P
Phil Hobbs
Guest
On 11/26/20 11:52 AM, John Larkin wrote:
Sure, but we\'re calibrating with this big complicated DDG thing.
We\'re using a 20-ns RC time constant on the ramp
Around 1 GHz. The pHEMT charges up an 8-pf cap connected directly to
the input of a 6-channel simultaneous-sampling ADC, so effectively the
sampling circuit is a switch driving 8 pF, a very short trace, and the
ADC input, which (apart from pad capacitance) looks basically like a
40-ohm resistor and a 12-pF cap. So there\'s a fast bit and a slow bit
that have to be fixed in software.
> It\'s a change from using diodes.
Sure is. I\'m trying to use SAV-5xx+es in everything, to help persuade
MCL to keep making them. (They\'re generally very good about doing that
anyway.)
BTW I notice that NTE is selling the NTE2403, a near-replica of the
BFT92 5-GHz PNP. They cost $3, and have a sucky datasheet, but what can
you do? Digikey claims they\'re in active production.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510
http://electrooptical.net
http://hobbs-eo.com
On Thu, 26 Nov 2020 11:29:34 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:
On 11/26/20 10:48 AM, John Larkin wrote:
On Wed, 25 Nov 2020 11:34:36 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:
On 11/25/20 10:14 AM, jlarkin@highlandsniptechnology.com
wrote:
On Tue, 24 Nov 2020 23:07:57 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:
On 11/24/20 10:33 PM, jlarkin@highlandsniptechnology.com
wrote:
On Tue, 24 Nov 2020 17:58:44 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:
On 11/23/20 6:54 PM, jlarkin@highlandsniptechnology.com
wrote:
On Mon, 23 Nov 2020 15:11:05 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:
On 11/23/20 2:43 PM,
jlarkin@highlandsniptechnology.com wrote:
On Mon, 23 Nov 2020 13:58:26 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:
On 11/21/20 6:48 PM, Phil Hobbs wrote:
On 11/21/20 11:05 AM,
jlarkin@highlandsniptechnology.com wrote:
On Sat, 21 Nov 2020 07:05:11 -0500, Phil
Hobbs
pcdhSpamMeSenseless@electrooptical.net
wrote:
On 11/21/20 12:39 AM,
jlarkin@highlandsniptechnology.com
wrote:
On Fri, 20 Nov 2020 22:35:52 -0500,
Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net
wrote:
On 11/19/20 10:44 AM,
jlarkin@highlandsniptechnology.com
wrote:
On Thu, 19 Nov 2020 10:30:17 -0500,
Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net
wrote:
On 11/19/20 8:22 AM, George
Herold wrote:
On Wednesday, November 18, 2020
at 2:27:39 PM UTC-5, Phil Hobbs
wrote:
Hi, all.
So I have this SiPM/MPPC
front end. It has pop options
to use either an On Semi
MicroFC-10010 1-mm SiPM chip
or a packaged Hamamatsu
S13362-3050DG 3-mm MPPC with
integral TE cooler, both
bootstrapped by a SAV-551+
running at 20 mA. So far, it
all works.
(The SAV-551+ is amazingly
stable--I\'ve got a shipping
product that runs a very
similar bootstrap across a
2-inch FFC cable. Bandwidth
suffers a bit, but it shows
no tendency to oscillate.)
The mystery is in the TIA
stage. It\'s a vanilla op amp
TIA made from either an
ADA4899 (600 MHz, 300 V/us)
or AD8045 (1 GHz, 1300 V/us @
Av=1), which are pin
compatible in the 3-mm LFCSP
package. Both are voltage
feedback amps.
I\'m seeing a 3 dB bandwidth
of 220 MHz, together with a
faster rolloff than I expect:
-3 dB @ 220 MHz, -9 dB @ 320
MHz. It\'s not slew limiting,
because the waveform looks
pretty good on a 3-GHz scope
(TDS 694C) and the rolloff
stays the same when I drop
the input by 6 dB.
The layout is pretty tight
(the whole board is only an
inch square), so getting
enough stray capacitance
across R_F to account for it
is implausible--it would need
about 1.4 pF.
DecouplingBypassing is
good--
For test, I removed the 0-ohm
jumper that connects the
bootstrapped SiPM to the
summing junction, and added a
1k input resistor, forming an
inverting amp with a nominal
gain of -0.5.
That\'s connected to the
terminated end of an RG-174/U
cable going to a PTS-500
synthesizer. The output goes
via a 10-ohm resistor into a
properly-terminated 50-ohm
cable (the TDS 694C is 50-ohm
only).
Here I\'m expecting a
bandwidth somewhere between
the datasheet\'s 1 GHz @ Av=1
and 400 MHz @ Av=-1, but it\'s
way off. There\'s no visible
change when I put the jumper
back in, on account of the
swoopy bootstrap.
I was going to suggest looking
at the \'speed\' of the light
source. But the above seems to
point to something \'in\' the amp
stage... (Is that right?) (And
maybe check the light source
rise time anyway?)
George H.
So where do you suppose the
missing factor of ~3 in
bandwidth went?
This has the SiPM and bootstrap
disconnected (0 ohm jumper
removed) and a 1/20W leaded 1k
resistor bodged in to make an
inverting amp with a gain of
-0.5.
I\'m looking at the trace
capacitance to figure out if that
might be it. There\'s about 3/4
inch of 10-mil trace on the
summing junction, but that ought
to produce a high frequency peak
if anything. hard to find 1.4 pF
across the feedback resistor.
Once I\'m back in the lab I\'ll
measure a bare board with a
Boonton and see.
Cheers
Phil Hobbs
What\'s the board stackup? Not the
specified one, but the real one.
I\'ve been burned by what some of
the fast-turn proto houses do.
https://www.dropbox.com/s/p3vpbaofzqurebz/Z462_PCB_Way_2.png?raw=1
The SJ capacitance is 2.4 pF, as measured on a Boonton, about
twice what I expected. That seems to
be the issue--in simulation it
produces a pretty big gain peak,
which reduces the bandwidth.
Time to Dremel the ground plane.
How many layers?
Only four, but of course ground is L2 and
there\'s an L3 ground pour in that area.
It is PCBway, so maybe they did the same
thing to me. (A generally very good
outfit in many ways, especially price and
delivery.)
Several of the chinese quick-turn houses
make 4-layer boards with very thin (like 4
mil) outer dielectrics. Maybe they roll
process the outers and glue them to a core
or something.
Maybe section the board to see what the
stackup actually is.
Monday I\'ll look at it under the good
microscope. (About three years ago, I got
a beautiful Mitutoyo FS-110 with 2x-50x
long working-distance objectives for $2k
on eBay. Apparently the guy didn\'t know
what he had, because he shipped this
massive precision instrument in a
cardboard box with foam peanuts. The box
was a mess when it got here, but the
scope survived because it\'s a beast.)
I sheared and sandpapered the example I
posted, and shot it with my super-good
microscope
https://www.amazon.com/gp/product/B01IV0TV50/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&psc=1
I have one a bit like that, but with a better base:
https://www.ebay.com/itm/Andonstar-500X-2MP-USB-Digital-Microscope-Video-webcam-Magnifier-Camera-Stand/143844969460
It does need a rubber band to apply a rotational preload to the arm.
Here is a Dremel-optimized SMA
connector.
https://www.dropbox.com/s/gc5j45995nqiftu/Isola_Right_Trim.jpg?raw=1
https://www.dropbox.com/s/t0cnfe9do4slvvf/Isola_Trimmed_Right_TDR.jpg?raw=1
Nice.
We\'ve had that problem too--you need some
extra pad area for the solder fillet, but
that makes a nasty capacitive
discontinuity at the connector.
The impedance of that SMA edge-launch
connector, center pin to the four ground
pins, is about 100 ohms in free air. So the
PCB has to be about 100 in the pin region
too. We\'ve worked that out, cutting away
inners and paving over the bottom with
ground. We simulated the whole geometry
with ATLC to get the dims right. That was
cool.
The case I posted was testing a laminate
sample, 20 mils thick.
The $1.50 edge-launches are just as good as
the $12 microwave connectors if the layout
is right, at least as far as we can resolve
with 30 ps TDR.
I sometimes cut away layer 2 (or more)
under critical circuit nodes. My triggered
Colpitts oscillator has a driven guard
patch on layer 5.
You bootstrap photodiodes, so you might
bootstrap the PCB too.
I\'ve often done that in high-Z TIAs, like
10M. The bootstrap effectiveness is limited
by the capacitance to ground from the summing
junction.
I wasn\'t expecting it to be a big issue with
a 511-ohm TIA, but then I wasn\'t expecting
that much capacitance either. I\'ll have to
calculate whether the SAV-551+ can bootstrap
the pour as well as the SiPM.
The usual local feedback tricks such as the
PNP wraparound sort of need fast PNPs, which
are no longer made.
Renesas/Intersil ones are a sort-of
exception.)
Well, I took out the Dremel, with one of JL\'s
fave dental burrs on it, and got the SJ
capacitance down to 1.2 pF. It was actually
pretty easy, and the results are sort of
interesting-looking.
https://electrooptical.net/www/sed/SiPMdremelled2.png
Top-to-bottom dimension is 18 mm.
Is that a 2-layer board? They are usually about
15 pf per square inch, with a lot of fringing.
No, it\'s 4 layers, with ground on L2. The #40 wire
going down the middle of the canyon there replaces
an L3 trace.
If we get our pcb laser blaster going, we\'ll
mostly make 2-sided boards with all ground plane
on the back. It will also be a huge time saver if
we leave a lot of copper on the top, namely blast
away just enough copper to make the insulated
features. That will make a lot of capacitance,
and lots of coplanar waveguide.
I\'m envisioning secondary Dremel operations, or
maybe a lot of lasering.
Fun. BTW PCBway recently cut their assembly prices
by a lot.
Cheers
Phil Hobbs
Did you ever section that board? Maybe the L2 ground
plane is very close to the layer 1 traces. Seems like
a lot of c for some tiny features like that.
We got our own p+p line because kitting and logistics
were a hassle for using outside assemblers.
Yeah, I cut one with snips and sanded it. The prepreg
was a bit skinny, but not 0.1 mm. Mostly I got
snookered by the AD8045 datasheet\'s specsmanship--their
headline 1-GHz bandwidth number turns out to be a
result of the details of the frequency compensation,
and only applies in the noninverting configuration.
Otherwise it\'s basically a pretty nice 600-MHz GBW op
amp, but not something for sub-nanosecond work.
The next version will connect the SAV-551+ bootstrap
directly to the inverting input of an EL5166 CFA, and
add a well-filtered chopamp to take out the resulting
DC offset. The pHEMT\'s source won\'t even notice the
current noise of the CFA, whereas if it were connected
to the PD it would dominate the noise.
A nuisance, but quite a preventable one--my least
favourite kind.
Bootstrap into opamp inverting input? I don\'t understand
that.
Do you know about THS4303? It\'s a roughly current-mode
opamp with an internal g=+10 feedback network. 1.8 GHz
net bw, and noise is pretty good.
If you hang a pHEMT bootstrap on a photodiode, the gate and
source go up and down together to decent accuracy.
The classical bootstrap architecture is off to the side, so
that the TIA connects directly to the cathode of the PD.
That makes the DC and low-frequency behaviour of the
bootstrap device essentially irrelevant, which is nice.
The down side is that the current noise of the TIA stage
gets summed with the photocurrent, which can be a problem
in low light.
Alternatively, though, you can connect the op amp to the
source of the bootstrap device. There are op amps such as
the LM6171A, which is basically a CFA with a follower
driving its inverting input--this notion just uses the
follower to bootstrap the PD as well as driving the
inverting input. (Our QL01 nanowatt photoreceiver does
that.)
Since the bootstrap\'s source follows its gate, this is
pretty much equivalent except that (on the plus side) you
can use much gnarlier op amps, and (on the minus side) the
bootstrap has to have decent DC and low-frequency
behaviour.
I still don\'t understand. If the bootstrap output drives the
opamp, what is the input to the bootstrap?
Got a sketch?
https://electrooptical.net/www/sed/VFandCFbootstraps.pdf
Oh, OK, the fet is both the follower into the amp and the
bootstrap. I didn\'t understand where Rfb connected.
Some of our fast o/e conveters use an un-bootstrapped photodiode
right into the inverting input of a nasty gnarly cfb amp (opamps
have feelings too!)
\"nasty gnarly\" is a term of adulation in some circles.
current fave is the EL5166. Its gnarliness knows few bounds.
but we use pd\'s with under 1 pF capacitance, and we work with
milliwatts of light.
Nice when that happens. Then you can concentrate on making the
step response beautiful.
BTW we\'ve been using the P400 to calibrate out 24-channel time
stretcher, which uses an RC ramp and one FIN1002 per channel,
switching a SAV-551+ via a fast Schottky diode. (We talked about
that a month or so ago--it\'s the one where I\'m forward-biasing the
gate.)
Looks like we can get 25-ps accuracy and single-digit picosecond
jitter, which I would not have expected. We\'ve got it taking a
whole bunch of data over TxG so that we can look at the effects of
(1) ramp slope on propagation delay and (2) signal voltage on the
aperture time. (The pHEMT switches at V_GS ~ 0.3V, but that
happens near the top of the switching edge at high signal voltages
and near the bottom at low voltages.
Jitter in the 10s of fs RMS can be done nowadays with affordable, or
frankly cheap, parts.
Sure, but we\'re calibrating with this big complicated DDG thing.
We\'re using a 20-ns RC time constant on the ramp
Around 1 GHz. The pHEMT charges up an 8-pf cap connected directly to
the input of a 6-channel simultaneous-sampling ADC, so effectively the
sampling circuit is a switch driving 8 pF, a very short trace, and the
ADC input, which (apart from pad capacitance) looks basically like a
40-ohm resistor and a 12-pF cap. So there\'s a fast bit and a slow bit
that have to be fixed in software.
> It\'s a change from using diodes.
Sure is. I\'m trying to use SAV-5xx+es in everything, to help persuade
MCL to keep making them. (They\'re generally very good about doing that
anyway.)
BTW I notice that NTE is selling the NTE2403, a near-replica of the
BFT92 5-GHz PNP. They cost $3, and have a sucky datasheet, but what can
you do? Digikey claims they\'re in active production.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510
http://electrooptical.net
http://hobbs-eo.com