If this does not get back on track, those offending will no longer be allowed to post here. Back to the topic, gentlemen.Seems like the diversion was winding down on its own anyway.
I am doing my least favorite part of design, which is choosing electrolytic caps. Please, if someone wants to decide for me, do chime in. I was thinking of using Nichicon LKS. The 1000 hour endurance is alarming, but I realize the caps last a lot longer at ordinary temperatures and currents.
Seems like, since I am planning a dual-mono supply, that 4,700uF caps will suffice in lieu of the 10,000uF units specified for the single supply version.
I have absolutely no concept of the sonic tradeoffs involved in choosing, say, 105C or 10,000 hour caps over standard stuff. Ironically, caps marketed (whether or not for good reason) as "audio grade" all have the lowest temperature and endurance ratings. I've read every thread I can find on electrolytic sound quality, multiple times, and come away more confused than when I started.
I am doing my least favorite part of design, which is choosing electrolytic caps. Please, if someone wants to decide for me, do chime in. I was thinking of using Nichicon LKS. The 1000 hour endurance is alarming, but I realize the caps last a lot longer at ordinary temperatures and currents.
Seems like, since I am planning a dual-mono supply, that 4,700uF caps will suffice in lieu of the 10,000uF units specified for the single supply version.
I have absolutely no concept of the sonic tradeoffs involved in choosing, say, 105C or 10,000 hour caps over standard stuff. Ironically, caps marketed (whether or not for good reason) as "audio grade" all have the lowest temperature and endurance ratings. I've read every thread I can find on electrolytic sound quality, multiple times, and come away more confused than when I started.
I just ordered a pair of Talema 2x22V 25VA toroidal transformers. Stock on these things, in all ratings, is dwindling worldwide, but I found some at RS Components again, branded Talema. You may remember, last time I got some 2x15V units, same OEM as Talema, but with RS branding. This is my third transformer order from RS. Again repeating myself, after the first order they sent me an alarming email saying they wouldn't do business with me again unless I gave them a valid export license, or some such nonsense. I was convinced black vans filled with armed troopers were going to show up outside my house when I placed the second order, but nothing happened and the parts arrived in a week. Hopefully it will work out this time as well.
It's funny what a wuss I am. I actually feel guilty for cheating. As if someone is going to yell at me for not paying my taxes. It's dumb, I know. "Honest, your honor, I had no idea ordering online was illegal."
I'm doing this because I'm frustrated and don't feel like designing PC boards right now. I designed a raw supply board for the DCG3 prototype that I can etch quickly and use with the spare regulators I posted a little while ago. Or I may build up my last integrated Super Regulator board. This will be a lot more compact and potentially get me going in maybe a week and a half or so while I am waiting for my Toroidy transformer and working on the official supply.
I am curious to see how Jam's low-feedback supply compares sound-wise with the Super Regulator. So there is some value here.
It's funny what a wuss I am. I actually feel guilty for cheating. As if someone is going to yell at me for not paying my taxes. It's dumb, I know. "Honest, your honor, I had no idea ordering online was illegal."
I'm doing this because I'm frustrated and don't feel like designing PC boards right now. I designed a raw supply board for the DCG3 prototype that I can etch quickly and use with the spare regulators I posted a little while ago. Or I may build up my last integrated Super Regulator board. This will be a lot more compact and potentially get me going in maybe a week and a half or so while I am waiting for my Toroidy transformer and working on the official supply.
I am curious to see how Jam's low-feedback supply compares sound-wise with the Super Regulator. So there is some value here.
Usually 105deg. C caps (but not always) can have a lower ESR and the benefit of a much longer life, especially at at elevated temperatures.
Some light reading.............
https://resources.altium.com/p/which-type-capacitor-should-you-use
I have always believed that the power supply is half of the audio circuit and you should not skimp on it.........beware in this case complexity can have it's drawbacks like the amplifier stage.
You seem to be making great progress................😉
Jam
Some light reading.............
https://resources.altium.com/p/which-type-capacitor-should-you-use
I have always believed that the power supply is half of the audio circuit and you should not skimp on it.........beware in this case complexity can have it's drawbacks like the amplifier stage.
You seem to be making great progress................😉
Jam
Attachments
Last edited:
Thanks for the article. I found this document, which goes into great detail on electrolytic capacitor manufacturing, specs, and lifespan: https://www.nichicon.co.jp/english/products/pdf/aluminum.pdf
There's nothing in there, of course, to help you choose the best-sounding caps, but it gives formulas for estimating endurance. The gist of it is you can get fifteen years of useful life if you buy a capacitor with a long rated endurance and keep it cool, with low ripple current. I've also read that you can roughly double the lifespan by operating the caps well below their rated voltage. But don't go too far with that, because the improvement levels off, and higher-voltage caps have higher ESR, which means more heating for a given ripple current.
There is regulator "shootout" published at Linear Audio that gave the Jung regulator high marks: https://linearaudio.nl/sites/linearaudio.net/files/v4 jdw.pdf
Interestingly, for the listening comparisons, the test used a Borbely JFET line stage that's somewhat comparable to the HPA-1 circuit. It's too bad they didn't try something like your regulator. My expectation is that if I like the sound of this amp with the stock power supply, I will like it with the Super Regulator too, though I may well hear a difference. That's what we want to find out.
For years, I've been reading complaints that "engineers" are too smug and put too much faith in theory and measurements. I can't honestly call myself an audio designer until I try these things for myself. There are too many combinations to try without making a full-time career of it, so this is just a casual sampling. I always start out assuming I won't hear a difference. Either way, I will tell you honestly what I hear, even if it's opposite to what you say. What I won't do is tell you there's something wrong with your findings or preferences or design choices.
There's nothing in there, of course, to help you choose the best-sounding caps, but it gives formulas for estimating endurance. The gist of it is you can get fifteen years of useful life if you buy a capacitor with a long rated endurance and keep it cool, with low ripple current. I've also read that you can roughly double the lifespan by operating the caps well below their rated voltage. But don't go too far with that, because the improvement levels off, and higher-voltage caps have higher ESR, which means more heating for a given ripple current.
There is regulator "shootout" published at Linear Audio that gave the Jung regulator high marks: https://linearaudio.nl/sites/linearaudio.net/files/v4 jdw.pdf
Interestingly, for the listening comparisons, the test used a Borbely JFET line stage that's somewhat comparable to the HPA-1 circuit. It's too bad they didn't try something like your regulator. My expectation is that if I like the sound of this amp with the stock power supply, I will like it with the Super Regulator too, though I may well hear a difference. That's what we want to find out.
For years, I've been reading complaints that "engineers" are too smug and put too much faith in theory and measurements. I can't honestly call myself an audio designer until I try these things for myself. There are too many combinations to try without making a full-time career of it, so this is just a casual sampling. I always start out assuming I won't hear a difference. Either way, I will tell you honestly what I hear, even if it's opposite to what you say. What I won't do is tell you there's something wrong with your findings or preferences or design choices.
There is a recent tidbit about this test made by Jack Walton. Borbely's JFET line amp has 40 dB PSRR and explanation of perceived sound differences is that they were because of various harmonic content found on supply lines of different PS.
https://www.diyaudio.com/community/threads/shunt-hv-vs-series-hv.376889/post-6909093
I wonder how large those harmonics were, that it was possible to hear them attenuated by 40 dB. Not to mention that they were much more below signal level. IMO, another strong reason for the perceived differences was PS transient response, which was not measured.
Could be. I saw something last night by Walt Jung, measuring the effect on transient response of different capacitors in his regulator. I didn't read it carfefully, but IIRC he concluded they didn't make much difference in his tests.
I don't believe in voodoo. I'm pretty sure if measurements fail to predict sound quality it's either because we're making the wrong measurements (or doing them badly), or we just misunderstand the relationship between technical performance and what sounds good. I think the second is more likely.
I don't believe in voodoo. I'm pretty sure if measurements fail to predict sound quality it's either because we're making the wrong measurements (or doing them badly), or we just misunderstand the relationship between technical performance and what sounds good. I think the second is more likely.
The only calculus nobody tell you is how long a capacitor's capacitance and esr will survive if unused on the shelf.Most of audiophiles have more than one amplifiers and they choose some to play daily while others stay like a few years doing nothing.Guess what happens with most of the capacitors after one year without any voltage on them! More on this...if you buy a lot of capacitors who already spent 5 years in some basement...
You probably refer to this article: SR 2020 Dynamic Tester
Conclusion that choice of output capacitors for Jung regulator is not critical is not the same as that transient response is not very important. As I understand it, this capacitors test was only related to their suitability for this specific regulator and it was determined that capacitor type doesn’t affect much regulators excellent performance.
I’m sure that transient response of other regulators was nowhere close to the response of Jung regulator.
Conclusion that choice of output capacitors for Jung regulator is not critical is not the same as that transient response is not very important. As I understand it, this capacitors test was only related to their suitability for this specific regulator and it was determined that capacitor type doesn’t affect much regulators excellent performance.
I’m sure that transient response of other regulators was nowhere close to the response of Jung regulator.
Good information Henry.
An intersting experiment (with the Jung regulator) is to try attacling the sense line to different points on the power supply line.
Deductive reasoning suggets the load but is it? 🤔
Jam
An intersting experiment (with the Jung regulator) is to try attacling the sense line to different points on the power supply line.
Deductive reasoning suggets the load but is it? 🤔
Jam
Last edited:
I didn't include the remote sensing feature in my Super Regulator design. I reasoned that a few milliohms of resistance will make no difference to the amplifier, but extending the feedback loop off the regulator board could be an invitation for trouble.
You might consider a nested (two sense lines-local and remote) loop so in case of an accident you won't loose the blue smoke.
Just thinking out loud........🤓
Jam
Just thinking out loud........🤓
Jam
Attachments
Last edited:
I see your point and that's a very good idea -- if I were using remote sensing. Flakey connection on remote sense... blammo. My boards are hard-wired for local feedback and don't have the remote sense connections at all. It seemed like an unnecessary feature to me.
I do have a question for you, about supply voltage. Since the absolute maximum for the OPA604 is +-24V, would it be wise to set up the supply for something lower, like +-23V? I'm ordering parts now, was contemplating this eternal question.
I do have a question for you, about supply voltage. Since the absolute maximum for the OPA604 is +-24V, would it be wise to set up the supply for something lower, like +-23V? I'm ordering parts now, was contemplating this eternal question.
I don't think that is necessary. I have overvolted the 604 with my original prototype with about 27V for a few months with no problems.
I've been thinking about current feedback amplifiers and wanted to share some comments, though not without trepidation. (The HPA-1 is, of course, a current feedback amplifier.)
There is a long and contentious thread on the subject here, where people far smarter than me failed to reach any firm conclusions. I don't want to revisit that chaos.
There are four "canonical forms" of feedback, and CFAs use the very same form as VFAs. Some people are bothered (but not me, mostly) that the term "current feedback" has been appropriated from its classic meaning to describe this topology. In the older sense of the term, CFAs are precisely voltage feedback amplifiers.
The main advantage of CFAs seems to be that they can achieve very high slew rates. This is because the IPS output current comes externally via the feedback loop, instead of being constrained by the input differential pair tail current source. This is a boon to IC op-amp designers, where supply current and package dissipation are important considerations, and parts need to be designed for general applications. With a purpose-built discrete VFA design, we can get around these problems easily enough. IIRC, my A1 has a slew rate of 100 V/uS and the A2 measures 50 V/uS. Arguably, 5 V/uS is more than enough for audio work, and 200 V/uS (obtained with a CFA) is absurd overkill.
Another purported advantage of CFAs is that the closed-loop bandwidth is independent of the closed-loop gain G = (1 + Rf/Rg), where Rf and Rg are the values of the two feedback network resistors. This is true but misleading. In a CFA, the input stage transconductance, and hence the open-loop gain, depends on the value of feedback resistor Rg. In a VFA, open-loop gain does not depend on the values in the feedback network. So, if you increase the closed-loop gain of a CFA by lowering Rg, you raise the open-loop gain, and this compensates for the drop in bandwidth you would get doing the same experiment with a VFA. On the other hand, if you change the CFA's closed-loop gain by changing Rf, you will see the usual tradeoff between gain and bandwidth.
Looking at the input stage of the HPA-1, you could convert it to a VFA by the addition of two more JFETs buffering the feedback signal.
This would make it exactly into a complementary-differential input stage, e.g., as described by Cordell and many others. It's interesting to think about what effect this would have on both technical and subjective performance.
CFA proponents often talk about the ease of compensating these amplifiers, but they are just as constrained by output stage poles as VFAs. The difference is that CFAs tend to have lower open-loop gains, so require less compensation for stability. You can crank up the open-loop gain of a CFA and then you run into exactly the same problems with phase margin as with a VFA.
In general, aside from slew rate, CFAs tend to have lower performance specs than VFAs. And there are some newer integrated VFA chips that achieve slew rates and bandwidths comparable to CFAs.
There is some debate about what exactly defines a CFA. Some people think it's synonymous with having a diamond buffer input stage, but this isn't true. The HPA-1 doesn't use a diamond buffer, but is a true CFA. The actual definition is clear: The CFA inverting input is a low-impedance (ideally zero-impedance) node, and can source or sink current to maintain its voltage equal to that applied to the non-inverting input. In other words, in a CFA, there is a unity-gain buffer between the non-inverting and inverting terminals. In all other respects, CFAs and VFAs are conceptually identical.
This is a little counter-intuitive, that the inverting input is actually the output of a low-impedance buffer. It works because the transimpedance stage that follows is driven by the input/output current from that buffer.
In both the VFA and the CFA, the open-loop voltage gain is the product of the first stage transconductance and the second stage transimpedance. This is partly why some people argue the two amplifier types are essentially equivalent.
All this notwithstanding, the current feedback amplifier topology is interesting to think about. There are many applications where CFA chips are the best solution, especially where wide bandwidth and very high slew rate are required. To my thinking, the differences between VFAs and CFAs are topological rather than fundamental. In other words, choosing a CFA design is a matter of implementation tradeoffs.
The premise of the current turn of this thread is that specifications and theory take a back seat to listening performance. So, while Douglas Self and friends can argue convincingly that VFAs are the higher-performance solution, the proof of the pudding is in the eating, as Jeff "Dr. Low Mu" Medwin would say. I am thinking and posting about this now because it interests me, but I can't begin to tell you what it all means for my satisfaction when the needle hits the groove, metaphorically speaking.
Given infinite monkeys and typewriters, it would be interesting to build a version of the HPA-1 that differed only in having those two extra input stage JFETs, to see how converting it from CFA to VFA (complementary diff amps) would change the sound. Or we could substitute a bipolar diamond buffer in the input. Heck, we could do that, and replace the MOSFET output stage with the diamond buffer from my A1/A2. This would turn it into the "canonical" CFA circuit. It would be a completely different amplifier then, of course.
There is a long and contentious thread on the subject here, where people far smarter than me failed to reach any firm conclusions. I don't want to revisit that chaos.
There are four "canonical forms" of feedback, and CFAs use the very same form as VFAs. Some people are bothered (but not me, mostly) that the term "current feedback" has been appropriated from its classic meaning to describe this topology. In the older sense of the term, CFAs are precisely voltage feedback amplifiers.
The main advantage of CFAs seems to be that they can achieve very high slew rates. This is because the IPS output current comes externally via the feedback loop, instead of being constrained by the input differential pair tail current source. This is a boon to IC op-amp designers, where supply current and package dissipation are important considerations, and parts need to be designed for general applications. With a purpose-built discrete VFA design, we can get around these problems easily enough. IIRC, my A1 has a slew rate of 100 V/uS and the A2 measures 50 V/uS. Arguably, 5 V/uS is more than enough for audio work, and 200 V/uS (obtained with a CFA) is absurd overkill.
Another purported advantage of CFAs is that the closed-loop bandwidth is independent of the closed-loop gain G = (1 + Rf/Rg), where Rf and Rg are the values of the two feedback network resistors. This is true but misleading. In a CFA, the input stage transconductance, and hence the open-loop gain, depends on the value of feedback resistor Rg. In a VFA, open-loop gain does not depend on the values in the feedback network. So, if you increase the closed-loop gain of a CFA by lowering Rg, you raise the open-loop gain, and this compensates for the drop in bandwidth you would get doing the same experiment with a VFA. On the other hand, if you change the CFA's closed-loop gain by changing Rf, you will see the usual tradeoff between gain and bandwidth.
Looking at the input stage of the HPA-1, you could convert it to a VFA by the addition of two more JFETs buffering the feedback signal.
This would make it exactly into a complementary-differential input stage, e.g., as described by Cordell and many others. It's interesting to think about what effect this would have on both technical and subjective performance.
CFA proponents often talk about the ease of compensating these amplifiers, but they are just as constrained by output stage poles as VFAs. The difference is that CFAs tend to have lower open-loop gains, so require less compensation for stability. You can crank up the open-loop gain of a CFA and then you run into exactly the same problems with phase margin as with a VFA.
In general, aside from slew rate, CFAs tend to have lower performance specs than VFAs. And there are some newer integrated VFA chips that achieve slew rates and bandwidths comparable to CFAs.
There is some debate about what exactly defines a CFA. Some people think it's synonymous with having a diamond buffer input stage, but this isn't true. The HPA-1 doesn't use a diamond buffer, but is a true CFA. The actual definition is clear: The CFA inverting input is a low-impedance (ideally zero-impedance) node, and can source or sink current to maintain its voltage equal to that applied to the non-inverting input. In other words, in a CFA, there is a unity-gain buffer between the non-inverting and inverting terminals. In all other respects, CFAs and VFAs are conceptually identical.
This is a little counter-intuitive, that the inverting input is actually the output of a low-impedance buffer. It works because the transimpedance stage that follows is driven by the input/output current from that buffer.
In both the VFA and the CFA, the open-loop voltage gain is the product of the first stage transconductance and the second stage transimpedance. This is partly why some people argue the two amplifier types are essentially equivalent.
All this notwithstanding, the current feedback amplifier topology is interesting to think about. There are many applications where CFA chips are the best solution, especially where wide bandwidth and very high slew rate are required. To my thinking, the differences between VFAs and CFAs are topological rather than fundamental. In other words, choosing a CFA design is a matter of implementation tradeoffs.
The premise of the current turn of this thread is that specifications and theory take a back seat to listening performance. So, while Douglas Self and friends can argue convincingly that VFAs are the higher-performance solution, the proof of the pudding is in the eating, as Jeff "Dr. Low Mu" Medwin would say. I am thinking and posting about this now because it interests me, but I can't begin to tell you what it all means for my satisfaction when the needle hits the groove, metaphorically speaking.
Given infinite monkeys and typewriters, it would be interesting to build a version of the HPA-1 that differed only in having those two extra input stage JFETs, to see how converting it from CFA to VFA (complementary diff amps) would change the sound. Or we could substitute a bipolar diamond buffer in the input. Heck, we could do that, and replace the MOSFET output stage with the diamond buffer from my A1/A2. This would turn it into the "canonical" CFA circuit. It would be a completely different amplifier then, of course.
Lm6172 is a voltage feedback op amp achieving 3000v/us...pretty low noise and fantastically stable for op-amp swap on almost any pcb...its output can swing 9v into 100 ohms load too...works up to 100 Mhz...5...40 Megohm input impedance...
Henry,
I agree with you, about slew rates above 1 V/uS are barely detectable and above 5 or 6 V/vS becomes acedemic (unless you are a dog or bat 🙂) and can become a futile pursuit and can have siginificant down sides like (taken from another blog and I could not have stated better myself ......there are more ) Higher numbers are not necessarily better and can lead to problems.
Jam
I agree with you, about slew rates above 1 V/uS are barely detectable and above 5 or 6 V/vS becomes acedemic (unless you are a dog or bat 🙂) and can become a futile pursuit and can have siginificant down sides like (taken from another blog and I could not have stated better myself ......there are more ) Higher numbers are not necessarily better and can lead to problems.
- High slew rates can push design margin to the edge, decreasing amplifier stability.
- High slew rate can cause excessive high-frequency content, potentially damaging drivers. If the speakers cannot respond quickly enough, power turns to heat that can damage or destroy the drivers.
- The greater bandwidth required for high slew rates can cause high-frequency peaking and ringing, or even oscillation, with certain loads. Higher bandwidth also means more noise, and more (higher-frequency) distortion components.
Jam
Attachments
Last edited: