There is a similar issue with giant physical processes, like oil refineries. The usual operating crew can keep it going, but shutdown and startup needs real experts.
The consequences are big too.
Modern RF power amplifiers will always have a very low output impedance, very few ohms. Just like audio amplifiers. If it were different you'd have to upsize the power devices, cooling and the budget for electricity by a factor of two. A good broadcast transmitter has an efficiency well above 90% which wold not be possible with a design that had an output impedance equal to the transmission line.
IOW, the coax is driven hard from a low impedance source. Of course, when you load it down too much by using a lower impedance than designed for the current in the final transistors will become excessive or the power supply will not be able to provide this much juice. Then a protection loop kicks in (ALC).
Even small ones get 90%:
And someone has to do the designs "under the hood".
Yes, bad parents are a major problem with many kids.
[...]
If you are driving a feedline to a transmit antenna, it's good if the antenna mostly matches the feedline impedance. But there's no reason that the transmitter can't be a low impedance; it just sees the feedline as a 50 ohm load or something.
If you drive a 50ohm coax with and amplifier that has a 50ohms source impedance you will burn more than 50% of the energy in the amplifier. Causing global warming :-)
If the amp has just a few ohms output impedance it can drive the transmission line hard and the efficiency can be well above 80% for digital mode, morse code, FM, and so on. For AM and SSB it can also be high if you employ pulse-width control methods (rare these days).
For broadcast transmitters this really matters because their yearly electricity bill is often in the six figures.
That example was one of half a zillion. I had just picked the first RF power amp schematic from my shelf. Even at that time (1980's) I didn't know of any transistorized power amplifier (and even many tube ones) that would have a high output impedance. It just doesn't make any sense, never has.
None of the real power circuitry was ever integrated. In the <100W category, yes, but the ICs there were/are overpriced and often notoriously unreliable. Combining high amperage circuits with micropower isn't that great of an idea. Once you get into the kilowatt range it can only be done with discretes. The guy clearly didn't know practical design. I find it sad that professors can pass their habilitation without ever having designed something that was successful in the marketplace.
I remember PALs, have designed out many of them. The main reason was their cost but they were also bad power hogs. Got hot even while doing mundane jobs.
Absolutely. Otherwise your get a high standing wave raton (SWR) and then some of the RF sloshes back and forth on the transmission line. I've blown out a coax cable that way. And a balun. And ...
Yes, that's how it should be. Driving with the lowest feasible source impedance. Source termination may be ok in digital lines (though I do not generally use it) but it is not a good idea for a transmitter.
Well, what I saw in Germany wasn't that great in resolution. But no matter, the content was boring anyhow.
You guys don't have any mountains and then it can work. I lived at the base of the highest "mountain" in the Netherlands, a whopping 321.5 meters tall. They insisted on mentioning that extra half meter.
Ham radio is better. We had great discussions on how to build PIN-diode T/R switches, repairing power amps, beer brewing and so on. In CW. One time I couldn't understand some guys and was doubting my morse proficiency. Then I realized they were chatting in Spanish, discussing in great detail how to repair the transmission of a Chevy Colorado pickup truck. It was interesting.
The millisecond the signal become iffy that all falls apart. Or when the other guy uses a bug.
I don't store any movies. Youtube has many and then I can always rent at the library for $1.
They have to keep track as it sinks into the mud. ;)
The data rate on Youtube at 2.0x is higher. ;)
That's why I never got involved with it as a kid--seemed mostly to be a bunch of guys talking about their rigs, which I didn't find very compelling. Could have been a bad night on 20 meters, I suppose--I only tried it once.
Cheers
Phil Hobbs
Depends on where on the bands you go. I am usually with the ragchewers and that nearly always gets interesting. Often we don't even mention our rigs but dive right in. On the CW band there are interesting people and it's great to learn about them. Pilots who flew missionary aircraft in the African bush and so on.
My longest CW ragchew (so far) was two hours. It would have gone on longer but by then our younger dog, the one with a bladder issue, had to go badly and made that known in a very vocal manner.
Am 11.01.23 um 17:15 schrieb John Larkin:
That was an experiment for ISS where there was an ultraprecise clock Zero-G H-Maser controlled by a Cesium, no, not a fountain; in near zero-G the atoms tend to stay where they are. The hydrogen maser has the better phase noise, but Cesium is the law by definition. I also made the analog part of the dual mixer time difference system to lock the Maser.
Single event upsets are to be expected; not too many. 20 meters away, there are some people living who cannot tolerate speedy protons.
The assumption was: If a SEU hits, it will probably hit a FlipFlop. FPGA configuration memory was scrubbed every minute (reloaded), also from triple module redundant logic in the same FPGA. Configuration memory is a lot of Flipflops. That is much like exchanging the carpet under your feet.
I avoided pointing out that the clock tree consists of segments with amplifiers. There are things you simply cannot handle.
There were data that were precious, like the hi resolution time counter that kept the current position in the orbit. That had to be redundant, also the photon flight time from a laser ground station with 5 ps resolution.
We had a radiation hardened voltage regulator, but in the data sheet they admitted that it could go Berserk for some ms if the regulation amplifier got a direct hit. The proposed workaround was to have a large enough output capacitor, so that the CPU & FPGA would at least not explode in these milliseconds.
cheers, Gerhard
I had that objection early on. The Philips "Coolrunner" CMOS parts didn't draw much power if you didn't clock them fast. Philips eventually sold them to Xilinx and they were still around when I last looked.
On a sunny day (Wed, 11 Jan 2023 13:57:53 -0800) it happened Joerg snipped-for-privacy@analogconsultants.com wrote in snipped-for-privacy@mid.individual.net>:
You gotta dig for it, also it (TV) helps to learn the language better. My Spanish improved from watching Cuba TV :-) Some nice music they had too. Most US movies in Germany seem to be German voice over ... They do not normally supply an original language channel it seems.
....
Ha! Been driving up / down that hill with my Ford Mustang and some kids in it. We _do_ have high TV towers too!
Well, online we have
Some AI will fix that, your turn to program it?
Yea, but imagine after WW3 or was it WW13 ... archaeologists digging up my movie collection and writing books eeeeh no ? whatever they do to communicate about it.
I have never had to restart a really big system but I did once have to nursemaid a newly installed and unreliable at the time quad processor mainframe through a year end against an accounting deadline.
After 24 hours with only brief snatches of sleep your decision making is far from optimal, but the alternatives are often worse.
One of my colleagues woke up driving parallel to the road in a desert after a long observing stint - they decided it was time to stop and take a nap. Fortunately traffic was light (ie nil) and the terrain flat. Do that coming down from Mauna Kea observatory and you do not live to tell the tale (driving there on the mountain is one of the biggest risks).
It worries me that international treaties are invariably negotiated and signed by people who have been seriously sleep deprived for several days.
The procedures are all written down and tested but some of the wisdom may not be. There was a company edict that certain of us could not travel together on the same plane for fear of losing too much technical expertise at once in the unlikely event of a plane crash.
I've wondered how awful it is to restart a giant system, credit card
Good that you noticed that too. Political decisions seem to depend on who has the most stamina rather than who has the best arguments. Worrying indeed.
Jeroen Belleman
Sure they can be enforced in a design review. Simply ask the programmer what the Invariants are and how the relevant code achieves them.
This has nothing to do with whether the language has anything like Invariants - the compiler won't understand the domain of the app code being written, or know when and where to use which invariant.
Yep. Depends on what the code does.
In 1985 or so, I built an invariant into some operating system code used in a radar. At this level in the kernel, nothing like printf exists. Nor did it have a printer for that matter.
The problem was that the app guys were not ensuring that what had to be global-access buffers were in fact in hardware global memory; if not, the system's picture of reality would rapidly diverge. Subtle but devastating.
So, what to do? This has to be checked at full speed, so it must be dead simple: Add to the kernel a few lines of assembly code that verified that the buffer address fell in hardware global memory and intentionally execute an illegal instruction if not. This caught
*everybody* within 8 hours.
Deadline schedulers do exist and were widely touted for realtime, but never caught on outside of academia because they fail badly when the planets align badly and deadline cannot be met. This proved too fragile for real applications. Not to mention too complex.
In RT, one does not use malloc except once, to get the working buffers. After that, the RT code handles memory management.
Java-style garbage collection would cause random system hangs, and so cannot be used in realtime.
Not if it's coded correctly. In C and like languages, one either declares the relevant variable to be volatile, or hides a critical part of the mechanism in a subroutine, to prevent the compiler's code optimizer from making such assumptions. Or write it in assembler.
Joe Gwinn
Yes, and earlier there was also an Intel series. However, Intel seems to bungle a lot of potentially very successful product lines and this one was no exception.
It's usually parents making wrong choices. If they'd forgo buying another nice new car every 3-4 years, those regular $200 nail jobs, a $150/mo family cell plan, $100/mo for cable TV and so on they would not need so many jobs.
This is not just talking theory, I have helped people (successfully) dig themselves out of a pile of debt and then live in a much less stressful fashion. On family couldn't even afford healthcare. Now they have it.
Example: Even though I am a consultant that sometimes has to be in reach
24/7 I have a $10/mo cell phone plan and it's totally sufficient. Our cars are both >25 years old but in good repair. Of course, no cable TV.
I suspect that Intel and AMD bought FPGA companies for gigantic server applications. And they don't especially want small users. We've mostly switched to efinix for new designs.
There is an impressive list of things that Intel has bungled. DRAM. Bubble memory. Super-CISC. Super-RISC. ARM. Phones. EUV. All they can do is x86.
Only full immersion really does, which means speaking, communicating. This is why I made myself a big "Nederlands a.u.b." button to wear on my T-shirt when learning Dutch. Otherwise they'd hear an accent and come back in German or English. For some reason I didn't need one in Belgium, when I started in Vlaams they didn't switch. Being in a sports club there messed up my Dutch a bit.
Which is why I preferred Dutch TV as a teenager. For a while, until I had my ham license, from then on no TV just like now.
[...]
Dutch I understand. But I prefer on air versus online. One reason being that the "line" can go away at any time while the ether doesn't.
I am not a programmer. Also, when it's one of the usual ancient bugs in various states of oxydation the irregular contact bounce is something AI can't really deal with.
[...]Am 13.01.23 um 22:02 schrieb John Larkin:
The next or next-next generation of the Philips design is still alive. I still use them; they are nice garbage collectors for leftover logic or SPI-parallel converters or a frequency counter with SPI to check my Larmor frequency.
I think that was the EP300++ series. AFAIK they also invested some money into the company that continued them, but PAL- lookalikes had no growth path to larger sizes. It is always N product times M inputs times I FlipFlops. That gets quickly out of the hand. And every product term * input crossing draws continuously current? All those 180 mA-PALs! Have you ever seen a real world product that contained a MMI MegaPAL?
In FPGAs, the effort doubles for twice the size. Much more friendly.
That was my first FIR filter, in 1984: <
You can bet: it ran hot.
I really need to update my web site.
If you look at the Xilinx portfolio with these Zyncs, or the monster FPGAs with GHz ADCs: that is intended for the cell phone tower market. I worked on cell phone tower electronics, with 10M devices for the continent over 10Y. Then each € counts. Xilinx made sure they would not lose that on price. Dream conditions. I do not see that often.
When your horse is dead, get down. Before you rott with it.
DRAMs was Intel technology when they were young and costly. Then came NEC, Fujitsu, Hitachi.. They were cheaper, but where are they now? Now it's a cutthroat biz for pure process specialists.
Bubbles were funny, but with these pipeline delays and not sooo large capacity they were between a rock and a hard place.
Intel bought a ARM license, just to be sure. For some time, they dominated the RISC market with the 960, but when the party's over, turn off the light (Doors).
And X86 is internally everything but X86. After that 432 fiasco, that really was a HP design, they were quick with a X86 alternative with > twice the performance to secure their market share. Where is the Fairchild Clipper, the 68K, 88k, Alpha, National 32032, MIPS, SPARC, HP Snakes (my Agilent 17602B logic analyzer has one runing HPUX. Was a HP KING COBRA workstation really THAT slow?)
Internally, PentiYummies are now multiple RISCs with 100s of renaming registers that MAY be seen as EAX or whatever from the outside.
Cheers, Gerhard
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.