Two regulators in series

For a sensor which is sensitive to the ripple in its supply voltage I want to create a low noise LDO regulator. If you read the specs from LDO’s they mention the Power supply ripple rejection ratio (PSRR) of say 50 dB. My raw supply voltage is 10V. and the sensor needs 3V. I was wondering what happens if I connect a 5V LDO which feeds a 3V LDO and both have a PSRR of 50dB. Can I just add the dB (50dB + 50dB = 100dB ?) Or is this to good to be true ?
The supplier of the sensor recommends in the application note a 12V supply connected to a resistor of 120R in series with a 3V zener. This is quite a waist of current as my application is battery powered. So that’s why I am tinkering with to low quiescent current LDOs in series in order to get the lowest PSRR possible.

There is some possibility of control loop interaction with cascaded regulators. You’d probably be wiser to use one of the class of low noise, wide bandwidth regulators like the ADM7151 or similar parts from TI or Microchip. to get much more than 60 dB reliably you are going to have to deal with bunch of other extremely subtle coupling effects in your board design like surface leakage, static and magnetic coupling. This is a lot harder than just cascading two regulators in a stable fashion. The vendor recommended option has it’s own issues, like the noise of that zener, and the resistor is going to have a noise floor in the ~180 dBm/hz range. This could be a factor depending on the sensor noise floor and bandwidth, but I’m betting not, as I suspect is your unquantified expectation for better PSRR.

2 Likes

I’ve been using such a topology power supply that’s been integrated into one chip, the Micrel (now Microchip) MIC38150. It’s a switching regulator followed by a linear one, paired up for applications that require very low-noise power supply rails. In my case it was for an HDMI circuit.

1 Like

“Ripple” is a vague term, usually implying a relatively low frequency. Do you have any more specific info on what kind of noise the sensor doesn’t like, how much noise it can tolerate on the supply, and what the frequency response of the sensor is?

PSRR stands for power supply rejection ratio, not power supply ripple rejection ratio.

Not sure that a Zener will provide very good “PSRR”, apart from Rich’s good comments on noise - you can probably calculate it based on the Zener’s dynamic resistance.

2 Likes

Thank you Rich for your valuable insides.

Thank you for this tip. I was not aware that a combi of switching and LDO exist. The noise specs of that part look very good. Unfortunately the quiescent current is to high for my application. I forgot to mention limited current consumption restriction which adds to my problems.

Thank you Julia. The sensor is a 5Ghz radar sensor. These work on the bead frequencies if something moves in front of them. So I assume this is quite a low but not stable frequency. The supplier is very vague what ripple you must watch out for. They only give this resistor zener combination which they recommend without giving a exact part for the zener. As my application restricts power consumption I can not waist current flowing through such zener. So I have to use some low quiescent current LDO with a good PSRR rating. That’s why I thought to put 2 LDO’s in series with the hope that the sum of those parts give a better PSRR than using a single one. Rich made a good point that I might introduce other problems. I will build up some prototypes were I test a couple of configurations to test which is best.

Where does the 10V supply come from?

Be sure to test them over the full temperature range. You never know where things will fall apart.

The 10V supply is directly drawn from a 220V linear non isolated supply. This is one reason I can not draw a lot of current. The sensor uses about 70uA.

Sounds like you could just add capacitance to the 10V rail to get the ripple to wherever it needs to be post-LDO.

Yes that’s sounds like a good idea. I did not receive the sensors jet but that is one of the options I definitely will try.

Or a capacitor-inductor-capacitor (CLC) filter.

I’m currently working on a project that processes an analog video camera signal – there’s ripple from the switching regulator which is working its way into the video enough to be objectionably visible. The main noise contribution, the switching spike, was tamped down by adding an inductor in series and then a capacitor to ground in between the regulator and the connector going to the camera.

That got rid of the sharp 200mV+ spikes that was very noticeable in the resulting image. I still have the smaller 40mV ripple which I can pick out if I look closely, but it’s good enough for the time being. When I first tried to tackle it with capacitance alone, it wasn’t good enough.

Does the sensor have to operate at 100% duty cycle? Could you synchronize charging a big capacitor with a switching supply and then turn it off during measurements?

If you use a capacitor multiplier, that will be less noise than any LDO you can find

Capacitance multiplier - Wikipedia

2 Likes

Just need to be very sure that it’s over-damped - wouldn’t want it to ring on line transients, as the OP’s supply comes directly off the mains.

1 Like

Thank you for your reply. Unfortunatuly the sensor needs permanent power. It has some controls and timers onboard with an IIC bus. I see your point that if you could sample such sensor you could decouple from the supply and measure disconnected from a potential noisy supply. To avoid spikes from any switching regulator we will use an LDO and steer away from any switching close to the sensor.

By 220V linear non-isolated do you mean you have 220V DC being linearly converted to 10VDC? Or do you mean 220VAC → transformer → linear → 10VDC?

I understand the pushback against using switching supplies. Done wrong they completely kill an application like this… but I managed to measure human heartbeats through brick walls with a 5-10GHz radio and that had switchers in it. No pressure to use them, but done right they can be fine.

Avoid over-doing anything. Capacitance is good, but too much capacitance can de-tune a control loop. Inductance is also good, but makes you more sensitive to your own transients.

Make sure your components are effective at the frequencies you need them to be. PSRR/CMRR should be a graph, not a number. Caps are parallel plates of lies and deceit, so check they’ll actually perform at the voltage and frequency. Sprinkle 100pF C0G caps in as appropriate.

2 Likes

Would love to hear that story…

http://witrack.csail.mit.edu/vitalradio/

I was the uncredited contract worker that did all the PCBs. To be fair, all I really did was take their RF napkin sketch and make it real. They did all the interesting novel work on it. Also, I only did one of their versions; they’ve updated a couple times since then.

2 Likes

I would second this - Find a high quality switcher IC running at around 2 MHz, and all you should need is a simple LC pre-filter before the linear regulator. You may not even need a linear regulator!