reg + reference

I could use a 1.2 volt shunt reference, with a reasonable resistor from 3.3, and use an opamp to scale the 1.2 up to 3. More parts! The National low-dropout reference is ideal, if it doesn't oscillate.

The NXP LPC1768 we're going to use has a 12 bit mux'd ADC and a 10 bit DAC, and all the other stuff like ram, flash, SPI we need. It's a 100 MHz CPU with single-cycle multiply, pretty impressive for around $7. We're going to be running a couple of PID loops as fast as the ADC can feed us data, 100K hits a second maybe.

Some of the ST parts look nice, but we already have the compiler/jtag infrastructure in place for the NXP, and we're blinking LEDs, so we'll stick with that.

ARM seems to have won the embedded game. Freescale/Coldfire is rumored to be seriously ill.

John

On a sunny day (Fri, 26 Mar 2010 18:27:35 -0700) it happened John Larkin wrote in :

This circuit of yours is basically a rectifier (be junction). In good opamps there is a current source in the emitter. Indeed very few opamps are free from RF interference, Dave had some chip on his video blog that actually could stand GSM signals close up, most cannot.

If you do not believe that, get a GSM phone, and put it next to your circuit. Low pass the reference your create, and use your nnnn+ digit volt meter to see any changes,

As for creating the reference, you have plenty of volts on the input, why not use something like a MCP1525? Or simply a LM317, I use LM317 for everything, very stable output at low loads. Some PICs have an internal programmable reference that you can output on a pin too IIRC. Maybe it can replace some of the rest of your circuit :-)

I would have typed 'never to old to learn', but of course in the case of Alzheimer that is no longer valid.

snip

snip

I meant connect the 1.2V reference to a spare ADC channel, no scaling needed.

-Lasse

AFAIK the LPC1768 can run on lower voltages as well like down to 2.6V. Not at full speed ofcourse. 40mA may be too optimistic though.

Be sure to decouple the reference properly. The ADC sinks quite some current from the reference pin when sampling.

Sorry but that must be some calculating error on your side. All PABXes I've seen had press-fit backplanes for a very good reason (reliability). But just like soldering and crimping connections mounting press-fit is something that needs to be done right.

And nothing seems to be guaranteed. When I read page 9 in the datasheet I am beginning to feel sick.

Why does working off of the +15V have complications?

My client also did, and AFAIR some engineers had Irish blood in them. At least they fancied Guinness like I do. But then one fine day luck ran out, bus errors crept in. Initially they didn't believe me but after a lot of cajoling agreed to adjust the solder machine temp profile and to run dozens of boards through. That took some doing because it was a huge and thick backplane. The number of bus error messages dropped to zero but only on machines with soldered backplanes.

_Then_, Mr. sales dude came in touting a new and improved pressfit process ...

On the first series it had been done by the pros. Sorry, but my experience with pressfit is not positive. Not as bad as with wire-wrap but close.

Yup, and then the LM4120 datasheet says, quote "Care must be taken when using output capacitors of 1?F or larger. These application must be thoroughly tested over temperature, line and load."

That would give me the creeps.

Oh. Then we'd use 3.3 as the ADC reference, measure 1.2, and normalize measured values against the 1.2. That would work, with a bit more math... unfortunately a divide, but not too often. I could do that at

John

As any common-emitter amplifier is.

And this is a current source into the base! Since average base current is nearly constant, RF should not affect average collector current.

close up, most cannot.

any changes,

The TC would be terrible, espacially as it self-heats. I want overall channel accuracy of at least 0.1%, and there are other errors to deal with, so I can't use a rotten reference.

John

Oh, I see what you're getting at, fig 4

ftp://jjlarkin.lmi.net/ARM_power.JPG

if I catch your drift. Nice.

John

We figure on sleeping at least half the time in operating mode. I'm using a 1-watt unregulated Murata sip DC-DC converter to power a channel, 12 volts in and +-15 out. The +-15 becomes floating ground,

Luckily, VME does have a lot of air flow.

The LM4120 allows me a maximum of 0.047 uF!

The NXP documentation for this chip, around 1K pages of it, is horrible. We can find no mention of how much current the Vref pin uses. There's also a pin called, if I recall, Vcc(Reg)(3V3) whose function is unknown. Got any idea? Kyle must not either, since on their eval board they run it to a jumper header, making it our problem.

The NXP documentation has roughly 1e6 syntactically-challenged words but not a single application circuit. They should have the schematic and code for a minimal LED blinker.

Mixed-signal datasheets come in three sorts:

Terrible analog specs

Terrible digital specs

Terrible specs.

John

Doesn't have to always be custom. I can pay off to check what's in stock among these:

Doubtful that this is going to cut it. An option may be to run an RC filter with the R being just large enough to fulfill the LDO's stability criteria (might require a trip to a psychic reader to find out what those will really be ...). Then maybe, just maybe, you can place a nice big MLCC.

That warrants an inquiry to app engineering. I'd also would want to find out about spikes because if sort of irregular they can mess with the accuracy of your ADC channels. Since you go into sleep mode the surf on that rail will most likely be low frequency.

Often they are written by non-native English speakers. Japanese datasheet can be rather interesting there. It's usually ok if you get the impression that they knew what they were doing.

No surprise these days :-(

Cute, how they say the compensation capacitor is in the "50 pF range" and how they expect *me* to compensate and extensively temperature/line/load test *their* part to ensure stability. I'm expected to do a couple thousand dollars worth of testing on their behalf.

National is very sloppy as regards stability of lots of their parts.

I'd need a series reference that can tolerate unregulated +15, so I'd probably have to preregulate into it... more parts. Shunt regs have mediocre TCs and would get not-very-constant current if I just use a resistor from +15. I expect that my 4-20 mA (really 0 to 25 maybe) load and uP power patterns will whack the DC-DC output voltage all over the place.

There will be 12 channels per VME board, so I want low parts count and cheap parts!

John

On a sunny day (Sat, 27 Mar 2010 09:51:08 -0700) it happened John Larkin wrote in :

I did say this before, maybe nobody believes it, but I had the LM317 on the fluke with no load (or light load actually), and heated it up with a soldering iron, the display did not change.

loads.

Can you quantify the TC? National's spec is 1% over the temp range, whatever that means, which is maybe 20x worse than I'd like. The typical curve on the datasheet just about eats all my error budget.

John

Which is why I haven't designed in any of their power parts lately, except legacy stuff such as the LM317. LTC is pretty good, very responsive, great support engineers who really know electronics. Plus LTSpice models for just about anything and that alone ist worth a ton, has won them many design-ins on my part. Although I recently unearthed an ugly power-up bug in a part but that was a comparator, LT6700. But they went to the ground of it, confirmed it, fessed up, and apologized. Way to go.

An RC filter or capacitor multiplier should suffice. Input ripple rejection is usually huge as long as it doesn't contain high frequency stuff. Or use one preregulator per board. Or even better, one reference per board.

Yes, LTC is great. And LT Spice is a gift to the world.

Sadly, all 12 channels are individually floating.

It's looking like the safest thing to do is use an LM1117 to make about 3.4 volts - close enough to 3.3 - and use a ca 500 ohm resistor into an LM4040-3.0 or equivalent bandgap with a 10 uF ceramic bypass. Typical TC will be about 15 PPM (guaranteed is not as good, but not lethally bad), sensitivity to 1117 output is pretty good (500 ohms into Zd~~1 ohm), and it's simple and cheap and most important, it's safe.

That out of the way, my next task is to explode some SSRs. Certain user abuse modes will put, say, 50 volts across an ON-state SSR. The ARM will sense the hazard and turn off the SSR, so it will have to tolerate the abuse for a millisecond maybe. I'm considering some Clare

I love to test parts to destruction.

John

Ever thought about these to make brew a multi-output converter:

I still don't understand why you want seperated floating channels. Why not take the inputs through a differential amplifier with a huge common mode range and use one controller?

The amount of current on Vref depends on the ADC's sampling frequency. You can also look at the LPC2000 datasheet to get an idea.

Vcc(reg) powers the on-chip DC-DC converter to make 1.8 volts for the core.

Thats where the Keil boards + example code kick in. Although I must warn that the Keil example code needs a lot of brushing up before its actually usefull.

NXP is a bit slow on releasing final specifications. Did you download the latest datasheets?