Finding An Opamp For A Constant Current Load CircuitI mentioned how I thought that OP07 and OPA27 opamps were not good choices for constant current load.
Going back to opamp basics, you want an opamp to have as much gain as possible. The higher the gain, the lower the gain errors. But the gain will fall off with frequency, and at some point, the input-output phase shift will be 180 degrees. If the gain is still positive, the opamp can be come an oscillator.
So an ideal opamp has one single resistor-capacitor type roll off that dominates, so you have a 6dB/Octave (or 20dB/decade) roll-off. The idea is that all other parts of the opamp circuit should be so fast that they do not even start to roll-off until below the opamp's unity gain point. In terms of phase shift, a single RC roll off becomes a constant 90 degree phase shift. This idea opamp is very stable as it always has a 90 degrees phase margin from instability. The other great thing is that this phase shift is spot on 90 degrees for every opamp regardless of frequency, voltage, input offset, etc (we are still talking the ideal opamp here) so if external compensation is required, the opamp behaves totally predictably.
In the real world, for every opamp, the designers have pushed a few limits to make the opamps specs as good as possible. So when you look for an opamp, you need to look for one with the design choices that suit the application.
If we look at the OPA27 data, we see there is no open loop phase information, but here is the gain versus frequency plot:
The fall-off should be a ruler-straight 20dB/Decade. Above 200KHz though, the fall-off rate increases, so you know for sure that ugly things are happening with the phase shifts between 200Khz and 10Mhz. Even the slightest deviation usually means big diversions from the ideal 90 degrees shift.
The designers have made tho opamp stable at unity gain - just. This is an opamp probably optimized for really great performance as a x10 or x100 amplifier, but it is unknown and unspecified for its phase performance near unity gain. With this opamp, even if you get one module to be stable, the next one might not be stable.
Now if we look at the OP27 instead of the OPA27, we see an improved situation:
The deviation is much smaller from ideal and the phase shift is specified (at +/-15V at least). The unity gain phase margin is 70 degrees which is good, and that means that even if the mosfet driver circuit adds a 35 degrees phase shift, the circuit will still be stable. The phase shift plummets to 180 deg at a gain of about -7dB, so that indicates that the designers have pushed the design limits to get maximum gain and speed. Definitely, the OP27 is much better for a constant load regulator then an OPA27 or OP07.
Now if I go to a Microchip MCP6V26 opamp (2uV offset 5.5V maximum supply voltage, 6.5V absolute maximum), we see the Open Loop performance.
This looks like an excellent amplifier for unity gain. The gain rolls off at an ideal 20dB/Decade to well below unity gain, and the phase shift shows no bumps to indicate the designers have added zero's to stabilize the opamp. The IC is specified for a 60pF capacitive load, and the 180 deg phase shift happens at a gain of about -12dB. This would be a great IC to use at a price well below $2. Since it is an auto-zero opamp, you need to give the opamp a few hundred uSecs on startup to zero the offset before the output regulation is accurate to 2uV across the shunt resistor. The 2MHz maximum frequency is a help too - lower unity gain frequencies make the design easier, and 2MHz would still mean a fast response is possible. This IC an go up to 5.3V at 3mA output, so if you have a transistor emitter follower, plus a 1mOhm shunt resistor, you have about 4.5V left to drive the MOSFET. If you can source mosfets that only need 4.5V or less to reach the desired current, it can be used.
Back to my comments about the opamp maximum frequencies. Adding compensation to unity gain opamps is not great, as at best, as capacitor across the opamp can reduce the gain from about 1.1 to 1.0. The effect of any compensation is it slows down the opamps ability to respond dynamically - like a change in the load voltage.
The way to design this circuit is aim for inherent stability without compensation, then add parts to improve stability. If you can design the driver and mosfet circuit to have a phase shift of less then 35 degrees at the maximum opamp frequency, then the circuit will probably be inherently stable. Both the emitter follower and source follower will be naturally fast, so the speed limit will be due to the time constant of the transistor drive into the mosfet gate capacitance. If the gate capacitance is 10nF, and the source follower gain is above 0.9, then the effective gate capacitance is less then 1nF. It is then all about getting the RC time constant from the transistor output to the mosfet input to be fast enough to to get the transistor/mosfet phase shift to less then the 35 degrees (or whatever you choose).
If you can manage this, then you can probably get the circuit to be stable even with no compensation. That would be a great starting place.
Hope that helps.
Richard.