MOSFET Current Mirror with discrete components issues

[picky33]

I'm building a Wilson current mirror using discrete components. The input is a current source from a National Instruments PXI-6704 and the output is a 2 wire current sink from a PLC for a 4-20mA signal. I made a model using some cheap 2N7000 nmos I picked up on Amazon and a resistor I had from a kit. it worked pretty well for what I needed it for. The accuracy wasn't great but it was close enough that I can hopefully take a bunch of data points and come up with a linear function to generate the current I need.

The problem I'm encountering is when I change the current too quickly I have problems where the output current starts to multiply. For example if I go from 2mA to 18mA on the source, the output jumps from 2.05mA to 56mA and I am trying to find someway to limit this or make it more accurate. my stopgap solution is to add a PID loop to the current source side to prevent the current from jumping too high, but I'm looking for a better option.

I choose MOSFETs because I thought I would have a harder time finding BJT's with the same gain and therefore would lead to problems with current multiplying. I may try a BJT version. I need something that works cuz currently I have a really bad system where I use a relief resistor and the current source hooked in reverse to use Kirchhoff current law to sink current. it's bad and I keep losing Cards trying to fix it.

 

[FvM]

Even an integrated current mirror won't achieve the accuracy adequate for PLC analog input. BJT current mirror with massive resistor degeneration will probably achieve the best accuracy.

Different transistor heating will generate % error nevertheless. Using medium power transistors (e.g. TO-126 case) with tight thermal coupling can reduce the error. This is only to demonstrate the capabilities of a current mirror. Surely an active OP current source will perform better in any case.

 

[dick_freebird]

Since you look to be working at low current, might look at MOSFET quad arrays like ALD makes (or used to). These offer both "make" and temperature match (this latter, mismatching thermals between the D=G and the high drain voltage devices).

 

[crutschow]

I choose MOSFETs because I thought I would have a harder time finding BJT's with the same gain

MOSFETS have a large variation in Vgs(th) which tends to mess up current mirrors made with discrete MOSFETs.
It's not the gain, but the base-emitter voltage drop and transconductance that matters with BJT's, and they are reasonably well matched for discrete devices with the same part number and manufacturer, so I suggest making your Wilson mirror with those.
Adding emitter resistors as FvM suggested, will minimize any mismatch.

 

[dick_freebird]

There are also bipolar matched-quad chips in DIP14 packaging. THAT Corp makes NPN and PNP quads on dielectric isolation with 40V breakdowns, nice transistors. I think there may also be a 2 NPN, 2 PNP family member.

You could make any of the 3, 4 bipolar transistor mirror topologies and pick the one that best suits input and output care-abouts (like, do you need to swing within 1 or within 2 Vbe of ground, with fidelity?).

DigiKey is one distributor.

 

[picky33]

MOSFETS have a large variation in Vgs(th) which tends to mess up current mirrors made with discrete MOSFETs.
It's not the gain, but the base-emitter voltage drop and transconductance that matters with BJT's, and they are reasonably well matched for discrete devices with the same part number and manufacturer, so I suggest making your Wilson mirror with those.
Adding emitter resistors as FvM suggested, will minimize any mismatch.

I just tried a BJT version of Current Mirror with the emitter resistors that FvM suggested and it’s working better then my MOSFET version. My question is how/why do the emitter resistors help to deal with any mismatch?

 

[picky33]

There are also bipolar matched-quad chips in DIP14 packaging. THAT Corp makes NPN and PNP quads on dielectric isolation with 40V breakdowns, nice transistors. I think there may also be a 2 NPN, 2 PNP family member.

You could make any of the 3, 4 bipolar transistor mirror topologies and pick the one that best suits input and output care-abouts (like, do you need to swing within 1 or within 2 Vbe of ground, with fidelity?).

DigiKey is one distributor.

I did look into that, but unfortunately i need to keep the price per circuit low. I have to create 50-75 circuits so I can’t afford the cost of matched quads for each circuit.

 

[crutschow]

how/why do the emitter resistors help to deal with any mismatch?

They add a base-emitter voltage which reduces the effect of any transistor base-emitter voltage mismatch.

For a stiffer current-mirror (less variation of output current with output voltage change), you could do a 3-transistor Wilson version with emitter resistors added to the bottom two transistors.

Below is the simulated comparison between the two-transistor mirror and 3-transisor Wilson mirror (no emitter resistors) for a change in output voltage from 2V to 12V.
You can see the two-transistor mirror current (green trace) varies significantly with that voltage change (horizontal axis) whereas the Wilson mirror output (yellow trace) is nearly a horizontal line with little variation.
This may or may not be important in your application.

 

[dick_freebird]

Infineon makes dual NPN "arrays" which go for ~ $0.49 (qty=1)
which might be more to your liking. You could do two Wilson
mirrors with three pieces, or an emitter-ballasted pair using only
one (plus the resistors).

 

[FvM]

Infineon has also dedicated current mirror arrays, but as far as I understand, BCV61 and its PNP companion BVC62 aren't monolithic arrays, just two chips in a package, the same with all isolated double, triple and quad transistors in a package. The arrays have better matching and thermal coupling than current mirrors made from arbitrarily selected transistors, but performance without resistor degeneration is still poor, guaranteed current ratio 0.7...1.3. Selfheating of Q2 with Vcb > 0 will cause the major current error. Using Wilson configuration is almost useless with discrete transistors, I presume.