Improved LM317 ?

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Searching the archives I came up with this "better than datasheet" implementation of the LM317 regulator according to some...

I think the idea is to fixate and isolate Adj from noise of the output, so the LM317 only regulates to keep 1.25V difference between Vout and Adj ?
Or do you effectively break the feedback loop, making the regutor "worse" and have the output caps do all the AC work, but sonically better due to removing regulator induced sonics ?

I'm puzzled.

Same questions with TL431 shunt regulator and capacitor decoupled reference...



  • lm317.jpg
    11.2 KB · Views: 6,055
Many questions here are about how good is 7815, LM317 etc. If we deside to that the load are rather constant and the load regulation aren't very important nor the stability, then you only have to filter the voltage, before and after the regulator. If you do that you will have a "quite" voltage.

If you check my QSXM3 phono amp, you will see how I have done. I have C+R+C+R+C for smoothing with makes very smoothe incoming voltage.

Check also datashets and application notes at Linear Tech for LM317/LT317. They have some ideas to make it better but for unknown loads!

Check also my extreme regulator QSXPS for MC-amps.
AX tech editor
Joined 2002
Paid Member
improved 317


Since the 317 does its best to keep 1.25V between Adj and output, the quietness of the Adj has a direct effect on the final result.

In the diagram you posted, Adj is made as stable and quiet as possible by using a current source and a ref diode. That is at any rate better than just the customary resistive divider with decoupling cap. The 'resistive divider' formed by I1 and D1 - for noise etc - probably is something on the order of 1 million to one, which will be hard with an R and C, especially at low freqs.

The other question on the 'musicality', whatever that is, is another matter altogether. The 317 is there to approach an ideal volt source. After all, the amp is designed with the assumption that it is fed with a DC voltage. If you think a less ideal volt source makes the amp sound better, there is no overriding reason to use a 317 or whatever regulator you can think of.

Cheers, Jan Didden
Oh my - it's Jan Didden!

Just thought I'd say Hi!

As I'm in the process of building some Walt Jung regulators, based on his updated bootstrapped design, you're currently in my mind as I've been reading and studying the original TAA articles you contributed extensively to.

Any hints, tips or updates would be gratefully received by all, I'm sure ;)

Anyway, thanks for the work you all did on the originals.


AX tech editor
Joined 2002
Paid Member

Hi yourself,

There is one additional change I did in one of my implementations which as far as I know hasn't been documented, and that is replacing the current source on the input side (you know, the LED & transistors and couple of Rs) with a current reg diode (CRD), which actually is a FET connected as a current source. IIRC I used a J50x, try to find the highest value, like 10mA. This does further improve input noise and ripple rejection especially in the higher freq area. There is one caveat though: you need at least 7 or 8V difference between the input and output to have the CRD function correctly. But other than that, I find it worthwhile towards measurable better performance. Whether it's worth it sonically is something I cannot decide for you for obvious reasons.

Cheers, Jan Didden

Thanks Jan for taking the time to update.

I seem to remember that one of Walts design aims was to reduce dropout, hence the choice of LED-biased current source. In many POOGE-type applications though the extra dropout could be handled, so I will try it at some point in order to assess the sonic effect.

Certainly the lashed-together Sulzer style regulator, and my own modified version of the POOGE 5.x discrete reg's have astounded me with their sonic abilities.

I was surprised actually that no-one AFAIK saw what seemed an obvious mod to the POOGE reg's, that had been implemented in all the op-amp based reg's i.e. tailoring the feedback response to unity at AC. It effectively lowers output impedance by a factor of 10, and noise by a significant margin, for just one extra capacitor. I guess it was largely due to preoccupation with the better designs.

The real benefit being these can be made small and cheap, relative to the AD797 based units, making them a good 3-terminal replacement. They are also wonderfully stable.

Thanks again,

Jan, can I pick your brains a sec?

Just a couple of brief questions that have arisen from the succesful implementation of the EDN super-regs I've just built, hope you can shed a little light.

1. I'm very puzzled as to why the sense line decoupling capacitor was taken to the regulator o/p. Even with this implemented I've had low-level oscillation at HF with the remote sensing option.

Returning the decoupling capacitor to 0V though cures it completely under all conditions I've tried it in. I'm curious because the connection seems unusual and less effective than it could be, although there must have been good reason for it.

2. Did the sense plane over the board just connect to one single point (i.e. the -ve sense resistor point) with the LM329 and it's filter capacitor returned to this point via tracks, or were the LM329 and filter cap returned directly to the ground / sense plane?

I have the option for both, but need to do a bit of work to increase resolution of the T&M system in order to assess, and was hoping for a pointer based on your experience.

I'm though stunned by the sonic performance - a simple PSU, containing a single +ve regulator (@ 24V) outperforms a major name HiFi manufacturers very expensive PSU, containing seperate regulators for each channel, by a considerable margin (and costing some 800 ukp!).

The vanishingly low o/p impedance of the super-reg's allows just a single unit to be used, providing star-connection schemes are used, although I'm sure there will be benefits to seperate reg's.

Thanks for your part in this - must drop Walt an email too, assuming he's still contactable.

op-amp based regulators


I have built a regulator based on BB OPA548, a power amp chip. The OPA548 has an internal voltage reference that can be used to generate the first voltage, this then gives the reference for the second chip, to generate the negative voltage. I implemented this very straightforwardly using the datasheet (another version of OPA548's used like that can be found on It is driving my preamp prototype, still on breadboard :) This should be somehow similar to the Walt Jung concept, except for the power devices which are not needed, the OPA548 can deliver 3A cont.

Now my question: cranking up to full gain (15dB) I can still hear some faint line ripple from my woofers and some noise from the mids. How would I go about improving such an OPA548 based concept? Is "bootstrapping" possible w/o adding power transistors (that would defeat the purpose of using a power op amp in the first place) and w/o using another regulator, say 78xx/79xx, in front (the project uses LM317 before OPA548 - I find that is a bit much of brute force, elegance is always appreciated:D ).

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.