OPA695 current feedback amplifier

[Yannick]

I was thinking of using the OPA695 current feedback amplifier for my
transimpedance amp. This amplifier is capable of delivering much more
gain over a much larger bandwidth... I also see the equi current noise
at the input is much larger then with the FET input opamps like the
OPA657 but if the signal is large enough (lets assume you have an
avalanche photodiode) isn't it better to use such a current feedback
amp in stead of a voltage feedback?

The reason why i am considering this is because i never will be able
to measure 300-500Mhz with the voltage feedback amps. I want higher
frequencies because this will give me far better distance resolution
with phase measurement due shorter wavelength.

 

[John Larkin]

I was thinking of using the OPA695 current feedback amplifier for my
transimpedance amp. This amplifier is capable of delivering much more
gain over a much larger bandwidth... I also see the equi current noise
at the input is much larger then with the FET input opamps like the
OPA657 but if the signal is large enough (lets assume you have an
avalanche photodiode) isn't it better to use such a current feedback
amp in stead of a voltage feedback?

The reason why i am considering this is because i never will be able
to measure 300-500Mhz with the voltage feedback amps. I want higher
frequencies because this will give me far better distance resolution
with phase measurement due shorter wavelength.

I'm using an AD8014 (400 MHz) current-mode amp with a 1.3 pF silicon
photodiode. Feedback is 680r + 1 pF, and resulting risetime is about
1.5 ns. You could use a faster amp if your detector has lower
capacitance. Noise isn't super important to me (I have a lot of light)
and I'm getting about 12 ps RMS jitter for a step input.

Seems to work good. I started with an OPA686, but it oscillated, and
then TI eol'd the damned thing.

Why not drive a 50-ohm multi-GHz MMIC? That's a whole lot of cheap
gain-bandwidth.

John

 

[colin]

I was thinking of using the OPA695 current feedback amplifier for my
transimpedance amp. This amplifier is capable of delivering much more
gain over a much larger bandwidth... I also see the equi current noise
at the input is much larger then with the FET input opamps like the
OPA657 but if the signal is large enough (lets assume you have an
avalanche photodiode) isn't it better to use such a current feedback
amp in stead of a voltage feedback?

The reason why i am considering this is because i never will be able
to measure 300-500Mhz with the voltage feedback amps. I want higher
frequencies because this will give me far better distance resolution
with phase measurement due shorter wavelength.

Your resolution ultimatly depends on the resolution of your adc converter
asuming thats what ultimatly measures the phase, also any errors thermal
drift etc, and probably most importantly SNR, and how much filtering you do.
the signal strenght i get reflected back from the target various
enourmously, have you tried the signal strenght after relecting of a target
at your maximum range? i dont think its safe to assume your signal strenght
is large enough, this is the asumption i made at the start and found it to
be very wrong indeed. APDs are great for amplification, but you do get more
noise as you increase the amplification level (by increasing the bias
towards Vbr), upto a point where the noise sudenly increases tremendously.

Noise can be averaged out but it depdnds how long you are going to wait,
errors can be nuled out by calibration, as long as it doesnt drift too much
over the short term wich makes it a good idea to keep power disipation low
as posible.

If you use a much highr frequency, although you get more phase change per
distance you get a corespondingly lower SNR due to lower Zc so u dont
necesarily gain anything. of course using a large lense as possible for the
detector has a lot to offer indeed, as would be using a larger laser if this
was safe to do, but very expensive.

To get best noise performance of the input stage it wld be well to consider
an input amp that has a noise figure optimal for the impedance of the
detector+input capacitance. at higher frequencies this becomes quite low and
a bipolar input amp might be better, although input curent noise is higher,
input noise voltage becomes much more of an issue. input noise voltage
/input noise current = optimum input impedance for lowest noise.

But tbh i think i wld still strongly consider a discrete input stage such as
a DG mosfet or even low noise bipolar, as it potentialy has lower noise than
any op amp you are considering. At 500mhz a 10pf total capacitance has a zc
of under 50 ohms, so as someone already sugested even a 50ohm MMIC might be
wel worth considering, even with the opa695, it looks quite good on paper
with an optimum input impedance of 100ohms, especialy if you use this as
second stage of amplification.

A simple discrete first stage with no feedback or tuning would of course
have a frequency response with a simple slope but this could be corected by
a 2nd stage with response sloping the other way. this would be much simpler
and have less potential for many components introducing unforseen noise,
instability, emi pickup and maybe phase variance etc and give you more
choice of components and configurations to use. but its up to you to
evaluate all this in this aproach lol.

300-500mhz is quite high for the inexperienced, hard to see whats going on
as scope probes alter the circuit so considerably, and is pushing the limits
of many scopes anyway. If you already have it working at 20 mhz id see how
the rest of the circuit performs, then you can see what is the limiting
factor in your resolution.

At this high frequency it would be interesting to look into using multiple
striplines to provide a multiple tuned frequencies of the input stage, and
reduce the efects of the capacitance, posibly even a coax step up
transformer.

Colin =^.^=

 

[Yannick]

Your resolution ultimatly depends on the resolution of your adc converter

asuming thats what ultimatly measures the phase, also any errors thermal
drift etc, and probably most importantly SNR, and how much filtering you do.
the signal strenght i get reflected back from the target various
enourmously, have you tried the signal strenght after relecting of a target
at your maximum range? i dont think its safe to assume your signal strenght
is large enough, this is the asumption i made at the start and found it to
be very wrong indeed. APDs are great for amplification, but you do get more
noise as you increase the amplification level (by increasing the bias
towards Vbr), upto a point where the noise sudenly increases tremendously.

Noise can be averaged out but it depdnds how long you are going to wait,
errors can be nuled out by calibration, as long as it doesnt drift too much
over the short term wich makes it a good idea to keep power disipation low
as posible.

ok, another thing about noise... I calculated the whole thing in
mathcad and i let it calculated the best feedback resistor RF for
20Mhz for the best S/N ratio, and it comes to the conclusion for best
S/N ratio i don't have to use a feedback resistor. This also seems
logic if you see at the equivalent Input Noise wich is :
Ieq(Rf)=Root(IN^2+4kT/Rf+(EN/Rf)^2+(EN*2*pi*((Cd+Cin)*df)^2)/3)

soo the higher Rf the lower the noise, and the transimpedance
amplification Af is only decreased a little, soo S/N ratio is best
with Rf is infinite.

Soo why using feedback resistor(or is this conclusion a result of
wrong calculations?)

But tbh i think i wld still strongly consider a discrete input stage such as
a DG mosfet or even low noise bipolar, as it potentialy has lower noise than
any op amp you are considering. At 500mhz a 10pf total capacitance has a zc
of under 50 ohms, so as someone already sugested even a 50ohm MMIC might be
wel worth considering, even with the opa695, it looks quite good on paper
with an optimum input impedance of 100ohms, especialy if you use this as
second stage of amplification.

i will study the dual gate mosfet first, never saw the theory about
it, i will take your advice in consideration.

300-500mhz is quite high for the inexperienced, hard to see whats going on
as scope probes alter the circuit so considerably, and is pushing the limits
of many scopes anyway. If you already have it working at 20 mhz id see how
the rest of the circuit performs, then you can see what is the limiting
factor in your resolution.

ofcourse that's what i am going to do now.

At this high frequency it would be interesting to look into using multiple
striplines to provide a multiple tuned frequencies of the input stage, and
reduce the efects of the capacitance, posibly even a coax step up
transformer.

ok, thanks,

Yannick

 

[colin]

ok, another thing about noise... I calculated the whole thing in
mathcad and i let it calculated the best feedback resistor RF for
20Mhz for the best S/N ratio, and it comes to the conclusion for best
S/N ratio i don't have to use a feedback resistor. This also seems
logic if you see at the equivalent Input Noise wich is :
Ieq(Rf)=Root(IN^2+4kT/Rf+(EN/Rf)^2+(EN*2*pi*((Cd+Cin)*df)^2)/3)

soo the higher Rf the lower the noise, and the transimpedance
amplification Af is only decreased a little, soo S/N ratio is best
with Rf is infinite.

Soo why using feedback resistor(or is this conclusion a result of
wrong calculations?)

i dont know about your equation but for sure as the frequency goes up the
impedance of C falls and so your voltage for a given curent falls too, hence
your signal becomes lower compared to your amplifier voltage noise and so
your SNR falls as frequency rises.

at higher frequencies tho like i said you can chose a diferent amplifier
that has a lower noise voltage at the expense of higher noise curent becuase
your using a lower source impedance wich means noise voltage becomes more
more critical and noise curent less so. but a non feedback amplifier wil
have less noise generaly speaking, but then there are other consideratiuons
that feedback gives you. the feedback resistor is there to give you a flat
frequency response of current to voltage, it does this by reducing the gain
at lowe frequenies, dont forget a diferential stage has two noise generating
transistors on the input wich add together.

you may end up with less noise, but u end up with a lot less signal at
higher frequency, maybe this what you are missing in your equation? this is
what makes the SNR worse. remember if you convert it to thevinin eqv you end
up with a voltage and impedance that both fall with frequency.

i will study the dual gate mosfet first, never saw the theory about
it, i will take your advice in consideration.

if you look at the bf998 it has 1db of noise at 50ohm at 800mhz wich means
the total noise wil be marginaly more than from a 50ohm resistor, if im not
mistaken that is, they dont publish noise voltage and current wich is a
shame, but from my experience the noise voltage from the bf998 is incredibly
low, i have tons of overall gain and when i short the input to gnd the
output goes very quiet, of course being a mosfet the curent noise is low
too.

i havnt studied parametric amplifiers but i understand thats the way to go
for virtualy zero noise.

Colin =^.^=

 

[Rene Tschaggelar]

I was thinking of using the OPA695 current feedback amplifier for my
transimpedance amp. This amplifier is capable of delivering much more
gain over a much larger bandwidth... I also see the equi current noise
at the input is much larger then with the FET input opamps like the
OPA657 but if the signal is large enough (lets assume you have an
avalanche photodiode) isn't it better to use such a current feedback
amp in stead of a voltage feedback?

The reason why i am considering this is because i never will be able
to measure 300-500Mhz with the voltage feedback amps. I want higher
frequencies because this will give me far better distance resolution
with phase measurement due shorter wavelength.

A repeatedly discourraged solution in s.e.d. :
In case of sufficient light (few mW average) take a
100 Ohm and dump the photocurrent from the reverse
biased photodiode into it.
Take this signal and go capacitive coupled through a
MAR6 (20dB) and a MAR3 (13dB). Both have 2GHz bandwidth
and are available for 1$ or so.

Rene

 

[Winfield Hill]

colin wrote...

if you look at the bf998 it has 1dB of noise at 50ohm at 800MHz,
which means the total noise wil be marginally more than from a
50ohm resistor, if im not mistaken that is, they dont publish
noise voltage and current, which is a shame ...

Most of the bf998 manufacturers don't publish the test circuit
for the noise measurement either, but it's probably similar to
the gain circuits in the Philips datasheet, which being tuned
has a gate source-impedance certainly much higher than 50 ohms.
Siemens published their 200MHz and 800MHz noise test circuits
in an old datasheet. The 200MHz amplifier uses a tuned step-up
transformer that's resonated with about 7pF of capacitance
(from 15pF in series with a bb505 tuning varactor), implying a
gate impedance of Q times 113 ohms = say about 5k ohms (Q=50).

So the bf998's seemingly low 1dB noise figure likely corresponds
to 3 - 5nV of noise voltage, or whatever.

I've never been _that_ impressed by dual-gate mosfet noise.

 

[Jeroen Belleman]

Most of the bf998 manufacturers don't publish the test circuit
for the noise measurement either, but it's probably similar to
the gain circuits in the Philips datasheet, which being tuned
has a gate source-impedance certainly much higher than 50 ohms.
Siemens published their 200MHz and 800MHz noise test circuits
in an old datasheet. The 200MHz amplifier uses a tuned step-up
transformer that's resonated with about 7pF of capacitance
(from 15pF in series with a bb505 tuning varactor), implying a
gate impedance of Q times 113 ohms = say about 5k ohms (Q=50).

So the bf998's seemingly low 1dB noise figure likely corresponds
to 3 - 5nV of noise voltage, or whatever.

I've never been _that_ impressed by dual-gate mosfet noise.

I measured it back in 2001, and found 1.1nV/rtHz around 200MHz.
I used the Y-method with a 50 Ohm resistor at 77K as the cold
source and a room temperature one as the hot source.

That's pretty good for a MOSFET. I was impressed.

Cheers,
Jeroen

 

[colin]

colin wrote...

Most of the bf998 manufacturers don't publish the test circuit
for the noise measurement either, but it's probably similar to
the gain circuits in the Philips datasheet, which being tuned
has a gate source-impedance certainly much higher than 50 ohms.
Siemens published their 200MHz and 800MHz noise test circuits
in an old datasheet. The 200MHz amplifier uses a tuned step-up
transformer that's resonated with about 7pF of capacitance
(from 15pF in series with a bb505 tuning varactor), implying a
gate impedance of Q times 113 ohms = say about 5k ohms (Q=50).

So the bf998's seemingly low 1dB noise figure likely corresponds
to 3 - 5nV of noise voltage, or whatever.

I've never been _that_ impressed by dual-gate mosfet noise.

Yes i was suspcious of the way they leave out any reference to the test
conditions let alone a circuit, for the 1db nf, and i asumed it might well
be something like a step up configuration, wich is ok but dificult for
wideband.

However i am impresed with the results i get and the almost seemingly
complete abscence of noise if I short out the input. however likewise I am
also using a tuned circuit.

Are there any other similar devices dual gate or otherwise wich actualy give
the noise voltage and curent to compare perhaps?

Colin =^.^=

 

[Phil Hobbs]

A repeatedly discourraged solution in s.e.d. :
In case of sufficient light (few mW average) take a
100 Ohm and dump the photocurrent from the reverse
biased photodiode into it.
Take this signal and go capacitive coupled through a
MAR6 (20dB) and a MAR3 (13dB). Both have 2GHz bandwidth
and are available for 1$ or so.

Rene

(Since I regard myself as one of the chief disparagers of suboptimal
photodiode front ends, here and elsewhere....)

This approach is fine if speed and low cost are the primary concerns, or if
there's lots of light, just as you say. I built something almost exactly
like this a month ago, except that I just dumped the photocurrent right into
a MAR-3 with nothing else on the input side. I used a Thor Labs FC-ferruled
InGaAs photodiode, and build the whole thing dead-bug fashion on a piece of
FR-4 set in to the lid of a die-cast aluminum box.

On the other hand, I'm putting in about 1 mW peak power from a
picosecond laser--the circuit was part of a new triggering setup. It
has a 200 ps rise time, which is pretty good for the price. (Rings like
a SOB, too, but all I care about is the first negative-going edge.)

Cheers,

Phil Hobbs

 

[Rene Tschaggelar]

(Since I regard myself as one of the chief disparagers of suboptimal
photodiode front ends, here and elsewhere....)

This approach is fine if speed and low cost are the primary concerns, or
if there's lots of light, just as you say. I built something almost
exactly like this a month ago, except that I just dumped the
photocurrent right into a MAR-3 with nothing else on the input side. I
used a Thor Labs FC-ferruled InGaAs photodiode, and build the whole
thing dead-bug fashion on a piece of FR-4 set in to the lid of a
die-cast aluminum box.

On the other hand, I'm putting in about 1 mW peak power from a
picosecond laser--the circuit was part of a new triggering setup. It
has a 200 ps rise time, which is pretty good for the price. (Rings like
a SOB, too, but all I care about is the first negative-going edge.)

You mean 1mW average from a picosecond laser ?
That still makes 10W peak or so.
No terminating resistor ? No capacitive coupling ?
That's what I call blunt.
Actually also with the picosecond setup and 5mW average
or so on the diode, we're using an AEPX65 to phase lock
the pulse with a jitter somewhat below 1ps.
I wish I could use an FC coupled one.

Rene

 

[Yannick]

i dont know about your equation but for sure as the frequency goes up the

impedance of C falls and so your voltage for a given curent falls too, hence
your signal becomes lower compared to your amplifier voltage noise and so
your SNR falls as frequency rises.

ofcourse, but if you take everything for one fixed frequency, say
20Mhz then you can calculate the signal noise ratio in function of Rf.
then it seems that when rf is infinite only voltage noise of the
amplifier is in the game, and because the feedback impedance is then
only from the feedback capacitance you get highest S/N ratio. If i
calculate the optimal Rf for the largest transimpedance gain
(D(Af)/Drf = 0 => Rf = 62k) i get 62K but for S/N ratio Rf has to be
as high as possible. The only reason you need a lower value of Rf is
to get like you say a flat frequency response.

you may end up with less noise, but u end up with a lot less signal at
higher frequency, maybe this what you are missing in your equation?

No this isnt what i am missing, i calculated the Signal as Sout =
ID*Af(Rf) with ID = 177na (calculated this for 10meter distance with
avalanche photodiode...)

Then i calculated the noise as Nout = Ieq(Rf) *Af(Rf)

then i calculated the SIgnal noise ratio as S/N = 20*log(Sout/Nout)

i plotted this in a X-Y plot with S/N on the y as and Rf on the x as
and it this gives a graph wich keeps increasing (although flatter and
flatter) for an increase in Rf , and the formula for Af(Rf) is
correct, it matches perfect with the Pspice simulation , soo i cant
see any errors...

Yannick

 

[Phil Hobbs]

You mean 1mW average from a picosecond laser ?
That still makes 10W peak or so.
No terminating resistor ? No capacitive coupling ?
That's what I call blunt.
Actually also with the picosecond setup and 5mW average
or so on the diode, we're using an AEPX65 to phase lock
the pulse with a jitter somewhat below 1ps.
I wish I could use an FC coupled one.

Rene

No, it's 1 mW peak--I'm just sticking a single mode fibre behind a ND 2.5
filter, right in the unfocused beam. The actual laser peak laser power is
more like 10 MW. Fibre optics works great when your optical pulses are only
a quarter of an inch long.

My rep rate is only 20 Hz, and the flashlamp jitter is probably hundreds of
nanoseconds, so I have to synchronize on each pulse separately.

The MAR-3 has a low, resistive input impedance, and the photocurrent pulse is
unipolar and has a very low duty cycle, so the net effect is very nearly
identical to capacitive coupling (within 1 part in 10**10). Adding
additional components on the input would just have added stray capacitive
loading without changing anything.

Cheers,

Phil Hobbs

 

[John Larkin]

The MAR-3 has a low, resistive input impedance, and the photocurrent pulse is
unipolar and has a very low duty cycle, so the net effect is very nearly
identical to capacitive coupling (within 1 part in 10**10). Adding
additional components on the input would just have added stray capacitive
loading without changing anything.

Cheers,

Phil Hobbs

The Sirenza SGA-series (SiGe) mmics are very nice. One of them, the
SGA-3586 actually - are you sitting down? - has a 50 ohm input
impedance!

John

 

[Winfield Hill]

John Larkin wrote...

The Sirenza SGA-series (SiGe) mmics are very nice. One of them, the
SGA-3586 actually - are you sitting down? - has a 50 ohm input impedance!

Be still my beating heart.

 

[John Larkin]

John Larkin wrote...

Be still my beating heart.

Well, I was pleased. Most MMICS seem to run low, mid-high 30's often.
I wonder why... maybe that optimizes gain or noise figure? The 3586
can be tuned to exactly 50 ohms by fiddling the device current; it's
the only MMIC I've found that can. Add a small series RC from input to
ground, and it becomes a very good bounceless wideband match.

John

 

[Rene Tschaggelar]

No, it's 1 mW peak--I'm just sticking a single mode fibre behind a ND
2.5 filter, right in the unfocused beam. The actual laser peak laser
power is more like 10 MW. Fibre optics works great when your optical
pulses are only a quarter of an inch long.

My rep rate is only 20 Hz, and the flashlamp jitter is probably hundreds
of nanoseconds, so I have to synchronize on each pulse separately.

The MAR-3 has a low, resistive input impedance, and the photocurrent
pulse is unipolar and has a very low duty cycle, so the net effect is
very nearly identical to capacitive coupling (within 1 part in 10**10).
Adding additional components on the input would just have added stray
capacitive loading without changing anything.

Interesting that you get some signal at all from 1mWpk.
And with a straight input to a MAR3

Rene

 

[colin]

ofcourse, but if you take everything for one fixed frequency, say
20Mhz then you can calculate the signal noise ratio in function of Rf.
then it seems that when rf is infinite only voltage noise of the
amplifier is in the game, and because the feedback impedance is then
only from the feedback capacitance you get highest S/N ratio. If i
calculate the optimal Rf for the largest transimpedance gain
(D(Af)/Drf = 0 => Rf = 62k) i get 62K but for S/N ratio Rf has to be
as high as possible. The only reason you need a lower value of Rf is
to get like you say a flat frequency response.

No this isnt what i am missing, i calculated the Signal as Sout =
ID*Af(Rf) with ID = 177na (calculated this for 10meter distance with
avalanche photodiode...)

Then i calculated the noise as Nout = Ieq(Rf) *Af(Rf)

then i calculated the SIgnal noise ratio as S/N = 20*log(Sout/Nout)

i plotted this in a X-Y plot with S/N on the y as and Rf on the x as
and it this gives a graph wich keeps increasing (although flatter and
flatter) for an increase in Rf , and the formula for Af(Rf) is
correct, it matches perfect with the Pspice simulation , soo i cant
see any errors...

yes i agree, sory i misunderstood slightly, the resistor inevitably adds
noise so leaving it out means you have less noise,
wich is what i think i said as well anyway, but i thought you were still
meaning the higher frequency wich you mentioned in the previous paragraph. i
wasnt fully awake when i read it lol.

If i were you i would measure what signal you get when u reflect it off a
target, my estimate of 10% reflected back towards the lense was pure
gueswork. the real chalenge comes when u have a smooth dark surface that is
angled away from the detector.

177 na at 20 mhz acros 10pf = 140uv wich sounds ok compared to the 18uv
noise for a 20mhz bandwidth from the amplifier, but this wil stil give u
quite a bit of jitter, of course this can be averaged out over many
milliseconds, but ive found it quite dificult to get as good as results as i
wld expect from this simple calculation. i have quite narow bandwidth too,
but i find most of the problem lies in noise picked up. in particular my
high voltage bias generator frequency seems to apear a lot on the signal,
despite a sheild over the rf input section and a sheild over the hv
generator and using a 7 stage multiplier so i need a lower voltage of the
200khz squarewave to the step up transformer, however you may have les
problem here as my input is tuned so is very much higher impedance.

Also i mentioned in another post one day i had it resting on my keyboard
wich is wirless and so was transmiting constant keypresses at 27mhz wich
swamped the output and had me looking for the cuase for ages. i normaly
switch off my computer/monitor flourescent lamp etc. when i try to measure
low noise performance, but hadnt considered the keyboard.

you sugested using a sinewave wich i gues would make more sense as would not
puting the thing so dam close to the detector like i did lol, i just thought
best to have a short a track as posible at 250v.

Of course the rf signal that drives the laser also apears on the signal
despite this being further away and shelded also. Although I havnt however
fuly soldered the sheilds in place yet as then it would be hard to make
changes. it is however only noticable when the gain is turned up max and the
detector totaly blanked out, also as i think i said before stray reflections
albeit invisible were also an issue at one point.

Colin =^.^=

 

[John Larkin]

Interesting that you get some signal at all from 1mWpk.
And with a straight input to a MAR3

Rene

1 mW is a huge optical signal! It will get you 0.5 mA or better from
most any pin diode.

John

 

[Rene Tschaggelar]

1 mW is a huge optical signal! It will get you 0.5 mA or better from
most any pin diode.

That'd be 25mV into 50 Ohms. We're getting far less,
or at least that I measure.
When I'm lucky, I get 3mVpk from 1mW average. Well,
the diode is held into a reflex from a crystal or into the
leakage from a mirror. While the leak of 1mW may be 4mm in
diameter, the diode is not that big. The diode doesn't have
the bandwidth of the pulse and acts as integrator.
Then the dutycycle of the pulses is in the order of 10^-4.
And there are 2m of cable between the sensor and the
amplifier electronics.


Rene