Assuming that your modelling program isn't going to fall over once more complex models are introduced, it needs fuller models. Practical measurements are very rarely as squeaky-clean as simulations, but your results look so smooth, which makes me wonder. Are you certain the simulation is doing what you ask it to? If you can make a plausible model work, I've probably got some numbers to throw at it...
No, for the psychoacoustician. You've got to give the physicist some solid material to start with.
runeight, it might be fun to include some sample ground and trace resistances, and maybe see if there's any effect from the protection components necessary to make a real-world regulator work more than one time.
I am certain that the simulators are doing what we think they're doing given the perfect component models. Both of these programs are excellent simulators. The commercial version of PSpice (I have the demo) is widely used in the electronics industry for both analog and digital design. There are probably thousands of models for PSpice. There are a handful of vacuum tube models created by folks like us.
We can introduce almost anything into the simulation, including stay capacitances, ESRs, ESLs, etc.
There are several things that the simulators don't do well. One of them is noise creation. At some point the simulator is dealing very accurately with voltages and currents that are below any real noise floor. So, when the simulator shows a -200db line rejection ratio, I think this simply means that the signal is "in the noise".
Another thing they don't handle well is temperature, although there are temperature-based models. And they don't handle component drift at all.
I have played with the simulations using LRC models for the electrolytics. There isn't much change when I do this, even if I use inductances that are artificially large.
I've noticed too, that the simulations can be very accurate. For example, I have a program called TubeCad that uses the basic tube equations to calculate the behavior of various topologies for various tubes. If I simulate the same topologies with the same tubes I get almost identical results. I've simulated several of Bruce Rozenblit's public designs and get the results that he claims for his circuits. etc.
So, I think this means that you guys are correct. Not enough of the relevant features that effect the regulators' behavior are being accounted for.
If you can suggest where to add components that represent real, but unaccounted for features (like ESR for small caps), I'll give it a try.
We can introduce almost anything into the simulation, including stay capacitances, ESRs, ESLs, etc.
There are several things that the simulators don't do well. One of them is noise creation. At some point the simulator is dealing very accurately with voltages and currents that are below any real noise floor. So, when the simulator shows a -200db line rejection ratio, I think this simply means that the signal is "in the noise".
Another thing they don't handle well is temperature, although there are temperature-based models. And they don't handle component drift at all.
I have played with the simulations using LRC models for the electrolytics. There isn't much change when I do this, even if I use inductances that are artificially large.
I've noticed too, that the simulations can be very accurate. For example, I have a program called TubeCad that uses the basic tube equations to calculate the behavior of various topologies for various tubes. If I simulate the same topologies with the same tubes I get almost identical results. I've simulated several of Bruce Rozenblit's public designs and get the results that he claims for his circuits. etc.
So, I think this means that you guys are correct. Not enough of the relevant features that effect the regulators' behavior are being accounted for.
If you can suggest where to add components that represent real, but unaccounted for features (like ESR for small caps), I'll give it a try.
Better Models
Here is the BJT regulator with the addition of lumped circuit elements to represent more realistic components. I’ve used an LRC model for the electrolytics with guesses for the values of ESR and ESL. I’ve put 0.5pf shunt caps on resistors and 1R ESR on plastic caps. The zener/transistor models already have junction capacitance, etc. built in. The simulation schematic looks like this:
D1 doesn't look like a zener, but it is.
And here is the line noise rejection plot in DB:
This is different, but not much different from the “perfect” version.
One additional thing that can be done is to take into account the resistance of the wire. But, first we have to decide how the thing is actually wired. This may make the most difference as it will lift some of the points in the circuit above perfect ground. If you all have any further suggestions, I will try them. But at the moment, I’m beginning to believe that this may be more accurate than we think.
Here is the BJT regulator with the addition of lumped circuit elements to represent more realistic components. I’ve used an LRC model for the electrolytics with guesses for the values of ESR and ESL. I’ve put 0.5pf shunt caps on resistors and 1R ESR on plastic caps. The zener/transistor models already have junction capacitance, etc. built in. The simulation schematic looks like this:
An externally hosted image should be here but it was not working when we last tested it.
D1 doesn't look like a zener, but it is.
And here is the line noise rejection plot in DB:
An externally hosted image should be here but it was not working when we last tested it.
This is different, but not much different from the “perfect” version.
One additional thing that can be done is to take into account the resistance of the wire. But, first we have to decide how the thing is actually wired. This may make the most difference as it will lift some of the points in the circuit above perfect ground. If you all have any further suggestions, I will try them. But at the moment, I’m beginning to believe that this may be more accurate than we think.
Thanks Dave. What tool did you use to do that?
To decode this, the worst curve (Line Small CR) is a pure RC filter with only 10u caps. As expected, this configruation shows the worst line rejection.
The best curve is the tube series regulator with large caps (Line Series Large CR). It shows the highest line rejection.
In order of quality from lowest rejection to highest rejection we have -- RC only, tube shunt regulator, tube series regulator, and SS series regulator (one curve that is missing is the SS series regulator with large caps, it's the best).
So, I guess one question is why not use an SS regulator? It's cheap, uses only a few components, and works better than the others (at least on paper).
One other thing to notice is that when I added the lumped circuit elements for the Rs and Cs, the frequency response above 100KHz changed (the bold yellow curve called SS Series Lumped). This is because with better models, the regulator doesn't reject the high frequencies as well.
BTW, can anyone tell me how to make a negative supply voltage series pass tube regulator? My first guess is that it's not simply a positive regulator with the ground moved to another point in the circuit. Thanks.
To decode this, the worst curve (Line Small CR) is a pure RC filter with only 10u caps. As expected, this configruation shows the worst line rejection.
The best curve is the tube series regulator with large caps (Line Series Large CR). It shows the highest line rejection.
In order of quality from lowest rejection to highest rejection we have -- RC only, tube shunt regulator, tube series regulator, and SS series regulator (one curve that is missing is the SS series regulator with large caps, it's the best).
So, I guess one question is why not use an SS regulator? It's cheap, uses only a few components, and works better than the others (at least on paper).
One other thing to notice is that when I added the lumped circuit elements for the Rs and Cs, the frequency response above 100KHz changed (the bold yellow curve called SS Series Lumped). This is because with better models, the regulator doesn't reject the high frequencies as well.
BTW, can anyone tell me how to make a negative supply voltage series pass tube regulator? My first guess is that it's not simply a positive regulator with the ground moved to another point in the circuit. Thanks.
runeight said:
PhotoShop
Another question to ask is what does the reg do if it gets a burst of current from the circuit?
The Vacuum State SuperReg is a good example.
dave
Hi,
Major factors to series regulator performance are, in order of importance ( to me) :
The properties of the series pass device: I like to have high perveance here.
6CW5/EL86 performs better than a 6BQ5/EL84, for instance.
High µ for the comparator (error amp) triode(s), a µ of 140 for each triode is better than a µ of 100.
A two stage comparator performs better than a single stage one.
I much prefer a glow discharge tube over any zener except the 5.6V ones which are useless in this application anyway.
An example of a reg. you may want to run through PSpice:
The 12AX7A replaces a ECC807 which has becomes extremely hard to get. (µ 140 per triode section).
Other triodes with a µ of 140 exist but are just as hard to get, some of them being nuvistors.
The cap behind the reg. was 660 µF/385VDC, the reg is set for 305VDC output originally.
This makes the cap dominant leaving the reg as a trickle charger that quite like never has to work.
The stage fed by the PSU is a White CF using a 12BH7A.
(Which I'll post next)
It would be interesting to see the influence of the cap and the influence of the reg on the WCF.
Cheers,
Major factors to series regulator performance are, in order of importance ( to me) :
The properties of the series pass device: I like to have high perveance here.
6CW5/EL86 performs better than a 6BQ5/EL84, for instance.
High µ for the comparator (error amp) triode(s), a µ of 140 for each triode is better than a µ of 100.
A two stage comparator performs better than a single stage one.
I much prefer a glow discharge tube over any zener except the 5.6V ones which are useless in this application anyway.
An example of a reg. you may want to run through PSpice:
The 12AX7A replaces a ECC807 which has becomes extremely hard to get. (µ 140 per triode section).
Other triodes with a µ of 140 exist but are just as hard to get, some of them being nuvistors.
The cap behind the reg. was 660 µF/385VDC, the reg is set for 305VDC output originally.
This makes the cap dominant leaving the reg as a trickle charger that quite like never has to work.
The stage fed by the PSU is a White CF using a 12BH7A.
(Which I'll post next)
It would be interesting to see the influence of the cap and the influence of the reg on the WCF.
Cheers,
Frank, I'll be happy to simulate this to see what we get. There will be only one small problem. I don't have a model for the 6CW5. I do have one for the 6BQ5 and if you don't mind my using that, I'll give it a try.
I don't think I can get to it tonight, but I'll set it up in the next day or two.
I don't think I can get to it tonight, but I'll set it up in the next day or two.
Frank,
I set this up over my lunch hour. Instead of the EL84 I used a 6KG6, which is a high perveance pentode too. The only plate curves I could find, however, were for the PL509. These curves at Vg2=170V look very similar to the 6CW5 to me. All this to say that I'm not using the correct tube and results need to be understood with this in mind.
Here is the simulation schematic:
I've shown the currents for each half of the 12AX7 on the diagram. Notice that the current for the upper triode is only about 25uA.
I've swept the current source and here is the R of the regulator as it is drawn:
There is a spike around 45KHz.
If I change the 1M resistor to 33K, then the current in the upper triode goes to about 750uA (I think this is much better) and the lower triode sees about 1.3mA.
Here is the equivalent R for this regulator:
This looks like a better curve to me. I'll do the line side analysis later tonight.
However, given that the WC is a class A device and it has two big caps at its plate, I suspect that the regulator won't have much effect on it. But, I'll try it anyway.
BTW, when I replace the 6KG6 with an EL84, I get the same spike, although the resistance of the EL84 give it some width.
Hope this is useful.
I set this up over my lunch hour. Instead of the EL84 I used a 6KG6, which is a high perveance pentode too. The only plate curves I could find, however, were for the PL509. These curves at Vg2=170V look very similar to the 6CW5 to me. All this to say that I'm not using the correct tube and results need to be understood with this in mind.
Here is the simulation schematic:
An externally hosted image should be here but it was not working when we last tested it.
I've shown the currents for each half of the 12AX7 on the diagram. Notice that the current for the upper triode is only about 25uA.
I've swept the current source and here is the R of the regulator as it is drawn:
An externally hosted image should be here but it was not working when we last tested it.
There is a spike around 45KHz.
If I change the 1M resistor to 33K, then the current in the upper triode goes to about 750uA (I think this is much better) and the lower triode sees about 1.3mA.
Here is the equivalent R for this regulator:
An externally hosted image should be here but it was not working when we last tested it.
This looks like a better curve to me. I'll do the line side analysis later tonight.
However, given that the WC is a class A device and it has two big caps at its plate, I suspect that the regulator won't have much effect on it. But, I'll try it anyway.
BTW, when I replace the 6KG6 with an EL84, I get the same spike, although the resistance of the EL84 give it some width.
Hope this is useful.
This was easy enough to do so I took 5mins to upload these.
These are the load side rejection ratios. First one is with 1M plate resistor, second one is with 33k plate resistor.
I spoke too soon before. The 1M version does have better characterstics except for the spike (which can be taken out by some other way).
These are the load side rejection ratios. First one is with 1M plate resistor, second one is with 33k plate resistor.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
I spoke too soon before. The 1M version does have better characterstics except for the spike (which can be taken out by some other way).
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.