You da man!
I've done something similar with lower voltages. How did you protect the LMs from exceeding 40V?
I’m interested in that LM317 FFT. It looks like a 1kHz signal at about -51dB and H1 at about -72dB. That’s just 21dB difference. Is this correct? I’m curious about the details of this measurement, like the signal current amplitude and the actual circuit, with regard to estimating the distortion performance of this regulator.
A circuit which uses feedback can not be "unconditionally stable". That's why they specify a minimal load -- it reduces the output impedance and flattens the Q.
The regulator probably failed shorted and it still sounded OK.
This thread reminds me of 'Fear and Loathing in Las Vegas'
Some folks think running their rented convertible tires (tyres) at 90 psi gives them that extra good hi-way feel and car handling (feedback) on those long trips to Las Vegas.
"Unconditionally stable" is a perfectly well-defined term for a system using feedback. In fact, it has two definitions:jackinnj said:
1. It won't oscillate whatever passive load impedance you apply at the output.
2. It won't oscillate if the loop gain is reduced.
Definition 1 is the one people usually mean, unless the context suggests otherwise.
The 317 contains among other things, a current limit.
A current limit into a big capacitor becomes a slew rate limit.
Once you are slew rate limiting, the system is no longer linear.
So a lot of the analysis can quickly become void.
The rails of a class B amp are actually quite a tough role for a linear regulator, as each rail starts or stops drawing current every time the audio crosses zero.
Not a linear regime at all.
A regulator is generally trying to maintain a constant voltage so slewing is normally the last thing on its mind. Yes, current limiting may be an issue but is so even if there is no cap on the output. The cap helps the regulator by providing some of the dynamic load current, proportional to the ratio of the regulator output Z to the capacitor Z.
I think we have established that a large capacitor will not make an LM317 oscillate. That's not true of all regulators. Not all regulators are unconditionally stable.
For the capacitor to make a reasonable contribution to the regulation you want its impedance to be lower than the output impedance of the LM317 at the frequencies of interest. With no capacitor the LM317 would do all the regulation and with, say, an infinite capacitor the LM317 would make no contribution other than, after an infinite time, setting the dc level.
The question of what the optimal capacitor value, with appropriate series resistors, should be for a particular application is an interesting one. To what extent does the load benefit from capacitive regulation vs active regulation?
Last edited:
I've never seen an LM317 oscillate, but I've seen them ring and have poor settling time.
Is it possible to make one oscillate? I'm not convinced either way, but it's not necessary for a reg to actually oscillate, in order to generate spurious. The regulator is getting current impulses on every zero crossing in a class B system.
I used to think 'a regulated rail must always be better' but I've come around to the idea that it's possible for a regulated rail to make things worse.
Is it possible to make one oscillate? I'm not convinced either way, but it's not necessary for a reg to actually oscillate, in order to generate spurious. The regulator is getting current impulses on every zero crossing in a class B system.
I used to think 'a regulated rail must always be better' but I've come around to the idea that it's possible for a regulated rail to make things worse.
OH?
This circuit appears in the Nat Semi Power Management Handbook, but I picked it off a website. If you look at the impedance curve of an LM317 without any output cap you'll see it has a high-Q peak:
Attachments
All the app note circuits with external transistors to boost current seem to have much bigger caps on the output. They really need to be a separate discussion. Many of them are quite slow circuits.
Nobody is advocating zero output capacitance, at the very least there are normally several 100n's distributed around the regulated rail.
To a first approximation you can regard the output of a regulator (almost any series regulator) as consisting of:
a DC voltage source
a noise voltage source (mix of white noise and ripple etc.)
a series inductance (possibly with a resistor across it)
a series resistance
Add a capacitor to this and you have a low pass filter, but potentially one with a high peak near the corner frequency. The peak is a peak in both noise and output impedance. You get to choose where in the spectrum you want this peak to be.
a DC voltage source
a noise voltage source (mix of white noise and ripple etc.)
a series inductance (possibly with a resistor across it)
a series resistance
Add a capacitor to this and you have a low pass filter, but potentially one with a high peak near the corner frequency. The peak is a peak in both noise and output impedance. You get to choose where in the spectrum you want this peak to be.
by eliminating the capacitor on the output they have exploited the impedance peak in the LM317 to use it as an SMPS controller. You mentioned that you had not seen an LM317 oscillate. under these conditions it oscillates purposefully.
Well, I see IC regulators as akin to opamps; if you believe different op-amps can have sonic affects then regulators can too, including detrimental ones. The regulator datasheets that I have seen don't reveal much info that pertains to audio quality, understandably. Op-amp datasheets are often more helpful.
This is a little off topic.