HP 8012a...

[server]

On Tue, 26 Apr 2022 22:59:24 -0400, bitrex <user@example.net> wrote:

On 4/26/2022 9:49 PM, Don Y wrote:
On 4/26/2022 8:28 AM, bitrex wrote:
On 4/26/2022 11:18 AM, bitrex wrote:
The HP 8082 uses ECL you can\'t get no mo, and looks like a PITA to
repair:

https://youtu.be/09zhUbJl37w

Far as I can tell its slower smaller cousin the 8012a is all
discrete-on-card though, yeah? It looks that way from the service
manual but I haven\'t checked every thing.

https://www.keysight.com/us/en/assets/9018-03164/user-manuals/9018-03164.pdf


A 5ns rise time is still useful for a number of jobs, and I can get a
dodgy unit for next to nothing locally. I figure if I use it once a
year it might be worth trying to fix up if it doesn\'t require exotic
parts.

You can often pick up used kit from local tech firms, universities
(excellent
source of kit as they often dump everything they purchased for a GRANT
project
when the grant is over), etc.  Some firms just contract disposal of
their kit
to a firm; others sell at auction (which may or may not be \"public\").
Knowing
folks at these places is a great way to get the inside track...


There is a re-seller like that near me with several dozen \"tested\"
3478As bench DMMs available for $150 per (better test the AC section
though as it uses an unobtaninum part) and another place will calibrate
it for $50.

https://youtu.be/9v6OksEFqpA

I think I\'ll pick up a couple I used them in college & they\'re still
nice meters.


I don\'t think I own equipment fast enough to calibrate a refurbed 8082
anyway, even if it\'s not the ECL that\'s the problem

Yeah, I rescued an HP 3458A many years ago.  After the initial \"pride of
rescueship\" (v \"ownership\") wore off, I realized it was just an oversized
*4* digit DVM given that I never needed more accuracy/precision than that!
And, surely wasn\'t going to PAY to have it calibrated to 8+ digits!

(here, bench space is far more precious than quality of kit!)

Going off BW * tr = 0.34 as the fastest edge a DSO can even assign a
sample to, much less measure the rise time of accurately, I figure a 100
MHz scope is too sluggish to calibrate a 5ns pulser on its fastest
setting. /shrug

The 0.34 is a relic of the past, when scopes were gaussian. Most
scopes these days are peaked to meet their bandwidth claim, so even if
they do make their -3 dB spec they ring like bells.

If a scope is an honest 100 MHz, with a 3.5 ns Trr, it should be OK to
measure a 5 ns edge; just math it.

Or get an old Tek 20 GHz sampler.



--

Anybody can count to one.

- Robert Widlar

 

[bitrex]

On 4/27/2022 10:31 AM, jlarkin@highlandsniptechnology.com wrote:

On Tue, 26 Apr 2022 22:59:24 -0400, bitrex <user@example.net> wrote:

On 4/26/2022 9:49 PM, Don Y wrote:
On 4/26/2022 8:28 AM, bitrex wrote:
On 4/26/2022 11:18 AM, bitrex wrote:
The HP 8082 uses ECL you can\'t get no mo, and looks like a PITA to
repair:

https://youtu.be/09zhUbJl37w

Far as I can tell its slower smaller cousin the 8012a is all
discrete-on-card though, yeah? It looks that way from the service
manual but I haven\'t checked every thing.

https://www.keysight.com/us/en/assets/9018-03164/user-manuals/9018-03164.pdf


A 5ns rise time is still useful for a number of jobs, and I can get a
dodgy unit for next to nothing locally. I figure if I use it once a
year it might be worth trying to fix up if it doesn\'t require exotic
parts.

You can often pick up used kit from local tech firms, universities
(excellent
source of kit as they often dump everything they purchased for a GRANT
project
when the grant is over), etc.  Some firms just contract disposal of
their kit
to a firm; others sell at auction (which may or may not be \"public\").
Knowing
folks at these places is a great way to get the inside track...


There is a re-seller like that near me with several dozen \"tested\"
3478As bench DMMs available for $150 per (better test the AC section
though as it uses an unobtaninum part) and another place will calibrate
it for $50.

https://youtu.be/9v6OksEFqpA

I think I\'ll pick up a couple I used them in college & they\'re still
nice meters.


I don\'t think I own equipment fast enough to calibrate a refurbed 8082
anyway, even if it\'s not the ECL that\'s the problem

Yeah, I rescued an HP 3458A many years ago.  After the initial \"pride of
rescueship\" (v \"ownership\") wore off, I realized it was just an oversized
*4* digit DVM given that I never needed more accuracy/precision than that!
And, surely wasn\'t going to PAY to have it calibrated to 8+ digits!

(here, bench space is far more precious than quality of kit!)

Going off BW * tr = 0.34 as the fastest edge a DSO can even assign a
sample to, much less measure the rise time of accurately, I figure a 100
MHz scope is too sluggish to calibrate a 5ns pulser on its fastest
setting. /shrug

The 0.34 is a relic of the past, when scopes were gaussian. Most
scopes these days are peaked to meet their bandwidth claim, so even if
they do make their -3 dB spec they ring like bells.

If a scope is an honest 100 MHz, with a 3.5 ns Trr, it should be OK to
measure a 5 ns edge; just math it.

Or get an old Tek 20 GHz sampler.

Keysight claims DSOs with BW < 1 GHz tend to have Gaussian responses and
BW > 1 Ghz are maximally flat, but this may only apply for Keysight scopes:

<https://f.hubspotusercontent40.net/hubfs/281197/Keysight_Evaluating_Oscilloscopes_AppNote_CControls.pdf>

I don\'t know if anyone\'s evaluated what the response of the inexpensive
Rigol scopes that are my work horses are, it\'s probably not particularly
anything but the captures here don\'t make it seem like it\'s very peaked,
at least:

<https://www.eevblog.com/forum/testgear/rigol-ds1054z-bandwidth/50/>

For 10% accuracy the difference between a maximally flat and Gaussian
response don\'t matter much according to the first reference, for 3%
quite a bit more.

 

[bitrex]

On 4/27/2022 1:17 PM, bitrex wrote:

On 4/27/2022 10:31 AM, jlarkin@highlandsniptechnology.com wrote:
On Tue, 26 Apr 2022 22:59:24 -0400, bitrex <user@example.net> wrote:

On 4/26/2022 9:49 PM, Don Y wrote:
On 4/26/2022 8:28 AM, bitrex wrote:
On 4/26/2022 11:18 AM, bitrex wrote:
The HP 8082 uses ECL you can\'t get no mo, and looks like a PITA to
repair:

https://youtu.be/09zhUbJl37w

Far as I can tell its slower smaller cousin the 8012a is all
discrete-on-card though, yeah? It looks that way from the service
manual but I haven\'t checked every thing.

https://www.keysight.com/us/en/assets/9018-03164/user-manuals/9018-03164.pdf



A 5ns rise time is still useful for a number of jobs, and I can get a
dodgy unit for next to nothing locally. I figure if I use it once a
year it might be worth trying to fix up if it doesn\'t require exotic
parts.

You can often pick up used kit from local tech firms, universities
(excellent
source of kit as they often dump everything they purchased for a GRANT
project
when the grant is over), etc.  Some firms just contract disposal of
their kit
to a firm; others sell at auction (which may or may not be \"public\").
Knowing
folks at these places is a great way to get the inside track...


There is a re-seller like that near me with several dozen \"tested\"
3478As bench DMMs available for $150 per (better test the AC section
though as it uses an unobtaninum part) and another place will calibrate
it for $50.

https://youtu.be/9v6OksEFqpA

I think I\'ll pick up a couple I used them in college & they\'re still
nice meters.


I don\'t think I own equipment fast enough to calibrate a refurbed 8082
anyway, even if it\'s not the ECL that\'s the problem

Yeah, I rescued an HP 3458A many years ago.  After the initial
\"pride of
rescueship\" (v \"ownership\") wore off, I realized it was just an
oversized
*4* digit DVM given that I never needed more accuracy/precision than
that!
And, surely wasn\'t going to PAY to have it calibrated to 8+ digits!

(here, bench space is far more precious than quality of kit!)

Going off BW * tr = 0.34 as the fastest edge a DSO can even assign a
sample to, much less measure the rise time of accurately, I figure a 100
MHz scope is too sluggish to calibrate a 5ns pulser on its fastest
setting. /shrug

The 0.34 is a relic of the past, when scopes were gaussian. Most
scopes these days are peaked to meet their bandwidth claim, so even if
they do make their -3 dB spec they ring like bells.

If a scope is an honest 100 MHz, with a 3.5 ns Trr, it should be OK to
measure a 5 ns edge; just math it.

Or get an old Tek 20 GHz sampler.




Keysight claims DSOs with BW < 1 GHz tend to have Gaussian responses and
BW > 1 Ghz are maximally flat, but this may only apply for Keysight scopes:

https://f.hubspotusercontent40.net/hubfs/281197/Keysight_Evaluating_Oscilloscopes_AppNote_CControls.pdf


I don\'t know if anyone\'s evaluated what the response of the inexpensive
Rigol scopes that are my work horses are, it\'s probably not particularly
anything but the captures here don\'t make it seem like it\'s very peaked,
at least:

https://www.eevblog.com/forum/testgear/rigol-ds1054z-bandwidth/50/

For 10% accuracy the difference between a maximally flat and Gaussian
response don\'t matter much according to the first reference, for 3%
quite a bit more.

A pulser is a nice thing to have if just for checking loop stability
sometimes, it don\'t have to be perfect

 

[Phil Hobbs]

jlarkin@highlandsniptechnology.com wrote:

On Wed, 27 Apr 2022 04:15:50 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Don Y wrote:
On 4/26/2022 7:59 PM, bitrex wrote:
You can often pick up used kit from local tech firms, universities
(excellent
source of kit as they often dump everything they purchased for a
GRANT project
when the grant is over), etc.  Some firms just contract disposal of
their kit
to a firm; others sell at auction (which may or may not be \"public\").
Knowing
folks at these places is a great way to get the inside track...

There is a re-seller like that near me with several dozen \"tested\"
3478As bench DMMs available for $150 per (better test the AC section
though as it uses an unobtaninum part) and another place will
calibrate it for $50.

Resellers increase the price but usually don\'t add much \"value\".  You want
to find the guy *he* buys from (usually someone at ABC Tech Corporation).

I think I\'ll pick up a couple I used them in college & they\'re still
nice meters.

I don\'t think I own equipment fast enough to calibrate a refurbed
8082 anyway, even if it\'s not the ECL that\'s the problem

Yeah, I rescued an HP 3458A many years ago.  After the initial \"pride of
rescueship\" (v \"ownership\") wore off, I realized it was just an
oversized
*4* digit DVM given that I never needed more accuracy/precision than
that!
And, surely wasn\'t going to PAY to have it calibrated to 8+ digits!

(here, bench space is far more precious than quality of kit!)

Going off BW * tr = 0.34 as the fastest edge a DSO can even assign a
sample to, much less measure the rise time of accurately, I figure a
100 MHz scope is too sluggish to calibrate a 5ns pulser on its fastest
setting. /shrug

Most of the products I\'ve designed have intentionally NOT required close
tolerances on anything.  That adds cost -- in parts or calibration labor.
So, we design products that rely on other means to make sense of their
observations (ratiometric or other \"self calibration\" tricks).

[It\'s not uncommon to have manufacturing tolerances of -60%, +300%.
Easy to
accommodate if you are smart about it.]

A pity you didn\'t just replace NIST.


Another NOLA story: we\'d send drawings out to machine shops and get
stuff that didn\'t fit. They responded that we should send them
drawings with tolerances, and circle some in red with the note HOLD.

Sounds like my approach to modelmaking.

Life\'s too short to worry about the small stuff, I guess.

The Silicon Bayou and similar plans never seemed to work out. That\'s
one reason why I left.

Yeah, things don\'t go as well when the clean room guys are sneaking into
the tool core for a smoke.

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

 

[Don Y]

On 4/27/2022 2:50 AM, Don Y wrote:

On 4/27/2022 1:15 AM, Phil Hobbs wrote:
Don Y wrote:

[It\'s not uncommon to have manufacturing tolerances of -60%, +300%. Easy to
accommodate if you are smart about it.]

A pity you didn\'t just replace NIST.

Lots of things don\'t need to be qualified in \"engineering units\".

How do you test that the *hammer* you are building, today, is as good
as the hammers you\'ve built in the past? Or, a screwdriver? Just
how \"durable\" should the finish on a tape rule be?

\"Quest\" (TV Network) just aired a show, \"Modern Marvels\", in which
such a device was shown in use. Apparently it\'s a rerun of a show from
~6 mos ago. I can\'t seem to find it online (in a free venue).

<https://www.stanleyblackanddecker.com/article/stanley-black-decker-featured-upcoming-modern-marvels-machines-episode-airing-history>

It would have been more interesting if they\'d shown how tape rules are
\"exercised\", hammer handles broken, screwdriver tips sheared off,
tool finishes abraded, etc.

For folks used to seeing things done with \"conventional\" units of measure,
it would be eye-opening!

They *do* use some \"laboratory\" tools to augment their collected data.
E.g., rockwell hardness tester to determine the hardness of the steel
used in the tool, etc. And, the fixtures are calibrated to produce
reproducible tests -- so 1135 hammer strikes today represents the
same amount of wear that it did 30 years ago, and a tape rule dropped
from 50 inches experiences the same sort of impact shock that it did
30 years ago, and...

What\'s particularly interesting is how many tools are easily broken by
careless users. A \"nominal\" male can easily tear the tip off a #0
Philips screwdriver. And, most can do that on a #1 tip as well!
But, at #2 and larger, you really need to be a bit of a gorilla...
(and capable of holding the screwdriver\'s tip *in* the screwhead
lest it \"cam out\")

And, I suspect everyone has a cabinet tip screwdriver whose blade is
no longer flat (overtorqued, bending the ends out of the plane of the
flats). Or, a handle that \"spins\" on the shaft (cuz you tried to
apply excessive torque with a pair of vice grips)? Or, a wood
chisel that seemed like it would be a great tool for cutting
aluminum gutters? Or, a screwdriver handle that has been cracked because
someone thought it would make a good *chisel*?

[who, besides the manufacturer, would know how to interpret those data?
What units? How to evaluate relative to other vendors\' products?]

How do you verify that the (pharmaceutical) tablet that you produced NOW
is as good as the one you produced 5 milliseconds earlier?

Ans: you monitor something that correlates well with the parameters
of the tablet that are important to the regulators AND your own QC.

In addition to being concerned with the amount of actives in the tablet,
regulators are concerned primarily with *consistency* (!) of actives in
a tablet, overall weight, dissolution time and disintegration time.
They\'re not going to let you sell a bottle with 300mg tablets and
100mg tablets on the argument that the 300mg tablet just has 200 extra
mg of binders and lubricants in its excipients!

[Your QC will step in before the regulator ever sees such an abomination]

Actives can only be determined by assay -- which is destructive in nature.
As are dissolution and disintegration tests.

All of these tests are far too slow to keep up with the production rate
of a high speed tablet press (1M+ tablets/hour). But, you can get an
insight into some of them -- weight, disintegration time (and other
attributes that may be of interest in certain products -- like
hardness for effervescents) -- by monitoring the *forces* encountered
by the tablet at various stages of its formation and production.

[There\'s no need to express forces in engineering units -- as long as the
calibration is consistent from press to press, lab to production, etc.
Early controllers had an analog meter that showed the current state
of the process -- with \"percent deviation\" labels (percent of WHAT?)]

As you are dealing with a constant geometry (the cavity in which the
tablet is formed), higher forces indicate more material (granulation)
must be present in that cavity. Lower forces correspond with lower
amounts of material. I.e., you have a good predictor of *weight*
(part of the scale up process is characterizing the relationship
of force to tablet weight in the lab, before releasing to manufacturing).
And, as such, you can put limits on measured forces to determine which
tablets are likely over/underweight. And, can do this at production speeds,
dispatching \"bad\" tablets on the fly while keeping good tablets (\"swatting\"
the bad ones off to the side).

You can also learn a lot about the condition of the tooling that you are
using (each product has a different set of tools) to determine wear
and other \"maintenance\" issues (e.g., a broken punch tip is bad news
as it means there\'s likely a tablet with a shard of metal in it!).

[There\'s a second school of thought that applies a constant force to
the tablet during formation and monitors the resulting thickness, in
real time -- by monitoring the motion of the compression rollers
as the tablet\'s tooling passes between them (not using engineering
units but, rather, fixed thresholds determined in the lab, at scale up).

But, these require significant mechanisms to move at reasonably high
rates... I seriously wonder if they can be as sensitive as simply monitoring
forces.

They also suffer from not being able to address all issues that
force monitoring can detect -- like the force required to eject
a formed tablet from its die (to detect barreling of dies and
a crude indication of capping).]

[Does the consumer care if a particular batch of tablets have a friability
of 1.2%? Or, variations of hardness on the order of 2KP? Or, a dissolution
time that varies by 4% in a particular sample of tablets?]

If the tablets are going to be *coated*, then friability is hugely important
(as a coating pan subjects the tablets to high abrasive forces). So, while
not directly controlled for regulators, it eventually factors into the
quality of the tablet (if regulator notices your coated tablets are \"all
coating\" and \"little tablet\"!)

In many cases, you rely on some other (external) determinant of \"acceptable
quality\" and the goal is just to ensure repeatability of process. E.g.,
those \"good\" tablets were produced with 4.5 bogounits of force exerted during
the compression phase; make sure all of them experience a similar force. Or,
a sample of those hammers struck the test anvil 1825 times, on average, before
the handle snapped; the competitor\'s hammers broke at 1615 strikes, on average.
Etc.

Statistical Process Control becomes more important than traditional control
theory (use a conventional control loop and wrap SPC around that)

Because you are dealing with repetitive processes, you\'re concerned with
how repeatable/consistent that process is. Is swaging station #7 producing
a wider distribution of tip shear strengths than station #2? Why??

This is particularly important for a tablet press which can actually be ~100+
presses rolled into one! A 75 station press has 75 sets of tools -- upper,
lower punches and a die per station. Each has manufacturing tolerances and
experiences wear at different rates (based on its \"tableting history\").

So, a manufacturer makes sure he keeps each upper, lower, die together -- along
with noting which station they occupied -- so their results are more repeatable
on the next run (imagine the number of mix-and-match combinations possible!).

But, each station is measurably different in terms of physical characteristics.
So, any information you observe about the production of a tablet on station #1
doesn\'t directly translate to the production of a tablet on station #2 -- just
a few ms later! A longer punch can make the force higher (or lower -- depends
on which punch) for the same amount of \"fill\"

So, ANY CONTROL ACTIONS that you take to keep the tablet weight (as expressed
by compression force) based on observations of the formation event for station
#1 can be \"wrong\" for the tablet being formed by station #2! Repeat...

[We\'ve actually had customers disable the control system to determine if
it might be INCREASING variability in tablets! In practical terms, you end
up severely overdamping the loops to ensure the process doesn\'t oscillate
purely as a consequence of the variations in *tooling*! Hence the value of
looking at the distributions of forces (weights) per-tooling-station and
using THAT information to decide when the process is stable.]

A tablet press is often double-sided -- meaning a tablet is formed on the
front side of the machine while another is being formed on the back side.
So, stations #1 and 38 are each in the same state of tablet formation;
ditto #2 & 39...

But, the front and back sides of the machine have different cam profiles,
wear patterns, etc. -- due to mechanical tolerances. And, different
feeders, control loops, etc. So, a tablet formed by the punches in station
#1 on the FRONT side of the press can have different characteristics than a
tablet formed in that station when it\'s on the BACK side of the press.

And, if you\'re using a two (or three!) sided press to make a bi-layer
(tri-layer) tablet, the observations of one \"side\" of the tablet\'s
production directly factor into the observations of the second (third)
side!

It is a **delightful** control system problem! Someday, they\'ll be able
to weigh individual tablets at the 2M/hr rate and much of this will be
unnecessary (though likely retained as an inner control loop).