Analog thermostatic heater

[Birb]

Hi. I do not know if this is the correct category, but it seems ok. I am attempting to create a small thermostatic heater. It should have a variable setpoint, and the temperature should remain fairly constant (+-0.1C over time). The temperature doesn't need to be accurate, only precise.
Currently, I have constructed a wheatstone bridge to measure the resistance of a PT1000 internal sensor, which feeds into a subtracting op amp circuit. This provides gain necessary to drive a transistor and the heater.

Problems:
The op amps need -5V, which means the entire circuit will need to be a bit more complex.
The transistor dissipates a lot energy. (Could possibly ditch the heater and use the transistor)
The temperature might oscillate, as the circuit only increases the output whenever the temeperature decreases compared to the setpoint, rather than using a convergent method
The idea is to use a piece of nichrome wire as the heater circuit. However, the resistance of the nichrome is also influenced by temperature somwhat.

Questions:
How can this circuit be improved? (I.e. increasing temperature stability)

Is it better to use a piece of maganin wire of the same resistance, or is the effect too negligiable?

Circuit:

[Kleinstein]

Using the transistor as heater is a good idea.  The resistive heater has the problem that the power is proportional the voltage squared and thus the regulator gain changing with the power.
The transistor power is more linear in the current. The emitter resistor would be much smaller and could be used for a current limit (with an extra transtor) to limit the power without to much heat lost at the resistor.

The shown circuit is a protorional regulator only. This gives a limted accuracy as too much gain will end up in oscillation. Usually one would want more like a PI type regulator. In the simplest for this could be a capacitor in series to the resistor in the feedback (and likely higher resistors (more like 1 M instead of 1 K) to get the requited rather long time constant).

For easy regulation it helps if the response from the heater to the sensor is relatively fast. The right position of the sensor can do a lot. If really critical one could opt. for 2 separate sensors for the P and I part: the P (and if used D) part close to the heater and the I part more at the critical part to really keep at temperature.

The "zero" for the difference amplifier could be moved up, so that one could get away without a negative supply, and just a single supply OP-amp that can work near GND.
One can drop a few resistors at the difference stage, as the bridge part already provides some source resistance.  One can also drop the extra amplifier stage and do the regulation / integration at the difference stage.

[Ian.M]

If you aren't too concerned about EMI, to reduce transistor dissipation,  the final OPAMP can to be configured as a comparator with a tiny bit of positive feedback to give a little hysteresis so the transistor is eithe hard ON or hard OFF.  This makes it a bang-bang controller.  To further reduce the dissipation replace the transistor with a low Vgs threshold N-MOSFET, wired between the load and ground.  Add a 100 ohm series gate resistor.
 
To get back proportional control, you need to PWM the load.   The easiest way to upgrade a bang-bang controller to do that is to either dither the setpoint with a triangle or sawtooth waveform, or add the dither to the comparator input.  Ideally the dither amplitude should be chosen to be equivalent to a couple of degrees temperature difference.    To do this, connect another OPAMP as a RC relaxation oscillator, and use a high value resistor (at least 10x the osillator's feedback resistor) to couple the sawtooth at its capacitor to your differential stage's -In.  To keep it balanced, add the same resistance in parallel with the resistor to ground from its +In.

Finally, the need for a negative supply can be eliminated by using a rail splitter, or by replacing all resistors to ground with a pair of resistors of twice the value, one to 0V the other to 5V.  Obviously the OPAMP needs to be a 5V single supply RRIO or near RRIO type. Use a quad OPAMP and the spare channel can be used to buffer a potential divider for the rail splitter. 

N.B the load then returns to 0V rather than the analog ground, which is now at 2.5V

If you are concerned about EMI (e.g. ovenizing a sensitive circuit), keep it linear and follow Kleinstein's advice . . .

[Birb]

Using the transistor as heater is a good idea.  The resistive heater has the problem that the power is proportional the voltage squared and thus the regulator gain changing with the power.
The transistor power is more linear in the current. The emitter resistor would be much smaller and could be used for a current limit (with an extra transtor) to limit the power without to much heat lost at the resistor.

The shown circuit is a protorional regulator only. This gives a limted accuracy as too much gain will end up in oscillation. Usually one would want more like a PI type regulator. In the simplest for this could be a capacitor in series to the resistor in the feedback (and likely higher resistors (more like 1 M instead of 1 K) to get the requited rather long time constant).

Thanks for the prompt reply. I've seen this PID circuit before. So the circuit could look something like this?
Circuit form https://control.com/textbook/closed-loop-control/analog-electronic-pid-controllers/

Also, for better heat distribution, can more transistors be used? I.e. the enclosure is cylindrical, and three transistors are evenly spaced. Then the center should be heated evenly.
 (Though that might complicate the P and D components due to requiring more sensors?)

[SeanB]

I have done it using just a NTC thermistor and a power transistor, with the NTC robbing base current from the transistor, and the transistor, using a PNP power device, TIP131 IIRC, with the collector grounded, and soldered to the crystal can. Temperature was around 70C, selected with appropriate base resistor, and the NTC 4k7 bead stuck to the crystal, and the whole lot buried in some foam insulation. Select on test resistor to give around 200mA on power on, off a regulated power rail of 5V, and then measured the rise after an hour, and adjusted resistor to get to around 70C. Yes horrid tempco of transistor, but as it was all at same temperature it worked out well, and gave the crystal, a cheap one in a meter, a lot more stability.  5V rail could easily handle the 200mA of power extra on start up, and low overall thermal mass made it fast to heat up and stabilise.

[Kleinstein]

Depending on the requirements and setup one could get away with more heaters in parallel and still just 1 sensor and hope for symmetry to still get it good enough.
So the additional transistors would spread out the heat.

The circuit shown may work, but is quite complicated. One would only need the D part if the system is tricky or one needs high speed.

For a relatively fast (small, like 20 x 30 x 40 mm³) oven I have a rather simple circuit with just 1 OP and a transistor used as sensor, as it can be mounted conveniently. The P and I part can be combined with a single OP-amp.  A convenient feature is that the feedback is taken from the actual current and thus taking into acound saturation. So the integrator wind-up is limited at least somewhat.

The circuit can however be tricky with a slow system and thus a long time constant. In this case the separate integrator and divider after it may be more convenient.