Problems of cascading linear regulators

[cloidnerux]

Hello all,

in a design I have a 12V input, which I needed to drop down to 5V and 3,3V. So I choose the NCV1117 and cascaded the 5V regulator and the 3,3V regulator to distribute the heat a bit on the PCB(5V reg in a 3DPAK, 3,3V in SOT-223). Input and output had 100nF + 10µF MLCC. Testing this setup showed a really bad performance, I could hear the 12V SMPS ringing, probing the voltage rails showed really bad spikes and oscillation.
This is of course partly due to a design fault of me, reading the datasheet there is a table linking output capacitance and ESR and some recomendations. So I added 2 22µF tantal capacitors on the output of both 1117 and this helped a bit, but there were still pretty bad spikes in the voltage.
So in a last attempt I connected the 3,3V reg directly to the 12V rail and this removed a lot of the noise significantly.

To be sure I don't have some issues with the dropout voltage, I checked the datasheet from OnSemi and the maximum dropout voltage is 1.2V, so this is within margins.
The 3,3V drives a LCD display, so a digital load.

My question: Is there a known problem of cascading linear regulators(of same type) in connection with non analog loads? My inital guess would be that the cascading would result in a lower noise, but apperently the cascaded control loops counteracted each other somehow.

[MosherIV]

My inital guess would be that the cascading would result in a lower noise

Linear voltage regulators do not remove switching noise. Their control loop simply does not have the bandwidth to cope with the high freq noise.

I am afraid that I do not know why the SMPS was generating audible noise in your case.

[Chalcogenide]

Have you added enough decoupling capacitance at the input of the 5V regulator? You could also try to add a resistor in series between the SMPS and the input of the linear regulator, sizing it to have enough margin on the dropout at the maximum output load. This could help reducing the high frequency noise by acting as a low pass filter.

[void_error]

Since the MLCCs have a really low ESR you could have an undamped LC circuit with the 12V input wires/traces but that depends on the current spikes your circuit draws and the trace/wire length. If that's the case then a low value resistor in series with the LDO input might fix it.

[cloidnerux]

Since the MLCCs have a really low ESR you could have an undamped LC circuit with the 12V input wires/traces but that depends on the current spikes your circuit draws and the trace/wire length. If that's the case then a low value resistor in series with the LDO input might fix it.

Well, that could be something. The traces are short, like 3cm, but there is a common-mode choke in the path.

Linear voltage regulators do not remove switching noise. Their control loop simply does not have the bandwidth to cope with the high freq noise.

Well, this is a somewhat simple statement. At 100kHz frequency(MC34063) the NCV1117 has about 40dB of ripple rejection, so there is some filtering. Of course at 900kHz where my SMPS operates the rejection is not that great.

Have you added enough decoupling capacitance at the input of the 5V regulator?

I am up to 50µF, but it seems not to be enough.

[ConKbot]

Frequency compensation for the regulator is provided by
capacitor Cout and its use is mandatory to ensure output
stability. A minimum capacitance value of 4.7 -
uF with an
equivalent series resistance (ESR) that is within the limits of
33 m (typ) to 2.2  is required. 

Is your 5v Cout combined with 3.3v Cin violating this requirement?

Tdk component characteristic viewer shows the resistance (not impedance) of one of their 25v 10uf 0805 capacitors as  dropping below 30 milliohms at 10kHz or so. Got any 0.1-2 ohm resistors you can put in series with the decoupling caps?

[David Hess]

The low ESR of those ceramic output capacitors is screwing up the frequency compensation of the regulators; low dropout regulators tend to be sensitive to output capacitor ESR.  Read the datasheet and when you fix it, run a load transient response test with an oscilloscope and pulse generator.

Frequency compensation for the regulator is provided by
capacitor Cout and its use is mandatory to ensure output
stability. A minimum capacitance value of 4.7uF with an
equivalent series resistance (ESR) that is within the limits of
0.25 to 2.2 ohms is required. The capacitor type can be
ceramic, tantalum, or aluminum electrolytic as long as it
meets the minimum capacitance value and ESR limits over
the circuit’s entire operating temperature range. Higher
values of output capacitance can be used to enhance loop
stability and transient response with the additional benefit of
reducing output noise.