If I put my notes here, I might be able to find them again later!
Voltage regulators and ... something almost as good?
A real, honest-to-goodness voltage regulator has three parts: a fixed voltage reference, an error amplifier, and a pass element.
Most people only put eyeballs on the final, all-wrapped-up-in-a-tidy-IC-package version, typified by the LM7812, or with a couple of extra gain-set resistors, the LM317. These chips have a working bandwidth about about 2 kHz, as they are designed to 1) reduce 120 Hz ripple and 2) be rock stable no matter what abuse they are subjected to. As a result at audio frequencies and above they are pretty much noise generators...
Knowing this, many people have set out to build better regulators for audio work.
The most obvious route is to build a high performance LM7812 from discrete components. (Most excellent review here.) AD797 for the error amplifier, high stability, low noise voltage reference, etc. The trick though, is stability. The LM7812 is low bandwidth not because it's too cheap to manage anything better, but because that makes it stable. To get everything working, most of the "high-end" audio regulators have a large capacitor in parallel with the feedback loop to reduce the bandwidth, i.e. the frequencies at which the error correction is active.
As shown in the figures below (for the positive half of the X-reg, though the concept is non-specific), you can remove the pass element from the op amp feedback loop completely, halting the error correction for all frequencies. This by definition makes the system stable: if there is no correction, then there can be no over-correction that leads to instability. It also by definition means the circuit is not longer a voltage regulator. The op amp is just multiplying the reference voltage and applying it to the control node of the pass element. Isn't that pointless? Well, yes and no. There are still two benefits:
1. A small voltage reference can be amplified to the correct value. It may be a lower noise solution to multiply a quiet, small reference (as generated by heavy RC filtering for example) than use a noisy, high voltage reference. It might also be more convenient.
2. The op amp provides low impedance drive for the base of the pass transistor.
(Note that error correction is not required to reduce the noise/ripple, since the transconductance of the pass element will work to do just that regardless. Error correction does, however, increase the performance of the regulator in that regard.)
The whole concept is not unlike the basic Z-reg (Zener reference attached to base of pass transistor) but it solves the "the Zener is noisy so I want to heavily filter it, but that stifles it's output impedance and hence its ability to drive the pass transistor" problem.
This brings me to the catch of the day, Teddy Pardo's TeddyReg. (cute name! Schematic below.)]
It is not a regulator, as there is no error correction, no op amp, no error amplifier. It addresses the problem of driving the transistor base by buffering it with a JFET. The FET does the traditional FET job of providing a very high input impedance to voltage reference. With this configuration, the reference can be heavily filtered and you don't have to worry about current drain or driving the base of a big pass transistor directly. Downside is the time to start up ends up rather long, hence the need for the "accelerator" noted below. It's is conceptually mid way between the Z-reg and X-reg circuits, which is why it interests me.
In passing, I have a few comments re. the TeddyReg:
1. The voltage reference is the filtered output of an LM317, which is also used as a noise generator ... I mean, uh, pre-regulator. Killing two birds with one stone it may be, but.. yuck. The whole point of building your own regulator is that you never have to touch these ICs every again ... yet so many people invite them right back!
2. 1. above wouldn't be the end of the world I guess, but there is another issue: The JFET bridges the pass transistor's base and it's collector, which is the noisy input voltage. Any input noise is applied to the base, reduced only by the transconductance of the JFET. Any noise that reaches the base is dumped 1:1 into the regulator output.
3. The "accelerator" function is cute, but the RC filtered reference naturally gives you a nice "soft start" function of a few seconds or so. It seems a shame to actively go out of your way to kill that. The problem is if you try to heavily filter the reference and not drop the output voltage, the time constant becomes very large. In the X-reg, I got around that by dropping the reference down to 300 mV or so, then using the op amp to amplify it back up to 10 V. Then, the time constants end up about right for practical soft start functionality.
UPDATE:
Not trusting my instincts 100%, I ran a quick sim under ltspice to confirm point 2. above. Depending on the output current, I saw the ripple rejection drop from 1/5000-1/600 to a constant 1/300 when the FET was used ot buffer the transistor base. So best case it only costs you 6 dB, worst case you lose over 20 dB worth of "regulation" by interposing the FET.
Most people only put eyeballs on the final, all-wrapped-up-in-a-tidy-IC-package version, typified by the LM7812, or with a couple of extra gain-set resistors, the LM317. These chips have a working bandwidth about about 2 kHz, as they are designed to 1) reduce 120 Hz ripple and 2) be rock stable no matter what abuse they are subjected to. As a result at audio frequencies and above they are pretty much noise generators...
Knowing this, many people have set out to build better regulators for audio work.
The most obvious route is to build a high performance LM7812 from discrete components. (Most excellent review here.) AD797 for the error amplifier, high stability, low noise voltage reference, etc. The trick though, is stability. The LM7812 is low bandwidth not because it's too cheap to manage anything better, but because that makes it stable. To get everything working, most of the "high-end" audio regulators have a large capacitor in parallel with the feedback loop to reduce the bandwidth, i.e. the frequencies at which the error correction is active.
As shown in the figures below (for the positive half of the X-reg, though the concept is non-specific), you can remove the pass element from the op amp feedback loop completely, halting the error correction for all frequencies. This by definition makes the system stable: if there is no correction, then there can be no over-correction that leads to instability. It also by definition means the circuit is not longer a voltage regulator. The op amp is just multiplying the reference voltage and applying it to the control node of the pass element. Isn't that pointless? Well, yes and no. There are still two benefits:
1. A small voltage reference can be amplified to the correct value. It may be a lower noise solution to multiply a quiet, small reference (as generated by heavy RC filtering for example) than use a noisy, high voltage reference. It might also be more convenient.
2. The op amp provides low impedance drive for the base of the pass transistor.
(Note that error correction is not required to reduce the noise/ripple, since the transconductance of the pass element will work to do just that regardless. Error correction does, however, increase the performance of the regulator in that regard.)
The whole concept is not unlike the basic Z-reg (Zener reference attached to base of pass transistor) but it solves the "the Zener is noisy so I want to heavily filter it, but that stifles it's output impedance and hence its ability to drive the pass transistor" problem.
This brings me to the catch of the day, Teddy Pardo's TeddyReg. (cute name! Schematic below.)]
It is not a regulator, as there is no error correction, no op amp, no error amplifier. It addresses the problem of driving the transistor base by buffering it with a JFET. The FET does the traditional FET job of providing a very high input impedance to voltage reference. With this configuration, the reference can be heavily filtered and you don't have to worry about current drain or driving the base of a big pass transistor directly. Downside is the time to start up ends up rather long, hence the need for the "accelerator" noted below. It's is conceptually mid way between the Z-reg and X-reg circuits, which is why it interests me.
In passing, I have a few comments re. the TeddyReg:
1. The voltage reference is the filtered output of an LM317, which is also used as a noise generator ... I mean, uh, pre-regulator. Killing two birds with one stone it may be, but.. yuck. The whole point of building your own regulator is that you never have to touch these ICs every again ... yet so many people invite them right back!
2. 1. above wouldn't be the end of the world I guess, but there is another issue: The JFET bridges the pass transistor's base and it's collector, which is the noisy input voltage. Any input noise is applied to the base, reduced only by the transconductance of the JFET. Any noise that reaches the base is dumped 1:1 into the regulator output.
3. The "accelerator" function is cute, but the RC filtered reference naturally gives you a nice "soft start" function of a few seconds or so. It seems a shame to actively go out of your way to kill that. The problem is if you try to heavily filter the reference and not drop the output voltage, the time constant becomes very large. In the X-reg, I got around that by dropping the reference down to 300 mV or so, then using the op amp to amplify it back up to 10 V. Then, the time constants end up about right for practical soft start functionality.
UPDATE:
Not trusting my instincts 100%, I ran a quick sim under ltspice to confirm point 2. above. Depending on the output current, I saw the ripple rejection drop from 1/5000-1/600 to a constant 1/300 when the FET was used ot buffer the transistor base. So best case it only costs you 6 dB, worst case you lose over 20 dB worth of "regulation" by interposing the FET.
Total Comments 0
