 -256 288
WIRE -128 336 -128 288
WIRE 80 336 -32 336
WIRE 144 336 80 336
WIRE -32 352 -32 336
WIRE 176 368 176 352
WIRE 576 400 576 320
WIRE -32 448 -32 432
FLAG -128 336 0
FLAG 176 272 +V
FLAG -128 176 +V
FLAG 176 368 -V
FLAG -256 336 0
FLAG -256 176 -V
FLAG -32 448 0
FLAG 176 144 0
FLAG 80 336 IN+
FLAG 576 400 0
SYMBOL voltage -128 192 R0
SYMATTR InstName V1
SYMATTR Value 15
SYMBOL voltage -256 192 R0
SYMATTR InstName V2
SYMATTR Value -15
SYMBOL opamps\\LT1127 176 256 R0
SYMATTR InstName U1
SYMBOL res 288 208 R0
SYMATTR InstName R3
SYMATTR Value 10k
SYMBOL res 288 64 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R6
SYMATTR Value 2k
SYMBOL res 576 304 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R7
SYMATTR Value 10K
SYMBOL voltage -32 336 R0
WINDOW 123 24 124 Left 2
WINDOW 39 0 0 Left 2
SYMATTR InstName V4
SYMATTR Value SINE(0 10m 1meg)
SYMATTR Value2 AC 10m
SYMBOL cap 384 256 R0
SYMATTR InstName C1
SYMATTR Value 10pf
TEXT 408 424 Left 2 !.tran 30u

[Julian Arnold]

Okay, so here are my simulation results.
At first the configuration that, I am think, I schoul have on my board:

C_in = 7pf R_f = 330Ohm C_f = 4pf
http://imageshack.us/a/img849/9161/cin7pfcf4pfrf330.jpg

Next with increased C_f (It is still hard to believe, that it is really starting to oscillate with increased feedback cap, but it definitively is...).

C_in = 7pf R_f = 330Ohm C_f = 15pf
http://imageshack.us/a/img834/2748/cin7pfcf15pfrf330.jpg

Than, as suggested, with increased C_in to increase the closed-loop? gain(still a bit confused about the gain thing...).

C_in = 7pf R_f = 330Ohm C_f = 15pf
http://imageshack.us/a/img59/2427/cin17pfcf15pfrf330.jpg

I will try to increase the C_in on my board tomorrow and report if it is helping.
I can not believe, that my current C_in should be bigger than 10pf and so there is a difference between simulation, where oscillation starts at around 15pf, and reality. So maybe there is also something else wrong...

Cheers,
Julian

[patricia herold]

> C_in = 7pf R_f = 330Ohm C_f = 4pfhttp://imageshack.us/a/img849/9161/cin7pfcf4pfrf330.jpg

>
> Next with increased C_f (It is still hard to believe, that it is really starting to oscillate with increased feedback cap, but it definitively is...).
>

> C_in = 7pf R_f = 330Ohm C_f = 15pfhttp://imageshack.us/a/img834/2748/cin7pfcf15pfrf330.jpg

>
> Than, as suggested, with increased C_in to increase the closed-loop? gain(still a bit confused about the gain thing...).
>

> C_in = 7pf R_f = 330Ohm C_f = 15pfhttp://imageshack.us/a/img59/2427/cin17pfcf15pfrf330.jpg

>
> I will try to increase the C_in on my board tomorrow and report if it is helping.
> I can not believe, that my current C_in should be bigger than 10pf and so there is a difference between simulation, where oscillation starts at around 15pf, and reality. So maybe there is also something else wrong...
>
> Cheers,

> Julian- Hide quoted text -

>
> - Show quoted text -

Cool, Everyone should do their own measurements.

So I don't recommend the 10pF as a permanent solution... just to see
if it makes the oscillations stop.

I'd try the opamp as X10 buffer,
or get a bigger photodiode,
or find a different opamp,
There's a fast Fet one from AD.

George H.

[Jamie]

George Herold wrote:
> On May 15, 8:47 pm, George Herold <gher...@teachspin.com> wrote:
>
>>On May 15, 1:02 pm, Frank Miles <f...@u.washington.edu> wrote:
>>
>>
>>
>>
>>
>>
>>>On Wed, 15 May 2013 09:05:11 -0700, George Herold wrote:
>>>

>>>>So here�s a question. If I take the same opamp out of the TIA and

2k in series with the 10p..

Jamie

[George Herold]

On May 16, 9:51 pm, Jamie

<jamie_ka1lpa_not_valid_after_ka1l...@charter.net> wrote:
> George Herold wrote:
> > On May 15, 8:47 pm, George Herold <gher...@teachspin.com> wrote:
>
> >>On May 15, 1:02 pm, Frank Miles <f...@u.washington.edu> wrote:
>
> >>>On Wed, 15 May 2013 09:05:11 -0700, George Herold wrote:
>

> >>>>So here’s a question.  If I take the same opamp out of the TIA and

Huh? The 'thing' oscillates at about 23MHz, that's like a 7-8 ns time
constant, 2k and 10pF is only 20ns. You can change the 2k to 4k ohm
and it doens't change much. I guess it's some other pole in the opamp
transfer function, but I'm not sure how to think about it.

George H.
>
> Jamie- Hide quoted text -

[Jamie]

George Herold wrote:
> On May 16, 9:51 pm, Jamie
> <jamie_ka1lpa_not_valid_after_ka1l...@charter.net> wrote:
>
>>George Herold wrote:
>>
>>>On May 15, 8:47 pm, George Herold <gher...@teachspin.com> wrote:
>>
>>>>On May 15, 1:02 pm, Frank Miles <f...@u.washington.edu> wrote:
>>
>>>>>On Wed, 15 May 2013 09:05:11 -0700, George Herold wrote:
>>

>>>>>>So here�s a question. If I take the same opamp out of the TIA and

Look, I gave you an alternate to your sim to remove the obvious mess
it was generating. what I got as a results was nothing more than the
normal phase margin error between input and output..

I didn't see any oscillations after that with the sim you supplied ?

Maybe we're using version XYZ from planet mars!

Jamie

[George Herold]

On May 17, 6:27 pm, Jamie

<jamie_ka1lpa_not_valid_after_ka1l...@charter.net> wrote:
> George Herold wrote:
> > On May 16, 9:51 pm, Jamie
> > <jamie_ka1lpa_not_valid_after_ka1l...@charter.net> wrote:
>
> >>George Herold wrote:
>
> >>>On May 15, 8:47 pm, George Herold <gher...@teachspin.com> wrote:
>
> >>>>On May 15, 1:02 pm, Frank Miles <f...@u.washington.edu> wrote:
>
> >>>>>On Wed, 15 May 2013 09:05:11 -0700, George Herold wrote:
>

> >>>>>>So here’s a question.  If I take the same opamp out of the TIA and

> Jamie- Hide quoted text -

>
> - Show quoted text -

Oh sorry, you should have said that the file didn't work.
It was just an lt1127 with gain of 6, non-inverting, and 10pF in
parallel with the 10k ohm feedback resistor.

George H.

[josephkk]

Not really. Normally there are two or three dominant poles and often
enough one zero in the open loop response.

?-)

[Julian Arnold]

Am Dienstag, 14. Mai 2013 10:25:29 UTC+2 schrieb Julian Arnold:
> Hey Folks,
>
>
>
> I need to design a broadband TIA for photo current amplification.
>
> I am using a LMH6629.
>
> I simulated the circuit in Tina-TI an it is working pretty well for a feedback resistor of 330Ohm and a feedback cap of 4pf and a signal of 10MHz.
>
> However the simulation is showing oscillation if I increase the cap to something above 20pf. In my opinion it should be stable for higher caps and only start oscillating if the cap is to small.
>
>
>
> However, I developed the board and now it is oscillating at 1.2 GHz with a perfect sine, even without a signal applied!
>
> This oscillation is nearly independent of the feedback cap. I tried form 1pf up to 10pf.
>
> Only if the cap is missing totally (still around 2pf from resistor) the oscillation is getting worth.
>
> So, I think it has nothing to do with the feedback cap but some other issue.
>
>
>
> I really have no ideas left..
>
>
>
> Schematic: http://imageshack.us/a/img211/617/schematic.png
>
> Board-top: http://imageshack.us/a/img585/4955/boardtop.png
>
> Board-bottom: http://imageshack.us/a/img822/2739/boardbottom.png
>
>
>
> Cheers,
>
> Julian Arnold

Hey everyone,

sorry for the late reply but I had to first gain some more theoretical knowledge
and proceed with my project.

I do now understand, that my TIA is running out of gain but as I am not able to equip the very small capacitor needed, I simply live with the resulting oscillation.
This oscillation is very small in amplitude and thus, the applied signal is easily dominating.
I am now able to transmit data at rates above 100 Mbps.
However, a new problem has shown up.
It seems, that for some particular input power a second, even worse, oscillation arises.
I am not quiet sure if it is some saturation problem or something completely different like reflections in front of the receiver diode.
The following images show the received signal at different input power levels.

With very low power there is only the 'normal' high frequency oscillation.
http://imageshack.us/a/img62/1231/increasingoneszerososc1.png

When increasing the input power (by moving the receiver optics deeper into the beam) the additional oscillation appears.
http://imageshack.us/a/img23/1231/increasingoneszerososc1.png

When further increasing the optical power, the 'down' parts of the signal calm down (Maybe some saturation ?).
http://imageshack.us/a/img208/1231/increasingoneszerososc1.png

I do not think, that it is an transient effect because the following image shows that the oscillation appears nearly randomly when operating near the 'critical' power.
http://img692.imageshack.us/img692/1231/increasingoneszerososc1.png

The yellow curve is the signal applied to the transmitter.

I am thankful for any suggestion!

Cheers,
Julian

[George Herold]

> With very low power there is only the 'normal' high frequency oscillation.http://imageshack.us/a/img62/1231/increasingoneszerososc1.png
>
> When increasing the input power (by moving the receiver optics deeper into the beam) the additional oscillation appears.http://imageshack.us/a/img23/1231/increasingoneszerososc1.png
>
> When further increasing the optical power, the 'down' parts of the signal calm down (Maybe some saturation ?).http://imageshack.us/a/img208/1231/increasingoneszerososc1.png
>
> I do not think, that it is an transient effect because the following image shows that the oscillation appears nearly randomly when operating near the 'critical' power.http://img692.imageshack.us/img692/1231/increasingoneszerososc1.png

>
> The yellow curve is the signal applied to the transmitter.
>
> I am thankful for any suggestion!
>
> Cheers,

> Julian- Hide quoted text -

>
> - Show quoted text -

Maybe you could repost your schematic... as it is now.
Maybe it's a power supply 'thing'. (any sign of the oscillation on
the power line?) The original schematic showed the photodiode tied
directly to the power rail... maybe a little series resistor and
capacitor to ground as RC isolation.

I'd be worried about the continued HF oscillations. (If you build
more than one it's guaranteed that to be worse in some.)

George H.

[Julian Arnold]

Hey George,

here is the link to the actual schematic: http://img51.imageshack.us/img51/7702/schematicj.png

On the board, the diode is next to the power decoupling caps of the TIA. Is it OK? I have not measured the power supply yet but I will do it on Friday.
Do you think this could be related to the optical input power?

I only need to get that one somehow working for my project ;)

[George Herold]

A much nicer schematic.
Is the 0.4pF Cf the estimated stray capacitance?

>
> On the board, the diode is next to the power decoupling caps of the TIA. Is it OK?

I would think so, but I've rarely tread above 100MHz. So I'm not
much of a HF 'expert'

I have not measured the power supply yet but I will do it on Friday.
> Do you think this could be related to the optical input power?

Well more power means more output signal.. more current.
Maybe a little decoupling would help, like this...

V(+)----RRR-+---|<|---->to opamp(-)
| (PD)
C
|
GND

So perhaps 10 ohms for R and 0.1uF for C.

>
> I only need to get that one somehow working for my project ;)

Well is the Professor going to see the oscillations?
And stopping the HF oscillations may also take care of this other
issue.

George H.

[Julian Arnold]

Hey,

yes, the professor is going to see them but I do not have the time to make new circuit boards.

Your suggestion sounds good!
Maybe I will have the time left to apply it.
But I think, I will first measure if there is any noise on the power supply.

Cheers,
Julian

[Joe Gwinn]

In article <d6e866f3-261d-419f...@googlegroups.com>,

Julian Arnold <julian....@gmail.com> wrote:

> Am Mittwoch, 12. Juni 2013 19:50:37 UTC+2 schrieb George Herold:
> > On Jun 12, 12:43 pm, Julian Arnold <julian.arnol...@gmail.com> wrote:
> >

[snip]

> >
> > Well more power means more output signal.. more current.
> >
> > Maybe a little decoupling would help, like this...
> >
> >
> >
> > V(+)----RRR-+---|<|---->to opamp(-)
> >
> > | (PD)
> >
> > C
> >
> > |
> >
> > GND
> >
> >
> >
> > So perhaps 10 ohms for R and 0.1uF for C.
> >
> >
> >
> >
> >
> > >
> >
> > > I only need to get that one somehow working for my project ;)
> >
> > Well is the Professor going to see the oscillations?
> >
> > And stopping the HF oscillations may also take care of this other
> >
> > issue.
> >
> >
> >
> > George H.
> >
> >
> >
> > - Hide quoted text -
> >
> > >
> >
> > > - Show quoted text -
>
> Hey,
>
> yes, the professor is going to see them but I do not have the time to make new circuit boards.
>
> Your suggestion sounds good!
> Maybe I will have the time left to apply it.
> But I think, I will first measure if there is any noise on the power supply.

I bet the feedback path is from the opamp Vcc pin to the photodiode and
thence back into the opamp, so one needs to ensure that the RC filter
is between the Vcc pin and the diode, with R connected between Vcc pin
and diode, and C from diode to ground.

What also often helps is a ferrite bead on the opamp input line.

I would expand the circuit diagram to show these details, so the
possible (unintended) feedback paths are apparent.

Joe Gwinn

[George Herold]

Julian, I was walking around after work today. And I think I know
how to fix both your problems. You should attach the PD lead tied to
the supply volatge to 'some other' node in your circuit. This will
both increase the capacitance seen by the inverting input,
and eliminate the PS feedback path.
You get to figure out which node.

George H.

[gvi...@indiana.edu]

On Wednesday, June 12, 2013 12:43:11 PM UTC-4, Julian Arnold wrote:
> Am Mittwoch, 12. Juni 2013 18:27:50 UTC+2 schrieb George Herold:
>
> > On Jun 12, 10:26 am, Julian Arnold <julian.arnol...@gmail.com> wrote:
>
> >
>
> > > Am Dienstag, 14. Mai 2013 10:25:29 UTC+2 schrieb Julian Arnold:
>

...

...


Hello Julian,
What is the status of this issue? Did you resolve it? I am finding a very similar issue on a transimpedance amplifier design I am making with LMH6629, and I believe that I have uncovered the solution which is that there must be a minimum source impedance seen by the LMH6629 input pins. For instance, try a 64.9 Ohm resistor (value what I found to work) between pin 3 and the bias network (C1 etc. on your schematic). Depending on wiring length to the photodiode another series resistor may be needed there, i.e. between the photodiode and the junction of inverting input pin and the feedback R-C.
I believe the issue (which these resistors seem to mitigate) is an internal oscillation of the LMH6629 input stage. I see the oscillation dependent significantly on the power supply voltage and even on the history of power supply voltage, and although I did not try it I believe it may well be dependent on signal amplitude too.
In any case, I do not think it is wise to have any oscillation in your final design, no oscillation is small enough to ignore. At the very least, the unknown phase of the oscillation amounts to extra noise, and nonlinearities must mix it into your frequency bands of interest.
I hope these comments help, and I would really appreciate to hear if anyone else has encountered similar issues with LMH6629 in any circuit. I don't think this issue is about transimpedance applications only. By the way, the datasheet mentions 25 Ohm minimum source impedance, but IMHO this is wrong... Also by the way, the oscillations I saw were typically at ~1.23 GHz, with other satellite peaks too.
Sincerely,

Gerard Visser
Indiana University

[Phil Hobbs]

I doubt that's the issue, particularly since almost that exact circuit
is on the first page of the datasheet. Because of the differential-mode
input capacitance, puttting a resistor in series with pin 3 applies a
feedback zero, reducing the high frequency loop gain.

I suspect that the problem is one of four things.

1. Layout problems

2. Too small a feedback cap (uncontrolled noise gain)

3. Too large a feedback cap (the op amp isn't unity gain stable)

4. Layout problems.

(Did I mention layout?) You can't put photodiodes on cables in this
sort of circuit, for instance. It has to be a very tight layout with
excellent grounds and as close as humanly possible to zero lead length.
Then you have to tune the feedback cap for stability.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net