I first played with 3 terminal regs LM317/337 and the LT serires as well, searching for and trying the best implementations. Next I tried JLH gyrator / shunt, or perhaps more appropriately called ripple eater. Next I tried a commercial shunt kit. Next I tried a series / emitter follower. Next I tried the Jung Supereg (ALW). Then I tried Salas v1.1, Iko v5a, v5d, Salas v1.2, Iko v5k.
In my personal opinion the best are v5k and v1.2, which I will end up living with. There are subtle differences between the two, but you have to find out yourself which flavour you like. The final sound quality is less depending on which reg you use, but more depending on how well you implement it, such as how well your PCB is designed, what output caps (value, brand, type) to use, what Vref caps (value, brand, type) to use, whether you parallel film caps with electrolytic caps, the wire thickness and lengths to the load, remote sense shielding, etc. The implementation produces more difference in sound than the different sound of the regulators themselves.
Last edited:
For example, a 10cm 22aug tinted copper wire, the normal wire used for perfboard tracks, has 100nH inductance and 0.0054R resistance. This would produce higher impedance than the regulator. Higher impedance means higher ripples in real load. This is the simpliest example how implementation will swamp the difference between the regulators.
The Vref cap and the output cap can strongly influence the sound. And again they produce more difference in sound than the difference of the regulator.
I mentioned yesterday that I parallelled my Vref 1.1uF MKP with 47uF electrolytic and found the sound was cleaner. Further listening proved that, it was actually quite a bit cleaner and the sound became really open and completely flied out of the speakers. However, as I almost always found parallelling caps produce some subtle (nasty) effects, I later found the sound to be a tiny bit "colder" than without it. So it may be a trade-off. It may be strongly system dependant.
And as discussed before, if you run long wires that introduces sufficent inductance, and if your load's rail input has high capacitance, regardless how stable the regulators are, you will still get resonance because you can not change physics - LCR will resonant unless the R is sufficent for damping. In that case, the regulators won't sound good any more, and if you compare the sound quality of the regulators under that condition you will never come to the right conclusion.
Unfortunately, implementation is LOAD DEPENDANT. For example, we may not be able to fix an output cap for the regulator, because, local bypass caps, if necessary, will appear to be part of the regulator from the regulator point of view.
Best is to try it, get a scope to measure it if possible, and post your questions in the forum and I am sure Ikoflexer and Salas can help you through.
The Vref cap and the output cap can strongly influence the sound. And again they produce more difference in sound than the difference of the regulator.
I mentioned yesterday that I parallelled my Vref 1.1uF MKP with 47uF electrolytic and found the sound was cleaner. Further listening proved that, it was actually quite a bit cleaner and the sound became really open and completely flied out of the speakers. However, as I almost always found parallelling caps produce some subtle (nasty) effects, I later found the sound to be a tiny bit "colder" than without it. So it may be a trade-off. It may be strongly system dependant.
And as discussed before, if you run long wires that introduces sufficent inductance, and if your load's rail input has high capacitance, regardless how stable the regulators are, you will still get resonance because you can not change physics - LCR will resonant unless the R is sufficent for damping. In that case, the regulators won't sound good any more, and if you compare the sound quality of the regulators under that condition you will never come to the right conclusion.
Unfortunately, implementation is LOAD DEPENDANT. For example, we may not be able to fix an output cap for the regulator, because, local bypass caps, if necessary, will appear to be part of the regulator from the regulator point of view.
Best is to try it, get a scope to measure it if possible, and post your questions in the forum and I am sure Ikoflexer and Salas can help you through.
Last edited:
Yes I did, although at the time without the scope. The important
discovery I have is the exception with 5k and v1.2 that for opamp load
using remote sensing and very short but thick rail and ground tracks
removes the need for local bypass. There is no reasonance. This makes
life a whole lot simpler.
The LM317/337 sounded more musical than the LT series IMHO. The LT
series did not sound right to me, though there was much hype.
But the LM317/337 can never deliver the resolution 5k or v1.2 deliver.
It is a night and day difference. They have much higher impedance and
are much noisier. If you rely on using capacitors to lower the impedance
at higher frequencies, first you can easily get resonance when
interacting with a real load, secondly, when you manage to get it right
with no resonance, the capacitor sound dominates, making it coloured.
The Jung supereg has much better potential comparing to the LM317/337,
but it is even more difficult to implement comparing to 5k and v1.2. It
uses an opamp for the error amp, and I did not like the recommended
AD825. opa627 sounded better to me. Nevertheless, the opamp is powered
by the regulator itself. That is something tricky. Even if it is
implemented properly, I don't know soundwise how it compares to 5k and
v1.2. In terms of specification in simulations, the 5k and v1.2 fare
better. One advantage the Jung supereg definitely has is less heat. If I
have the time, I would be doing it again with my scope, and hopefully
get it sound optimal without resonance. I believe it will take a lot of
time, which I don't have. So I think I will give it a miss.
Last edited:
Hey Bill, this is a lot to chew on, for sure. I'm glad we got something good here, but I'm sure there's room for improvement. Have you heard of Paul Hynes' regulators? Not diy, a commercial product, highly recommended by many. You're a perfectionist, so you might be looking for your next target 
Anyway, the more I think about it the more it becomes clear that the universal shunt regulator that would work over a wide range of output voltage and current is simply not realistic. Not if we want the best performance. So then, I'm thinking to target specific and smaller voltages and currents and design the best regulator for it. A lot of people are interested in +-3.3V, or +-5V for DAC applications, for instance, but they don't need a lot of current. Then, as crazy as it seems, but I think we have a real target in lowish power amps, something up to 30 something watts or so (chipamps, t-amps, etc.). And so on...
Anyway, the more I think about it the more it becomes clear that the universal shunt regulator that would work over a wide range of output voltage and current is simply not realistic. Not if we want the best performance. So then, I'm thinking to target specific and smaller voltages and currents and design the best regulator for it. A lot of people are interested in +-3.3V, or +-5V for DAC applications, for instance, but they don't need a lot of current. Then, as crazy as it seems, but I think we have a real target in lowish power amps, something up to 30 something watts or so (chipamps, t-amps, etc.). And so on...
The regulator is already top notch. I don't know if we can expect more.
What I mean is that there is no universal regulator because there is no
universal load. So even the best regulator still needs to integrate with
the particular load to achieve the best, that is what I mean by
implementation dependance.
What I mean is that there is no universal regulator because there is no
universal load. So even the best regulator still needs to integrate with
the particular load to achieve the best, that is what I mean by
implementation dependance.
Last edited:
I don't know if you guys have seen this thread. http://www.diyaudio.com/forums/solid-state/109147-geddes-distortion-measurements-20.html#post1336279
To summarize, it's about removing the noise from THD+N plots to look at what's fundamentally going on with an amp along with how the amp distortion at low levels effects sound quality. At post 199, Michael plots very low level distortion measurements for his amp. The first plots are with his normal power supply, the second with a laboratory PSU. The results are significant. For what it's worth, my reading of the aforementioned thread lead me to this thread. I know that I have personally learned a lot and presume others have as well. Keep up the great work.
Ken
To summarize, it's about removing the noise from THD+N plots to look at what's fundamentally going on with an amp along with how the amp distortion at low levels effects sound quality. At post 199, Michael plots very low level distortion measurements for his amp. The first plots are with his normal power supply, the second with a laboratory PSU. The results are significant. For what it's worth, my reading of the aforementioned thread lead me to this thread. I know that I have personally learned a lot and presume others have as well. Keep up the great work.
Ken
Ikoflexer, it's not crazy, I may be nuts, but the idea is exactly what I'm looking for. I was thinking about using a regulator to drive a LME49811 chip and maybe the driver transistor, leaving the output transistors on a more conventional PSU, so, the current requirements wouldn't be over the top.
Ken
Yes, I'm hoping to power the output devices with a shunt reg as well.
@salas, hey, another nice example of a successful v1!
But for this the current design is too power-hungry (i have basically given up on that).
Powering drivers and predrivers is very good anyway
Yes, but how to get a stiff, unregulated supply? Chokes for high current are enormous and $$$.
Yes, but how to get a stiff, unregulated supply? Chokes for high current are enormous and $$$.
Are they more expensive than the heatsinks and possible multiple output mosfets necessary regulate the aforementioned high currents?
In my case they are definitely cheaper than the required heatsinks, especially because this regulator must be located close to the circuit that is going to regulate. And two class A amps dissipating about 150W per channel each in a single case requires some extreme metalwork skills that I do not possess.
For a class D or gainclone amp instead, full regualtion seems a very good idea
Really? Have a look:
Power Chokes: Large Transmitter Applications
In particular, let's consider the Basler choke near the end of the page. 1 Henry, DCR of 10 ohms, 4.5A, for only $325 plus shipping. These seem surplus, but it never bothered me buying used. Run psudii and see what ripple you get with this. Or simulate it in ltspice. Do a CLC with some large capacitors.
Oh, my favorite on that page would be the Magnatran filter reactor. 3 Hy @ 15 dca @ 1200v, 380 cycles. DC resistance = 6.6 ohm, operating volts = 25 kvdc. 59 gallon oil (may contain PCB). 36" x 59" max dimensions. New, never used. Yeah baby, groovy! Very shaggadelic for only $9500
Power Chokes: Large Transmitter Applications
In particular, let's consider the Basler choke near the end of the page. 1 Henry, DCR of 10 ohms, 4.5A, for only $325 plus shipping. These seem surplus, but it never bothered me buying used. Run psudii and see what ripple you get with this. Or simulate it in ltspice. Do a CLC with some large capacitors.
Oh, my favorite on that page would be the Magnatran filter reactor. 3 Hy @ 15 dca @ 1200v, 380 cycles. DC resistance = 6.6 ohm, operating volts = 25 kvdc. 59 gallon oil (may contain PCB). 36" x 59" max dimensions. New, never used. Yeah baby, groovy! Very shaggadelic for only $9500
In non regulated power stage psus sometimes a low regulation massive transformer combined with very high value capacitors is called ''stiff''. Though author Ben Duncan has argued that some slack there comes across subjectively more dynamic due to the compression effect before clip and psychoacoustics.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.