DC Power Voltage Regulators

Automotive Power@ Batterys@ Switching Regulators@

Ultra Basic DC Regulated Power Supply using LM7805 or others

I would suggest the MAX667. About a year and a half ago we did quite a bit of research for a low-power project we were working on (several weeks off of four AAs), and these are great low dropout regs - 20uA of current draw just sitting there, and they have a ultra-low power shutdown mode, though by your application I don't think this would probably be of use to you. In addition, the voltage drop over them is much lower than a 7805 or other conventional linreg, so you'll be able to discharge the 9V to a deeper state. In addition, it has a built-in low battery comparator, so you could conceivably signal the PIC to stop taking readings after the power became unreliable. At least this was an issue I had to deal with...
One thing I might suggest, though, is changing from a 9V to either four AAAs or AAs. The reason for this is because you are using a linear regulator, any extra voltage is just being burned as heat. Yet a 9V battery sacrifices mAh of storage to produce this potential.
With a 9V system and a linreg, you burn 4/9 of the battery's capacity as heat. According to what I've read, the usable ampacity of a 9V is about 350mAh. So usably, you're extracting 5V*350mAh=1.75Wh of energy. That lasts you a week, meaning you us about 0.25 Wh/day. And that's with a standard 7805, which suck down unbelievable amounts of power when a load is placed on them...
A 6V system of AAAs would give you 600mAh of capacity, so 5*600m=3Wh. Assuming a 7805 would work with four AAAs (which isn't true, but a low dropout reg like a 667 will), you'd still get 71% more run time - almost another week - in close to the same space and weight. Plus AAAs are cheaper...
With AAs, the gains would be even greater, but then the size and weight are getting much larger. For reference, an AA I believe is 1500mAh.

William Kitchen suggested:

It should be possible to have a PIC16F87x processor regulate one or more supply voltages in software, using it's A/D converters for feedback, and using either pwm or a simple on/off from an I/O pin to control the power supply. How well this would work depends on how much ripple you can tolerate vs. how much processor overhead you can tolerate. I think it could work fairly well even with simple on/off control and long polling intervals if these processor regulated outputs were used as references for linear regulators, rather than bearing the load directly. Now that I think about this a little more, this arrangement is really just a D/A. Any kind of D/A could be used to provide reference for a linear regulator, so there are several ways to do this. Having several fully programmable voltages would allow for target modules for just about anything. It would also allow the programmer to meet Microchip's requirements for a "production" programmer.
While on the topic of power supply, having the adjustable voltages readable with the A/D's (via resistor dividers, of course) would also be useful even with a simple manually adjustable power supply because they could help you set the voltages without having to use a voltmeter. Enther the front end software could display the voltages, or the programmer itself could tell you when the voltages are right by lighting LED's. This might even eliminate the need for the front end software to be aware of whether a programmable supply is available. The software would attempt to set the voltages, then the programmer will either go agead and program the part if what it sees on the A/D's is already right, or set the voltages if it has a programmable supply, or wait for the user to adjust the voltages.
If the op-amp has a high enough input impedance, a capacitor on the input, charged via a resistor by a PIC I/O pin while checking the value with an A/D, would be able to serve as an analog sample and hold that would only have to be refreshed at long intervals, thus requiring very little software overhead. Might even be able to get away with charging the cap only once at the start of programming a chip, thus having no overhead at all while actually programming. It may be doable with only a single PIC pin by switching it between digital output, analog input, and high impedance input. The output of the op-amp would then serve as reference for a linear regulator. Since there are more A/D pins than PWM pins, you could also control more voltages this way.

Byron A. Jeff aggreed:

It certainly wouldn't be a problem using PWM with an op-amp amplifier driving the adjust pin of a voltage regulator.
That's brilliant! One question: Do we measure the voltage on the cap and trust the op-amp multiplier, or do we measure the actual output of the regulator? In the latter case we'd then need another resistor voltage divider to scale the output voltage for the A/D.
And just to make sure I got it right:

A PIC pin charges a cap that serves as the input to a op-amp.
The op-amp scales the input voltage to a higher output voltage which is fed into the adjust pin of a LM317 type regulator.
The output of the regulator will correspond to op-amp output.

So with 2 resistors, a cap, a op-amp, and a regulator we get precise regulated, settable voltages.

William Kitchen responded:

The latter has the advantage of better accuracy, the former has the advantage of using fewer pins on the PIC. Tough choice. A compromise would be to have a trimpot to calibrate the gain of the op amp. Temperature changes might be a problem, but maybe not too much if the components are chosen with this in mind. Temperature sensitivity of the R and C on the input side of the op amp won't matter since that's regulated by the PIC.
Another idea that comes to mind as I type this is to make all the op-amps unity gain and use this as an "offset" to the regulators. A trimpot can then be used to calibrate the linear regulator relative to the offset voltage that is provided by the op-amp. Each of the regulators can be set such that each provides a different 5V range. The advantages to this are that the op-amp wouldn't have to have such a high supply voltage, and the resolution would be better (though I don't think that will really matter much with the 16F87x's 10 bit A/D's). Don't know if that's really a better way to go, but something to think about. An obvious disadvantage is that you couldn't change a pin from 5V to 13V (or any other change bigger than 5V) without switching the pin to a different supply output.

Brusque aggrees:

The sugestion of using A/D + PWM to control the voltages is very high-end. I love the idea.
The 'F877 can be configured for two PWM modules without wasting processing resources. I sugest one fixed +5V and two PWM regulated voltages from (say) +2 to +25V. We could configure the 'F877 for 3 A/D, so we can monitor and set the two PWM regulated voltages and also monitor the +5V.
If people think 3 PWM regulated voltages is necessary, It can also be done, but we will need a timer interrupt to do that. I've done 12 syncronized PWM on a 'F877 using the timer0 interrupt.
We have to measure the actual output of the regulator, so we can be sure (with a good precision) that the voltage is where it needs to be. If we measure the voltage on the cap, we would need to make some calculations to know how is the output of the regulator. We will also have a compromisse with the precision of the comercial components value. Two 78L05 give slight different voltages output. Two diferent discrete circuits could have two very different voltages output.
The op-amp scales the input voltage (say, 25VDC) to the lower required voltage (say, 12.5VDC). I feel more secure if it works this way, so we can have a bigger voltage swing.
I was thinkin in using a switching voltage regulator, but I'm now looking at the Burr-Brown's "Designing with Operational Amplifiers" book by Jerald G. Graeme, and the simplest circuit uses 1 op-amp, 1 transistor, 1 zener, 2 diodes, 1 cap, 5 resistors and an inductor. It's too much complex for our task, so I the voltage multiplier seens to be a more interesting aproach.