For those who are thinking about building their own customized version of a Super Regulator, which does NOT use the diyAudio Store PCB . . . . AND which happens to output a voltage Vout_reg that is at least 4.5 volts greater than Vref . . . . then my little experiment yesterday might be of interest. I'm fooling around with preliminary ideas for a Super Regulator that outputs Vout_reg = 15.0 volts and uses a Vref around 7 volts. Since (15 - 7) is comfortably greater than 4.5 volts, yesterday's experiment is applicable to my situation.
However I realize that many builders want a Super Regulator with output voltage waaaaay less than 15 volts, and therefore (Vout - Vref) is waaaay less than 4.5 volts. So the little contrivance shown here is not useful for them and I am very sorry.
Figures 1 and 2 (attached below) show the basic idea. Resistors R5 and R12 on the official Store schematic, provide bias current to the super excellent Zeners-On-Steroids D5 and D10. I plan to replace R5 and R12 (on a custom PCB not the Store PCB) with constant current sources. Indeed, with cascoded constant current sources.
I built the Figure 1 circuit on a protoboard, and connected a potentiometer in the R1 position. Then I dialled the knob up and down until the current was 3 milliamps. It happened to require 260 ohms for this particular J113 JFET; no doubt other J113's will require different resistances than 260 ohms to give 3 mA. That's an annoying reality of cheap analog switch JFETs like the J113 -- their analog properties vary dramatically. Fortunately I had two 130 ohm resistors in my box, so with them in series I was able to achieve 260 ohms. That gave 3 mA for this particular J113.
When I attached the Figure 1 circuit to the "Locky Z Curve Tracer" I got the plot in Figure 3. The trace is just a single J113 current source below ~ 4 volts on the plot. Then the magic of cascoding takes over, the J112 starts to help out, and suddenly the plot goes flatline above ~ 4 volts.
The LockyZ curve tracer can barf out its measured data as an ascii text file, so I did that and slurped the X,Y datapoints into a spreadsheet. That produced the plot in Figure 4. It's just like Figure 3 except we're not using LockyZ's hard-to-reconfigure plot software any more; now we're using a spreadsheet. Again you can see the single-JFET-only current source behavior below ~ 4 volts, transitioning to J112+J113 cascode current source above 4 volts.
Zooming in the plot view, we get Figure 5. Zooming in even more, we get Figure 6. It's plain and evident that the DACs and ADCs inside the LockyZ Curve Tracer do not have unlimited precision - - - but still, the shape of the plot tells a story. And it's a happy story.
Figure 7 is the same as Figure 6 except I have drawn a straight line among the measured data, deliberately exaggerating the slope (output impedance) of the data. It's a conservative under-estimate of the output impedance of this cascoded current source. I encourage others to draw their own straight line upon Figure 6 and calculate their own slope (output impedance). It will be fun!
Conclusion: by replacing one component (4.99K resistor R5) with a three component CCS, I was able to increase the impedance of the Vref bias network by a factor of 320. (math: 1.6E6 / 5E3 = 320). This increases isolation between the regulated output and the Vref; it also opens the door to using different Vref components which may not be as good as the venerable but expensive LM329. Ahem.
_
However I realize that many builders want a Super Regulator with output voltage waaaaay less than 15 volts, and therefore (Vout - Vref) is waaaay less than 4.5 volts. So the little contrivance shown here is not useful for them and I am very sorry.
Figures 1 and 2 (attached below) show the basic idea. Resistors R5 and R12 on the official Store schematic, provide bias current to the super excellent Zeners-On-Steroids D5 and D10. I plan to replace R5 and R12 (on a custom PCB not the Store PCB) with constant current sources. Indeed, with cascoded constant current sources.
I built the Figure 1 circuit on a protoboard, and connected a potentiometer in the R1 position. Then I dialled the knob up and down until the current was 3 milliamps. It happened to require 260 ohms for this particular J113 JFET; no doubt other J113's will require different resistances than 260 ohms to give 3 mA. That's an annoying reality of cheap analog switch JFETs like the J113 -- their analog properties vary dramatically. Fortunately I had two 130 ohm resistors in my box, so with them in series I was able to achieve 260 ohms. That gave 3 mA for this particular J113.
When I attached the Figure 1 circuit to the "Locky Z Curve Tracer" I got the plot in Figure 3. The trace is just a single J113 current source below ~ 4 volts on the plot. Then the magic of cascoding takes over, the J112 starts to help out, and suddenly the plot goes flatline above ~ 4 volts.
The LockyZ curve tracer can barf out its measured data as an ascii text file, so I did that and slurped the X,Y datapoints into a spreadsheet. That produced the plot in Figure 4. It's just like Figure 3 except we're not using LockyZ's hard-to-reconfigure plot software any more; now we're using a spreadsheet. Again you can see the single-JFET-only current source behavior below ~ 4 volts, transitioning to J112+J113 cascode current source above 4 volts.
Zooming in the plot view, we get Figure 5. Zooming in even more, we get Figure 6. It's plain and evident that the DACs and ADCs inside the LockyZ Curve Tracer do not have unlimited precision - - - but still, the shape of the plot tells a story. And it's a happy story.
Figure 7 is the same as Figure 6 except I have drawn a straight line among the measured data, deliberately exaggerating the slope (output impedance) of the data. It's a conservative under-estimate of the output impedance of this cascoded current source. I encourage others to draw their own straight line upon Figure 6 and calculate their own slope (output impedance). It will be fun!
Conclusion: by replacing one component (4.99K resistor R5) with a three component CCS, I was able to increase the impedance of the Vref bias network by a factor of 320. (math: 1.6E6 / 5E3 = 320). This increases isolation between the regulated output and the Vref; it also opens the door to using different Vref components which may not be as good as the venerable but expensive LM329. Ahem.
_
Attachments
The prices, ouch! I bought a bag of 100 from DK a century ago at a fraction of that price. There are other regulators from ADI with better noise and drift stats -- but not TO92 (and you'll learn the pleasure of soldering MSSOP).
Hello all,
If I am targeting +9-0-9V from super regulator I plan to build and had some query on the input side. Can I short the pos return and neg return pins on the input of super regulator board? Since the final connection will have a common "zero" or ground, I think it should be ok but wanted someone to confirm the same. Please let me know.
See the attached picture for reference.
Thanks
Balaji
If I am targeting +9-0-9V from super regulator I plan to build and had some query on the input side. Can I short the pos return and neg return pins on the input of super regulator board? Since the final connection will have a common "zero" or ground, I think it should be ok but wanted someone to confirm the same. Please let me know.
See the attached picture for reference.
Thanks
Balaji
