non-pipelined fast ADC

By concidence, i'm currently writing libraries and app to drive the Prologix adapter. The idea is to make an instrument agnostic system, driven by a plain text strings in a command file, thus bypassing the need for instrument specific drivers. It's up to the user to decide what to send to the instrument as command strings.

The Prologix has a large command set, some of which take args and some don't and often needs to be in a particular mode for the command. Have the network layer written, with a Prologix layer that normalises the commands into a single list. It also remembers what mode it's in to avoid repetitious mode commands. So far, that's all working and currently working the top level command language design.

This was prompted by having a load of old HP and other kit in the lab, much of it with gpib interfaces. An ongoing requirement for data logging capability and more. Did have a look at the hp software packages, but felt like bloatware from the start, Gb on disk and expensively protectionist to boot. It's a bit of work, but keeps coding skills alive in semi retirement and can make it do just what I want, rather than hp's idea. It will be opensourced eventually...

Chris

Given a charged capacitor, with some tempco of capacitance, is charge conserved as the temperature changes? If so, the voltage across the cap is immediately modulated by temperature wobbles.

The XO situation is made worse by the dissipation of the oscillator. The air only has to change velocity to mess up the frequency.

If it's running open-circuit, sure.

If so, the voltage across the

Yup. The parts advertised as capacitors for that sort of job tend to be NP0/C0G, which are pretty blameless, but when you start getting down to parts in 1E11, things get less simple.

That too. Styrofoam covers a multitude of sins. ;)

Cheers

Phil Hobbs

Sounds super useful for what we do. Please keep us posted on your progress!

Cheers

Phil Hobbs

Just a thanks for your page on restoring tubes, "for the brave experimenter". Gave me the confidence to fix the tubes in 85xx series analyser display sections here. Didn't take notes, but the setup was a 250v tube ht transformer for isolation, a bridge rectifier, a current limiting resistor and a variac on the tx primary. Fed unsmoothed dc between the tube grid and cathode. Heater voltage from DC psu, turned up to bright orange, the dangerous bit :-). Turned up the variac initially to 20mA for around 10s, but not much change, so turned up to 50mA limit or so and saw sparks inside the tube. After a few seconds, current drops down. A couple of sessions like that, 10 secs or so each and voila, the tube works with perfect focus, though the intensity pot needs to be at 60 or

Chris

After experimenting with small phenolic potting-shell type covers, and little drawn-aluminum ones, I concluded that the best insulation inside a component cover is air.

Cheapest, too.

Assuming that the cover is small enough that convection doesn't occur.

For a given geometry there's a critical temperature gradient at which convection starts to dominate.

Long ago at UBC I took a stellar structures class taught by the estimable Prof. Jason Auman, one of the pioneers of computer modelling of stellar structure and evolution.

In stars, near the core the energy transfer is dominated by radiation. At some point nearer the surface, a convective instability starts, which rapidly achieves the speed of sound, which at a million degrees C in a mostly-hydrogen ambient is on the order of 100 km/s, enough to homogenize the convective layers pretty thoroughly. (Magnetic effects start to become important nearer the surface.)

All of which is to say, that a sufficiently-small air-filled insulator is as good as an ideal solid insulating material. ;)

Cheers

Phil Hobbs

If the cover is small. The Rayleigh number matters, and this depends linearly on the volume of air enclosed by the cover.

If it 's less than 500, there's relatively little convective heat transfer. If it gets up 100,000 you get turbulent convection currents and they transfer a lot more heat, and do it erratically.

If the cover is less than a inch (25mm) across, you don't get much convection at all, and what you've got isn't turbulent.

Still air, yes. You can keep it still with a bit of fiber (like, a blanket) which is why glass fiber insulation shows up a lot.

For cinderblock, it was once determined that you could fill it with peanut butter, and it'd be a better insulator than empty (because the large-ish void allows air circulation).

Peanut butter, at $2 per pound, is not widely recommended.

Am 01.09.20 um 01:12 schrieb Chris:

I had some problems with my Agilent 89441A control program that became unreliable over the years. It used the BNC-Ethernet interface via an Amazon converter box to the normal 1 GB LAN.

You simply open port 5002 on 192.168.178.38 or whatever and then you can dump / read ordinary GPIB strings to/from that port. Last year, this communication often got stuck and finally it was unusable.

I decided to circumvent this by using real GPIB via a Prologix USB GPIB dongle. The program got an option -488 and I inserted a multiplexer software layer just above the Ethernet access. That was easy since the de/composition of control strings stayed the same.

But I never could make the last 5% of the Prologix interface work. BTW it's nowhere specified what happens in the case of a timeout and the Prologix documentation refers mostly to third parties on the net.

I then found out what the problem was with the LAN/socket interface.

If you switch data direction, you need to fseek() even if this is just an empty call. The newer Linux kernels seem to be more picky about that, with older kernels it just used to work.

Now that the LAN works again I have no motivation to complete that Prologix option. I think I have given the program to someone on s.e.d. The update is available.

Cheers, Gerhard

I'm writing this system in layers, with a test harness for each to give the code a good thrashing to detect errors. I'm developing under Solaris 10, gcc, makefiles etc and had some weirdness with the networking layer as well. Nprmally, one would just open a socket to connect to the device and port, then a read to get the return data, but this didn't work. Tried various variations on that, but eventually settled on a select() call to check for data, then a read() which works fine most of the time, Rewrote the code as a little state machine with retries which made it nearly 100% good. Slight mods to the makefile and it builds and runs fine under FreeBSD 12 as well. Not tried Linux, but should build ok there as well. No windows version is planned, unix only for now. Have had a test harness running for weeks, reading an old Keithley dvm and has done over 10 e6 reads with just 2 errors, so may be good enough, but may revisit that later. Network loading does seem to affect it,

Of course, fseek() is an os level call, not Prologix, but suspect the Prologix firmware may have corner cases and may occasionally drop command frames, but can't be sure...

Chris

Am 01.09.20 um 13:50 schrieb Chris:

Nonono, the fseek() was needed to make the LAN socket interface work again, it has nothing to do with the Prologix.

I have a backburner project with some LTC2500-32 ADCs and a BeagleBone Black playing FFT analyzer (because of this sh*tty

BTW, You can compile FFTW (the fastest FFT in the west) without any problems on the BBB with its NEON floating point unit. That wasn't much more than downloading and "make, make install".

cheers, Gerhard

Yes, that's what I was saying. Normally, if you open a unix socket, it stays open until closed, but needing an fseek to refresh access sounds like a network tuning parameter, or bugs in the os network layer, perhaps keepalive timeout ?. Developing under Solaris 10, which worked fine, but needed to add retries and timeouts to the network code for FreeBSD. Stevens books on network programming are the bible here, but it certainly didn't work out of the box as described in the book. Suspect Linux may have simalar issues, but would want to know why that was happening. Assume you use Wireshark or tcpdump to debug ?.

Application is long term characterisation of voltage reference drift. The LM199 / 399 is out of production fwir, but some Ebay sellers have the mil / 883 qualified 199 parts for sale, which are in theory prescreened. The 399's are noisy and drift a lot initially, but settle down after a couple of months running. The

Chris

There was a class a A/D called folded cascode by my recollection but not c alled that anymore. Front end would flash the input into 2-bits and then a fast pipeline would take it from there to get 6-bits or more total. They we re uncomplicated. National had some in their last days, so did Maxim. Maybe they fizzled away.

Likewise, big fan of your book! Still waiting on the sequel. :)

-- john, KE5FX

You're welcome -- that sounds about like what I've seen, and what others have reported. 20 mA isn't enough to do anything, and 70 mA is probably more than necessary/advisable.

I will say that I've always found it necessary to tweak the trimpots for best results, rather than just readjusting the intensity control.

Definitely have never seen sparks in the tube during this process. Wonder what does that? I guess there's enough material flaking off of the cathode surface to bridge the K-G1 gap momentarily?

-- john, KE5FX

I sent a copy of Phil's book to the co-founder of a giant laser lithography company. He hated me for doing that because it absorbed him for a week and he couldn't get anything else done.

I did tweak the pots as well, but suspect the tube is low on normal emission, due to hours powered up in a past life.

As for the sparks, while the current was limited via series resistor, there could have been as much as 100 volts off load, depending on the Variac setting on the primary. Had the idea that the use of unsmoothed rectified dc at a reasonable voltage might cause vibration in the tube elements at 2 x line frequency, but can't verify that. Whatever, but it cured the problem completely...

Chris

On Wednesday, September 2, 2020 at 6:40:59 AM UTC-7,

Yep, it's like the AoE X-chapters volume in that regard. If I open it for any reason, that's it for the rest of the afternoon.

-- john, KE5FX