LFLNA-80

[iMo]

FYI

[Andreas]

Hmm,

how do they get a lower edge frequency of 0.1 Hz with 1000 uF * 1000 Ohms?
And this is not a single high pass. (so the frequencies will add geometrically).

with best regards

Andreas

[iMo]

The 5 parallel opamps are the ADA4528-2 (A32), the sixth is the servo, it seems.
The low pass opamps are the OPA342 (B42).

PS: below my current understanding of the wiring..
The R7 should be 10ohm.

[iMo]

It should work from 0.1Hz with 2200uF input cap.
Not sure how exactly are the three 10Meg resistors in the servo wired, though..

[Andreas]

Shure with R6 = 1K? I think it should be much higer ohmic.
AC-Wise it is parallel to R1.

with best regards
Andreas

Edit: or we should perhaps remove R1 when R6 is populated?

[Kleinstein]

The video of the internal shows an 1 K resistor and 1000 µF electrolytic capacitor. It however does not show R1. So I don't think that R1 is there in the real circuit and ideally there should be an extra voltage divider at the output of the DC servo OP-amp to reduce the voltage range and noise of the OP-amp. The divider would also provide extra resistance to add to the 1 K. However I don't see such a divider.

R1 and R6 would make sense also acting as a divider. However this would be with different values, like 2 K and 22 K  (or even higher resistors to have the first cross over intentionally lower) and than maybe an small capacitor or resistor at the integrator to compensate for a reduced gain.
The DC servo acts like an additional AC coupling and it adds the amplifier noise in the transition region. A divider at the amplifier output can reduce the noise of the DC servo at the cost of a reduce compensation range. It can also reduce the need for the very long time constant at the integrator.

[KT88]

R6 being 1k basically defies the niose reduction due to paralleling five amps. The noise output of the servo is only divded by two...
A higher value for R6 as Andreas implies makes more sense.
The only benefit from the low resistance of R6 is is the quite fast settling time.
Another oddity is the high resistance (780 Ohms) in front of the non-inverting inputs.

[iMo]

This is with only R1=1k as seen on the pcb and with C1=1000uF (my previous wirings above are wrong around the input).
With 2200uF the AC is peaking around 0.1Hz.
Still the 3x 10Meg (R4, R5, R6) resistors in the servo are not fully clear, what I can see there the R5=10Meg is in parallel with C2.

PS: there is the R109 on the pcb not populated, it is wired from the servo opamp's output to ground (not a divider with the R1 then). Also the pcb is from 2021, thus my understanding is it is an early prototype.

[Kleinstein]

At least the filter reponse with only the 1 K from the DC servo looks reasonable sharp and with the expecte 40 db per decade slope. Still the drop at 0.1 Hz is still more than 3 dB, more like -4 dB.

With many filters the noise BW limit is different from the -3dB limit. So the shifted cross over of the filter may actuall be OK to the 0.1 to 10 Hz noise.
The definition of an effective noise cut off is however tricky with non white noise. The noise BW is usually defined for white noise, which is usually OK for the upper limit. There it is normal that the noise BW is larger than the -3dB BW.

[Rico Sonderegger]

Hello all,
I am Rico from Euler Precision.
As we have seen, there is a lively discussion about the function of the "DC-Servo" circuit.
Here now the solution of the mystery:

The circuit consists of the components C101, R103 and the DC Servo loop U203, R105, C103, R106 and R107. These represent an integrator which is limited in bandwidth (C103 // R105 and C103*(R106+R107) and gain (R105 / (R106+R107).
As some have already noticed, the theoretical cutoff frequency is above 100mHz. However, since the large capacitors also have a larger tolerance, we were able to measure that this cutoff frequency is 105mHz (-3dB) for our product over all stages.
The input filter consists of the components C101 and R103 (-20dB/dec) and the integrator (-20dB/dec) and is therefore a 2nd order filter.

R109 is not equipped. This resistor is intended for service purposes.

We hope to have brought a little more light into the darkness with our explanation.

Many greetings Rico

[Kleinstein]

R109 and R108 should obviously not be populated at the same time with 0 ohms. It may however make sense to populate them as a divider (e.g. 1:5) to reduce the noise of the DC servo. This would than need a change (e.g. 5 x lower) to R106+R107 too.  Using an Electrolytic capacitor in the DC servo could be tricky: if used with no or possibly even negative voltage can lead to drift (up) in the capacitance.

[iMo]

While staring again at the pcb picture I see the R109 is wired to the output of the opamp against gnd, thus not creating a divider. But perhaps the picture is too blurry..
Elyt caps - I was taught in past the elyt capacitors need some "minimal polarization" to work properly, not sure it is still valid, however..

[Rico Sonderegger]

Hello all,

An other explanation of the LFLNA-80.

All these 1mF capacitors are non-polarized electrolytic capacitors!
You can see this in the schematic because there is no "+" or "." attached to the schematic symbol.
That's why we can use them without "bias" in the circuit.
The disadvantage is that they cost more than normal types.

In addition, we have spent a lot of time evaluating the best capacitors. Many types were measured over weeks and statistics were created.

Using R108 and R109 as voltage divider may not be such a good idea, because the integrator then needs a bigger gain as it is part of the 100mHz filter.

R109 is not populated.

[iMo]

With Kleinstein's mods:
PS: the max noise at ~100mHz has dropped from 90 to 67uV/rHz..

Rsource = 1ohm.

[KT88]

Without deviding the output of the servo it's (broadband-) noise would be coupled directly into the input. It would be shunted by the source impedance though...

...and I still don't see a need for the 392Ohms resistors other than converting the input current noise into voltage noise...

[Rico Sonderegger]

These 392R resistors are required to protect the inputs of the op-amps when the user applies a high input voltage,
as a large current can flow through the 1mF capacitor, which could damage the input protection diodes of the op-amps.

All of these "non-essential" components are necessary when the product goes into production.
There are so many different users or EMC interference, the product simply must be robust.

Our measurements have shown that removing the resistors reduces the inherent noise by max. 0.3nV (see PDF)

The calculated voltage noise from current noise is -> In (Current Noise Density) x sqrt(10Hz - 0.1Hz) x 392R / sqrt(Ampl. count) =
0.7pA x 3.146 x 392R / 2.236 = 386pV= 0.386nV (without broadband RMS noise current)

[iMo]

The simulation below fits your measurements (002_REV_0_Noise_compare.pdf) pretty well..

Sim:
392ohm 133.08uV rms from 0.1Hz-10Hz
10ohm 127.56uV rms from 0.1Hz-10Hz

Aprox 0.55nV rms diff (referenced to input)..

PS: Kleinstein's mod above (with C1=C2=1150uF in order to have the same AC char):
Sim:
392ohm 129.87uV rms from 0.1Hz-10Hz
10ohm 124.23uV rms from 0.1Hz-10Hz

PPS: KL's mod with all 392 ohm resistors shorted
Sim:
0ohm 91.8uV rms from 0.1Hz-10Hz

All with 50ohm input source.

[Kleinstein]

The 392 ohm resistors at the individual OP-amps are kind of OK. They only compete with the individual OP-amps and see the OP.amps current noise. The main point is the thermal noise of some 2.5 nV/Sqrt(Hz), which is still quite a bit less than the OP-amps noise.

The not so good one is the 390 ohm from the input to the protection diodes. This would compete with the OP-amps in parallel and see the combined current noise of the 5 amplifiers.
The resistor should cause about as much noise as the amplifiers.

Some resistance or current limit is still needed for 2 reasons:
1) protection of the amplifier when connecting a DC voltage or shorting with a charged capacitor.
2) protection of a sensitive DUT (e.g. some LTZ1000 based ref. circuits) that can not safely handle to much load. Here even the 390 ohm may be too much and a switchable high resistor may be nice for a soft start.
If ultimate low noise is wanted one may get current limiting with less noise / lower resistance from some pairs of JFETs or depl. MOSFETs or MOSFETs with PV gate drive and a fast turn off.

[checksum]

I spent the last week working with an LFLNA-80 that was ordered a few days after seeing Marco Reps YouTube review. It's proving to be simple, reliable and repeatable as a measurement tool for trouble shooting low noise supplies.

Shortly before I bought an Alpha LNA-10 but found the AC coupled noise floor was that bit too high (~5uV p-p) for the work I had in mind. Despite this, the LNA-10 has more than proved it worth for its ability to reject common mode noise whilst making measurements. Its the first differential probe I have owned, so getting clean measurements with the LNA-10 where CM noise between DUT and scope / analyser would previously have ruined them still feels very novel  .

The LFLNA-80 feeding the LNA-10 probe in differential mode gives almost no drift at all and resolution to measure down to ~100nV. The main benefit for me is that this spares the complication of isolating (with batteries etc) the DUTs I am working with from their mains powered environments. It works a treat, both the scope and DUT are mains connected and the measurements are rock solid down to the 100nv noise floor.

The build feel and operation of the LFLNA-80 are really good too. IMHO a good product, very pleased to have taken the plunge.     

[ZhuraYuk]

Does anyone figured out schematic and values for output filter?

[ZhuraYuk]

So I am trying to replicate this for few month now but, it looks like this DC Servo circuitry is a bit tricky to make it working.
In my case the op amp is always railing out when input is connected to 9V battery.  The reason for that is unequal voltage in op amp input. The inverting input has around 0 V but non inversing has some strangely shaped of 50mV offset. Have no idea where it is coming from. All electrolytes in schematic are bipolar Nichicon same series as in original device. Does anyone have a guess?