Don't take my finding as a rule. It may very well depend on the load circuits.
Also, I only tested with 470uF and 100uF on Iko's 5k. 470uF obviously sounded worse.
But who knows if a 47uF would sound better than the 100uF? I have not tried it. I will do eventually.
4.7uF is sufficient and stable. What I found is that increasing it made the sound smoother, possibly by reducing noise. The result was less edges in female voices, for example. The value can be tricky as Salas indicated that up to a certain point it may lose too much gain.
I think you are spot-on. I don't doubt your earlier report that the Rubycon ZA sounded better than a film. It must be true in your particular implementation. But I won't generalize that a Rubycon ZA sounds better than a film cap. In my opinion it does not. If both are properly implemented, I believe a film cap sounds better than an electrolytic in general.
If used in AC coupled stages, an electrolytic can sound better than a film only if the film cap value is too small causing early rolloff.
If used in power supply decoupling, it is most likely that the film cap, due to its low ESR hence lack of damping in the LCR circuit, cause excessive ringings. I modelled this a lot using LTSpice and I have actually measured it with a scope, and observed that a 0.01uF to 1uF MKP on the rails of a load can easily cause resonance between 800kHz to 10MHz with ripples up to a few hundred mV. I have to be extremely careful with using film cap on the rails and ensure that either the circuit inductance is extremely low. Or the safest way is to pad a 1R resistor to increase the "ESR" of the film cap.
So I guess in your case of the ZA sounding a whole lot better than the film cap, it was possible that your film cap caused exessive ringing. Not that your film cap sound quality was lesser and the ZA superior.
You use 1.1uF on the Vref and 5k is 30dB stronger gain than 1.2. If you don't filter enough at the root, you probably need filter at the exit.
Bill, I would not draw these conclusions so quickly. One, the ltspice simulations are to be taken with a large grain of salt. Always. Have you tried to include the wire parasitic and trace inductance in your simulations?
Using a 1R in series with the output cap can have a very bad effect on this regulator.
Iko, he said "and I have actually measured it with a scope".
Film caps needs to be investigated more deeply IMO. There are big differences between brands and models (not considering the obvious - the dieletrcic).
Hmm... Are you referring to film capacitor only PS or electrolytic cap with film bypass too?
To confirm Bill's findings I hooked up rev 5k to the spectrum analyzer to check for high frequency ripple (checked up to 40MHz). These are the results. Without remote sensing there was millivolt ripple at very high frequency even with a 22uF electrolytic capacitor. 47uF and up there was no ripple.
With remote sensing there was HF ripple with 1uF, 1uF in series with 0R33, as well as with 10uF electrolytic. No ripple with 22uF or higher, and no ripple with 470nF polypropylene bypassing 22uF electrolytic.
With remote sensing there was HF ripple with 1uF, 1uF in series with 0R33, as well as with 10uF electrolytic. No ripple with 22uF or higher, and no ripple with 470nF polypropylene bypassing 22uF electrolytic.
Ikoflexer,
I was not talking about your regulator. I was merely talking about general on board local bypass.
Your regulator is the only exception to all other regulators because anything else I have seen so far require a high ESR at the output. And yes, I have confirmed that your regulator works well with very low ESR film caps directly at the output.
Regards,
Bill
Last edited:
Thanks for the information. This seems to parallel my subjective experiences.
One clarification is needed though - is the capacitor you referred to the regulator output cap or local bypass cap?
With regards to the film local bypass at the circuit (not at the regulator output), to support what I previously mentioned, I have just quickly drawn up a simple circuit in LTSpice.
Note that I am using only the simplest way and assume I am only modelling the LCR circuit, nothing else. I have not created current source as load, etc. The purpose is only to identify the resonance frequencies and magnitudes and what sort of resistance required to pad the ESR of the capacitors to damp the resonances.
Note that the 1R resistor is not optimized. It is only a safe value. Indeed, after padding the 1R, the capacitor barely does anything below the MHz region, but above that, it is still useful.
I added 15nH to 20nH to the capacitor. Note that 1cm PCB track gives about 10nH inductance.
Note that I am using only the simplest way and assume I am only modelling the LCR circuit, nothing else. I have not created current source as load, etc. The purpose is only to identify the resonance frequencies and magnitudes and what sort of resistance required to pad the ESR of the capacitors to damp the resonances.
Note that the 1R resistor is not optimized. It is only a safe value. Indeed, after padding the 1R, the capacitor barely does anything below the MHz region, but above that, it is still useful.
I added 15nH to 20nH to the capacitor. Note that 1cm PCB track gives about 10nH inductance.
An externally hosted image should be here but it was not working when we last tested it.
If you just use 47-100uF locally at the output of a 5k or a 1.2 without further resistance in series than its own esr (both will work with enough own phase margin) doesn't it swamp enough the interactions so to simplify the local decoupling?
What I have found is this:
Use a +/-15VDC regulator, connect the output to an opamp buffer with 10cm wires (about 100nH inductance) (note that the opamp has input capcitance), put a film cap 0.1uF or 0.01uF at the opamp supply pin directly and connect the other end to ground, just as all manufacturers' datasheets recommend, put your scope probe there, you will find hundreds of mV ripples at the frequency region as indicated in the above LTSpice model.
No wonder it never sounds good wtih opamps!
The only cure to me so far is to use either Ikoflexer's v5k or Salas v1.2 with remote sensing and the shortest possible wires, and this allows ditching the local bypass completely. No ripples found.
Use a +/-15VDC regulator, connect the output to an opamp buffer with 10cm wires (about 100nH inductance) (note that the opamp has input capcitance), put a film cap 0.1uF or 0.01uF at the opamp supply pin directly and connect the other end to ground, just as all manufacturers' datasheets recommend, put your scope probe there, you will find hundreds of mV ripples at the frequency region as indicated in the above LTSpice model.
No wonder it never sounds good wtih opamps!
The only cure to me so far is to use either Ikoflexer's v5k or Salas v1.2 with remote sensing and the shortest possible wires, and this allows ditching the local bypass completely. No ripples found.
The good news is that with the method I mentioned above, you don't even need a 47uF to kill the ripples. Even 4.7uF with Iko's v5k, or 2.2uF + 1R with your v1.2, I saw no ripples with remote sensing and very short wires. but of course, as reported, on Iko's v5k a 100uF at the output reduces the noise. I will try 47uF and see if it makes any differences.
Last edited:
But if you up the output caps of those regs say to 47-100uF can't you have longer cables?
P.S. If you do that 47-100 on 1.2 delete the 1R.
P.S. If you do that 47-100 on 1.2 delete the 1R.
I am not sure. Each time I shorten the distance between the regulator output and the load, I got sound improvement - reduced noise, increased dynamics, increased smoothness, openness and transparency (Oh, I shouldn't use those subjective terms).
Actually Iko's post this morning made me think the same thing as you suggested - perhaps an increased output cap (reduced noise) would do the same trick?
Unfortunately, my load is now right next to the regulator output with so many wires coming in and out in some tiny space it would be time consuming for me to do further tests. I am very content with what I have and am very keen to do the real final build of my Active XO/EQ/Preamp board, in which for the tweeter at least I will build the circuit and the regulator on the same board to even shorten the distance.
I think the last thing I will do is, as you suggested, select the jFETs for my DCB1 and see if it betters the opa627, and I will wrap up the whole thing.
Oh I see, you were talking about another cap locally on the load. I'll test that tomorrow. Thanks Bill!
We better find a recipe to make sure people without an oscilloscope at home can build it and be sure it works well. This is the problem when we're after the last drop of performance. It becomes harder to build right.
We better find a recipe to make sure people without an oscilloscope at home can build it and be sure it works well. This is the problem when we're after the last drop of performance. It becomes harder to build right.
Because you see the 1uH on his Spice pic? Hmm not sure he means a remote lytic considering the reg. A photo will settle it.
From this, I think Bill calls "local bypass cap" the capacitor that's at the load. I think.
- Status
- Not open for further replies.