Author Topic: Marco Reps circuit LT7805 ultra low noise : what is the circuit & caps values  (Read 2903 times)

0 Members and 1 Guest are viewing this topic.

Offline Maxoverdrive

  • Contributor
  • Posts: 16
  • Country: fr
Hello
In the very interesting video on YouTube


Marco Reps have built a circuit base on LT7805 with electrolytic cap and ceramic caps to reach amazing low noise result ( for a LT7805 ) and even more compare to costly DC/DC IC

Question : Does that possible to have the schematic of this circuit with BOM

Thanks in advance
MOD
 
The following users thanked this post: imo, Grandchuck

Offline imo

  • Super Contributor
  • ***
  • Posts: 2998
  • Country: li
In that video that is a standard 7805 voltage regulator, typical values for ceramics are 100nF and he uses electrolytics like 1000uF.
There is also LTC7805 - it is a switcher (not in that video)..
 

Offline MiDi

  • Frequent Contributor
  • **
  • Posts: 510
  • Country: de
I do not recommend building that circuit.
The OPs that are responsible for low noise are ADA4523.
I rather recommend building Andreas type of low frequency LNA.
See lengthy thread DIY low frequency noise meter
 
The following users thanked this post: BU508A

Offline guymo

  • Regular Contributor
  • *
  • Posts: 87
  • Country: gb
More details of this circuit are available on Marco Reps's github:
https://github.com/marcoreps/low_noise_amplifier
 

Offline Maxoverdrive

  • Contributor
  • Posts: 16
  • Country: fr
Thanks for your answer
For the L7805 the purpose was to be able to compare with Marco design ( same circuit + same compoenets = same noise ou close )

For the noise amplifier the Marco files on Github are in process so I prefer to work on a stable version rather to launch PCB and order components to generate troubles
the article  on Analog is detailled
https://www.analog.com/en/technical-articles/775-nanovolt-noise-measurement-for-a-low-noise-voltage-reference.html

In both design , the use of precision capacitor is mandatory

Marco Reps works are impressive as his reverse engineering on Fluke Voltage zener reference

If some one on this forum have a link on a validated / finished circuit for noise measurement I will happy to built it

Regards
MoD



 

Offline julian1

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: au
the topology of the two cascaded/higher order sallen key filters in the marco reps schematic appear different to those I get by googling randomly.

google mostly shows input from the "resistor divider" going to to the non-inverting input.
while the marco reps goes to inverting input. and the cap placement is different.

But perhaps it is a valid inverting variation - good when two are coupled together?.

Edit. Looks like its a MFB low-pass filter.
 
« Last Edit: September 22, 2021, 06:26:29 am by julian1 »
 

Offline MiDi

  • Frequent Contributor
  • **
  • Posts: 510
  • Country: de
If some one on this forum have a link on a validated / finished circuit for noise measurement I will happy to built it

See Andreas post and following.
« Last Edit: September 22, 2021, 04:17:28 am by MiDi »
 

Offline julian1

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: au
Actually the schematic looks like a MFB low pass filter, rather than sallen key.
Plugging values into a calculator here, http://sim.okawa-denshi.jp/en/OPttool.php
I get,
 

  first stage,
  100k,100k,100k and 270n,270n
  fc = 5.9Hz.
  Q=0.33

  second stage
  100k,100k,100k and 680n, 39n
  fc = 9.7Hz.
  Q=1.4



 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 10034
  • Country: de
A cross over frequency of 6 Hz and a Q or 0.33 looks like a mistake. To get an overall 10 Hz upper limit the cross over of the individual stages should be normally a littler over 10 Hz and for a steep role off the Q should be more like in the 1-2 range. A Qof 0.33 hardly needs an active filter stage. So maybe the 2nd 270 nF cap should be 27 nF ?

I don't know how the ADA4523 behaves on large input differentials, but if it has a hard limit like many types (excepts the Ti mux friendly ones) there could be quite some current spike when connecting such an amplifier. One should also habe a may to carefully charge the input cap (to protect the DUT), and to keep the input cap charged when not in use. 

The 100 Ohms resistor at the input should be better separate for the 4 amplifiers - there is a theoretical chance the amplifiers can interact via input current spikes. individual resistos would reduce that chance, though it would not 100% prevent it. When unlucky one may have to swap OPs if they are too close in chopper frequency.

Much is already said in the other threads about low frequency amplifiers.
 
The following users thanked this post: julian1

Offline KT88

  • Regular Contributor
  • *
  • Posts: 188
  • Country: de
The block diagram of the ADA4523-1 shows antiparallel diodes across the input (DS Fig.64 p.20).
 
The following users thanked this post: julian1

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2797
  • Country: de

  first stage,
  100k,100k,100k and 270n,270n
  fc = 5.9Hz.
  Q=0.33

  second stage
  100k,100k,100k and 680n, 39n
  fc = 9.7Hz.
  Q=1.4

Ok that partly explains why the measured value for the LTZ1000 is too low.
(I never had below 1 uVpp for a 10 sec measurement duration).

If I have understood it right the whole 1000:1 amplification is done in the first stage.
But 1000:1 is too low for most of the oscilloscopes with 1-2 mV/Div lowest input range.
Usually for a 0.1-10 Hz LNA a 10000:1 amplification is needed.

I usually also need a metal cookies box for the whole cirquit including the batteries to get useful results.
(and a more or less gaussian looking noise on the screen).
Also a sanity check (noise floor + FFT for mains frequency) is necessary before each "real" measurement.

I would simulate the circuit before building it wether it gets the right bandwidth and amplification.
And of course do a final check with a bode plot.

with best regards

Andreas
« Last Edit: September 22, 2021, 06:46:22 pm by Andreas »
 
The following users thanked this post: julian1

Offline julian1

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: au
It's a bit hard to read the Marco schematic, but the fb gain is 100k/10 Ohm, so 10000x.  Which should be ok from an op GBW point of view due to the low frequency. the ops are paralleled 4x - which if i understand the theory - should mean half the op noise.  interaction of chopper/frequency behaviour is something that requires testing.

The most interesting bit in the video presentation for me, was to use a signal gen - with a resistor divider 1M/? - to sanity check the (AC) entire setup/ observe the noise floor.  Without that test/ the whole thing is indistinguishable from magic/voodoo.  It should be be possible to create a bode plot, and confirm filter passives, with at least a few test frequncies, following the same approach.

I read the original the zylmex thread but never understood parts of that circuit/schematic. The comments about having some switching in the front end, to avoid overloading the DUT during cap charge, and to store the cap charged to reduce DA/ hysteresis help explain what's going on.
 
« Last Edit: September 22, 2021, 08:55:49 pm by julian1 »
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2797
  • Country: de
I read the original the zylmex thread but never understood parts of that circuit/schematic. The comments about having some switching in the front end, to avoid overloading the DUT during cap charge, and to store the cap charged to reduce DA/ hysteresis help explain what's going on.

Hello,

The switch for precharging is necessary since the 2 inputs of the OP-Amp are connected by clamping diodes.
Otherwise with a 10V input and a uncharged capacitor the initial current through the 100R input resistor is ~90 mA which is above the maximum rating of the OP-Amp (10 mA per Input). And you can easyly age (very fast) a unbuffered LTZ1000 by this load.

The input resistors cannot be increased without increasing the noise floor due to current noise.

Without keeping the input capacitor at nominal input voltage it may need 1-2 days until the minimum leakage current (corresponding to minimum noise floor) is reached.

It's a bit hard to read the Marco schematic, but the fb gain is 100k/10 Ohm, so 10000x.  Which should be ok from an op GBW point of view due to the low frequency. the ops are paralleled 4x - which if i understand the theory - should mean half the op noise.   

Ok I overlooked the 10Ohm and read 100R.
Paralleling OPs reduces only the "voltage noise" but increases the "current noise"
The datasheet states only that the current noise is not above the voltage noise for input impedances (of a single OP) of <1K Ohm.
So I cannot tell wether the 2K at the input are really a good choice.

The most interesting bit in the video presentation for me, was to use a signal gen - with a resistor divider 1M/? - to sanity check the (AC) entire setup/ observe the noise floor. 

Yes you need a resistor divider to generate signals on the uV level while the Scope needs the mV level. But 1M seems to be on the high side. The 2K input resistor of the LNA is loading the voltage divider.  So I would not go above 50 Ohms for the output of the divider.
The noise floor is clueless at 0V input voltage since the leakage current through the input capacitor is only included when the input voltage is near the test voltage (e.g. 10V). I recommend to use (fresh) NiMH batteries (8 AA cells) for the noise floor.

with best regards

Andreas
 
The following users thanked this post: julian1

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 10034
  • Country: de
The 2 K resistor to ground at the input is not the relevant resistor for the current noise. In the pass band the impedance of input capacitor is smaller than the resistor. So most of the noise current flows to the DUT and not through the 2 K.  It is more than the 2 K resistor gives extra noise current. So for lowest noise it is better to have a larger resistor at the input (e.g. 100 K) and than define the lower frequency limit in a later stage after the amplifier.
The voltage noise if the resistor is only relevant in the open circuit case.

A high input current noise requires a large inout capacitor and limits the use to very low frequencies. The 0.1 to 10 Hz band is standard test, but real world use may also have interest also in even lower frequencies (e.g. 0.01 Hz) and this than gets increasingly difficult.
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 2998
  • Country: li
FYI - Marco's LN preamp with 4xADA4523 - Noise and AC.
 
The following users thanked this post: Andreas, julian1

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2797
  • Country: de
Hello,

so at 10 Hz gain = 60 dB instead of 77 dB. (80-3 dB).

The 2 K resistor to ground at the input is not the relevant resistor for the current noise.

mhm at 0.1 Hz the capacitor should have the same impedance as the 2K resistor. right?

with best regards

Andreas


 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 10034
  • Country: de
The 2K resistor has the same impedance as the input capacitr, if the input capacitor also defines the lower frequency limit. At higher frequencies und thus inside the band of interest the capacitor and DUT is the lower impedance path. There normally is at least 1 more high pass stage to suppress noise from below the passband.  One still needs the size of the capacitor, but not necessary the low resistor.  The resistor also does not have to be a specially low noise or low TC one.

From the resistors own Johnson noise there will be more noise current from a smaller resistor and thus inside the passband more resistor noise contribution. It is only outside the passband of the input coupling that a lower resistor reduces the noise. I would consider it a real option to use a larger resistor (but still the same size cap) and than a 2nd high pass fitler stage after the amplifier. So the input AC coupling would no longer define the lower frequency limit. The difference is not very large, mainly the small cross range that changes. It could still be convenient to have the higher input impedance and thus less need to correct for the loading.
 

Offline MiDi

  • Frequent Contributor
  • **
  • Posts: 510
  • Country: de
The most interesting bit in the video presentation for me, was to use a signal gen - with a resistor divider 1M/? - to sanity check the (AC) entire setup/ observe the noise floor. 

Yes you need a resistor divider to generate signals on the uV level while the Scope needs the mV level. But 1M seems to be on the high side. The 2K input resistor of the LNA is loading the voltage divider.  So I would not go above 50 Ohms for the output of the divider.
The noise floor is clueless at 0V input voltage since the leakage current through the input capacitor is only included when the input voltage is near the test voltage (e.g. 10V). I recommend to use (fresh) NiMH batteries (8 AA cells) for the noise floor.

I used 10k/1R divider for verification of gain & bandwidth.
As I showed, there is no significant difference if input is biased or not, though it is useful to check for.
See the other thread, there are measurements of different voltage sources, best is big capacitor.
« Last Edit: September 23, 2021, 07:13:57 pm by MiDi »
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2797
  • Country: de
As I showed, there is no significant difference if input is biased or not, though it is useful to check for.
See the other thread, there are measurements of different voltage sources, best is big capacitor.
Hello,

the "best" has the lowest voltage so the lowest leakage current through the input capacitor.
Try different numbers of NiMh cells as sanity check.

Wether you have 120 nVpp or 200 nVpp makes a difference when testing a low noise reference like ADR1000A with 600nVpp.
The noise floor should be  a factor 4-5 lower than the DUT to get a error in the 5% range.
After 2 days precharging the input capacitor I usually get 120nVpp with 8 NiMh cells.

with best regards

Andreas
 
The following users thanked this post: 2N3055, MiDi, MegaVolt

Offline julian1

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: au
I made a simple board following the Reps lna.  At the moment, only one of the 4x ADA4523 is populated (gain =10000x).

Both filters are the same - but the caps were constrained to parts I have on hand (approx c1=850n,c2=33n,  fc = 9.5, q = 1.7 ),
end to end freq response with sig-gen  1V p2p into 10^4x attenuation (10k/1R) is,

    0.02Hz    0.12V p2p
    0.05Hz    0.40V
    --
    0.1Hz     0.84V
    0.5Hz     1.12V
    1Hz       1.12V
    5Hz       1.20V
    10Hz      1.10V
    --
    20Hz      0.03V

- For lt1021. 5V
  measure around 3.5-4.5uV  p2p on 12sec rolling scope (10mV scale)
  which matches lt1021-5 datasheet spec.   <1ppm P-P (0.1Hz to 10Hz)

- For a deadshort input, I measure.  (only 1 of 4 ADA4523 populated)
  260-280nV p2p,   on 1mV or 2mV horiz resolution.
  (I understand it's not a noise floor, without a potential across cap, to include leakage current).

- Also tested 10M/10R (thick-film) divider.  Was unsure what to expect with such a high impedance source.
  With 1V p2p 1Hz sig-gen source. the 1uV is clearly impressed on the scope, with about 2.5uV extra resistor noise.

other details,

Cap is Nichicon   UKL 1000uF / 50V.
susumu rr resistors
4 layer board, solid signal agnd plane.  BNC connectors/ and cables.
power linear bench supply +-15V.
There is an input series resistor, with shorting toggle switch to protect the DUT.
no shielding (emi or thermal).  it seems reasonably immune/robust regardless. I can touch/ bump boards and leads etc, without too much interference. led light ok.
Using a rigol 1054z in rolling scope mode works very well, and the 1mV and 2mV ranges are available on 1x probe setting. Though not sure the 1mV is a true range.

I am quite keen on a board revision - (even before adding the other ada4523)
Things to add,
  - extra switches to disconnect the cap to enable it to be stored/ and adjust to charged.
  - power switch to more easily run off batteries. Mostly convenience, to free up the bench supplies.
  - larger cap footprint - to support 63V version (currently 50V) of the UKL.
  - support separate 100ohm resistors to each op.

Is there anything else that could be useful?
 

Offline julian1

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: au
Covering the board loosely with a cardboard box to reduce air thermals, and turning on the scope 20M bandwidth filter, reduces input short noise to 220-230nV p2p. (only one ADA4523 populated).

I need to change the caps in the 100k/10r feedback dividers to bypass high-freq. from unknown mlcc to c0g or film.   

I found some Nimh eneloops, to do a bias voltage test, but they have been left in a discharged state for several years, so probably better to purchase new.
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2797
  • Country: de
I found some Nimh eneloops, to do a bias voltage test, but they have been left in a discharged state for several years, so probably better to purchase new.
Hello,

absolutely. Old defective cells may have several 10 uVpp noise which is not related to the amplifier input capacitor.

with best regards

Andreas
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf