Very fast rise time generator...

[Klaus Vestergaard Kragelund]

On 16-02-2023 02:50, John Miles, KE5FX wrote:

On Tuesday, February 14, 2023 at 7:16:45 AM UTC-8, Klaus Kragelund wrote:
Anyone got other ideas?

An amusing one to think about:

http://www.ke5fx.com/Nanoplasma_spark_gap_picosecond_switches_for_ultrafast_electronics.pdf

From my reply to an earlier (2020) thread where someone mentioned that paper:

--------snip--------
I tried cutting a slot in the top layer of a copper PCB strip using
a scalpel, the idea being to pinch it shut manually until it fires
somewhere near the bottom of the Paschen curve:

http://www.ke5fx.com/microgap/gap.jpg (closeup)
http://www.ke5fx.com/microgap/10x.jpg (test setup for HV edge)
http://www.ke5fx.com/microgap/50R.jpg (test setup for 50 ohms)

Driving it at -350V through a 100K resistor makes the gap fire somewhat
randomly at about 100 V/ns, limited by the probe and various strays.
The recovery time is also swamped by the scope probe RC:

http://www.ke5fx.com/microgap/mso_10x_probe_1us_div.png
http://www.ke5fx.com/microgap/mso_10x_probe_1ns_div.png

A 50-ohm series tap is faster, but likely still limited by strays:

http://www.ke5fx.com/microgap/mso_50R_1ns_div.png (MSO6054A, 500 MHz BW)
http://www.ke5fx.com/microgap/tds_50R_1ns.gif (TDS 694C, 3000 MHz BW)

I did see some edges closer to .35/3000 = 117 ps, but 164 ps was the
fastest one that I saved before the PCB strip finally broke.
--------end snip--------

So there\'s your 30 volts, and then some, I guess. Good luck in the EMC lab...

Very interesting. I need something with longer lifetime, but nice
information anyway.

Often when you reverse engineer Chinese products, they will use a spark
gap arrester in the form of two pointy pads between Line and Neutral.

Cheap, but will probably not take many hits.

 

[Anthony Stewart]

On Thursday, February 16, 2023 at 3:41:26 a.m. UTC-5, Klaus Vestergaard Kragelund wrote:

On 16-02-2023 02:50, John Miles, KE5FX wrote:
On Tuesday, February 14, 2023 at 7:16:45 AM UTC-8, Klaus Kragelund wrote:
Anyone got other ideas?

An amusing one to think about:

http://www.ke5fx.com/Nanoplasma_spark_gap_picosecond_switches_for_ultrafast_electronics.pdf

From my reply to an earlier (2020) thread where someone mentioned that paper:

--------snip--------
I tried cutting a slot in the top layer of a copper PCB strip using
a scalpel, the idea being to pinch it shut manually until it fires
somewhere near the bottom of the Paschen curve:

http://www.ke5fx.com/microgap/gap.jpg (closeup)
http://www.ke5fx.com/microgap/10x.jpg (test setup for HV edge)
http://www.ke5fx.com/microgap/50R.jpg (test setup for 50 ohms)

Driving it at -350V through a 100K resistor makes the gap fire somewhat
randomly at about 100 V/ns, limited by the probe and various strays.
The recovery time is also swamped by the scope probe RC:

http://www.ke5fx.com/microgap/mso_10x_probe_1us_div.png
http://www.ke5fx.com/microgap/mso_10x_probe_1ns_div.png

A 50-ohm series tap is faster, but likely still limited by strays:

http://www.ke5fx.com/microgap/mso_50R_1ns_div.png (MSO6054A, 500 MHz BW)
http://www.ke5fx.com/microgap/tds_50R_1ns.gif (TDS 694C, 3000 MHz BW)

I did see some edges closer to .35/3000 = 117 ps, but 164 ps was the
fastest one that I saved before the PCB strip finally broke.
--------end snip--------

So there\'s your 30 volts, and then some, I guess. Good luck in the EMC lab...

Very interesting. I need something with longer lifetime, but nice
information anyway.

Often when you reverse engineer Chinese products, they will use a spark
gap arrester in the form of two pointy pads between Line and Neutral.

Cheap, but will probably not take many hits.
John Miles has made some nice accurate signal captures. I discovered similar test results with a sparkplug with slow ramped HV excitation in Transformer oil. I was trying to measure the reduction of PDIV over time [hrs] (Partial Discharge Inception Voltage) or breakdown thresholds which when prolonged induce transformer failure from H2 outgassing.

I learned many things which may or may not be relevant to Klaus\'s purpose about these pulses, and the effects of geometry, ionization time, dielectric constant, BIL rating peak impulse/ rms breakdown ratio, effects of coax quality. Also I recall an Engineer showing me photos of picosecond captures of finger static discharge. It was an epidemic field ($) problem on 5 MVA transformers HVAC where H2 dissolved in oil was being generated by internal invisible contamination particles. When a oil sample gas analysis shows H2> 2% the transformer must be removed from service before a large scale detonation of a long hydrocarbon chain versus the nanoscale event of Partial Discharge (PD).
- The threshold will reduce with contamination ions, particles, dust, oxide, moisture.
- The random repetitive discharge rate and threshold deviation is a key indicator of contamination of a pure insulator line frequency . The ns risetime I measured was limited by the BW of the DSO using 50 ohm matched link.
- These are all negative resistance avalanches where the L/R=T is much smaller than the RC=T for the ESR during breakdown. Thus the length/width ratio must be small. The arc ESR resistance decreases with discharge current ( as verified by Faraday ? or Coulomb ?)
- Semiconductors with forward switching tend in the same family to increase Coss with the reduction of RdsOn, whereas BJTs have better results and why they are used in HV IGBTs.

There are tens of thousands of research papers on Google Scholar search that might help by changing the keywords.

https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=avalanche+pulse+high+voltage&btnG=

 

[Klaus Vestergaard Kragelund]

On 14-02-2023 18:21, Martin Rid wrote:

Klaus Kragelund <klaus.kragelund@gmail.com> Wrote in message:r
HiI am working on a very fast rise time generator. I need to supply 30V or more into a capacitive load of 50pF in less than a nanosecond, sourcing about 2A during the transient. It\'s for a prototype test, so no real concerns of cost and availability.So far I have been using a fast GaN gatedriver and a EPC2018. That performs well, about 600ps to 30V.I would like to push it even further.One idea I was thinking about was to use LVDS driver, or any other driver with below 100ps rise time. That can create a fast edge, but not to 30V. I could then trigger 10 of these circuits, and capacitive couple them in a rolling sequence (diodes in series to prevent back flow), each carefully triggered so they would build on each other.Other concept is to use a HF transistor, but I don\'t think it would be faster than the GaN, all though I have no idea, never done real HF stuff before.I was also thinking bringing a transistor into avalance, like a circuit done by Jim Thomson:https://electronicprojectsforfun.files.wordpress.com/2017/10/jimwilliamsan47applicationnode.pdf(Page 94)He states below 300ps risetime. It\'s been 30 years, surely something exists that is faster?ADCMP580 has 37ps risetime:https://www.analog.com/media/en/technical-documentation/data-sheets/ADCMP580_581_582.pdfA regular inverter switches around 1ns:https://www.ti.com/lit/ds/symlink/sn74auc2g06.pdf?HQS=dis-dk-null-digikeymode-dsf-pf-null-wwe&ts=1676384474996&ref_url=https%253A%252F%252Fwww.ti.com%252Fgeneral%252Fdocs%252Fsuppproductinfo.tsp%253FdistId%253D10%2526gotoUrl%253Dhttps%253A%252F%252Fwww.ti.com%252Flit%252Fgpn%252Fsn74auc2g06Seems about 500ps is possible with ultra cheap inverter.Here\'s a 6 USD 20ps switch:http://ww1.microchip.com/downloads/en/DeviceDoc/sy58051u.pdfAnyone got other ideas?

Have you looked at analog.com app note an94f.pdf ?

I have not, so thanks for the hint

 

[Klaus Vestergaard Kragelund]

On 14-02-2023 22:54, Uwe Bonnes wrote:

John Larkin <jlarkin@highlandsnipmetechnology.com> wrote:

The diodes Inc/Zetex avalanche transistors are still available.

Higher rates are hard to reach with avalache transistors. What frequency
does the original poster need?

I can do with below 1MHz, just need the very high risetime

 

[John Larkin]

On Thu, 09 Mar 2023 07:28:05 +1100, \"Rod Speed\"
<rod.speed.aaa@gmail.com> wrote:

On Thu, 09 Mar 2023 06:20:58 +1100, John Larkin
jlarkin@highlandsnipmetechnology.com> wrote:

On Thu, 09 Mar 2023 06:03:24 +1100, \"Rod Speed\"
rod.speed.aaa@gmail.com> wrote:

On Thu, 09 Mar 2023 02:18:15 +1100, John Larkin
jlarkin@highlandsnipmetechnology.com> wrote:

On 8 Mar 2023 03:16:12 GMT, rbowman <bowman@montana.com> wrote:

On Tue, 7 Mar 2023 12:47:32 +0000, The Natural Philosopher wrote:

Tommy Flowers just about managed to make a crude computer out of
them...

https://hackaday.com/2021/12/27/single-bit-computer-from-vacuum-tubes/

I consider myself luck to have joined the workforce at the very tail
end
of vacuum tube logic. You can implement a NOR gate and you can build
anything from NOR gates if you don\'t mind going insane.

The first DTL and TTL parts were nand and jk flops. Before that, we
had RTL which was mostly nor. Both were horrible, slow and expensive
and unreliable.

Bullshit on the unreliable.

The Fairchild and Motorola RTL parts were horrible,

Bullshit and I did the maintenance on a DEC PDP9 which
used it extensively.

as were the early
TI TTL parts.

Bullshit and I did the RTL/TTL interface for the mag tape
for the PDP15 which was the TTL version of the PDP9

You are really into poop. A lot of guys are.

They hadn\'t got the plastic packaging right yet and
temperature cycling would break wire bonds.

Never had a single failure like that.

Maybe the ceramic
flat-pack parts were OK, but we couldn\'t afford them.

I used the plastic packaging exclusively and
so did DEC and we never had any failures at all.

The PDP11/10AD didn\'t either.

By the time the PDP-11 came out, or the PDP-8I for that matter, things
were much better.

There was Purple Plague in those days too.

Never saw that either.

Look it up.

 

[Rod Speed]

On Thu, 09 Mar 2023 09:28:38 +1100, John Larkin
<jlarkin@highlandsnipmetechnology.com> wrote:

On Thu, 09 Mar 2023 07:28:05 +1100, \"Rod Speed\"
rod.speed.aaa@gmail.com> wrote:

On Thu, 09 Mar 2023 06:20:58 +1100, John Larkin
jlarkin@highlandsnipmetechnology.com> wrote:

On Thu, 09 Mar 2023 06:03:24 +1100, \"Rod Speed\"
rod.speed.aaa@gmail.com> wrote:

On Thu, 09 Mar 2023 02:18:15 +1100, John Larkin
jlarkin@highlandsnipmetechnology.com> wrote:

On 8 Mar 2023 03:16:12 GMT, rbowman <bowman@montana.com> wrote:

On Tue, 7 Mar 2023 12:47:32 +0000, The Natural Philosopher wrote:

Tommy Flowers just about managed to make a crude computer out of
them...

https://hackaday.com/2021/12/27/single-bit-computer-from-vacuum-tubes/

I consider myself luck to have joined the workforce at the very tail
end
of vacuum tube logic. You can implement a NOR gate and you can build
anything from NOR gates if you don\'t mind going insane.

The first DTL and TTL parts were nand and jk flops. Before that, we
had RTL which was mostly nor. Both were horrible, slow and expensive
and unreliable.

Bullshit on the unreliable.

The Fairchild and Motorola RTL parts were horrible,

Bullshit and I did the maintenance on a DEC PDP9 which
used it extensively.

as were the early
TI TTL parts.

Bullshit and I did the RTL/TTL interface for the mag tape
for the PDP15 which was the TTL version of the PDP9

You are really into poop. A lot of guys are.


They hadn\'t got the plastic packaging right yet and
temperature cycling would break wire bonds.

Never had a single failure like that.

Maybe the ceramic
flat-pack parts were OK, but we couldn\'t afford them.

I used the plastic packaging exclusively and
so did DEC and we never had any failures at all.

The PDP11/10AD didn\'t either.

By the time the PDP-11 came out, or the PDP-8I for that matter, things
were much better.

The PDP 15 was well before that and didnt have that problem at all.

There was Purple Plague in those days too.

Never saw that either.

Look it up.

I didn\'t need to.

 

[Jasen Betts]

On 2023-03-08, John Larkin <jlarkin@highlandSNIPMEtechnology.com> wrote:

On Thu, 09 Mar 2023 07:28:05 +1100, \"Rod Speed\"
rod.speed.aaa@gmail.com> wrote:

On Thu, 09 Mar 2023 06:20:58 +1100, John Larkin
jlarkin@highlandsnipmetechnology.com> wrote:

On Thu, 09 Mar 2023 06:03:24 +1100, \"Rod Speed\"
rod.speed.aaa@gmail.com> wrote:

On Thu, 09 Mar 2023 02:18:15 +1100, John Larkin
jlarkin@highlandsnipmetechnology.com> wrote:

On 8 Mar 2023 03:16:12 GMT, rbowman <bowman@montana.com> wrote:

On Tue, 7 Mar 2023 12:47:32 +0000, The Natural Philosopher wrote:

Tommy Flowers just about managed to make a crude computer out of
them...

https://hackaday.com/2021/12/27/single-bit-computer-from-vacuum-tubes/

I consider myself luck to have joined the workforce at the very tail
end
of vacuum tube logic. You can implement a NOR gate and you can build
anything from NOR gates if you don\'t mind going insane.

The first DTL and TTL parts were nand and jk flops. Before that, we
had RTL which was mostly nor. Both were horrible, slow and expensive
and unreliable.

Bullshit on the unreliable.

The Fairchild and Motorola RTL parts were horrible,

Bullshit and I did the maintenance on a DEC PDP9 which
used it extensively.

as were the early
TI TTL parts.

Bullshit and I did the RTL/TTL interface for the mag tape
for the PDP15 which was the TTL version of the PDP9

You are really into poop. A lot of guys are.

You\'re the one who sees it where it aint.

--
Jasen.
🇺🇦 Слава Україні

 

[Peeler]

On Thu, 09 Mar 2023 12:14:05 +1100, cantankerous trolling geezer Rodent
Speed, the auto-contradicting senile sociopath, blabbered, again:

<FLUSH the abnormal trolling senile cretin\'s latest trollshit unread>

--
williamwright addressing Rodent Speed:
\"This is getting beyond ridiculous now. You\'re trying to prove black\'s
white. You\'re arguing with someone who has been involved with the issues all
his working life when you clearly have no knowledge at all. I think you\'re
just being a pillock for the sake of it. You clearly don\'t actually believe
your own words. You must have a very empty life, and a sad embittered soul.
MID: <j08o6bFeqc1U1@mid.individual.net>

 

[John Larkin]

On 9 Mar 2023 05:08:10 GMT, rbowman <bowman@montana.com> wrote:

On Wed, 08 Mar 2023 09:33:49 -0800, John Larkin wrote:

What is megaML? It doesn\'t google well.

Not an actual term outside of my imagination. TensorFlow is a Google
endeavor and they provide a web interface called Colab where you can
execute TF in Python.

https://colab.research.google.com/notebooks/intro.ipynb

Google also builds their own specialized hardware:

https://en.wikipedia.org/wiki/Tensor_Processing_Unit

Others use GPUs like NVidias. After you construct a model you train it
with a large set of labeled data over many iterations to optimize the
weights and it\'s very processor intensive. As you might guess from tensor
you\'re deep into linear algebra.

After training the model and getting the error down to an acceptable level
you can export a .tflite file that\'s relatively small. The metaphor is
when you\'re learning something new you\'re very busy but when you get
around to applying the knowledge it\'s nowhere near as intensive so it can
be handled by an embedded processor.

The Arduino nano 33 BLE Sense is popular. It uses the Nordic nRF52840 32-
bit ARM processor and has an inertial unit, magnetic sensor, temperature,
humidity, and other sensors in a very small package.

At least that\'s the theory. In the \'80s I was programming MCUs lke the
8048 family for devices like handheld pH/ion concentration meters. I took
a course in neural networks which was going to be the Next Great Thing. It
was oversold since the hardware didn\'t exist to make it feasible.

40 years later neural networks were reborn as machine learning except now
you have racks and racks of TPUs churning away to make it work. Whether it
can be used to do useful work on MCUs is another question.

For a sense of scale Facebook (Meta) managed to leak their large language
model. It\'s only 288 GB. Nadella recently said Siri, Alexa, Cortana and so
forth are as dumb as rocks. We\'ll see what the next generation of AI buzz
brings.

OK, machine learning. In a hard embedded electronic instrument, we
want predictability and zero errors. ML and NNs and that sort of thing
seems to me to continually generate dangerous bugs.

 

[rbowman]

On Thu, 09 Mar 2023 08:56:02 -0800, John Larkin wrote:

OK, machine learning. In a hard embedded electronic instrument, we want
predictability and zero errors. ML and NNs and that sort of thing seems
to me to continually generate dangerous bugs.

It concerns me that it isn\'t too clear how it works once past the
conceptual level. The linear algebra, gradient descent, and other
algorithms are straightforward, but after you feed in a few hundred
thousand labeled data values it\'s hard to tell what you have. I read an
article a couple of days ago that some researchers are trying to apply
Fourier analysis to the models to get a better understanding.

 

[Peeler]

On 10 Mar 2023 06:10:22 GMT, lowbrowwoman, the endlessly driveling,
troll-feeding, senile idiot, blabbered again:


> It concerns me that

....that there seems to be no means to shut up a pathological grandiloquent
bigmouth and braggart like you, you self-important and self-admiring senile
shithead who just CAN\'T get enough of hearing himself talking. LOL

--
More of the senile gossip\'s absolutely idiotic senile blather:
\"I stopped for breakfast at a diner in Virginia when the state didn\'t do
DST. I remarked on the time difference and the crusty old waitress said
\'We keep God\'s time in Virginia.\'

I also lived in Ft. Wayne for a while.\"

MID: <t0tjfa$6r5$1@dont-email.me>

 

[Mike Monett VE3BTI]

Klaus Kragelund <klaus.kragelund@gmail.com> wrote:

NLTLs, shock lines, are another way to get fast high voltage pulses.
McEwan did a lot of that.

Yes, seems Thomas McEwan had a lot of patents:

https://radaris.com/f/Thomas/Mcewan/Inventor

Unfortunately, most were stolen from Larry Fullerton, who invited McEwan to
his workshops where he discussed his achievements, including the fact that
he had patented them. This didn\'t matter to McEwan, who went ahead and re-
patented them anyway.

I posted the following note on one of McEwan\'s patents:

Unfortunately, most of Tom McEwan\'s ideas were stolen from Larry
Fullerton. Here\'s some notes I took of a government inquiry into McEwan\'s
patents and the resulting cancellation of all his claims due to prior art
by Fullerton.

McEwan was also very active in stealing Fullerton\'s work in other areas.

Here are the notes, in no particular order.

The Development & Commercialization of Micropower Impulse Radar at
Lawrence Livermore National Laboratory

A Report by the Democratic Staff Committee on Science

U.S. House of Representatives

April 9, 1999

The original is available at:
http://www.house.gov/science_democrats/archive/mirrpt99.htm

LLNL/UC and the MIR inventor, Thomas McEwan, were aware of
Fullerton\'s inventions, but did not cite the inventions or other
publications describing them to the Patent Office as is required by law;

TDC inventor Larry Fullerton invented and patented the same
technology 7 years prior to LLNL/UC;

In the late 1980s, claims were made regarding the ability of UWB radar
to detect and identify stealth aircraft. The utilization of such a wide
portion of the frequency spectrum to transmit information would
reportedly enable the detection of stealth aircraft skins which absorb
conventional radar, and the use of impulses to transmit information would
reportedly allow the delineation of the sharp edges of stealth aircraft
to a much higher degree than continuous-wave radar could. These claims
were described in several press articles in Aviation Week & Space
Technology in 1989 and 1990.

Two presentations on UWB radar were given by collaborators of
Larry Fullerton at the March, 1990 LANL meeting. One presentation listed
Mr. Fullerton as a co-author, and the other referred to his proprietary
UWB radar equipment in the text of the paper.

Also in attendance at that meeting were Thomas E. McEwan of LLNL and 9
other LLNL employees. All known press reports
of the meeting highlighted Fullerton\'s work,
mentioning that he had secured several patents on the technology and
describing his inventions.

Records show that Thomas McEwan and other LLNL employees began
targeted UWB radar R&D immediately upon their return from the March, 1990
LANL meeting. Internal LLNL memos obtained by Democratic Staff indicate
that Thomas McEwan had read at least one of the press reports surrounding
the meeting which contained a description of Fullerton\'s inventions. In
September, 1990, Thomas McEwan and David Christie (at that time also of
LLNL) submitted an internal funding proposal entitled <U>Ultra-wideband
Time Domain Imaging Radar</U>. According to Mr. Christie\'s recollections,
the proposal was funded. By February, 1991, LLNL was making UWB radar
proposals to prospective industrial and governmental licensees,
advertising a cheap, sensitive, low power, ultra-wideband radar that
could fit on a single microchip.

The first LLNL Invention Disclosure forms for MIR were filed in
August, 1992. In 1993, LLNL/UC filed their first UWB radar patents on
what would become known as MIR technology, listing Thomas E. McEwan as
the inventor and UC as the patent owner. LLNL/UC did not cite any of
Larry Fullerton\'s patents, publications, or articles describing TDC\'s
inventions as prior art on their early MIR patent applications. MIR was
characterized as a cheap, sensitive, low power, ultra-wideband radar
technology that could fit on a single microchip and that was ready to be
licensed for use in a wide variety of applications including heartbeat
monitors, power tools, and automotive collision sensors. Like Fullerton\'s
radar inventions, MIR detects and identifies targets by relying on the
reception and interpretation of very short, randomly-spaced impulses of
electromagnetic energy that are reflected off the targets.

In September, 1995, after learning about the LLNL/UC patents, TDC
contacted both LLNL and DOE and conveyed their beliefs that the
inventions described in the MIR patents were extremely similar to
Fullerton\'s inventions. This led to extensive correspondence between TDC
(and its affiliated entity Pulson), LLNL/UC and DOE. LLNL/UC denied all
of TDC\'s allegations, and maintained that the inventions were patentably
different, and moreover, that LLNL/UC did not even begin targeted UWB
radar work until 1992, 2 years after the March, 1990 LANL meeting. On
June 19, 1997, at the request of Dr. C. Bruce Tarter, Director of LLNL,
TDC submitted extensive documentation to LLNL that summarized the dispute
and suggested a potential settlement.
This settlement was not accepted by LLNL/UC, no
counter offer was made, and TDC\'s invitation to resolve the dispute using
alternate dispute resolution was also turned down.

2. The Patent Reexamination

In May, 1998, the U.S. Patent and Trademark Office (PTO) rejected
the 4 core claims on LLNL\'s MIR motion sensor patent, as well as 8 of the
remaining 16 claims on the basis of the Fullerton patents.

Here are the notes. The url is 404.

The Development & Commercialization of Micropower Impulse Radar at

Lawrence Livermore National Laboratory

A Report by the

Democratic Staff Committee on Science

U.S. House of Representatives

April 9, 1999

The original is available at:

http://www.house.gov/science_democrats/archive/mirrpt99.htm

LLNL/UC and the MIR inventor, Thomas McEwan, were aware of
Fullerton\'s inventions, but did not cite the inventions or other
publications describing them to the Patent Office as is required by law;

TDC inventor Larry Fullerton invented and patented the same
technology 7 years prior to LLNL/UC;

In the late 1980s, claims were made regarding the ability of UWB radar
to detect and identify stealth aircraft. The utilization of such a wide
portion of the frequency spectrum to transmit information would
reportedly enable the detection of stealth aircraft skins which absorb
conventional radar, and the use of impulses to transmit information would
reportedly allow the delineation of the sharp edges of stealth aircraft
to a much higher degree than continuous-wave radar could. These claims
were described in several press articles in Aviation Week & Space
Technology in 1989 and 1990.

Two presentations on UWB radar were given by collaborators of
Larry Fullerton at the March, 1990 LANL meeting. One presentation listed
Mr. Fullerton as a co-author, and the other referred to his proprietary
UWB radar equipment in the text of the paper.

Also in attendance at that meeting were Thomas E. McEwan of LLNL and 9
other LLNL employees. All known press reports
of the meeting highlighted Fullerton\'s work,
mentioning that he had secured several patents on the technology and
describing his inventions.

Records show that Thomas McEwan and other LLNL employees began
targeted UWB radar R&D immediately upon their return from the March, 1990
LANL meeting. Internal LLNL memos obtained by Democratic Staff indicate
that Thomas McEwan had read at least one of the press reports surrounding
the meeting which contained a description of Fullerton\'s inventions. In
September, 1990, Thomas McEwan and David Christie (at that time also of
LLNL) submitted an internal funding proposal entitled Ultra-wideband
Time Domain Imaging Radar. According to Mr. Christie\'s recollections,
the proposal was funded. By February, 1991, LLNL was making UWB radar
proposals to prospective industrial and governmental licensees,
advertising a cheap, sensitive, low power, ultra-wideband radar that
could fit on a single microchip.

The first LLNL Invention Disclosure forms for MIR were filed in
August, 1992. In 1993, LLNL/UC filed their first UWB radar patents on
what would become known as MIR technology, listing Thomas E. McEwan as
the inventor and UC as the patent owner. LLNL/UC did not cite any of
Larry Fullerton\'s patents, publications, or articles describing TDC\'s
inventions as prior art on their early MIR patent applications. MIR was
characterized as a cheap, sensitive, low power, ultra-wideband radar
technology that could fit on a single microchip and that was ready to be
licensed for use in a wide variety of applications including heartbeat
monitors, power tools, and automotive collision sensors. Like Fullerton\'s
radar inventions, MIR detects and identifies targets by relying on the
reception and interpretation of very short, randomly-spaced impulses of
electromagnetic energy that are reflected off the targets.

In September, 1995, after learning about the LLNL/UC patents, TDC
contacted both LLNL and DOE and conveyed their beliefs that the
inventions described in the MIR patents were extremely similar to
Fullerton\'s inventions. This led to extensive correspondence between TDC
(and its affiliated entity Pulson), LLNL/UC and DOE. LLNL/UC denied all
of TDC\'s allegations, and maintained that the inventions were patentably
different, and moreover, that LLNL/UC did not even begin targeted UWB
radar work until 1992, 2 years after the March, 1990 LANL meeting. On
June 19, 1997, at the request of Dr. C. Bruce Tarter, Director of LLNL,
TDC submitted extensive documentation to LLNL that summarized the dispute
and suggested a potential settlement.

This settlement was not accepted by LLNL/UC, no
counter offer was made, and TDC\'s invitation to resolve the dispute using
alternate dispute resolution was also turned down.

2. The Patent Reexamination

In May, 1998, the U.S. Patent and Trademark Office (PTO) rejected
the 4 core claims on LLNL\'s MIR motion sensor patent, as well as 8 of the
remaining 16 claims on the basis of the Fullerton patents.




--
MRM

 

[John Larkin]

On Wed, 15 Feb 2023 07:13:39 -0800 (PST), Klaus Kragelund
<klaus.kragelund@gmail.com> wrote:

On 14-02-2023 17:19, John Larkin wrote:
On Tue, 14 Feb 2023 07:16:41 -0800 (PST), Klaus Kragelund
klaus.kragelund@gmail.com> wrote:

Hi

I am working on a very fast rise time generator. I need to supply 30V or more into a capacitive load of 50pF in less than a nanosecond, sourcing about 2A during the transient. It\'s for a prototype test, so no real concerns of cost and availability.


I\'ve been thinking about fast high-voltage pulsers lately too.

So far I have been using a fast GaN gatedriver and a EPC2018. That performs well, about 600ps to 30V.

Is that a sim or a real circuit? All my EPC2018 sims have been
optimistic on speed.
It a real circuit. Even with non-optimal layout, we got 600ps to 30V.
What are you driving the gate with?
LMG1020. Pretty much the fastest gatedriver I can find

I drive EPC ganfets from Tiny Logic gates. They only need 5 volts.
They accumulate some sort of gate damage at higher voltages.

I\'ve simulated as fast as 100 ps edges from an EPC2018, but never seen
anything that fast in real life. Your 600 ps sounds pretty good.

https://www.dropbox.com/sh/8pu9afypd9lxkd2/AACLpsruC3_X8p5vuFuV60vTa?dl=0

I would like to push it even further.

One idea I was thinking about was to use LVDS driver, or any other driver with below 100ps rise time. That can create a fast edge, but not to 30V. I could then trigger 10 of these circuits, and capacitive couple them in a rolling sequence (diodes in series to prevent back flow), each carefully triggered so they would build on each other.

Other concept is to use a HF transistor, but I don\'t think it would be faster than the GaN, all though I have no idea, never done real HF stuff before.

I was also thinking bringing a transistor into avalance, like a circuit done by Jim Thomson:

https://electronicprojectsforfun.files.wordpress.com/2017/10/jimwilliamsan47applicationnode.pdf

(Page 94)

He states below 300ps risetime. It\'s been 30 years, surely something exists that is faster?

Zetex sells cool avalanche transistors, but I\'ve not seem then switch
really fast. Last I heard, they were made on an ancient diffusion fab
line in Russia.

ADCMP580 has 37ps risetime:

https://www.analog.com/media/en/technical-documentation/data-sheets/ADCMP580_581_582.pdf

A regular inverter switches around 1ns:

https://www.ti.com/lit/ds/symlink/sn74auc2g06.pdf?HQS=dis-dk-null-digikeymode-dsf-pf-null-wwe&ts=1676384474996&ref_url=https%253A%252F%252Fwww.ti.com%252Fgeneral%252Fdocs%252Fsuppproductinfo.tsp%253FdistId%253D10%2526gotoUrl%253Dhttps%253A%252F%252Fwww.ti.com%252Flit%252Fgpn%252Fsn74auc2g06

Seems about 500ps is possible with ultra cheap inverter.

Here\'s a 6 USD 20ps switch:

http://ww1.microchip.com/downloads/en/DeviceDoc/sy58051u.pdf


There are lots of gates and comparators with edges below 40 ps,
including the GigaComm parts, but they have small swings and need to
be amplified. MC10EL/EP89 are pretty fast and swing more, but would
still need an amp.

One Tiny cmos flop has a rising edge below 100 ps.

https://www.dropbox.com/s/7gajbmt923oesli/NC7SV74_2.JPG?raw=

That shot is actually 1v/div.
Very fast rising edge

How do you hook up the 11802 to do that measurement. A 1k resistor at the board into the coax cable and 50ohms termination?

Here are some pix. We assign a prototype/experiment number to
experimants and document them on a server.

https://www.dropbox.com/sh/gyn0nz486fmqp1s/AAB5kwDWJ1VR8EXMjGRHs4iEa?dl=0

Reviewing that, it looks like the rise time from clock to Q may be
faster than fall time from clear.

I\'ve made a 1400 volt pulse with 3 ns risetime with a SiC fet.
Can you share how, which FET and gatedriver?

That was my ill-fated Pockels Cell driver, 1400 volt pulses at 4 MHz.
Ill-fated because the customer didn\'t buy any and stole my personal
P400 Delay generator too.

https://www.dropbox.com/sh/vwzonrz9ymev5yb/AAA22Tj8RhBHzPq21uB978qma?dl=0

I used a CREE C2M0280120D driven by some EPC Gan fets in a truly
bizarre circuit. I suspect that S stole the idea and did it
themselves. That\'s standard operating procedure in Silicon Valley.

Anyone got other ideas?

The classic brutal driver is a step-recovery diode.

Or a contact closure.

Distributed amplifiers are interesting but difficult. They are usually
linear amps but there is no fundamental reason why they couldn\'t work
switchmode.

A radial, as opposed to inline, DA might be fun.

NLTLs, shock lines, are another way to get fast high voltage pulses.
McEwan did a lot of that.

Yes, seems Thomas McEwan had a lot of patents:

https://radaris.com/f/Thomas/Mcewan/Inventor

He spun a business out of LLNL but it didn\'t turn out well.

 

[John Larkin]

On Sat, 11 Mar 2023 06:27:38 -0000 (UTC), Mike Monett VE3BTI
<spamme@not.com> wrote:

Klaus Kragelund <klaus.kragelund@gmail.com> wrote:

NLTLs, shock lines, are another way to get fast high voltage pulses.
McEwan did a lot of that.

Yes, seems Thomas McEwan had a lot of patents:

https://radaris.com/f/Thomas/Mcewan/Inventor

Unfortunately, most were stolen from Larry Fullerton, who invited McEwan to
his workshops where he discussed his achievements, including the fact that
he had patented them. This didn\'t matter to McEwan, who went ahead and re-
patented them anyway.

I posted the following note on one of McEwan\'s patents:

Unfortunately, most of Tom McEwan\'s ideas were stolen from Larry
Fullerton. Here\'s some notes I took of a government inquiry into McEwan\'s
patents and the resulting cancellation of all his claims due to prior art
by Fullerton.

McEwan was also very active in stealing Fullerton\'s work in other areas.

Here are the notes, in no particular order.

The Development & Commercialization of Micropower Impulse Radar at
Lawrence Livermore National Laboratory

A Report by the Democratic Staff Committee on Science

U.S. House of Representatives

April 9, 1999

The original is available at:
http://www.house.gov/science_democrats/archive/mirrpt99.htm

LLNL/UC and the MIR inventor, Thomas McEwan, were aware of
Fullerton\'s inventions, but did not cite the inventions or other
publications describing them to the Patent Office as is required by law;

TDC inventor Larry Fullerton invented and patented the same
technology 7 years prior to LLNL/UC;

In the late 1980s, claims were made regarding the ability of UWB radar
to detect and identify stealth aircraft. The utilization of such a wide
portion of the frequency spectrum to transmit information would
reportedly enable the detection of stealth aircraft skins which absorb
conventional radar, and the use of impulses to transmit information would
reportedly allow the delineation of the sharp edges of stealth aircraft
to a much higher degree than continuous-wave radar could. These claims
were described in several press articles in Aviation Week & Space
Technology in 1989 and 1990.

Two presentations on UWB radar were given by collaborators of
Larry Fullerton at the March, 1990 LANL meeting. One presentation listed
Mr. Fullerton as a co-author, and the other referred to his proprietary
UWB radar equipment in the text of the paper.

Also in attendance at that meeting were Thomas E. McEwan of LLNL and 9
other LLNL employees. All known press reports
of the meeting highlighted Fullerton\'s work,
mentioning that he had secured several patents on the technology and
describing his inventions.

Records show that Thomas McEwan and other LLNL employees began
targeted UWB radar R&D immediately upon their return from the March, 1990
LANL meeting. Internal LLNL memos obtained by Democratic Staff indicate
that Thomas McEwan had read at least one of the press reports surrounding
the meeting which contained a description of Fullerton\'s inventions. In
September, 1990, Thomas McEwan and David Christie (at that time also of
LLNL) submitted an internal funding proposal entitled <U>Ultra-wideband
Time Domain Imaging Radar</U>. According to Mr. Christie\'s recollections,
the proposal was funded. By February, 1991, LLNL was making UWB radar
proposals to prospective industrial and governmental licensees,
advertising a cheap, sensitive, low power, ultra-wideband radar that
could fit on a single microchip.

The first LLNL Invention Disclosure forms for MIR were filed in
August, 1992. In 1993, LLNL/UC filed their first UWB radar patents on
what would become known as MIR technology, listing Thomas E. McEwan as
the inventor and UC as the patent owner. LLNL/UC did not cite any of
Larry Fullerton\'s patents, publications, or articles describing TDC\'s
inventions as prior art on their early MIR patent applications. MIR was
characterized as a cheap, sensitive, low power, ultra-wideband radar
technology that could fit on a single microchip and that was ready to be
licensed for use in a wide variety of applications including heartbeat
monitors, power tools, and automotive collision sensors. Like Fullerton\'s
radar inventions, MIR detects and identifies targets by relying on the
reception and interpretation of very short, randomly-spaced impulses of
electromagnetic energy that are reflected off the targets.

In September, 1995, after learning about the LLNL/UC patents, TDC
contacted both LLNL and DOE and conveyed their beliefs that the
inventions described in the MIR patents were extremely similar to
Fullerton\'s inventions. This led to extensive correspondence between TDC
(and its affiliated entity Pulson), LLNL/UC and DOE. LLNL/UC denied all
of TDC\'s allegations, and maintained that the inventions were patentably
different, and moreover, that LLNL/UC did not even begin targeted UWB
radar work until 1992, 2 years after the March, 1990 LANL meeting. On
June 19, 1997, at the request of Dr. C. Bruce Tarter, Director of LLNL,
TDC submitted extensive documentation to LLNL that summarized the dispute
and suggested a potential settlement.
This settlement was not accepted by LLNL/UC, no
counter offer was made, and TDC\'s invitation to resolve the dispute using
alternate dispute resolution was also turned down.

2. The Patent Reexamination

In May, 1998, the U.S. Patent and Trademark Office (PTO) rejected
the 4 core claims on LLNL\'s MIR motion sensor patent, as well as 8 of the
remaining 16 claims on the basis of the Fullerton patents.

Here are the notes. The url is 404.

The Development & Commercialization of Micropower Impulse Radar at

Lawrence Livermore National Laboratory

A Report by the

Democratic Staff Committee on Science

U.S. House of Representatives

April 9, 1999

The original is available at:

http://www.house.gov/science_democrats/archive/mirrpt99.htm

LLNL/UC and the MIR inventor, Thomas McEwan, were aware of
Fullerton\'s inventions, but did not cite the inventions or other
publications describing them to the Patent Office as is required by law;

TDC inventor Larry Fullerton invented and patented the same
technology 7 years prior to LLNL/UC;

In the late 1980s, claims were made regarding the ability of UWB radar
to detect and identify stealth aircraft. The utilization of such a wide
portion of the frequency spectrum to transmit information would
reportedly enable the detection of stealth aircraft skins which absorb
conventional radar, and the use of impulses to transmit information would
reportedly allow the delineation of the sharp edges of stealth aircraft
to a much higher degree than continuous-wave radar could. These claims
were described in several press articles in Aviation Week & Space
Technology in 1989 and 1990.

Two presentations on UWB radar were given by collaborators of
Larry Fullerton at the March, 1990 LANL meeting. One presentation listed
Mr. Fullerton as a co-author, and the other referred to his proprietary
UWB radar equipment in the text of the paper.

Also in attendance at that meeting were Thomas E. McEwan of LLNL and 9
other LLNL employees. All known press reports
of the meeting highlighted Fullerton\'s work,
mentioning that he had secured several patents on the technology and
describing his inventions.

Records show that Thomas McEwan and other LLNL employees began
targeted UWB radar R&D immediately upon their return from the March, 1990
LANL meeting. Internal LLNL memos obtained by Democratic Staff indicate
that Thomas McEwan had read at least one of the press reports surrounding
the meeting which contained a description of Fullerton\'s inventions. In
September, 1990, Thomas McEwan and David Christie (at that time also of
LLNL) submitted an internal funding proposal entitled Ultra-wideband
Time Domain Imaging Radar. According to Mr. Christie\'s recollections,
the proposal was funded. By February, 1991, LLNL was making UWB radar
proposals to prospective industrial and governmental licensees,
advertising a cheap, sensitive, low power, ultra-wideband radar that
could fit on a single microchip.

The first LLNL Invention Disclosure forms for MIR were filed in
August, 1992. In 1993, LLNL/UC filed their first UWB radar patents on
what would become known as MIR technology, listing Thomas E. McEwan as
the inventor and UC as the patent owner. LLNL/UC did not cite any of
Larry Fullerton\'s patents, publications, or articles describing TDC\'s
inventions as prior art on their early MIR patent applications. MIR was
characterized as a cheap, sensitive, low power, ultra-wideband radar
technology that could fit on a single microchip and that was ready to be
licensed for use in a wide variety of applications including heartbeat
monitors, power tools, and automotive collision sensors. Like Fullerton\'s
radar inventions, MIR detects and identifies targets by relying on the
reception and interpretation of very short, randomly-spaced impulses of
electromagnetic energy that are reflected off the targets.

In September, 1995, after learning about the LLNL/UC patents, TDC
contacted both LLNL and DOE and conveyed their beliefs that the
inventions described in the MIR patents were extremely similar to
Fullerton\'s inventions. This led to extensive correspondence between TDC
(and its affiliated entity Pulson), LLNL/UC and DOE. LLNL/UC denied all
of TDC\'s allegations, and maintained that the inventions were patentably
different, and moreover, that LLNL/UC did not even begin targeted UWB
radar work until 1992, 2 years after the March, 1990 LANL meeting. On
June 19, 1997, at the request of Dr. C. Bruce Tarter, Director of LLNL,
TDC submitted extensive documentation to LLNL that summarized the dispute
and suggested a potential settlement.

This settlement was not accepted by LLNL/UC, no
counter offer was made, and TDC\'s invitation to resolve the dispute using
alternate dispute resolution was also turned down.

2. The Patent Reexamination

In May, 1998, the U.S. Patent and Trademark Office (PTO) rejected
the 4 core claims on LLNL\'s MIR motion sensor patent, as well as 8 of the
remaining 16 claims on the basis of the Fullerton patents.

Don\'t you have some patents?

 

[John Larkin]

On Sat, 11 Mar 2023 08:15:26 -0800, John Larkin
<jlarkin@highlandSNIPMEtechnology.com> wrote:

On Wed, 15 Feb 2023 07:13:39 -0800 (PST), Klaus Kragelund
klaus.kragelund@gmail.com> wrote:

On 14-02-2023 17:19, John Larkin wrote:
On Tue, 14 Feb 2023 07:16:41 -0800 (PST), Klaus Kragelund
klaus.kragelund@gmail.com> wrote:

Hi

I am working on a very fast rise time generator. I need to supply 30V or more into a capacitive load of 50pF in less than a nanosecond, sourcing about 2A during the transient. It\'s for a prototype test, so no real concerns of cost and availability.


I\'ve been thinking about fast high-voltage pulsers lately too.

So far I have been using a fast GaN gatedriver and a EPC2018. That performs well, about 600ps to 30V.

Is that a sim or a real circuit? All my EPC2018 sims have been
optimistic on speed.
It a real circuit. Even with non-optimal layout, we got 600ps to 30V.
What are you driving the gate with?
LMG1020. Pretty much the fastest gatedriver I can find

I drive EPC ganfets from Tiny Logic gates. They only need 5 volts.
They accumulate some sort of gate damage at higher voltages.

I\'ve simulated as fast as 100 ps edges from an EPC2018, but never seen
anything that fast in real life. Your 600 ps sounds pretty good.

https://www.dropbox.com/sh/8pu9afypd9lxkd2/AACLpsruC3_X8p5vuFuV60vTa?dl=0




I would like to push it even further.

One idea I was thinking about was to use LVDS driver, or any other driver with below 100ps rise time. That can create a fast edge, but not to 30V. I could then trigger 10 of these circuits, and capacitive couple them in a rolling sequence (diodes in series to prevent back flow), each carefully triggered so they would build on each other.

Other concept is to use a HF transistor, but I don\'t think it would be faster than the GaN, all though I have no idea, never done real HF stuff before.

I was also thinking bringing a transistor into avalance, like a circuit done by Jim Thomson:

https://electronicprojectsforfun.files.wordpress.com/2017/10/jimwilliamsan47applicationnode.pdf

(Page 94)

He states below 300ps risetime. It\'s been 30 years, surely something exists that is faster?

Zetex sells cool avalanche transistors, but I\'ve not seem then switch
really fast. Last I heard, they were made on an ancient diffusion fab
line in Russia.


ADCMP580 has 37ps risetime:

https://www.analog.com/media/en/technical-documentation/data-sheets/ADCMP580_581_582.pdf

A regular inverter switches around 1ns:

https://www.ti.com/lit/ds/symlink/sn74auc2g06.pdf?HQS=dis-dk-null-digikeymode-dsf-pf-null-wwe&ts=1676384474996&ref_url=https%253A%252F%252Fwww.ti.com%252Fgeneral%252Fdocs%252Fsuppproductinfo.tsp%253FdistId%253D10%2526gotoUrl%253Dhttps%253A%252F%252Fwww.ti.com%252Flit%252Fgpn%252Fsn74auc2g06

Seems about 500ps is possible with ultra cheap inverter.

Here\'s a 6 USD 20ps switch:

http://ww1.microchip.com/downloads/en/DeviceDoc/sy58051u.pdf


There are lots of gates and comparators with edges below 40 ps,
including the GigaComm parts, but they have small swings and need to
be amplified. MC10EL/EP89 are pretty fast and swing more, but would
still need an amp.

One Tiny cmos flop has a rising edge below 100 ps.

https://www.dropbox.com/s/7gajbmt923oesli/NC7SV74_2.JPG?raw=

That shot is actually 1v/div.
Very fast rising edge

How do you hook up the 11802 to do that measurement. A 1k resistor at the board into the coax cable and 50ohms termination?


Here are some pix. We assign a prototype/experiment number to
experimants and document them on a server.

https://www.dropbox.com/sh/gyn0nz486fmqp1s/AAB5kwDWJ1VR8EXMjGRHs4iEa?dl=0

Reviewing that, it looks like the rise time from clock to Q may be
faster than fall time from clear.

So clocking Q low might be even faster.

One might balance the Q and Qbar loads too.

 

[whit3rd]

On Saturday, March 11, 2023 at 8:15:42 AM UTC-8, John Larkin wrote:

On Wed, 15 Feb 2023 07:13:39 -0800 (PST), Klaus Kragelund
klaus.k...@gmail.com> wrote:

On 14-02-2023 17:19, John Larkin wrote:
On Tue, 14 Feb 2023 07:16:41 -0800 (PST), Klaus Kragelund
klaus.k...@gmail.com> wrote:

He states below 300ps risetime. It\'s been 30 years, surely something exists that is faster?

Zetex sells cool avalanche transistors, but I\'ve not seem then switch
really fast. Last I heard, they were made on an ancient diffusion fab
line in Russia.

The thing about avalanche is that it is a local and positive-feedback process;
the \'switch\' time of a transistor isn\'t. The avalanche possibilities include laser
Q-switching, and some very fast pulses are indeed achieved with
simple photoconductors and an avalanche out of a pumped laser medium...

At the high-power end, photoSCRs are the way to generate megawatt-scale
power inverters, and those also use an avalanche of photoelectrons for clean
triggering, and a (gas-tube?) breakdown avalanche to make the light.

Only the smallest transistors are similarly speedy switches to an avalanche process.

 

[John Larkin]

On Sun, 19 Mar 2023 20:35:00 -0000, \"Commander Kinsey\"
<CK1@nospam.com> wrote:

On Wed, 08 Mar 2023 16:52:42 -0000, Cindy Hamilton <hamilton@invalid.com> wrote:

On 2023-03-08, rbowman <bowman@montana.com> wrote:
On Wed, 08 Mar 2023 10:19:48 GMT, Cindy Hamilton wrote:


Herb sounds affected to me. OTOH, I pronounce the H in herbivore,
herbicide, etc. \"A foolish consistency is the hobgoblin of little
minds.\"

herbal? I often hear the h dropped there too, possibly as an extension of
erb. Like you say anyone who wants consistency better learn German.

Yes, \"erbal\". Isn\'t that how it\'s pronounced in the commercials?
https://www.youtube.com/watch?v=aGaXOQH8vok

That sounds so utterly stupid. Merkins, do you really want to sound thick?

We can\'t imitate the language of every tiny obscure impoverished
island on the planet.

 

[Commander Kinsey]

On Thu, 09 Mar 2023 16:56:02 -0000, John Larkin <jlarkin@highlandsnipmetechnology.com> wrote:

On 9 Mar 2023 05:08:10 GMT, rbowman <bowman@montana.com> wrote:

On Wed, 08 Mar 2023 09:33:49 -0800, John Larkin wrote:

What is megaML? It doesn\'t google well.

Not an actual term outside of my imagination. TensorFlow is a Google
endeavor and they provide a web interface called Colab where you can
execute TF in Python.

https://colab.research.google.com/notebooks/intro.ipynb

Google also builds their own specialized hardware:

https://en.wikipedia.org/wiki/Tensor_Processing_Unit

Others use GPUs like NVidias. After you construct a model you train it
with a large set of labeled data over many iterations to optimize the
weights and it\'s very processor intensive. As you might guess from tensor
you\'re deep into linear algebra.

After training the model and getting the error down to an acceptable level
you can export a .tflite file that\'s relatively small. The metaphor is
when you\'re learning something new you\'re very busy but when you get
around to applying the knowledge it\'s nowhere near as intensive so it can
be handled by an embedded processor.

The Arduino nano 33 BLE Sense is popular. It uses the Nordic nRF52840 32-
bit ARM processor and has an inertial unit, magnetic sensor, temperature,
humidity, and other sensors in a very small package.

At least that\'s the theory. In the \'80s I was programming MCUs lke the
8048 family for devices like handheld pH/ion concentration meters. I took
a course in neural networks which was going to be the Next Great Thing. It
was oversold since the hardware didn\'t exist to make it feasible.

40 years later neural networks were reborn as machine learning except now
you have racks and racks of TPUs churning away to make it work. Whether it
can be used to do useful work on MCUs is another question.

For a sense of scale Facebook (Meta) managed to leak their large language
model. It\'s only 288 GB. Nadella recently said Siri, Alexa, Cortana and so
forth are as dumb as rocks. We\'ll see what the next generation of AI buzz
brings.

OK, machine learning. In a hard embedded electronic instrument, we
want predictability and zero errors. ML and NNs and that sort of thing
seems to me to continually generate dangerous bugs.

Programmers do that all by themselves.

 

[Peeler]

On Sun, 19 Mar 2023 13:44:58 -0700, John Larkin, another obviously brain
dead, troll-feeding senile asshole, blathered:

We can\'t imitate the language of every tiny obscure impoverished
island on the planet.

But you can take every single idiotic bait set out for you by the trolling
Scottish wanker, sociopath and attention whore! <BG>

 

[John Larkin]

On Wed, 22 Mar 2023 16:49:10 +0000, Max Demian
<max_demian@bigfoot.com> wrote:

On 21/03/2023 21:53, John Larkin wrote:
On Tue, 21 Mar 2023 19:04:15 -0000, \"Commander Kinsey\"
CK1@nospam.com> wrote:
On Thu, 16 Mar 2023 23:14:56 -0000, SteveW <steve@walker-family.me.uk> wrote:

More to the point, railway tracks are normally fenced off and it is
illegal for the public to access them except at level crossings

A law to prevent you hurting yourself is silly.

Of course we all went on the tracks as kids.

I got busted for putting coins on a track to be smashed flat. They
tooks us into the station office. We got a lecture, catch-and-release.

They should have fined you forty shillings. Taken from your pocket money.

What\'s 40 shillings worth in real money?

We were also changing the signals by shorting the links in the track
with coins. That\'s probably what really annoyed them.