Negative Supply Rejection Sucks

first hit on google is

I don't know the voltage ranges you need but Analog Devices usually makes quality stuff:

Thanks.

It costs a buck more than a 337M, but has a noise reduction pin that seems to give an extra 20 dB or so. That ‘s enough to be quite useful without being a complete solution.

I can solve the problem with a PNP cap multiplier supervised by an LM4041 shunt regulator, but that takes a fair number of parts and gives up a lot of comfort features.

Cheers

Phil Hobbs

LT3094 (LT3045).

TPS723xx claims ultra low noise and high PSRR.

I like to put an RC at the input of a lin reg. That improves hf psrr, shares some power dissipation, and gives me a way to measure the current.

I like to bypass the top resistor in a feedback divider too.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

About 20 dB at 10 kHz. :( They're almost all like that.

I've got a four-pole RLC LPF on each input that's -3 dB at 12 kHz and (theoretically) -100 dB at 300 kHz, before veering off into spherical meadows of -150 to -220 dB from 1 to 50 MHz. ;) Simulation shows no peaking and no overshoot.

The caps are all CL31B475KAHNNNE (1206, X7R, 5 cents) and the inductor is ANR6045T470M (6mm x 6mm, 0.25 ohm ESR, also 5 cents). I'm splurging on the polyswitch (RXEF030, about 12 cents). That's seven parts on each polarity, amortized over all the rails, not counting the RF bypass cap and TVS (SMBJ15A, about 8 cents).

The regulators typically have 50 dB-ish rejection near DC, some considerably less, but then go into the tank above 1 kHz or so, like

So it's looking like a cap multiplier supervised by an LM4041 shunt reference as the next stage. That's about a dozen parts per rail, which isn't brilliant. It's hard to share them, because two of the rails need to be near the negative input supply.

Humph.

Cheers

Phil Hobbs

Me neither, for much the same reason. ;)

Cheers

Phil Hobbs

That's a very nice part, which would be perfect if it didn't cost $6 per rail. :(

Thanks

Phil Hobbs

I guess that the substrate is the unreg input on a negative regulator, so it couples into everything.

Maybe get some 150 uF polymer caps?

Maybe make a regulator with an opamp and a mosfet or bipolar follower?

I'm lucky, I guess, that I don't have to sweat over pennies. Our selling price over parts cost ratio is pretty high. I'd hate to be making stuff where a gang of competitors are battling over price.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

I have used them, and I like them. If you want better PSSR or better noise, you'll have some work ahead. I've measured the nV/rtHz. Exactly as they say.

cheers, Gerhard

I think that's right. N-type Si substrates certainly exist, but IIRC they aren't common except for some photodiodes. One might wish that the old saying "reverse polarity and use PNPs" were more applicable to low-cost bipolar processes. ;)

Re: Regulator

After doing something fancier, and then Muntzing it down, I managed to make a reasonable regulator (60ish dB near DC, 70ish from a few kilohertz up) with only 10 cheap parts. No current limit, but oh well--the quiet supply doesn't have to run any external outputs. It piggybacks a bit from a -5V rail generated by an LM337L, which saves a zener diode for biasing the cascode transistor on the LM4041-Adj.

I had to fudge the modeal a bit, because none of the LM4041-Adj models I can find actually work in LTspice. (Dunno about pspice.) Even the TLV431 models don't seem to work in AC mode--they don't sink any current for some reason.

The LM4041 and LM385 adjustable refs look like a PNP transistor--feedback makes the reference voltage appear between ADJ and cathode, so if you want 1.2V, you connect ADJ to anode. That's super useful for negative supplies, because you can ground the anode and connect the feedback divider between output and ground.

The more common TLV431 and its ilk are NPN-style, where the reference voltage appears between ADJ and anode, and connecting ADJ to cathode makes a 1.2V reference.

To make the NPN-style work like a PNP, there's a voltage-controlled voltage (e) source to effectively move the reference terminals. Hopefully the AC characteristics of the two chips are similar enough--the mid- and high-frequency behavior is mostly determined by the BJT and passives anyway.

Here's the current iteration, including a screen shot. Unfortunately the LM4041 model doesn't seem to work in .ac mode, so it's several .tran simulations instead.

Cheers

Phil Hobbs

I just did some positive PSR measurements on a TCA0372. Conditions were a single +10V supply with 200 mV p-p sinusoidal ripple, courtesy of my trusty Siglent arb. I had to use one of our LA-22 100x precision low noise lab amps to be able to see the feedthrough on a scope. (I could also have broken out the lock-in, of course.)

Amp is wired as a follower, with a 3.3V lithium battery providing its input. PSR is calculated as

PSR = 20*log(p-p ripple/200 mV)

Verdict: Awesome.

Freq Ripple PSR

This is a lot better than an LM317.

My 25-cent RLC filter is 100 dB down at 300 kHz, so I should be laughing as far as the positive rails are concerned.

Cheers

Phil Hobbs

Are you planning to use the TCA power opamp as the voltage regulator? That makes sense, since I think you want two regs anyhow.

We did some testing on that amp. The current limiting mostly doesn't work so you can blow it up. It can be made c-load stable.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

This guy is going to be loafing, so a 1-ohm series resistor and a cap from the amp output to SJ should work well, I think. Barefoot, it doesn't like capacitance one bit--the datasheet graph shows instability starting at low-nanofarad loads even at a closed-loop gain of 100. (The thing's A_VOL is only 3000 or so.)

I wish I only needed two rails. On account of the limitations of available cheapish parts (which have virtues good enough to make this worthwhile), I actually need ten, counting the reference.

VCC/VEE: +-13.5ish, to run the output amps. These don't need much PSR, because of the RLC filters and the low-frequency PSR of the LM6171A output amps. AP2205 for VCC, probably LM337M for VEE, with attention to its dropout voltage.

VQ+ / VQ-: +-12ish, super quiet for the discrete front end (These have to be strictly between VEE and VCC.) Both come from the TCA0372.

+6.6V for the photodiode current monitors (ZXCT213), running on VQ and

+3.3 reference for the ADS7054 ADC. I'm using a REF1933 for this, which produces a half-scale reference for the full-diff amp as well.

+3.3 digital for MCU and so forth. AP2205.

-2.5 analog for a couple of op amps with 6V abs max. (I'll get the +2.5 from +3.3 with a diode in series.) TCA0372.

Fun stuff.

Cheers

Phil Hobbs

More capacitance can leapfrog the island of instability. A 56uF polymer is good.

Incidentally, the TCA works fine at +5 volts supply.

That's about 12 power supplies!

My new gadget, the PPG (Precision Pulse Generator) only needs about 7.

Interestingly, the most critical one, for noise and drift, is the 1.2 volt FPGA core supply. 100uV changes the prop delay more than I like.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Thanks, that’s helpful. I reproduced parts of it earlier today. A 15 uF ceramic made it oscillate when connected with very short leads, but not with 3/4 inch of #24 stranded in between, which is only a couple of milliohms. Any old alpo seems to be fine, whereas it sings with an ohm in series with 220 nF.

The supply current is diagnostic—it goes from a few mA up to some hundreds when oscillation starts.

Well, if you count putting a diode in series as making a new supply. ;)

Mixed signal life is fun that way. ;)

Cheers

Phil Hobbs

john larkin snipped-for-privacy@glen--canyon.com wrote: ...

Consider some stabilization loop. Some FPGA TDC design use similar approaches.

I'm designing basically an economy DDG, digital delay generator. So I'm not investing engineering hours or parts cost on extreme performance.

So I'll just make the 1.2v supply as good as we reasonably can and sell whatever it does.

Prop delay in the FPGA is about inverse on core power supply voltage. If the supply is noisy, I don't think we could compensate for that in real time.

We will know the PCB temperature so maybe we'll compensate for that a little. That would be a quick test and a bit of code.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Using a backwards PLL to regulate the supply would be pretty amusing, though, you have to admit.

Cheers

Phil Hobbs