Is PWM more efficient than linear operation?

[Aeternam]

I'd like to build a microprocessor controlled dew heater for a telescope. (For the non-astronomers among us, if ambient temperature falls below a certain temperature - the dew point - dew starts to form and the optics cloud up. This is why you put dew heaters on your scope. These things are really just resistive heating elements http://www.teleskop-express.de/shop/product_info.php/language/en/info/p7906_AST-Optics-DH150---Heizmanschette-L-150-cm.html) The idea is to measure ambient temperature and relative humidity and then calculate the dew point. When the ambient temperature falls below this point, the heater should kick in in order to keep the glass above the dew point and keep it there. I plan on using a MOSFET to control the heater (heater specs: 30W, 12V at 2.5 amps). I was wondering why most commercially available dew controllers use PWM to control output power instead of using linear operation.

Long story short: is PWMing the gate of a MOSFET more efficient than using linear operation?

If so, why?

[tszaboo]

You dont really need PWM, on-off control is even easier to do. It is just heating.
Yes it is more efficient, because you barely have any dissipation happening in the FET.

[madires]

PWM means that you switch the load on for some specific time (duty cycle) in a given time window (PWM frequeny). Therefore the MOSFET is used as switch and has only to dissipate the loss caused by RDS_on (usually a few milli-Ohms) and the current, when switched on. WIth linear regulation the MOSFET acts as a variable resistor and would have to dissipate all the power which doesn't go into the load.

[Ian.M]

Most MOSFETs are designed for switching, not linear operation, and can only handle a small fraction of their rated power if kept in their linear region.  Also if you reduce the heater voltage by 50% you'd be dissipating as much power in the MOSFET as in the heating element, but the heat from the MOSFET is not likely to be in a useful place.

Especially if you are using a MCU with a PWM module, it is therefore far easier to simply PWM it, via a gate driver chip if you need fast control to get a more stable temperature.   Simple, slow bang-bang (on/off) control can be done with a MCU and a logic compatible (low gate threshold) MOSFET without a gate driver, but this will result in some temperature cycling which you may not want.

[rstofer]

Technically, bang-bang control is PWM with a long time base.

For all the reasons given above, you will want PWM at a frequency of 10 kHz or less.  Consider a trapezoid which is what the MOSFET current will look like.  The transition regions at turn-on and turn-off have the MOSFET in a linear region and it is dissipating heat.  You want to move through these transition regions as quickly as possible (hint: use a MOSFET driver) to steepen the slope.  You also want the transition regions short compared to the total on time - lower frequency.

Since PWM is where uC's excel, there should be no problem creating it.  Now all you need to do is create a PID loop to take the measurements and control the percentage of on time.  These projects are all over the Internet:

http://programmers.stackexchange.com/questions/186124/programming-pid-loops-in-c

This covers the topic in broad terms.

[jeroen79]

You dont really need PWM, on-off control is even easier to do. It is just heating.
Yes it is more efficient, because you barely have any dissipation happening in the FET.

Does that matter for heating?
A hot transistor should provide heat just as well as a hot resistor.

[tszaboo]

You dont really need PWM, on-off control is even easier to do. It is just heating.
Yes it is more efficient, because you barely have any dissipation happening in the FET.

Does that matter for heating?
A hot transistor should provide heat just as well as a hot resistor.

The transistor is located in the controller, which does not need to be heated. The resistors are in contact with the optics which needs to be heated.

[CatalinaWOW]

The heater tape or resistor array will be designed to heat the optics uniformly to avoid thermal distortion.  Any heat from the transistor/MOSFET should be isolated from the telescope and will be wasted. 

I do not see a good reason for PWM at 10 kHz or above.  This application doesn't require precision control, and the thermal time constant will be measured in seconds, not milliseconds.  While the difference is small, the efficiency will be better at lower frequencies.  Most power is wasted in the transition time between on and off, and the higher your PWM frequency the more time you spend in transition. (More transitions per second).    There is nothing horribly wrong about the higher frequencies, just no requirement to be there.  I  would pick a frequency above roughly one Hertz that is convenient for you.

[Cerebus]

There is nothing horribly wrong about the higher frequencies, just no requirement to be there.  I  would pick a frequency above roughly one Hertz that is convenient for you.

I'd be inclined to pick something in the 10s of Hertz region just to be sure that the time constant of the heater was acting as an effective low pass filter. The heater is likely to have quite a short time constant and waves of heating passing through the glass might have a similar distortion effect to waves of heat passing across the silicon of op amps. Without precise data on the heater I'd be inclined to play safe.

[CatalinaWOW]

I agree, that would be safer, with no meaningful downside.  While I suspect the thermal time constant will be quite a few seconds, none of us can be sure without knowing the construction of both the telescope and the heater.

[Aeternam]

Lots of good info here guys, thank you.  I'll report back with a schematic as soon as I have something tangible.