Is building a discrete voltage regulator worth it?

[Tjj226_Angel]

Hello all,

I am weighing the pros and cons of building a discrete voltage regulator. Normally I just use an off the shelf IC, but I need one that can handle 2 amps with around 100mv drop out and have an output noise level of 10uv or below.

There are some off the shelf ICs that get close, but since we have a lot of low noise opamps and jfets for making the feedback network, I am curious to know if it is plausible to being in the sub 10uv noise region with a discrete design.

[magic]

For 2A it may be sort of necessary if you don't want to parallel multiple low noise LDOs and pay for them.

Also, 10μV what and over what sort of bandwidth? RMS, 20-20kHz?

[temperance]

It would help if you state the input/output voltage requirements.

[SiliconWizard]

Yes, the LT3045 for instance, but it's only 500 mA max, so you'd have to parallel 4 of them. That would be a hefty ~ 4x10 (per 1) = $40. Also, the dropout voltage at 2 A would exceed 100 mV for sure (260 mV typ.)

More specs would be needed for sure, but even with the little the OP gave, that's going to be a pretty tricky one to design.

[Tjj226_Angel]

Sorry guys, I do this every time.

Input voltage would be from a lithium ion battery. So 4.2V when fully charged, 3.7v for a majority of the operation, and 3.5v would be the cutoff.

The output would be 3.4v ideally.

Bandwidth would be 10hz to 50Khz.

The idea would be to cut back on the noise made by the lithium ion battery when draining 2A and stabilize the voltage. Then feed into a constant current source to bring the noise down below 1uv peak to peak (ideally).

[AnalogTodd]

Sorry guys, I do this every time.

Input voltage would be from a lithium ion battery. So 4.2V when fully charged, 3.7v for a majority of the operation, and 3.5v would be the cutoff.

The output would be 3.4v ideally.

Bandwidth would be 10hz to 50Khz.

The idea would be to cut back on the noise made by the lithium ion battery when draining 2A and stabilize the voltage. Then feed into a constant current source to bring the noise down below 1uv peak to peak (ideally).

Yes, the LT3045 for instance, but it's only 500 mA max, so you'd have to parallel 4 of them. That would be a hefty ~ 4x10 (per 1) = $40. Also, the dropout voltage at 2 A would exceed 100 mV for sure (260 mV typ.)

More specs would be needed for sure, but even with the little the OP gave, that's going to be a pretty tricky one to design.

Why not use the LT3042 with an external transistor? Lower cost than multiple LDOs, same noise performance, and it can get the voltages and dropout desired. Page 27 of the LT3042 data sheet if you're interested.

[magic]

Modifying this circuit to achieve 100mV dropout would bring similar complexity to a DIY regulator.

[David Hess]

There are some off the shelf ICs that get close, but since we have a lot of low noise opamps and jfets for making the feedback network, I am curious to know if it is plausible to being in the sub 10uv noise region with a discrete design.

Plausible?  Sure, a discrete design can achieve that performance with low noise.  Noise will be primarily determined by the reference.  Even the best reference will have higher noise than a general purpose operational amplifier.

The design will be a little bit tricky because a PNP or p-channel MOSFET pass element will be required in common emitter or common source configuration which will make frequency compensation more difficult.  All low dropout regulators have to deal with this.

The example shown below is pretty close to what you need.  The current limit part of the circuit could be removed for lower dropout.

[PCB.Wiz]

The diodesinc AP7179D would meet those specs.
It is 3A 4.4uV LDO

[Tjj226_Angel]

The idea would be to cut back on the noise made by the lithium ion battery when draining 2A and stabilize the voltage. Then feed into a constant current source to bring the noise down below 1uv peak to peak (ideally).

Why do you think you can build a regulator that would have lower noise than a battery?

At 2 milliamps of current draw, it would be a fools errand to beat an 18650 cell. You could do it, but it's going to be a pain.

At 2 amps of current, the 18650 cell can have a bit of noise (20uv of noise rms is what I have seen). Even a very well implemented LT1085 would have lower noise.

[magic]

Hmm, does it need to be stable or would it be OK if it's just "100mV less than Vbat, and without noise"?

For the latter, maybe one could come up with some completely discrete contraption similar to cap multiplier, but LDO.

Another thought is that maybe you could actually use LT3042 with an external transistor, but not like it's shown in the datasheet (of course). Say, load the output to ground and treat it as a reference + opamp to control a P-ch MOSFET. Compensation left as an exercise for the reader, but you are going to need something anyway for any kind of DIY regulator. Sorry, not gonna work, wrong feedback polarity.

[Tjj226_Angel]

Guys, I think I am having a stroke.

I am on Texas instruments website and I click on this LDO here: https://www.ti.com/product/TPS7A52

If you look at the data sheet it clearly says that the dropout voltage is 65mv.

However, if you scroll down to select a part for purchase, click on one, and then open up it's datasheet, it is a completely different set of specs.

What am I missing? Did TI goof or am I being blind? If I can get the device in the main datasheet, I would call it a day.

[Tjj226_Angel]

There are some off the shelf ICs that get close, but since we have a lot of low noise opamps and jfets for making the feedback network, I am curious to know if it is plausible to being in the sub 10uv noise region with a discrete design.

Plausible?  Sure, a discrete design can achieve that performance with low noise.  Noise will be primarily determined by the reference.  Even the best reference will have higher noise than a general purpose operational amplifier.

The design will be a little bit tricky because a PNP or p-channel MOSFET pass element will be required in common emitter or common source configuration which will make frequency compensation more difficult.  All low dropout regulators have to deal with this.

The example shown below is pretty close to what you need.  The current limit part of the circuit could be removed for lower dropout.

Yeah, I am quickly learning that a voltage reference is either going to have to be made or I have to spend $$$$$.

I am looking into using something similar to a walt jung LED reference. I would love to use a Jfet as a constant current source for the LED, but options are scarce.

I am slowly working my way through your posted circuit. There are some cute ideas in there. I like it.

[PCB.Wiz]

If you look at the data sheet it clearly says that the dropout voltage is 65mv.

Those parts have a BIAS option that interacts with VIN, the curves are the best indications of dropout.

[SiliconWizard]

Guys, I think I am having a stroke.

I am on Texas instruments website and I click on this LDO here: https://www.ti.com/product/TPS7A52

If you look at the data sheet it clearly says that the dropout voltage is 65mv.

However, if you scroll down to select a part for purchase, click on one, and then open up it's datasheet, it is a completely different set of specs.

What am I missing? Did TI goof or am I being blind? If I can get the device in the main datasheet, I would call it a day.

No, they just stated the most favorable case when using an external Vbias and for Vin = 1.1 V. You can't blame them for stating the most favorable value in the product brief.

That said, with 100 mV typ. for Vin = 5.3 V @2 A, that's still very good. (But yes, 170 mV max. Still impressive specs.)
Doing better with a discrete design is going to take some skills.
Or maybe having a bit more leeway in terms of Vin would help as well.

[temperance]

If quiescent current is not an issue I would double the battery voltage with a charge pump with an RC filters on the input to make the current seen by the battery continuous. The output would need an other RC low pass filter.

From this doubled battery voltage you can power an op amp to drive a logic level or low Vgs-th N-ch MOSFET. The Ref. can be filtered with some RC network with "low noise" capacitors as much as required to cut down the noise. Compensation of such a circuit is straight forward and it can be build dead bug style in a less than an hour.

[magic]

The classic common source LDO may be possible to compensate with nothing more than a right output capacitor. There was a thread about it which shows what the open loop gain looks like.

Ignore all the talk about adaptive bias, MB, IB and MBA. The important part is this Bode plot. With most IC opamps, p2 will be below p1 because their frequency response starts to decrease at a few Hz. Then you have p1, when reactance of Cout becomes lower than all resistive loads on the output. You must ensure that loop gain doesn't fall down to unity at this point. Later still, at z1 reactance becomes lower than ESR and the capacitor turns into an AC coupled resistor and loop gain is only determined by opamp response again. It can safely fall down to unity here. Opamp's GBW must of course be faster than unity gain frequency of the whole loop, otherwise you hit p3 which is the second pole of the opamp. Gain increases with load current (the dashed line) and p3 must be high enough for worst conditions (in this example, it's not). Shouldn't be a problem, we are talking <1MHz with large output electrolytics.

[David Hess]

Yeah, I am quickly learning that a voltage reference is either going to have to be made or I have to spend $$$$$.

I am looking into using something similar to a walt jung LED reference. I would love to use a Jfet as a constant current source for the LED, but options are scarce.

The supply voltage will not support a zener reference, but low noise bandgap references are available.

I am slowly working my way through your posted circuit. There are some cute ideas in there. I like it.

All designs are going to come down to a reference and an operational amplifier driving a p-channel power MOSFET unless you move regulation to the negative side, and then an n-channel power MOSFET can be used which will be less expensive and perform better.  I might add a transistor buffer to drive the capacitance of the power MOSFET gate better.

[Someone]

I am slowly working my way through your posted circuit. There are some cute ideas in there. I like it.

All designs are going to come down to a reference and an operational amplifier driving a p-channel power MOSFET unless you move regulation to the negative side, and then an n-channel power MOSFET can be used which will be less expensive and perform better.  I might add a transistor buffer to drive the capacitance of the power MOSFET gate better.

As hinted at above with bias supplies, if the positive side regulation is important (charging of battery from mains referenced source such as USB, while operating, for example) another choice that could still meet power/noise/efficiency concerns would be a small boost regulator or charge pump to provide enough headroom for a N-FET pass element. Which may well be back to an off the shelf part.

Negative side switching or regulation is great when one side is known to be isolated.

[Tjj226_Angel]

I am slowly working my way through your posted circuit. There are some cute ideas in there. I like it.

All designs are going to come down to a reference and an operational amplifier driving a p-channel power MOSFET unless you move regulation to the negative side, and then an n-channel power MOSFET can be used which will be less expensive and perform better.  I might add a transistor buffer to drive the capacitance of the power MOSFET gate better.

As hinted at above with bias supplies, if the positive side regulation is important (charging of battery from mains referenced source such as USB, while operating, for example) another choice that could still meet power/noise/efficiency concerns would be a small boost regulator or charge pump to provide enough headroom for a N-FET pass element. Which may well be back to an off the shelf part.

Negative side switching or regulation is great when one side is known to be isolated.

The idea is that the battery will be switched from one bank to another for total isolation.