Big capacitors can cause instability especially since their ESR is lower. The LM317 has a small inductance on it's output which can form a resonant circuit with the output capacitor if there is not enough ESR to damp the resonance.
I've used 1000uF caps on the output, but with added series resistance of around 0.3 ohms from memory. It's probably safer not to go bigger than 200 - 300uF. The bigger the cap value the lower the frequency of resonance. 1000uF is right in the audio range (I can't remember exactly where).
Tony.
I've used 1000uF caps on the output, but with added series resistance of around 0.3 ohms from memory. It's probably safer not to go bigger than 200 - 300uF. The bigger the cap value the lower the frequency of resonance. 1000uF is right in the audio range (I can't remember exactly where).
Tony.
It's very rarely a good thing to add that much, since you are using a regulator with negative feedback to filter ripple and provide decent control of changing loads. It's not designed for those kinds of capacitive loads.
Various power supply turn off / on sequences can damage semi-conductors with that much energy (0.5C*V squared) stored at an output. Use diodes and such to discharge the thing.
I'd say much more to lose than can be gained, I can't think of any good reason to add a ton of C at the output of any series V regulator. Any capacitive loading should only be device bypassing / local PCB decoupling and smaller capacitor parts works much better there, Using multiples of ~0.01uF and 10uF for audio. Then after some point its probably better to add the regulator devices on each PCB instead of wiring it to various points. Id say no more than a couple of hundred micofarads combined TOTAL at most and you still may not have the data sheet performance.
Last edited:
I can think of some advantages. But it depends on what you are trying to achieve. What are you trying to achieve?
On the LM317 application notes, they show a 1uF capacitor on the output. Is that what you are recommending?
Yes, I am using the protection diodes recommended in the LM317 datasheet---one around the input/output and one across Vref.
Such as something chosen from the Linear Tech range according to what your actual needs are?
More current?
Less dropout volts?
Lower noise?
Better step response?
There are plenty of other suppliers too.
I am trying to achieve the lowest amount of ripple and noise on a regulated supply of +/- 18 volts that will provide 500mA of current. All this at a very minimal cost.
If you want to reduce ripple may I suggest that putting a CRC or CRCRC on the input (and a higher voltage transformer to cope with the drop from the CRC will do a lot more than putting a large cap on the output 
I do not think that the output cap is going to have any major effect on the noise level.
Have a look here YARPS, finally some progress! - diyAudio Note that this was my first attempt at designing things. An excercise in overkill, and whilst it works I would do things differently today.
There comes a point where chasing lower ripple becomes a bit pointless because the noise output of the LM317 will swamp it
For some insight into the mind of an obsessive person you could check out this thread From LM317 to LT3080 an LTSpice Journey It's all sims, and I haven't got around to starting a layout for a circuit board for the circuit as yet, but it *may* be of interest.
Tony.
I do not think that the output cap is going to have any major effect on the noise level.
Have a look here YARPS, finally some progress! - diyAudio Note that this was my first attempt at designing things. An excercise in overkill, and whilst it works I would do things differently today.
There comes a point where chasing lower ripple becomes a bit pointless because the noise output of the LM317 will swamp it
For some insight into the mind of an obsessive person you could check out this thread From LM317 to LT3080 an LTSpice Journey It's all sims, and I haven't got around to starting a layout for a circuit board for the circuit as yet, but it *may* be of interest.
Tony.
As a design goal, that has a few obvious limitations.
What bandwidth would you like the noise to low in?
In real circuits, the change in voltage, ringing etc, from a change in load are often much more significant than ripple under a steady load.
Aiming for 'lowest' often brings no advantage. Your amplifiers should have a resonable PSRR. A better goal might be to work out what your amps can reject before the supply makes an audible contribution, then aim for XdB better than that. It may be worth thinking about that across different frequency bands.
Alternately, if you know what your supplies really look like in frequency and time domains, you can optimise your amplifier to work with them.
The regulator is there to cut out low frequency variations.
Decoupling capacitors work at higher frequencies.
It's the transistion between the two that can be tricky....
I'd tend to agree with Wintermute to get rid of it upstream of the regulator.
It depends how far you want to go...how small you want the ripple to be.
Here's a LCR solution (an inductor is your friend) although it might be overkill and more costly.
Example values for 60Hz mains:
C1 = C2 = 4700uF
R = 7.5 ohms, L = 10mH
I estimate a ripple rejection of about 40dB at 500mA.
You have to make sure there is enough voltage across C2 for the 18V regulator...maybe allow 5V. And then there will be a drop across R, say 4V in the example. So you need at least 27V across C1. A 22Vac transformer would do. Make R >3W for margin...it will get hot at 500mA continuous current.
Mind you, an LM317 is specified to have 64dB of 120Hz rejection with a 10uF Adj cap. so you might end up with overkill rejection at 120Hz. However, the ripple isn't pure 120Hz and contains a lot of HF hash. The LCR would definitely crush this.
It depends how far you want to go...how small you want the ripple to be.
Here's a LCR solution (an inductor is your friend) although it might be overkill and more costly.
Example values for 60Hz mains:
C1 = C2 = 4700uF
R = 7.5 ohms, L = 10mH
I estimate a ripple rejection of about 40dB at 500mA.
You have to make sure there is enough voltage across C2 for the 18V regulator...maybe allow 5V. And then there will be a drop across R, say 4V in the example. So you need at least 27V across C1. A 22Vac transformer would do. Make R >3W for margin...it will get hot at 500mA continuous current.
Mind you, an LM317 is specified to have 64dB of 120Hz rejection with a 10uF Adj cap. so you might end up with overkill rejection at 120Hz. However, the ripple isn't pure 120Hz and contains a lot of HF hash. The LCR would definitely crush this.
Attachments
Linear regs should not have any problems with ripple. Low frequencies are what they are designed to handle and they do it well.
High-frequency noise, on the other hand, passes right through them. Perhaps that is where you need to focus your attention if you see issues.
If your load pulls too much current and you see modulation on the reg and you want to avoid it because it's shared for example, then use better local decoupling caps close to your load.
High-frequency noise, on the other hand, passes right through them. Perhaps that is where you need to focus your attention if you see issues.
If your load pulls too much current and you see modulation on the reg and you want to avoid it because it's shared for example, then use better local decoupling caps close to your load.