Super Regs

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Oh, then unless you used the same reference device for each of them I'm afraid we're comparing apples with oranges. No?

Well, the 78L15 and the LM317 have their own internal references, the reference for the LME49600 regulator is the LM4040 and the WJ Regulator uses an LM329DZ. All the regulators were powered by the same battery setup, and all used the same load configuration.
 
I note, for the record, that the measurement cited for the Super-Reg was with an LT1028 from Linear, not an AD797 or AD825.

For jollies, I measured the noise (again A-weighted) with the SMT version of WJ's Super Reg -- same load and connection to the AP Analyzer -- it measures 602nV noise RMS and 968nV Quasi Peak. That's nano-s.

This uses the SMT versions of the transistors discussed in the article, and the MJD series transistors which are D-Pak versions of the TO-220 used in the 1995 article, and one and only one LM329DZ as reference.
 
You can measure nanoV? Wow! Nice results. How's the output impedance of the smt one?

The AP Measures nV -- but I want to repeat the results -- here in NJ there is so much environmental electronic pollution it is just unbelievable -- so when I measure all the regs at the same time during the day. It means that I have to set up all the regulators at the same time and solder them into the test jig, let them cool off etc., etc.

The preamplifier I have when I use the HP3581 has a noise floor in the low nano-Volts -- it is basically Dennis Colin's from AudioXpress without the filters.
 
Disabled Account
Joined 2007
Well, forget about the LME49600 Regulator -- here are some results --

A Weighted level measurements -- a.c. coupled with 15mA load -- shielded by my Oreo Cookie Tin:

78L15 (ST Micro) RMS 26.9uV, Quasi-Peak 44.3uV
LM317 (National, No adj bypass) RMS 181uV, Q-Peak 289uV
Super-Regulator (LT1028 error amp) RMS 2.2uV, Q-Peak 3.6uV

LME49600 (LM4040 Ref) RMS 347uV, Q-Peak 564uV

My advice to you, young man, get the Super-Reg boards from Old Colony and stop wasting your time.

The Super Reg is only slightly noisier than the battery.


Excellent work :)
 
I want to take the numbers in post 42 back -- I can't replicate them this morning for the SMT version -- it is coming up a little worse than the LT1028 super-reg on the Old Colony Board and I can't figure why -- I checked the noise floor of the AP with a 50 ohm Tektronix attenuator and it is the same as last evening when I did the measurement.

Perhaps it was the aroma of roasting chicken which influenced the measurements, I don't know...
 

iko

Ex-Moderator
Joined 2008
Measurement of such low values is quite a pain to do reliably. Lots of things can be questioned about the number that comes out. Just like you don't want to hear the word "proof" during a phd defense, so you don't want to hear "instrument calibration" in a home setup for low noise measurement. :)

I've given up on absolute value measurements, instead, I go for ranking. I find that suits my purpose much better. Of course, it brings no laurels or papers or articles, it's just good for my personal conclusions of how something is worse or better than something else.

Jack, do you intend to measure the output impedance of your regs? That to me is just as important as the intrinsic noise of the device. I don't have the setup to run full spectrum measurements of output impedance. Instead, I just measure the p-p modulation on the output of the regulator with a sine wave of some mA load, at a few frequencies, 100Hz, 500Hz, 1kHz, 3kHz, 5kHz, 10kHz, 15kHz, 20kHz. That gives me a very good idea of what's going on.
 
Proof or "poof"?

Jan sent me a reference to a white-paper on the AP website for impedance measurement. I appreciate the work that went into the 1995 articles.

Decades ago my university had a shielded room for this kind of stuff. One of the instructors said that when you ran experiments the Penn Central trains (now Metro North -- about 2000 feet away) could be seen on the detectors if you weren't in the shielded room! For the life of me I don't know why they took the structure down.
 
Well, forget about the LME49600 Regulator -- here are some results --

A Weighted level measurements -- a.c. coupled with 15mA load -- shielded by my Oreo Cookie Tin:

78L15 (ST Micro) RMS 26.9uV, Quasi-Peak 44.3uV
LM317 (National, No adj bypass) RMS 181uV, Q-Peak 289uV
Super-Regulator (LT1028 error amp) RMS 2.2uV, Q-Peak 3.6uV

LME49600 (LM4040 Ref) RMS 347uV, Q-Peak 564uV

My advice to you, young man, get the Super-Reg boards from Old Colony and stop wasting your time.

The Super Reg is only slightly noisier than the battery.

I know this is a pretty old thread, but, I finally got around to measuring my layout of the LME49600 regulator. The DC supply to my board is an old HP lab supply, I measured it un-loaded, with with a HP3562 spectrum analyzer. The board was out in the open, no shielding, LM4040 Ref running at 6.6v out. I saw 3.2uV at 1k, have a spike of 6.6uV around 10k, which I suspect is environmental noise and some spikes at 60,120 adn 180hz. Different from Jack's layout, I have bypass caps on the LME49600, though none for the LME49710.

I will put a load on it and put it inside a tin and measure again.

Ken
 
The following are my measurements of the LME49600 regulator as shown in the National (not TI) application note for the LME49600. I'm going to try making a new version of this with a hopefully improved layout. I hope the new layout will kill the 10kHz spike.
 

Attachments

  • LME49600_regulator_0-2k.png
    LME49600_regulator_0-2k.png
    38 KB · Views: 337
  • LME49600_2k-12k.png
    LME49600_2k-12k.png
    41 KB · Views: 321
  • LME49600_0-100k.png
    LME49600_0-100k.png
    37.6 KB · Views: 322
K,
is there a post detailing your measurement system?
1uV to 10uV is mighty impressive.
I'm using an Hp3562 spectrum analyzer. It's connected to my computer via HPIB. The screen shots are from a Vee program I wrote to capture and display the data. The setup for the test is to take the output of the voltage regulator into the 3562 which is set for AC & Gnd coupling, power spectrum, Mag - linear, vlts/sqrtHz (so, the vertical units are in uVlts/sqrtHz). As usual, Andrew's questions get me thinking. Just ran it again with the units changed to Vrms and the scale jumped up. So, measured in Vrms I get around 9.9uVrms and 89uVrms at the 10kHz spike. I will repost plots in a bit.

Ken
 
Plots with vertical scale at Vrms. First plot is overall bandwidth 0 to 100kHz second plot is limited bandwidth 0 to 2kHz. I have put a common mode choke and film cap in line prior to the regulator which killed any 60hz related artifacts.

Ken
 

Attachments

  • lme49600_vrms_0-100k.png
    lme49600_vrms_0-100k.png
    38.3 KB · Views: 341
  • lme49600_vrms_0-2k.png
    lme49600_vrms_0-2k.png
    34.7 KB · Views: 344
Finish my build of Per-Anders SSR-02. Results at 0-100k look very nice, note the vertical scale is 0 to 10uVrms. But, when looking at 0-2k, not so nice. Regulator is fed from identical supply configuration as the LME49600 regulator. It is not placed in a shielded box, but, neither was the LME regulator. Maybe I have something wrong...

Ken
 

Attachments

  • SSR-02_0-100k_7.8v.png
    SSR-02_0-100k_7.8v.png
    34.9 KB · Views: 274
  • ssr-02_0-2k_7.8v.png
    ssr-02_0-2k_7.8v.png
    40.7 KB · Views: 273
The very high gain of the SR means that they are susceptible to excitation. That's why some have had issues with the AD797 in this application.

For noise measurements, you should really use a battery. The battery and regulator and any local amplifier should be shielded and grounded.

You should also baseline the measurements -- to get a handle of the SNR of your system.

attached is a snapshot of a run I did for the Sjostrom regulator last March

Left scale is nV/RtHz
 

Attachments

  • SjostromNoise.gif
    SjostromNoise.gif
    18.9 KB · Views: 274
The very high gain of the SR means that they are susceptible to excitation. That's why some have had issues with the AD797 in this application.

For noise measurements, you should really use a battery. The battery and regulator and any local amplifier should be shielded and grounded.

You should also baseline the measurements -- to get a handle of the SNR of your system.

attached is a snapshot of a run I did for the Sjostrom regulator last March

Left scale is nV/RtHz

But Jack, if it can't filter out 60hz fundamentals, why use it? The LME48600 setup is quite clean relative to 60hz fundamental. And it was tested exactly the same way. It has other problems... which I will try to address with a new layout.

Best,

Ken
 
But Jack, if it can't filter out 60hz fundamentals, why use it? The LME48600 setup is quite clean relative to 60hz fundamental. And it was tested exactly the same way. It has other problems... which I will try to address with a new layout.

Best,

Ken

Because, that's a separate measurement -- "Line Rejection" - in which you inject a swept sine voltage into the input of the regulator and measure how much the circuit attenuates the perturbation.

Take a look at Walt Jung's articles -- archived on his site Home
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.