Wenzel finesse voltage regulator (smoothing voltage regulator output)

[ftransform]

So I was interested in making a low noise dc power supply. I came across the following website:
http://www.wenzel.com/documents/finesse.html

I built a linear regulator on a breadboard with the following configuration:

12-0-12 Transformer > Diode Bridge (with 2200uF smoothing)  > LM317 voltage regulator (with 10uF/100n on output, 10uF on adjust, 100n on input + protection diodes) > figure 3 from the website.

I used a LT1001 as a replacement for the LM883 and a 2n2222 instead of a 2n4401 and I used a 100 ohm 26 turn trimpot as a replacement for the 15 ohm resistor (for trimming). Bourns brand.


Is this type of circuit testable on a breadboard, with such low noise levels?
When I measure it the LM317 output gives me a noise reading of 0.002mV rms while the output of the wenzel shunt is 0.003mV rms, which is slightly worse. Measurements made by a agilent 34401a.

Is the lt1001 responsible for this? The slew rate and gbp are 1/20th of the lm833, a direct contradiction to the recommendations given on the wenzel website.. I do not have any large bandwidth high slew rate amplifiers so I decided to just throw it together using "inappropriate parts" for fun (lol new years!).

I have a more suitable op-amp coming in the mail but do you think there will be a detectable difference between the output and input on a breadboard? Or must circuits of this type be soldered with solid ground planes before they work?

Or is the transistor replacement choice inappropriate? I am not sure what transistor parameters are important for this type of circuit. I tried a s9018 with little difference.

I never experimented with low-noise circuitry so any tips are appreciated.

[SeanB]

Make the board, you need the low resistance and low impedance of fat tracks along with the shielding offered bt copper pours.

[ftransform]

copper pours?

Also, what type of capacitors are recommended? Are tantalums recommended for this job?

Right now I am using regular electrolytics from a midway manufacturer.

[SeanB]

Multiple parallel caps, along with a decent LC filter to the regulator to reduce noise on the input that will show on the output reduced by the PSRR of the regulator. Big areas of copper ground plane to make sure common is actually common.

[ftransform]

Multiple parallel caps, along with a decent LC filter to the regulator to reduce noise on the input that will show on the output reduced by the PSRR of the regulator. Big areas of copper ground plane to make sure common is actually common.

In which part should I include parallel capacitors? Everywhere? Or only the 2200 uH ones? 100uh ones? 10uh ones?

[Shuggsy]

In which part should I include parallel capacitors? Everywhere? Or only the 2200 uH ones? 100uh ones? 10uh ones?

All the big bypass caps (the 100uH and 22uH might benefit from the combined parallel capacitors... each smaller value will perform a little better at higher frequencies. Capacitor physical size also makes a difference at higher frequencies. See this app note from Intersil for some further explanation on why smaller capacitance values and smaller capacitors can make a difference at high frequencies. I have a feeling that the stray capacitance on breadboards may start to dominate for the higher frequencies though.

The LT1001 also has 2-4 times the input noise voltage and, as you said, about 1/20 of the LM833 they used. You should see some performance gains with a better opamp provided the opamp is the biggest source of error/noise in the circuit at the moment.

I'd say that the biggest performance gain will be accomplished through building the circuit on a copper-clad PCB. Keep components close to the PCB to make the most of the big common ground plane you have and minimize lead lengths between components to reduce the inductance created between components. Any signal paths with lower-frequency content don't need to be so close to the ground plane as the benefits will be extremely small -- the two 10K resistors connected to the non-inverting input of the opamp would be an example.

[ftransform]

Sounds good. The lt1001 has lower offset currents and better CMRR and PSRR and a few other better specs but I guess that the slew rate and gbp dominates in this application?

Since the capacitors are doubled up I will have to half the values I guess.

What is the "formula" for multiple capacitors? When do diminishing returns occur from parallel capacitors? I.e. 5 22uF in parallel as a replacement for a 100uF seems ridiculous and 2 50uF seems more correct.. or is my assumption false?

And how about adding 100 nanofarad capacitors in parallel with the electrolytics in wenzel.com circuit? Would this not help too? Same as is done for the LM317 circuit.

I'm going to try testing with a different load on the circuit.

[codeboy2k]

breadboard it Manhattan style on a piece of blank copper clad fr4, and you'll get a better idea of your power supply's performance before you go to a PCB.

If it's a one-off, you can just run with the Manhattan breadboard as your finished product.  This guy does great stuff here;

https://aa7ee.wordpress.com/tag/manhattan-construction/

[ftransform]

breadboard it Manhattan style on a piece of blank copper clad fr4, and you'll get a better idea of your power supply's performance before you go to a PCB.

If it's a one-off, you can just run with the Manhattan breadboard as your finished product.  This guy does great stuff here;

https://aa7ee.wordpress.com/tag/manhattan-construction/

Do I really need to do crazy ass shit like that to test a DC power supply? That looks like radio-grade stuff?
The most I ever did is put a big grounded metal object near my breadboard to reduce noise.

Making a whole PCB is easier then that..

[codeboy2k]

Do I really need to do crazy ass shit like that to test a DC power supply? That looks like radio-grade stuff?
The most I ever did is put a big grounded metal object near my breadboard to reduce noise.
Making a whole PCB is easier then that..

I like to build stuff on copper clad like that guy, although for my breadboards I don't do it nearly as pretty and I often do dead bug style or flying connections in the air without going to a pad.

Yes, that guy is doing RF , and a power supply is DC more or less. You don't need to do it like his work of art, that's for sure, but you will get better noise performance from your PS when its on copper clad, and if you are incorporating amplifiers and shunting away noise  like that on that web page, then it seems like you would want to try to breadboard it as best you can to verify that you get what you want from it, and have an opportunity to tweak it as needed, before you commit to a PCB.

I work from home, so for me a ordering a PCB is a delay on my project, and a personal expense, or perhaps an unnecessary expense for my client.  For others who work in corporate environments building the next project that will sell 10 million units, a PCB **IS** the breadboard, and it's not unusual to go through 3,4,5 iterations before it's finalized.

[robrenz]

From this link from this thread you are expecting to much from your 34401A.  I am not dissing your meter it just does not give accurate rms measurements at the levels you are trying to use.
From the manual AC specs:
[ 4 ] Specifications are for sinewave input >0.3% of range and > 1mVrms.
Add 30 µV error to AC voltage specification for frequencies < 1kHz.

[ftransform]

From this link from this thread you are expecting to much from your 34401A.  I am not dissing your meter it just does not give accurate rms measurements at the levels you are trying to use.
From the manual AC specs:
[ 4 ] Specifications are for sinewave input >0.3% of range and > 1mVrms.
Add 30 µV error to AC voltage specification for frequencies < 1kHz.

Yea you are defiantly right I need a better noise measurement solution.
But in this case I think it clearly tells me that the circuit is not working how it is intended to work on a breadboard. Just from a relative measurement sense.
And you can defiantly perform the tuning as described on the webpage using just the mulitmeter. YOu can see the noise reading drop down as the trimpot is adjusted so the gain ratio matches.

[ftransform]

I'm working on a PCB right now I'm going to etch it and populate it with parts. I will use a double sided PCB and utilize the entire bottom side as a ground plane and do the top routing using traces and wires (if it becomes too dicey). I figure that it will be better to run a wire on top then to break the ground plane in this type of circuit.

Gonna try this out with eaglecad. Wish me luck.

If anyone wants to resolve my other questions:
-what is important while considering the transistor
-would tantalums provide any benefit over electrolytics
-How many capacitors to parallel up



Also, what do you guys recommend for the shunt resistor (0.05ohms)? In my breadboard I actually paralleled up two large sand-bar type 0.1 ohm resistors (smallest ones I could find in my scrap box)...
What type of resistor is commonly used for this application?
Would using resistance wire be appropriate instead of a large sand bar?

That manhattan style of construction does seem appealing though. I may order some of those pads and copper plate when I have some spare money but right now I have presensitized PCB board so I will just work with that.

[KE5FX]

Thread necromancy time! 

I finally got tired of the AC line spurs in my HP 5087A distribution amp and decided to do something about it.  At -100 dBc at 120 Hz, it was hard to make usable ADEV measurements at short-term taus.  Seemed like a good excuse to try the circuit discussed above.  It works nicely and isn't sensitive to aesthetics.

I modified two 5087As this way with identical results.  Didn't bother using a Darlington for the PNP stage in the Sziklai pair, but just grabbed whatever was handy -- an NTE374 power transistor for one amp and a PN2907A in another.  The circuit is insensitive to both transistor type and current drain (same residual performance observed after removing 9 of the 10 amplifier modules).  Big two thumbs up to Charles W. for this idea.



Residual plots show 24 dB of improvement in the 120 Hz spur and more-or-less complete suppression of the 60 Hz spur:





Circuit from figure 2 of Wenzel's article with the values I ended up using: