RF Black Magic- Eliminating Fringing Capacitance...

[server]

On Tue, 17 Nov 2020 05:36:14 +0100, Gerhard Hoffmann <dk4xp@arcor.de>
wrote:

Am 16.11.20 um 17:48 schrieb Brent Locher:



First off, this calibration \"error\" is going to only effect phase readings on either S21 or S11. Is this several degrees of phase error something that is important to you? If not then move on.


This here is an amplifier made by a regular of s.e.d. who is currently
badly missed.

https://www.flickr.com/photos/137684711@N07/49922062651/in/album-72157662535945536/


The amplifier is potentially unstable. Between 100 KHz and 50 MHz, the
real part of the input impedance is negative. The rest of the input
impedance is a small capacitor. If you connect anything inductive to
the input, the amplifier will oscillate. The measured phase angle will
tell the oscillation frequency.

The proper measurement to diagnose this would be a measurement of S11.
From a different amplifier that shares the same problem:

https://www.flickr.com/photos/137684711@N07/34701106245/in/album-72157662535945536/

Around Marker2, the trajectory leaves the unit circle, meaning that
if you send a wave into that port at these frequencies, a bigger wave
will come back, the effect of the negative real part.



BTW, I have given up to cure that kind of amplifier. I have made one
without feedback around the FET. Everything else either loses a lot
of bandwidth to the point of being uninteresting or plays games with
parameters that are badly defined. :-(

Gerhard

I always put a title block in my sims: title, author, date. It\'s an
engineering document.



--

John Larkin Highland Technology, Inc

Science teaches us to doubt.

Claude Bernard

 

[Gerhard Hoffmann]

Am 17.11.20 um 16:00 schrieb Phil Hobbs:

But, that all does not answer the question: DOES IT INVERT???
I admit, 270° is hard to decide, even when one assumes an error band.

Cheers, Gerhard


Nah, if it were 270 degrees noninverting right at its output, you\'d have
a 3-pole rolloff.

It could also be a 1 THz gain block immersed
in 23.23 mm of 50 Ohm coplanar waveguide on Rogers TMM6
or any weird combination. Or multiple parallel amplifiers
force-matched with quadrature hybrids.
No assumptions on the innards. Just a black box.

The question is still unanswered: is 270° inverting,
or for goddess\' sake 45° or 90° or whatever.
Somewhat inverting?

For gain blocks, JL could always peek at the phase of
S21 at 5 MHz or so.


Darlingtons fall at 12 dB/octave which makes them unfriendly to
wrap feedback around unless used well below their bandwidth.
Gain is the product of 2 transistors with similar transition frequency.


> Cheers

Gerhard

 

[Gerhard Hoffmann]

Am 17.11.20 um 17:28 schrieb jlarkin@highlandsniptechnology.com:

I always put a title block in my sims: title, author, date. It\'s an
engineering document.

No, it\'s a screen dump, made on 2020.05.22 at 04:36, definitely from MY
virtual machine, with filename, put on flicker for a discussion.

The engineering document is 35 MBytes of .odt files with pics and
descriptions that have all the history, including the fails and
the whys of the fails.

In avionics, you must document the entire history. You cannot
present any results out of nothing. Even if they look good.
They don\'t like it when things fall from the sky. Even results.

Gerhard

 

[server]

On Tue, 17 Nov 2020 18:21:05 +0100, Gerhard Hoffmann <dk4xp@arcor.de>
wrote:

Am 17.11.20 um 16:00 schrieb Phil Hobbs:

But, that all does not answer the question: DOES IT INVERT???
I admit, 270° is hard to decide, even when one assumes an error band.

Cheers, Gerhard


Nah, if it were 270 degrees noninverting right at its output, you\'d have
a 3-pole rolloff.

It could also be a 1 THz gain block immersed
in 23.23 mm of 50 Ohm coplanar waveguide on Rogers TMM6
or any weird combination. Or multiple parallel amplifiers
force-matched with quadrature hybrids.
No assumptions on the innards. Just a black box.

The question is still unanswered: is 270° inverting,
or for goddess\' sake 45° or 90° or whatever.
Somewhat inverting?

For gain blocks, JL could always peek at the phase of
S21 at 5 MHz or so.

One simple Spice model, a small block of text, would tell us
everything about a MMIC, except maybe noise. It would tell us about
bias, impedances, temperature effects, gain, small and large-signal
behavior, voltage swing, everything.

Thousands of s-parameter numbers give us a peek at a few points of
small-signal AC response at a few frequencies and usually one bias
current. The mmic folks spend tons on developing s and x parameters
and refuse to provide a simple Spice model.

I think that s-params are a relic of pre-computer days when people
designed with slide rules and graph paper.



--

John Larkin Highland Technology, Inc

Science teaches us to doubt.

Claude Bernard

 

[John Larkin]

On Tue, 17 Nov 2020 18:38:40 +0100, Gerhard Hoffmann <dk4xp@arcor.de>
wrote:

Am 17.11.20 um 17:28 schrieb jlarkin@highlandsniptechnology.com:


I always put a title block in my sims: title, author, date. It\'s an
engineering document.

No, it\'s a screen dump, made on 2020.05.22 at 04:36, definitely from MY
virtual machine, with filename, put on flicker for a discussion.

The engineering document is 35 MBytes of .odt files with pics and
descriptions that have all the history, including the fails and
the whys of the fails.

In avionics, you must document the entire history. You cannot
present any results out of nothing. Even if they look good.
They don\'t like it when things fall from the sky. Even results.

Gerhard

There are people who don\'t comment code either.

 

[Cursitor Doom]

On Tue, 17 Nov 2020 11:10:25 -0800, John Larkin
<jlarkin@highland_atwork_technology.com> wrote:

>There are people who don\'t comment code either.

Then sack \'em.

 

[John Doe]

Cursitor Doom <cd@noreply.com> wrote:

John Larkin
jlarkin@highland_atwork_technology.com> wrote:

There are people who don\'t comment code either.

Then sack \'em.

For my only program, a little C++ systemwide keyboard and mouse hook,
originally written by Paul DiLascia...

{//comment about what happens
...code
}//comment about what happens (duplicate)

Might be ugly, but that\'s neat IMO because it makes use of aligned braces.

Less important...

And as the braces are nested, the prior comments get abbreviated.

LRESULT CALLBACK KeyboardHook(int nCode,WPARAM wParam,LPARAM lParam)
{//key action
if(UnknownAction)
{//K, unusable message
return::CallNextHookEx(hKeyboardHook,nCode,wParam,lParam);
}//K, unusable message
if(GetTitle)
{//K, reading input for window switching
if(KeyUp)
{//K-READ, upstroke
if((wParam==VK_OEM_5)&&(GetKeyState(VK_SHIFT)&0x8000))
{//K-READ-UP, switch windows
....

 

[Phil Hobbs]

On 11/17/20 11:24 AM, jlarkin@highlandsniptechnology.com wrote:

On Tue, 17 Nov 2020 10:00:32 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 12:58 AM, Gerhard Hoffmann wrote:
Am 17.11.20 um 06:34 schrieb Steve Wilson:

A phase shift of 270 degrees at 5GHz could be caused by a
propagation delay of 50 picoseconds through the amplifier.

Calculations:

1/5e9 = 0.2ns = 200ps 180 degrees = 100ps 270 degrees = 100ps +
50ps

50ps is reasonable for two transistors, (emitter follower plus
inverter), at 5GHz.

Probing a circuit to minimize time shift error is difficult at
5GHz.

Matching two probes at 5GHz is difficult.

No, not really. My, nowadays somewhat historic 2.5 GHz 1152A
probes are the pure joy (apart of running somewhat hot for my
fingers). Having the deeper pockets of my customers, something
faster should be possible.

But, that all does not answer the question: DOES IT INVERT??? I
admit, 270° is hard to decide, even when one assumes an error
band.

Cheers, Gerhard


Nah, if it were 270 degrees noninverting right at its output, you\'d
have a 3-pole rolloff.

Cheers

Phil Hobbs


Here\'s the TDR/TDT response of a GALI-59, a classic Darlington mmic:

https://www.dropbox.com/s/uyvhwi5jplo36ym/Gali-59_TDR.JPG?raw=1

Nice edge, 200-350 ps depending on where you measure. It\'s also
unusually well matched to 50 ohms--your average amplifier has a VSWR
around 2:1. Settling time, meh. Still a few percent off at ~2 ns.

Of one wants maximum pulse amplitude for narrow pulses, you can pull
the input up or down to bias the output off-center.

Folks are often surprised when (after a bit of mental math) I tell them
that they\'ll get to the shot noise limit by just connecting their
photodiode directly to the input of a $10 MMIC. Happens fairly often.

It may surprise the RF boys that not all waveforms are symmetric.
S-parameters are small-signal linear, which isn\'t terribly
interesting for pulse work. S11 cares a lot about the load too, since
the input impedance (and the DC bias!) is mostly set by the feedback
resistor from the output.

Well, that\'s S21, to be fair.

This is a GVA-63, which is a newfangled self-biasing part:

https://www.dropbox.com/s/dhmmakm4xs1stuo/GVA-63_TDR.jpg?raw=1

Blech. Looks like the device is turning completely off. If it were the
biasing loop or AC coupling, I\'d expect the drooly settling transient to
slope the other way. I suppose the bias loop might just be slow.

The s-param charts for these typically cut off on the low-frequency
end to hide what\'s going on. Just apply 5 volts through an inductor
and don\'t ask questions.

Maximum voltage swing is another thing that has to be found by
experiment.

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

 

[Phil Hobbs]

On 11/17/20 12:21 PM, Gerhard Hoffmann wrote:

Am 17.11.20 um 16:00 schrieb Phil Hobbs:

But, that all does not answer the question: DOES IT INVERT???
I admit, 270° is hard to decide, even when one assumes an error band.

Cheers, Gerhard


Nah, if it were 270 degrees noninverting right at its output, you\'d
have a 3-pole rolloff.

It could also be a 1 THz gain block immersed
in 23.23 mm of 50 Ohm coplanar waveguide on Rogers TMM6
or any weird combination. Or multiple parallel amplifiers
force-matched with quadrature hybrids.
No assumptions on the innards. Just a black box.

The question is still unanswered: is 270° inverting,
or for goddess\' sake 45° or 90° or whatever.
Somewhat inverting?

For gain blocks, JL could always peek at the phase of
S21 at 5 MHz or so.

Right. An inverting amp is still an inverting amp if you hang a long
cable on it.

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

 

[server]

On Wed, 18 Nov 2020 08:10:11 -0500, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 11:24 AM, jlarkin@highlandsniptechnology.com wrote:
On Tue, 17 Nov 2020 10:00:32 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 12:58 AM, Gerhard Hoffmann wrote:
Am 17.11.20 um 06:34 schrieb Steve Wilson:

A phase shift of 270 degrees at 5GHz could be caused by a
propagation delay of 50 picoseconds through the amplifier.

Calculations:

1/5e9 = 0.2ns = 200ps 180 degrees = 100ps 270 degrees = 100ps +
50ps

50ps is reasonable for two transistors, (emitter follower plus
inverter), at 5GHz.

Probing a circuit to minimize time shift error is difficult at
5GHz.

Matching two probes at 5GHz is difficult.

No, not really. My, nowadays somewhat historic 2.5 GHz 1152A
probes are the pure joy (apart of running somewhat hot for my
fingers). Having the deeper pockets of my customers, something
faster should be possible.

But, that all does not answer the question: DOES IT INVERT??? I
admit, 270° is hard to decide, even when one assumes an error
band.

Cheers, Gerhard


Nah, if it were 270 degrees noninverting right at its output, you\'d
have a 3-pole rolloff.

Cheers

Phil Hobbs


Here\'s the TDR/TDT response of a GALI-59, a classic Darlington mmic:

https://www.dropbox.com/s/uyvhwi5jplo36ym/Gali-59_TDR.JPG?raw=1

Nice edge, 200-350 ps depending on where you measure. It\'s also
unusually well matched to 50 ohms--your average amplifier has a VSWR
around 2:1. Settling time, meh. Still a few percent off at ~2 ns.

Of one wants maximum pulse amplitude for narrow pulses, you can pull
the input up or down to bias the output off-center.

Folks are often surprised when (after a bit of mental math) I tell them
that they\'ll get to the shot noise limit by just connecting their
photodiode directly to the input of a $10 MMIC. Happens fairly often.

My latest 1-GHz o/e converter dumps one end of the photodiode into a
MMIC and the other end into the DC path, only it uses a 99 cent MMIC.

It worked fine for the longwave version but was crazy slow for the 850
version. Turns out that the longwave photodiode has 20 pF from the pd
cathode to the can! We just got some new Appointech diodes that are
below 1 pF all around, and Jonathan is trying them.

I was tempted to dump my photodiode current into a resistor to ground,
maybe peak that a tad, then voltage amplify it. But a certain book
trashed that architecture.

It may surprise the RF boys that not all waveforms are symmetric.
S-parameters are small-signal linear, which isn\'t terribly
interesting for pulse work. S11 cares a lot about the load too, since
the input impedance (and the DC bias!) is mostly set by the feedback
resistor from the output.

Well, that\'s S21, to be fair.

This is a GVA-63, which is a newfangled self-biasing part:

https://www.dropbox.com/s/dhmmakm4xs1stuo/GVA-63_TDR.jpg?raw=1

Blech. Looks like the device is turning completely off. If it were the
biasing loop or AC coupling, I\'d expect the drooly settling transient to
slope the other way. I suppose the bias loop might just be slow.

I guess they could have done a real DC bias loop. The slow loop is
probably great for RF, but not for pulses.

The s-param charts for these typically cut off on the low-frequency
end to hide what\'s going on. Just apply 5 volts through an inductor
and don\'t ask questions.

Maximum voltage swing is another thing that has to be found by
experiment.

Cheers

Phil Hobbs

Does anyone remember h-parameters? They were an attempt to analyze
transistors with algebra, before computers and Spice were born.

My always-humble prediction is that time-domain (Spice and such)
analysis will eventually kill off s-params, x-params, z-params, load
pull testing, all that slide rule/graph paper stuff. We would need a
lot of compute power; maybe Mike is working on that. Nvidia?



--

John Larkin Highland Technology, Inc

Science teaches us to doubt.

Claude Bernard

 

[Phil Hobbs]

On 11/18/20 10:57 AM, jlarkin@highlandsniptechnology.com wrote:

On Wed, 18 Nov 2020 08:10:11 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 11:24 AM, jlarkin@highlandsniptechnology.com wrote:
On Tue, 17 Nov 2020 10:00:32 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 12:58 AM, Gerhard Hoffmann wrote:
Am 17.11.20 um 06:34 schrieb Steve Wilson:

A phase shift of 270 degrees at 5GHz could be caused by a
propagation delay of 50 picoseconds through the amplifier.

Calculations:

1/5e9 = 0.2ns = 200ps 180 degrees = 100ps 270 degrees = 100ps +
50ps

50ps is reasonable for two transistors, (emitter follower plus
inverter), at 5GHz.

Probing a circuit to minimize time shift error is difficult at
5GHz.

Matching two probes at 5GHz is difficult.

No, not really. My, nowadays somewhat historic 2.5 GHz 1152A
probes are the pure joy (apart of running somewhat hot for my
fingers). Having the deeper pockets of my customers, something
faster should be possible.

But, that all does not answer the question: DOES IT INVERT??? I
admit, 270° is hard to decide, even when one assumes an error
band.

Cheers, Gerhard


Nah, if it were 270 degrees noninverting right at its output, you\'d
have a 3-pole rolloff.

Cheers

Phil Hobbs


Here\'s the TDR/TDT response of a GALI-59, a classic Darlington mmic:

https://www.dropbox.com/s/uyvhwi5jplo36ym/Gali-59_TDR.JPG?raw=1

Nice edge, 200-350 ps depending on where you measure. It\'s also
unusually well matched to 50 ohms--your average amplifier has a VSWR
around 2:1. Settling time, meh. Still a few percent off at ~2 ns.

Of one wants maximum pulse amplitude for narrow pulses, you can pull
the input up or down to bias the output off-center.

Folks are often surprised when (after a bit of mental math) I tell them
that they\'ll get to the shot noise limit by just connecting their
photodiode directly to the input of a $10 MMIC. Happens fairly often.

My latest 1-GHz o/e converter dumps one end of the photodiode into a
MMIC and the other end into the DC path, only it uses a 99 cent MMIC.

It worked fine for the longwave version but was crazy slow for the 850
version. Turns out that the longwave photodiode has 20 pF from the pd
cathode to the can! We just got some new Appointech diodes that are
below 1 pF all around, and Jonathan is trying them.

I was tempted to dump my photodiode current into a resistor to ground,
maybe peak that a tad, then voltage amplify it. But a certain book
trashed that architecture.

It may surprise the RF boys that not all waveforms are symmetric.
S-parameters are small-signal linear, which isn\'t terribly
interesting for pulse work. S11 cares a lot about the load too, since
the input impedance (and the DC bias!) is mostly set by the feedback
resistor from the output.

Well, that\'s S21, to be fair.

This is a GVA-63, which is a newfangled self-biasing part:

https://www.dropbox.com/s/dhmmakm4xs1stuo/GVA-63_TDR.jpg?raw=1

Blech. Looks like the device is turning completely off. If it were the
biasing loop or AC coupling, I\'d expect the drooly settling transient to
slope the other way. I suppose the bias loop might just be slow.

I guess they could have done a real DC bias loop. The slow loop is
probably great for RF, but not for pulses.


The s-param charts for these typically cut off on the low-frequency
end to hide what\'s going on. Just apply 5 volts through an inductor
and don\'t ask questions.

Maximum voltage swing is another thing that has to be found by
experiment.


Does anyone remember h-parameters? They were an attempt to analyze
transistors with algebra, before computers and Spice were born.

My always-humble prediction is that time-domain (Spice and such)
analysis will eventually kill off s-params, x-params, z-params, load
pull testing, all that slide rule/graph paper stuff. We would need a
lot of compute power; maybe Mike is working on that. Nvidia?

Well, small-signal analysis will still be linear. You need your amp to
be stable with zero input, after all.

H-parameters maintain a tenuous and shadowy existence because a capital
beta is just B, so there\'s no good nomenclature for distinguishing
large- and small-signal current gain except H_FE vs h_FE. Otherwise I
know of no good reason to use them. Nobody I know of uses impedance or
admittance matrices any more. (I actually never did.)

S-parameters are useful because 50-ohms in two leads of any device is
generally enough that it won\'t oscillate, unlike shorts and opens! They
also have a natural connection with Smith charts, which are sort of an
RF version of Karnaugh maps--you can easily visualize what sort of
matching network you\'re going to need. (Or I used to, back when I was
an RF guy long ago.)

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

 

[Cursitor Doom]

On Wed, 18 Nov 2020 13:55:06 -0500, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/18/20 10:57 AM, jlarkin@highlandsniptechnology.com wrote:
On Wed, 18 Nov 2020 08:10:11 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 11:24 AM, jlarkin@highlandsniptechnology.com wrote:
On Tue, 17 Nov 2020 10:00:32 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 12:58 AM, Gerhard Hoffmann wrote:
Am 17.11.20 um 06:34 schrieb Steve Wilson:

A phase shift of 270 degrees at 5GHz could be caused by a
propagation delay of 50 picoseconds through the amplifier.

Calculations:

1/5e9 = 0.2ns = 200ps 180 degrees = 100ps 270 degrees = 100ps +
50ps

50ps is reasonable for two transistors, (emitter follower plus
inverter), at 5GHz.

Probing a circuit to minimize time shift error is difficult at
5GHz.

Matching two probes at 5GHz is difficult.

No, not really. My, nowadays somewhat historic 2.5 GHz 1152A
probes are the pure joy (apart of running somewhat hot for my
fingers). Having the deeper pockets of my customers, something
faster should be possible.

But, that all does not answer the question: DOES IT INVERT??? I
admit, 270° is hard to decide, even when one assumes an error
band.

Cheers, Gerhard


Nah, if it were 270 degrees noninverting right at its output, you\'d
have a 3-pole rolloff.

Cheers

Phil Hobbs


Here\'s the TDR/TDT response of a GALI-59, a classic Darlington mmic:

https://www.dropbox.com/s/uyvhwi5jplo36ym/Gali-59_TDR.JPG?raw=1

Nice edge, 200-350 ps depending on where you measure. It\'s also
unusually well matched to 50 ohms--your average amplifier has a VSWR
around 2:1. Settling time, meh. Still a few percent off at ~2 ns.

Of one wants maximum pulse amplitude for narrow pulses, you can pull
the input up or down to bias the output off-center.

Folks are often surprised when (after a bit of mental math) I tell them
that they\'ll get to the shot noise limit by just connecting their
photodiode directly to the input of a $10 MMIC. Happens fairly often.

My latest 1-GHz o/e converter dumps one end of the photodiode into a
MMIC and the other end into the DC path, only it uses a 99 cent MMIC.

It worked fine for the longwave version but was crazy slow for the 850
version. Turns out that the longwave photodiode has 20 pF from the pd
cathode to the can! We just got some new Appointech diodes that are
below 1 pF all around, and Jonathan is trying them.

I was tempted to dump my photodiode current into a resistor to ground,
maybe peak that a tad, then voltage amplify it. But a certain book
trashed that architecture.

It may surprise the RF boys that not all waveforms are symmetric.
S-parameters are small-signal linear, which isn\'t terribly
interesting for pulse work. S11 cares a lot about the load too, since
the input impedance (and the DC bias!) is mostly set by the feedback
resistor from the output.

Well, that\'s S21, to be fair.

This is a GVA-63, which is a newfangled self-biasing part:

https://www.dropbox.com/s/dhmmakm4xs1stuo/GVA-63_TDR.jpg?raw=1

Blech. Looks like the device is turning completely off. If it were the
biasing loop or AC coupling, I\'d expect the drooly settling transient to
slope the other way. I suppose the bias loop might just be slow.

I guess they could have done a real DC bias loop. The slow loop is
probably great for RF, but not for pulses.


The s-param charts for these typically cut off on the low-frequency
end to hide what\'s going on. Just apply 5 volts through an inductor
and don\'t ask questions.

Maximum voltage swing is another thing that has to be found by
experiment.


Does anyone remember h-parameters? They were an attempt to analyze
transistors with algebra, before computers and Spice were born.

My always-humble prediction is that time-domain (Spice and such)
analysis will eventually kill off s-params, x-params, z-params, load
pull testing, all that slide rule/graph paper stuff. We would need a
lot of compute power; maybe Mike is working on that. Nvidia?

Well, small-signal analysis will still be linear. You need your amp to
be stable with zero input, after all.

H-parameters maintain a tenuous and shadowy existence because a capital
beta is just B, so there\'s no good nomenclature for distinguishing
large- and small-signal current gain except H_FE vs h_FE. Otherwise I
know of no good reason to use them. Nobody I know of uses impedance or
admittance matrices any more. (I actually never did.)

S-parameters are useful because 50-ohms in two leads of any device is
generally enough that it won\'t oscillate, unlike shorts and opens! They
also have a natural connection with Smith charts, which are sort of an
RF version of Karnaugh maps--you can easily visualize what sort of
matching network you\'re going to need. (Or I used to, back when I was
an RF guy long ago.)

You\'ve both forgotten about Y-parameters. But in any event, S\'s are
the only ones worth paying attention to today. The rest have fallen by
the wayside...

 

[Phil Hobbs]

On 11/20/20 6:17 PM, Cursitor Doom wrote:

On Wed, 18 Nov 2020 13:55:06 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/18/20 10:57 AM, jlarkin@highlandsniptechnology.com wrote:
On Wed, 18 Nov 2020 08:10:11 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 11:24 AM, jlarkin@highlandsniptechnology.com wrote:
On Tue, 17 Nov 2020 10:00:32 -0500, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 11/17/20 12:58 AM, Gerhard Hoffmann wrote:
Am 17.11.20 um 06:34 schrieb Steve Wilson:

A phase shift of 270 degrees at 5GHz could be caused by a
propagation delay of 50 picoseconds through the amplifier.

Calculations:

1/5e9 = 0.2ns = 200ps 180 degrees = 100ps 270 degrees = 100ps +
50ps

50ps is reasonable for two transistors, (emitter follower plus
inverter), at 5GHz.

Probing a circuit to minimize time shift error is difficult at
5GHz.

Matching two probes at 5GHz is difficult.

No, not really. My, nowadays somewhat historic 2.5 GHz 1152A
probes are the pure joy (apart of running somewhat hot for my
fingers). Having the deeper pockets of my customers, something
faster should be possible.

But, that all does not answer the question: DOES IT INVERT??? I
admit, 270° is hard to decide, even when one assumes an error
band.

Cheers, Gerhard


Nah, if it were 270 degrees noninverting right at its output, you\'d
have a 3-pole rolloff.

Cheers

Phil Hobbs


Here\'s the TDR/TDT response of a GALI-59, a classic Darlington mmic:

https://www.dropbox.com/s/uyvhwi5jplo36ym/Gali-59_TDR.JPG?raw=1

Nice edge, 200-350 ps depending on where you measure. It\'s also
unusually well matched to 50 ohms--your average amplifier has a VSWR
around 2:1. Settling time, meh. Still a few percent off at ~2 ns.

Of one wants maximum pulse amplitude for narrow pulses, you can pull
the input up or down to bias the output off-center.

Folks are often surprised when (after a bit of mental math) I tell them
that they\'ll get to the shot noise limit by just connecting their
photodiode directly to the input of a $10 MMIC. Happens fairly often.

My latest 1-GHz o/e converter dumps one end of the photodiode into a
MMIC and the other end into the DC path, only it uses a 99 cent MMIC.

It worked fine for the longwave version but was crazy slow for the 850
version. Turns out that the longwave photodiode has 20 pF from the pd
cathode to the can! We just got some new Appointech diodes that are
below 1 pF all around, and Jonathan is trying them.

I was tempted to dump my photodiode current into a resistor to ground,
maybe peak that a tad, then voltage amplify it. But a certain book
trashed that architecture.

It may surprise the RF boys that not all waveforms are symmetric.
S-parameters are small-signal linear, which isn\'t terribly
interesting for pulse work. S11 cares a lot about the load too, since
the input impedance (and the DC bias!) is mostly set by the feedback
resistor from the output.

Well, that\'s S21, to be fair.

This is a GVA-63, which is a newfangled self-biasing part:

https://www.dropbox.com/s/dhmmakm4xs1stuo/GVA-63_TDR.jpg?raw=1

Blech. Looks like the device is turning completely off. If it were the
biasing loop or AC coupling, I\'d expect the drooly settling transient to
slope the other way. I suppose the bias loop might just be slow.

I guess they could have done a real DC bias loop. The slow loop is
probably great for RF, but not for pulses.


The s-param charts for these typically cut off on the low-frequency
end to hide what\'s going on. Just apply 5 volts through an inductor
and don\'t ask questions.

Maximum voltage swing is another thing that has to be found by
experiment.


Does anyone remember h-parameters? They were an attempt to analyze
transistors with algebra, before computers and Spice were born.

My always-humble prediction is that time-domain (Spice and such)
analysis will eventually kill off s-params, x-params, z-params, load
pull testing, all that slide rule/graph paper stuff. We would need a
lot of compute power; maybe Mike is working on that. Nvidia?

Well, small-signal analysis will still be linear. You need your amp to
be stable with zero input, after all.

H-parameters maintain a tenuous and shadowy existence because a capital
beta is just B, so there\'s no good nomenclature for distinguishing
large- and small-signal current gain except H_FE vs h_FE. Otherwise I
know of no good reason to use them. Nobody I know of uses impedance or
admittance matrices any more. (I actually never did.)

S-parameters are useful because 50-ohms in two leads of any device is
generally enough that it won\'t oscillate, unlike shorts and opens! They
also have a natural connection with Smith charts, which are sort of an
RF version of Karnaugh maps--you can easily visualize what sort of
matching network you\'re going to need. (Or I used to, back when I was
an RF guy long ago.)

You\'ve both forgotten about Y-parameters. But in any event, S\'s are
the only ones worth paying attention to today. The rest have fallen by
the wayside...

Y-parameters are the admittance parameters I mentioned.

In equivalently relevant news, I used to have a Gump Worsley hockey card.

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