Yes, the high dropout of circa 3.5v or so was going to mean a lot of heat dissipation at peak spec load - for any of the 3 rails. Hence the focus on lowering Vdrop as best as possible (level shift etc). Worst case was dropping about 9V aka 45W for each rail. Hence mounting on the sidewall of a 120x400mm heatsink-walled enclosure. Of course worse case is worse case.
I took a look at the circuit in #159. You must have a rather different TL431 model to me as I could not replicate your output (even when stripping out ESR for the caps so that all components were plain vanilla LTspice native). In any event I still had problems applying it to also provide current for the level shift. I will have another look in the morning when I am less tired.
This is becoming a challenge for my health
although I feel like I am going around the traps of a large number of the things I need to learn. I'll take a look at Darlington BJT as well although I thought the issue with a BJT pass device was that the op amp had to sink a lot of current (although presumably less with a Darlington pair).
I took a look at the circuit in #159. You must have a rather different TL431 model to me as I could not replicate your output (even when stripping out ESR for the caps so that all components were plain vanilla LTspice native). In any event I still had problems applying it to also provide current for the level shift. I will have another look in the morning when I am less tired.
This is becoming a challenge for my health
although I feel like I am going around the traps of a large number of the things I need to learn. I'll take a look at Darlington BJT as well although I thought the issue with a BJT pass device was that the op amp had to sink a lot of current (although presumably less with a Darlington pair).Attachments
Last edited:
Another option is discussed in post #84. Let Linear Technology do the heavy lifting. Overcurrent shutdown and thermal shutdown are built into their ICs. If you've got plenty of input voltage, put two LT integrated circuits in series and gloat about your excellent regulation specs. If you don't, use one and lick your wounds.
The people who built these ICs, had (proprietary) spice models that were extremely accurate AND that tracked the tolerance guardbands of FabProcess, VoltageVariation, and TemperatureVariation (the holy trinity PVT) 800% more accurately than you, a lowly customer, can ever hope to realistically achieve. Let the pros do what the pros do, for a living.
The people who built these ICs, had (proprietary) spice models that were extremely accurate AND that tracked the tolerance guardbands of FabProcess, VoltageVariation, and TemperatureVariation (the holy trinity PVT) 800% more accurately than you, a lowly customer, can ever hope to realistically achieve. Let the pros do what the pros do, for a living.
Yup, although:
- I thought I was getting close to being done with the MFET pass regulator
- two LT1084 still require 3.5 volts or more of headroom in order to not degrade PSRR, so about the same as the MFET setup from a heat dissipation point of view
Agree on the other points; certainly easier to implement.
One general question regarding PSRR: obviously the focus is on the mains ripple frequency, in my case 100Hz. But what about the other frequencies? How much should one care about levels of PSRR out to, say, 10kHz or 100kHz? (Nice to have but not worth worrying about, get it if you can easily or strive hard for it?)
- I thought I was getting close to being done with the MFET pass regulator
- two LT1084 still require 3.5 volts or more of headroom in order to not degrade PSRR, so about the same as the MFET setup from a heat dissipation point of view
Agree on the other points; certainly easier to implement.
One general question regarding PSRR: obviously the focus is on the mains ripple frequency, in my case 100Hz. But what about the other frequencies? How much should one care about levels of PSRR out to, say, 10kHz or 100kHz? (Nice to have but not worth worrying about, get it if you can easily or strive hard for it?)
Last edited:
I got lost in much of the technical details so I'm not sure this is going to be helpful.
The interference and harmonics of the mains supply seem to go way beyond 100kHz and probably beyond 1GHz.
PSRR above 10kHz is important because these higher frequencies interfere with the wanted audio.
If a reg/PSU allows most of the HF stuff to pass through without much attenuation, then there is a risk that the audio processing will be adversely affected.
Two cascaded PSU/regs each with poor attenuation of the above audio interference still leaves the supply contaminated with the HF stuff.
Somewhere in the cascade of stages, good HF attenuation is required. Passive with appropriate components with appropriate screening may be as good as, or maybe better, than adding another active stage with poor HF attenuation.
The interference and harmonics of the mains supply seem to go way beyond 100kHz and probably beyond 1GHz.
PSRR above 10kHz is important because these higher frequencies interfere with the wanted audio.
If a reg/PSU allows most of the HF stuff to pass through without much attenuation, then there is a risk that the audio processing will be adversely affected.
Two cascaded PSU/regs each with poor attenuation of the above audio interference still leaves the supply contaminated with the HF stuff.
Somewhere in the cascade of stages, good HF attenuation is required. Passive with appropriate components with appropriate screening may be as good as, or maybe better, than adding another active stage with poor HF attenuation.
Hi. No, I mean in the context of in-rush. (I have one of Mark's Quasimodos for optimising snubbers.) I can use ratios of Lp and Ls to get secondary voltages for all my simulation work, but when it comes to in-rush, as I understand it, the transformers' magnetic fluxes need to charge (plus caps etc) and this is a function of their inductance. So here it is the absolute value of inductance that becomes important. Hence to properly set up my in-rush limiter I need a reasonable estimate of transformer inductance. Or do I have this wrong?
Thanks... Current depends on size of transformer and filter caps; the larger these are the bigger the in-rush as they initially charge. To understand what resistance you want to place in the path of this in-rush to slow the charging current to a desirable level, surely you have to have a sense of the size of these two. Caps are easy. How do you determine the charging profile of the transformer without an understanding of its inductance?
I suppose if you got really desperate, you could make some assumptions and apply some engineering mathematics. An illustration of this type of thinking is shown below. It requires no special knowledge beforehand, merely a willingness to think.
FIRST write down the transformer's Volt-Ampere rating. For fun let us assume 500 VA, which gives a very comfortable safety margin for (12 + 5 + 3.3) x 5 amperes.
SECOND decide the input voltage and input frequency. For fun let us assume 230VAC and 50 Hertz.
THIRD calculate the AC current flowing in the primary when max-rated-VA power is delivered to the load on the secondary. 500VA / 230V = 2.17 amperes.
FOURTH make a wild guess estimate (in Silicon Valley we call this "chosen by rectal extraction") of the ratio between good primary current and bad primary current. Good primary current provides the power that is delivered to the load on the secondary. Bad primary current arises because the inductance of the primary is not infinity. With no load on the secondary, the current flowing in the primary is NOT zero; instead it is I = V / Z where Z is the impedance of the primary (2 * pi * freq * Lprimary). If Lprimary is infinity, Iprimary is zero. If Lprimary is finte, Iprimary is greater than zero. This is "bad primary current".
Let us assume the transformer maker has decided that it is okay to "waste" or "squander" 2% of the primary current because the inductance is not infinite. We are assuming the ratio of good primary current to bad primary current is 50 to 1. 100 to 2, i.e., 2%.
Then our assumed bad primary current is 2% of 2.17 amperes, which is 43.4 milliamps.
FIFTH calculate the primary inductance. I = V / Z, and Z = (2 * pi * freq * L), thus 43.4mA = 230V / (2 * pi * 50Hz * L). Working out the arithmetic, L = 17 henrys.
It is pleasant to observe that this number is near the 20 henrys used by Rod Elliott to model a 500VA transformer primary in a DC-blocker appliation (link 1).
I will also remark that the Antek transformer company (recommended by Nelson Pass for several of his high current Class-A DIY projects) puts measured data on their website which allows fearless readers to make a rough estimate of the ratio: good primary current vs bad primary current. No I will not quote it chapter and verse, nor tell you how to manipulate the buttons on your calculator to interpret it. If such things are truly important to you, you'll figure it out all by yourself. If not, at least it didn't cost any money to read this paragraph.
It will be amusing to see whether the (silly) phrase "bad primary current" starts popping up elsewhere, spouted imperiously by self-important loudmouths, as though they invented it themselves or derived it from first principles. Same for the rectally extracted 2% squanderous waste assumption. Let's watch.
FIRST write down the transformer's Volt-Ampere rating. For fun let us assume 500 VA, which gives a very comfortable safety margin for (12 + 5 + 3.3) x 5 amperes.
SECOND decide the input voltage and input frequency. For fun let us assume 230VAC and 50 Hertz.
THIRD calculate the AC current flowing in the primary when max-rated-VA power is delivered to the load on the secondary. 500VA / 230V = 2.17 amperes.
FOURTH make a wild guess estimate (in Silicon Valley we call this "chosen by rectal extraction") of the ratio between good primary current and bad primary current. Good primary current provides the power that is delivered to the load on the secondary. Bad primary current arises because the inductance of the primary is not infinity. With no load on the secondary, the current flowing in the primary is NOT zero; instead it is I = V / Z where Z is the impedance of the primary (2 * pi * freq * Lprimary). If Lprimary is infinity, Iprimary is zero. If Lprimary is finte, Iprimary is greater than zero. This is "bad primary current".
Let us assume the transformer maker has decided that it is okay to "waste" or "squander" 2% of the primary current because the inductance is not infinite. We are assuming the ratio of good primary current to bad primary current is 50 to 1. 100 to 2, i.e., 2%.
Then our assumed bad primary current is 2% of 2.17 amperes, which is 43.4 milliamps.
FIFTH calculate the primary inductance. I = V / Z, and Z = (2 * pi * freq * L), thus 43.4mA = 230V / (2 * pi * 50Hz * L). Working out the arithmetic, L = 17 henrys.
It is pleasant to observe that this number is near the 20 henrys used by Rod Elliott to model a 500VA transformer primary in a DC-blocker appliation (link 1).
I will also remark that the Antek transformer company (recommended by Nelson Pass for several of his high current Class-A DIY projects) puts measured data on their website which allows fearless readers to make a rough estimate of the ratio: good primary current vs bad primary current. No I will not quote it chapter and verse, nor tell you how to manipulate the buttons on your calculator to interpret it. If such things are truly important to you, you'll figure it out all by yourself. If not, at least it didn't cost any money to read this paragraph.
It will be amusing to see whether the (silly) phrase "bad primary current" starts popping up elsewhere, spouted imperiously by self-important loudmouths, as though they invented it themselves or derived it from first principles. Same for the rectally extracted 2% squanderous waste assumption. Let's watch.
One of your most serious problems is going to be: how to employ the AD797 opamp (which requires at least 10 volts between Vcc and Vee) in your 3.3 volt regulator.
One option is to use the wonderful Linear Technology LT1498 opamp in all three regulators, since its supply voltage range extends from 2.2V to 36V (!!) and its other specs are reasonably nice. But this would mean abandoning the AD797 and you seem to be enamored of it.
Another option is to use 3 different opamps in the 3 different applications; use the "best available" opamp for each supply. And then hope they are pin compatible. And then hope that the frequency compensation schemes for each one, can be mapped into a common PCB footprint.
Well, here is yet one more option: provide a low current "super voltage" supply to each board, which is at least 10 volts higher than that board's regulated output voltage. Now you've got 13.3 volts, 15 volts, and 22 volts. You can use the exact same AD797 opamp, with the exact same frequency compensation, in all three boards. Another benefit: no level shifter! Even with the non-rail-to-rail AD797 driving a high-threshold-voltage enhancement mode MOSFET, the opamp output pin can swing higher than necessary.
You could set Vsupervoltage = 15V for both the 3.3V and the 5.0V boards. You could set Vsupervoltage = 24V for the 12V board.
You could generate these with a little dinky auxiliary transformer (24VAC @ 10 volt-amperes), an LM7824 for +24V, and an LM7815 for +15V. Small and simple.
Conceptual schematic below.
_
One option is to use the wonderful Linear Technology LT1498 opamp in all three regulators, since its supply voltage range extends from 2.2V to 36V (!!) and its other specs are reasonably nice. But this would mean abandoning the AD797 and you seem to be enamored of it.
Another option is to use 3 different opamps in the 3 different applications; use the "best available" opamp for each supply. And then hope they are pin compatible. And then hope that the frequency compensation schemes for each one, can be mapped into a common PCB footprint.
Well, here is yet one more option: provide a low current "super voltage" supply to each board, which is at least 10 volts higher than that board's regulated output voltage. Now you've got 13.3 volts, 15 volts, and 22 volts. You can use the exact same AD797 opamp, with the exact same frequency compensation, in all three boards. Another benefit: no level shifter! Even with the non-rail-to-rail AD797 driving a high-threshold-voltage enhancement mode MOSFET, the opamp output pin can swing higher than necessary.
You could set Vsupervoltage = 15V for both the 3.3V and the 5.0V boards. You could set Vsupervoltage = 24V for the 12V board.
You could generate these with a little dinky auxiliary transformer (24VAC @ 10 volt-amperes), an LM7824 for +24V, and an LM7815 for +15V. Small and simple.
Conceptual schematic below.
_
Attachments
Mark, you're pretty darned good at explaining stuff in a way that sticks....thank you
I agree! Thanks Mark.
Thanks again Mark. At the moment I am feeling rather beaten in this fight and the dual LT1084 has gained a very high level of appeal.
This afternoon I have been trying to understand the implementation of remote sense shown in the data sheet.
'Normally' one samples the output via a resistor divider and feeds that into the inverting pin of an op amp, the output of which drives the pass the device. The LT1084 standalone does similar with the resistor divider feeding the Adj pin. The above all seems a bit proverbial-backwards... I'm particularly struggling to understand how the Adj pin can sense the (lower) voltage at the load when the upper resistor in the divider is connected directly to the Out pin...
This afternoon I have been trying to understand the implementation of remote sense shown in the data sheet.
An externally hosted image should be here but it was not working when we last tested it.
'Normally' one samples the output via a resistor divider and feeds that into the inverting pin of an op amp, the output of which drives the pass the device. The LT1084 standalone does similar with the resistor divider feeding the Adj pin. The above all seems a bit proverbial-backwards... I'm particularly struggling to understand how the Adj pin can sense the (lower) voltage at the load when the upper resistor in the divider is connected directly to the Out pin...
Fortunately, post #92 in this thread offers some advice: "Sprinkle in some ferrite beads, with or without resonance killing resistors"
I plunked down a few of the beads whose models are built-in, presupplied with LTSPICE, interspersed among SGK's circuitry that sits between the rectifier bridge and the input to his (first) regulator. Simulator results attached. The beads seem to do a lot of good below 200 MHz or so. For attenuation above 200 MHz he will need to find a way to shunt out those equivalent series inductances; the standard ploy is to install SMD ceramic bypass caps. I'd start by putting one at the downstream side of each resistor.
_
Attachments
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.