doing a PCB resistor

[Simon]

I'm trying to sense current 20ish amps and have been using a 1 milliohm resistor although I think I can afford to take it to 2 mR. now I also wanted the current limit on the circuit to be a little temperature dependant so that in warm weather the value would tend to be higher than it really is letting less current through and in cold weather it will be low and let a bit more current through. After poking about some ideas I had a thought: resistors usually have a thermal spec as well, they will normally increase resistance slightly when warm, then it hit me: why not make the resistor out of PCB ? in fact copper has a 0.393% increase in R/C so might just work.

Now before I go thinking myself really clever here perhaps somebody can tell me if I'm mad and how I may come a cropper on this one

[alm]

I wouldn't expect very good tolerances, the variation in trace width and copper weight is probably a lot more than the 0.4%/degC tempco. Plus .4%/degC is a really lousy tempco, can't imagine that 'real' current sensing resistors are any worse. Make sure to use Kelvin sensing, since resistance between the voltage sensor and the shunt is likely to exceed 1mOhm. For reliable results, I would prefer real resistors. If you don't care about accuracy and can cal variation in resistance out, it seems to me that it would work fine. Dissipating 20mW shouldn't be an issue either, I would try to make the trace fairly long and wide, to make variations less significant. But I've never done it either.

[marianoapp]

DN-71 Using Copper PCB Etch for Low Value Resistance

[Simon]

I wouldn't expect very good tolerances, the variation in trace width and copper weight is probably a lot more than the 0.4%/degC tempco. Plus .4%/degC is a really lousy tempco, can't imagine that 'real' current sensing resistors are any worse. Make sure to use Kelvin sensing, since resistance between the voltage sensor and the shunt is likely to exceed 1mOhm. For reliable results, I would prefer real resistors. If you don't care about accuracy and can cal variation in resistance out, it seems to me that it would work fine. Dissipating 20mW shouldn't be an issue either, I would try to make the trace fairly long and wide, to make variations less significant. But I've never done it either.

0.001 R * 20*20A = 400 mW with 2 mR it will be 800 mW

[Zero999]

What about making the trace slightly wider than you need and drill a hole in it?

To trim, you increase the size of the hole slightly using a countersink, until the resistance is just right.

[Simon]

Well from what the app note say's a PCB resistor is just the ticket as I'm after current limiting at the end of it all. I could easily trim in software, I don't know how good using the discrete milliohm resistor would be as I'd have to be careful with soldering etc so verification is going to be necessary anyhow so I can trim it in software, or drill a hole if it is way off the mark but I'd rather avoid that.

My aim is that in cold weather a bit more current can flow but in hot weather it will start to limit it a bit more, basically by doing this more or less accurately I will get out of using a separate temperature sensor which costs money and I've run out of inputs on my MCU to feed it into, basically I'm very crudely combining to the the two parts of data into one rather than having two inputs and making compensation in software, of course I also save myself the costs of the resistor and pain of soldering an SMD part

[Neilm]

I use this free application which will help if you are planning to do make a PCB resistor. It can be downloaded from herehttp://www.miscel.dk/MiscEl/miscel.html

I have used PCB traces as resistors in the past but generally the precise resistance for this was unimportant. I have certainly not used one for precision current measurement.

Neil

[alm]

0.001 R * 20*20A = 400 mW with 2 mR it will be 800 mW

You're right, brain fade.

Soldering a current sense resistor wouldn't really matter for the result, if it uses proper Kelvin connections (four-terminal). The extra resistance will cause extra voltage drop and dissipation, which might be an issue.

One issue with high-current shunts (including those especially designed for that purpose, but even more for improvised ones) is that the self heating of the high current might change the resistance, which makes resistance a function of the current. But you can compensate for this in software. If the tempco happens to match your requirements, that would be a lucky coincidence, and might save you two parts.

[Simon]

yea I've been over power calculations time and time again, that's how i ended up with a crazy 1 mR value.

In a way the whole thing being thermo regulated by the resistor in just the right way would be very handy, self heating is not a big issue depending on the extent as it means that not only do I have ambient temperature conditioning but that the circuit will output a bit more until it warms up after which it will drop to nominal, of course thorough thermal calculations for the trace must be made as it could go horribly wrong. I'll probably get a first prototype running with a fixed resistor before playing this game.

[RayJones]

Perhaps this may be of use for you Simon?

http://www.pcbco.com.au/tracecalc.html

[Simon]

thanks ray, I'm wondering if it might be easier to get a poor discrete resistor, as I'm using the max4172 the output voltage depends on the voltage in and the output resistor so if I have a resistor with "bad" temperature characteristics it would do the same thing

[alm]

Note that if the manufacturer species wide (bad) tolerances, that does not mean that the part will necessarily be at the outer limit of this tolerance. That the tempco sucks doesn't mean the manufacturer didn't try to minimize it, they just didn't try very hard. You might get parts from a better series some day, or they might improve the process. It's dangerous to rely on non-guaranteed specs, especially if it's for production. But if it's for hobby, it doesn't hurt to experiment. Otherwise, something specified as thermistor might be better, although I doubt they make those in 1mOhm, so that would require extra parts.

[Simon]

well I have a 1-2 mR resistor as a current sense and then the MAX4172 chip that is a current mirror so for 20A in a 1mR resistor I get 4 uA out, I choose my resistor so as to get the voltage scale I want and feed that to an MCU, now I have no idea what the temperature coef's of resistor are and what sort of variation I'd get naturally,

I've looked at thermistors but frankly their data is confusing and often missing on farnell

[Simon]

Well I think I'll end up going and getting a proper sensor chip from farnell, it will just simplify life and enable accurate predictions of behavior

[RayJones]

What do you mean by "proper" current sense?

In my experience so far, using opamps on the low side I experienced the same vagueness you seem to be experiencing, but this all went away when I went to using a LT6100 on the high side. (only gotcha is Vcc *must* be below Vsense)

I would have hoped that the Maxim parts are equally stable?

As others have alluded though, you must use kelvin sensing from as close to the leads of your shunt as possible, and no load current shall pass down these traces.

Are you thinking of hall effect sensors instead?

[Simon]

I meant a proper temperature sensor with separate input to the pic (well a larger one as I've exhausted this one)

I'm using the MAX4172 which is a high side current sense chip and yea you can have the inputs as high as you like

[RayJones]

Yeah OK, it certainly wasn't clear what sensor you were speaking of 

[Simon]

yea sorry, I think I was getting carried away with making the hardware do what software was made to do