Most LDOs containg a PNP as regulating transistor show an intrinsic instability. Only by the adding of decoupling cap at output the urgently needed phase shift can be provided, to make the regulator stable. Such a cap can only do this job, if the ESR is within a certain range. Like Oliver already mentioned, this ESR must not be too low, but not be too high either. Unfortunately, this otpimum range highly depends on actual output load and input voltage.

Normally, manufacturers recommend a certain output cap for achieving optimum stability. But what always is ignored is, that this cap is not the only cap in an application powered by this LDO regulator. There are also the many decoupling caps or low pass filters at the chips. To make the situatuion even more difficult, these additional decoupling caps are far away from the LDO regulator, so that inductances of connections come into play. All these capacitances, ESRs and inductances of this whole output voltage routing never looks like this recommended decoupling cap at output of regulator any more, but forms a complex network. Complex here means, that there are poles and resonances in the associated impedance versus frequency curve.

So, you can use this recommended output cap and nevertheless result in instability!

Have a look at this nice application note, which discusses this stability stuff of LDO regulators:

http://www.onsemi.com/pub/Collateral/SR003AN-D.PDF

To me, the only way to find an answer to your question is to measure the output voltage of LDO by the help of a scope: Is there any ringing, or any noise? Then, there's something wrong. If not, you will be lucky. Please read the given link very carefully, it gives you proper help and useful information on testing and checking.


I would use the following scheme for the LM2937 and LM2990: A tantal of 10µF directly at the output of LDO regulator. Also a 10µF tantal at input of the pi-filter, instead of 47µF (aluminium electrolytic)//100nF combination. At output of pi-filter you can use the given combination or a 10µF tantal, too. Here, the ESR doesn't play any role, because it's isolated by the 100R resistance of pi-filter.
With this scheme you get two 10µF tantals in parallel, which yields a total ESR of 0.1 Ohm. So, LM2937 and LM2990 should be stable.
If further decoupling caps are needed in front of pi-filter, at the LDO side, then use RC-filters containing a 1...2.2R resistor and any cap you want. Then, this 1...2.2R resistor isolates the ESR of cap.

The situation looks different with the LM2935! Here the minimum ESR of output capacitance is 0.3Ohm. But only one 10µF tantal alone can provide an ESR of only 0.2Ohm, so that instability will be the consequence. LM2935 is a good example for a LDO which is very very difficult to handle and should not be used today any longer, because there are tons of improved LDOs on the market today.

Kai