Questions on interfacing to current sense transformer

Yeah, and I stated 20W (and cheap). SOL?

All I get are journal articles that I can't access from here.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

In quantity, easy. Once you've set up the punch/die or paid for a photoetch run, and bought a lot of sheet mangain rolled to your specs, the shunt blanks can cost 50 cents maybe. Epoxy to a heat sink and you're done.

That's life these days. Publicly-funded research behind monopoly private firewalls.

John

Or fold a strip into a tight U shape, with a bit of mica or kapton as an insulator. That's pretty good for minimizing loop area. Or get clever and reduce it to zero.

John

3nH

I tried getting clever on this problem, once. Some geometries (like a spherical shell with current injected at small radius, removed at large radius) have, by symmetry, no associated magnetic field and therefore no inductance. The problem is, any resistor has some ohmic heating, which means Poynting's vector over the surface can not vanish- the whole circuit of any resistor has to create some magnetic field, i.e. inductance. So, the connecting wires have to be carefully considered. The whole circuit, all the wiring, not just the 'shunt part'.

The best you can accomplish, is a pair of Kelvin taps to a shunt that, by their geometry, do not sense an inductive component of the shunt's voltage drop.

Sure. It needs engineering.

That's a good shunt.

John

Well, ten is a quantity, it's just not much of a quantity.

Indeed. Doesn't help me find out about coaxial shunts though...

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Soooo... I could, like... take 8" of 18AWG copper (what's that, 4mohm?), and as long as the kelvin leads are close enough to the shunt (ideally speaking, inside, which would be coaxial), it's fine?

Another way might be this: at one end, the first kelvin lead comes off; the other bends to travel parallel with the shunt, tracking down the length until the other connection point; inbetween, ferrite beads could cover the pair of wires, to maximize coupling between shunt and kelvin lead / minimize stray inductance. Nah, that wouldn't be good, the shunt should be folded in a hairpin; then four pieces travel (shunt up, kelvin up, shunt down, kelvin down) and those could go inside ferrite beads I suppose. Or if the shunt can be hairpinned tight enough, who even needs the kelvin lead following. That depends on how much area is enclosed by the shunt, and how much of that can be copied by the kelvin lead...

More or's: intentionally construct a slightly inductive shunt element, then attach inversely polarized windings for the kelvin lead. Use a pot or something to vary between them, nulling apparent inductance. Since the shunt might be 1 turn, you'd be varying between 0 and -2 turns, looking for the ideal -1. Hey, I think that would work...

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

I go to a nearby University of Californa Medical School facility that has a small library. I can access all the journals on a public "guest" PC (meaning that I have to stand up) and for about 25 cents a page I can print stuff and carry it away. Actually, I think I type better standing up.

Or if I have an intern during the summer, as I do now, they can often use their student account, through their school, to access journals.

It's really stupid that research is kept buried like this. The abstracts are deliberately dumbed down so that you can't tell that a paper is useless until after you've purchased it online for $30 or so.

John

Copper has a rotten positive TC. Manganin is good but hard to get. Nichrome has a good tc and is easy to get, but hard to connect to. Sigh.

A tight hairpin is good to minimize loop area. And it's possible to design geometries that cancel the effects of bulk inductance at the Kelvin connections. But I can't tell you how we do it.

I have done multi-trimpot-tweaked equalizers to null out transient errors, especially eddy current effects, but we have since developed geometries that need little or no compensation. Another problem is to avoid hum pickup from stray 60 Hz fields and thermocouple transients, but the solutions track.

John

The IEEE at its useless best (*)

I inquired for some sort of membership that would allow access to ALL the "digital libraries" without having to join each individual group.

I was ignored.

I cancelled my IEEE membership.

Fortunately, "sharing" abounds ;-)

I've grabbed several nice books, magazines and articles from

alt.binaries.e-book.technical

Are there any other such usenet groups?

==

(*) Ever priced their insurance? More costly than _anywhere_ else.

...Jim Thompson

--
| James E.Thompson, P.E.                           |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
 I love to cook with wine     Sometimes I even put it in the food

[snip]

I always bought Nichrome with spot-welded leads.

...Jim Thompson

--
| James E.Thompson, P.E.                           |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
 I love to cook with wine     Sometimes I even put it in the food

I happen to be just a few blocks from Beloit College, although I'm not sure if they have IEEE access. MSOE has access, but I'm not on their intranet and I'm not sure if they have a proxy I can use.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

and

ng,

Not what I had in mind; the intention was not to minimize the inductance at all, but to negate the small voltage variation it causes.

en

or

Yep, that's it. The shunt and its associated connecting wires ALWAYS have inductance (resistive heating and Poynting vector proves that).

So the shunt's delta-V is

Vshunt =3D I * R + L* dI/dt

and that's what shows up on a Kelvin tap wire connection. The inductive part is unlikely to be a big effect on the power calculation because it's out of phase, but it IS an error. The solution I came to, is to note that the magnetic flux, phi, near the shunt is proportional to I, so a winding (there is very low flux coupling here, so we're talking about fractional-turns transformers) will pick up

Vxformer =3D K * d(phi)/dt =3D k * dI/dt

and there's a 'k' value that makes the series combination of the Kelvin probe wires and the aircore winding into a true source of the I*R signal we were looking for. No need for a pot, just very slightly bend the wire loops that the Kelvin taps connect through.

It might make more engineering sense to minimize L, or to note that the effect doesn't change the revenue equation, or even to calculate the effect and subtract in postprocessing, but I sleep best if Ohm's law seems reliable.

Hmm, so let's say, vary K by varying the diameter of the kelvin turn with respect to the shunt turn? That could be done, in small part, by leaving one part of the kelvin lead loose, so it can be bent in or out of the loop the required amount. Or, it could go around the loop slightly too far (coming off perpendicular before returning to the other lead at a larger radius, to let the extra flux leak by) and adjusting that. 'Course, the whole thing should be shielded and potted to avoid stray fields that might mess with settings.

Yeah, maybe an RC network could compensate or something, but getting it right the first time is easier, and it looks better anyway. :-)

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

So, I put together most of my project and it works, though I'm sure the output is subject to all the errors you guys have discussed so far and then some.

- I used a CT and burden resistor to generate the current signal (Triad CSE187L), which I biased up so that it went almost 0-5V when my tea kettle is plugged into it (~1400W). For the voltage, I used a

300mA 12.6V transformer (Radio Shack cheapie), again, with the output divided and biased to 0-5V w.r.t.the AVR and loaded a bit, to about 10mA.

- I put a simple low-pass filter on both signals, but it probably is not necessary; the signals were relatively clean, at least with the nearly resistive loads I am testing with

- I had a limited set of plug loads to test with. I used a crock pot, a waffle iron, and an electric tea kettle. I assumed that all of these are more or less resistive, but I expect that's not quite true.

- getting decent performance out of the AVR takes care. I ended up using a timer interrupt to kick off the ADC and return. The ADC itself calls an interrupt when ready with a complete sample. I did not plan on writing two handlers, but this was the most straightforward way to avoid polling and allow as much time to the non-time-sensitive code. The timer starts an I-sample, and when that is ready, I immediately start a V-sample. On the next timer interrupt, I start with V first. Right now I'm sampling at about 2kHz, but I can go faster.

- I added a simple software low pass filter of the type: out[i] = (in [i-1] + in[i] > 2. It does not appear to make much difference.

- My AVR is clocked at 16MHz. This is annoying as the clock divider options for the ADC are all 2^n, so I can't ever get a rate that matches 60Hz exactly. Who knew 60 doesn't go into 16,000,000?

- The AVR has 2k bytes of ram. That makes delaying by 90 degrees for directly calculating the imaginary power an expensive proposition. I have not done so, but may experiment with it. In the meantime, I can work it back from the apparent and real.

- I tried to avoid using floating point math, which is pretty slow on the AVR. Sticking with all integers and using only shifts to divide requires much more careful management of precision that I was bargaining for. I spent a lot of time tweaking to get this right. I still might go back to FP, though, as the code is so much simpler.

- Phase error. To a first order, this does not seem to be a problem. I put my current and voltage signals up on the scope, and to the degree I could see, they lined up pretty nicely. (again, waffle iron, kettle, crock pot). Assuming these were all supposed to be near 0 degrees, the appeared to be so.

- linearity. This is indeed pretty bad for the CT. Using my VOM for comparison, current as measured is not linear with load. I can fix this with a lookup table and interpolation, but being as I don't really have the equipment (calibrated loads, calibrated measurement equipment), will probably leave this for another day.

- isolation. yeah, two transformers for perfect isolation seems like overkill now that I've done it. I probably would let one or the other float. It's easier to take the voltage off a divider than the current. For the current, I'd need a calibrated low-resistance shunt, probably making a voltage that would require amplification before the ADC. no big deal, I guess, but I was trying to keep this project very simple and rely on the uC + passives. I was also thinking it would be nice to AC-couple using only passives, but I think at 60Hz this would require unreasonably big caps.

So, if one looks at this as an attempt at a revenue-quality or lab- quality power meter, it fails. But as a simple hack on the level of Arduino, it's not too bad. I get real power, v-rms, i-rms, apparent power, and frequency. I can get phase, too, but have not done so yet.

On the flip side, the whole point of this project was to put the info on the internet, where I can access it remotely. For that, I plan on using the wiznet ethernet module with the AVR. Getting that to work will be just a software problem, so it's not much interest here. My plan is to have the AVR act as an http client, initiate POSTs to some a cg script running on a server which will store the data. Other cgi scripts will draw from that db. (I wanted originally to hook into the Google power meter API. It turns out that API is still not published, and it requires SSL, which I think will be too much for the poor little AVR)

If I feel like it, I might wire up a couple of optoisolators+SCR to allow me to switch loads on and off remotely. If I do that, I'll need to make the AVR act as a server...

-- dave j