Tracking ultra low drop low noise regulator

[blueskull]

I'm in the process of designing a very high specced audio DAC, and I'm looking for a circuit to power the OPAMPs and headphone driver.

Unlike most audio designs where Vs=+-15V or +-18V since they don't have a stringent power requirement, my design is battery powered and I need long running life.

Therefore, I want the regulator to be:
1. Programmable, from +-4V to +-12V (no digital pots, I prefer an external reference voltage from DAC)
2. Of low noise, lower than 15uV RMS
3. High PSRR, shooting at 80dB at 20kHz
4. Very low dropout, looking at <400mV while keeping a high PSRR
5. Tracking, negative rail tracks positive rail
6. I don't have a quiescent current requirement, as long as it's not very high (say, 10mA+)

The input will be a bipolar boost converter, generating programmable Vout+0.4V. A 3.6V clean voltage is also available.

Here is my draft, I would like to hear suggestions and critiques from you guys.

Thanks in advance.

[T3sl4co1l]

Q4 is supposed to be NPN, right?

I don't get making noisy (bootstrap powered) CCSs for a "low noise" circuit.  Maybe it'll be okay.  Or it may need a lot of filtering there already.

Big no-no: the op-amp is followed by a very high gain stage (probably ~1000).  You've got a power oscillator!

Reducing EMI from the boost converter will go a long way towards making filtering easier.  100MHz+ squigglies are pernicious and don't like to stay in wires (and do like to upset bipolar op-amps).  Junk in the 10s of MHz is pretty easy to filter.  Junk in the <1MHz range isn't as easy to filter, and pass transistors with high VA (Early effect parameter) and low Ccb will be very helpful.

You can also use a capacitor multiplier topology, with the boost reg's loop referenced to the final output, so you don't need extra control loops.  (This wouldn't get you balanced supplies, though, as cross-regulation would be limited by stray inductance between the two secondaries.)

Or don't bother with tight voltage control at all, as it's just DC, something op-amps have magnificent PSRR for.

Tim

[T3sl4co1l]

Idunno... seems like way too much trouble to go to if you're only concerned about <100kHz.  The above circuit and discussion seems applicable to VHF noise problems (with the difference that, an op-amp won't do anything up there).

Would the 3.6V "low noise" supply happen to be the same as the charge pump supply?  Don't neglect induced supply ripple; the charge pump needs to be very smooth and/or very well filtered to keep from affecting that.

Cap multipliers are excellent, as far as I know.  This guy uses them religiously: http://electrooptical.net/ he's fond of using 2 or even 3 transistors cascaded, with a common RCRCRC filter chain.  (Note that you only incur one Vbe in this configuration, plus a modest Vce for each transistor.)  The base filtering, plus double or triple duty against Ccb and Early effect, kills noise hard (80dB+).

But again, that's in the MHz, where these sorts of steps are necessary.

Regarding the gain, you've got a CCS (common base) feeding a CCS, so the voltage gain of that node is tremendous (set by (load resistance) * (pass transistor hFE), which will be pretty high).  Agree, common base is a good level shifter.  In light of the gain problem, though, better would be an inverting configuration with shunt feedback, so it has a CV output instead of CC, and the voltage gain is fixed.  Then swap the op-amp inputs to get the right feedback polarity, and all is well.

A low output impedance also has more immunity against ripple in the CCS.  And note the CCS will have ripple for the same reason that the pass transistor allows noise through: Ccb and finite VA.  Also a small input from base supply voltage, which will be relatively small because of the zener.  Hmm, if that's a low voltage zener (under 5V), it won't be too much less ripple than the same circuit using two diodes in series as bias, really.  A red LED might be good to use, they're about as stable as a zener diode (or more stable, maybe?), but have lower voltage drop (~2V).

If truly your only problem is under 100kHz, would it not be easier to run the switcher(s) fairly fast (some MHz), and use a tight control loop there?

Tim

[David Hess]

BJTs are current controlled devices, and I'm essentially making a controllable current sink with a common base amplifier, to divert current from CCS, in order to control the BJT. It would be bad for FETs, but for BJTs, which are current controlled devices, I don't see a big problem.

That does not mean that you are not adding voltage gain within the control loop.  And unless you have a bias supply, low dropout will require a common emitter/source output transistor which is going to add voltage gain anyway.

Since I need level shifting anyway (to control 12V from 3.3V OPAMP), I have to choose one from common base and common emitter, and I think in this case, to get rid of the extra miller capacitance zero, I should better use a common base.

Bypassed zener diodes work well for level shifters while adding no gain.

Did you want the negative output to track the positive output or both outputs to track each other?  I have only seen this done with integrated regulators but there are some dual operational amplifier phase splitter designs which use feedback from both output together so they really track.  I have never tried this but have consider making a low dropout low noise tracking regulator this way for essentially the same type of application.

Check out the datasheet for the LT1033 and this EDN link.

The issue with voltage gain within the feedback loop from a common emitter output stage can be handled and low dropout regulators built this way can be low noise but it is much better to use LC filtering to remove switching noise than to try and do it actively.

It might be worth considering a switching regulator design optimized for low noise.  Linear Technology application note 70 discusses switching regulator noise in detail.