SR560 Low noise amp -- Noise Calculation

[sky2city]

Hi Guys，
I have two SR560 Low noise amplifiers. Different series number.
After some teardown work, I found the difference.
JFET of first amp stage on new one is LSK389B, the older one is NPD5564.


I tested the noise floor with them by APX515.
(Of cause CMMR and OFFSET adjustments is done according the service manual.)

For the older one, the output noise of about 4.0nV√Hz at 1KHz (SR560 Gain is x1000). That is make sense since there is a NPD5564 JFET inside. The Noise of NPD5564 is about 3.5nV√Hz at 1KHz.

But for the new one, the output noise of about 3.5nV√Hz at 1KHz (SR560 Gain is x1000). I am confused, It is a LSK389B inside. The noise of LSK389B is about 1.6nV√Hz at 1KHz at all. The Noise of this unit (with LSK389B) should be close to 1.6nV√Hz, but it is not.  : -(


(1)   Noise calculation:
I tested most major DC value of components of SR560, make sure it’s working in good condition.



The gain of JFET stage is about x10（~20ms（JFET gfs） x 499Ω( R131 ）. The input noise of OP37 is about 3nV√Hz, so the equivalent noise of OP37 at input is about 3nV√Hz / 10 = ~0.3nV√Hz

Looking at 499 Ω resistor R131(or R117), it’s the load of JFET, the noise of 499Ω is about 1.8 nV√Hz, so the equivalent noise of R131 at input is about 1.8nV√Hz / 10 = ~0.18nV√Hz
Summing LSK389 noise and R131 and OP37 noise together.
Total noise at input  = √ 1.6 x 1.6 + 0.18 x 0.18 + 0.3 * 0.3  = 1.64 nV√Hz.



(2)   2nd amp stage
The 2nd amp stage is LT1028.
I found that the U101A Pin2 and Pin3 is shorted when setting the SR560 Gain to x50000.
After setting Gain to x50000, it’s possible to reduce the noise downgrade by divider (R108,R109,R110).



Test the noise from SR560  50 Ω output BNC connector is show below.
The noise as show in pic is 138.8 nV, so the equivalent noise at SR560 input A is about
138.8 uV/ Gain ( 50000) = 2.78 nV√Hz。
It is closer to the calculation 1.64 nV√Hz, but still not corrective.


Is there something wrong or something I miss? 
I am so confused.
From my test, That’s mean，to enhance noise performance of SR560, changing NPD5564 JFET to a LSK389B only, it will not work at all.

Hei guys， it's appreciated let me know if you have any idea.

Roger.

[moffy]

I believe that the gain of the JFET stage, differential, is less than 5, (500+500)/(110+110) plus whatever the effective JFET source resistance added to the 110 source resistors. The quick LTSpice simulation I did got a gain of about 3 to 3.5 but my bias point might be a little off.

[Kleinstein]

As an additional noise source there is the noise of R113+R144, that should be about as much or a little more than the noise if the LSK389.
Changing this would need a much more dramatic change to the amplifier, as R113 and R114 are also setting the gain.

The gain of the JFET stage is indeed quite low, which is also not ideal, but could be still OK.

[sky2city]

To Improve noise performance, changed the Gain of first amp. Stage.
As show below, add 2 parallel 10R resisters (R113,R114). The gain change to x110.



Now the noise performance seems to better. About 2.7nV√Hz @ 1KHz.

And then I try to replace NP5564 with LSK389, JFE2140,2SK389.

Yeah, the noise is better and better.

Looking at 1/f noise, it seems to be another problem and it’s confused me again.

The noise of JFE2140 and LSK389 are:
at 1Hz is about 8nV.
at 10Hz is about 3nV/√Hz.

As my measure data show, it’s 65 nV/√Hz@1Hz.
The spec. of SR560 at 1Hz is 30nV/√Hz.

Where is this 1/f noise come from? 

[Kleinstein]

For the 1/f noise, there is a chance that the LF411 in the current source contributes. Ideally there should not be much sensitivity to the amplifier current, but the LF411 has lot's of 1/f  noise and the suppresion from symmetry is limited.  Exchanging the LF411 for something less noisy would be relatively easy.

[Kleinstein]

There is another, even worse point in how the amplfier current is regulated: there is a low noise zener (1N753) as the reference, but the reference voltage is from the wrong side. The current is sensed from the votlage over R117,R131+R119. So Ideally the ref. voltage would have to be from the positive side (or at C113).   As shown the noise of the +10 V regulator (filtered by R119 and C113) would also enter to some degree.

For a first test one could add more capacitance (e.g. some 100 µF) in parallel to C113. This would shift the filtering of supply noise to lower frequencies.

[sky2city]

It looks so strange. This is over my head. 

[KE5FX]

They are apparently counting on R125/C112 to filter the supply noise that the Zener doesn't suppress.  I'd be surprised if it made any difference where R126 is being fed from, given that there is no hump or breakpoint near 30 Hz, but it would be easy enough to check.  Does anything interesting happen if you move the +10V side of R126 to C113's positive terminal?

It does sound like the noise is influenced by the +10V bus to some extent, which means that more of the +10V noise is making it into one side of the signal path versus the other.  Additional capacitance at C113 sounds like it made this differential effect even worse.  What happens to the noise if you remove C113 altogether?  Does disconnecting R121 make any difference?

[Kleinstein]

I am a bit surprized that the additional capacitance increases the noise. It should reduce the effect from the postive supply. It would make noise from the LF411 more relevant (the effective resistance for the current control changes from 250+165 ohm to only 250 ohm. Still this would be only a factor of 1.66 for part of the noise.

The circuit is reasonable simple and a simulation could show at least many of the noise sources. Parts like the effect of the LF411 will however depend on the FET matching. Also the pot settings for the offset and CMR adjustment may have an influence.

As an additional 1/f noise source there could be thermal fluctuations / thermal drift. P103 is not only effecting the offset, but also the TC and the thermal drift. Still I would have expected even lower frequencies for the thermal effects.
Another 1/f noise souce is excess noise of some resistors. So the resistor type can matter and especially thick film resitars can have additional noise, if there is a significant voltage. Here R117,R131 and P103 could be candidates to look at.  A large capacitor from the wiper of P103 to ground could change things.

[sky2city]

Many thanks to John and Kleinstein for valuable comments and information.

I found some issues of my test bench, which made my measured data unstable.
1)   The JFE2140 is not a good one. 
As show below, data is rescaled to ENI (equivalent noise at input) noise.


I changed JFE2140_#A to another one #B, The 1/f noise at 1Hz are reduce from ~70 to ~11 nV/√Hz. Maybe that JFE2140 #A is black sheep.
That is my fault, didn’t double check it. 

2)   The original battery of my SR560 is dead, so I powered SR560 by HP3620A (+12V and -12V) with power lines about half meter long. It might be the couple path for 1/f noise. So, I powered up SR560 from a 4 x 18650 Li battery with short wires now, and tested it again. As show the red trace in above pic, the hump or breakpoint near 50 Hz is very small now.


3)   Ensure the SR560 Cover is mounted and let SR560 warm up 15 mins every test.

The following data is base on change list above:
Test #0 : add 220uF cap at C113 Positive side.


Test 1: Add 220uF cap at C113 + side and C112 + side.


Test A:
[Regarding to John's suggestion: Does anything interesting happen if you move the +10V side of R126 to C113's positive terminal?]



It seems that the effect of noise couple from power supply line is eliminated now.

Moreover, just replace OP37(U102) with AD797.


Replace OP37(U102) with AD797 and replace LF411(U103) with AD797.


It seems that everything is back to right position.
Replacing OP37(U102) and LF411(U103) did not make big difference to noise.

The noise at 1 KHz is 1.53 nV/√Hz.


Recalculate the noise:
Assume that JFE2140 working at 5mA, checking JFE2140 datasheet, noise floor at 1KHz is about 0.9nV/√Hz. Since SR560 is a Difference Input Amplifier. Both JFET at positive and negative terminal would contribute noise to output.

So the EIN is calculated:
EIN (JFE2140) = √(0.9 x 0.9 + 0.9 x 0.9) = 1.3 nV/√Hz.

It’s close to my measured data., 1.53 nV/√Hz. 

My next step, is plan to make a JFET noise test kit, to get the measured/actual noise data of JFE2140, so it’s possible to figure out how much improvement can be done on SR560.

JFET noise test kit
https://www.mvaudiolabs.com/diy/modern-jfet-noise-measurements/

[Kleinstein]

The AD797 is not a good replacement for the LF411. The LF411 works with a rather high impedance signal and one thus need an OP-amp with low current noise. So more like an OPA134 (could be available in DIP), OPA1641 or OPA205.  With well matched FETs there may still be little effect from the LF411.

[sky2city]

The AD797 is not a good replacement for the LF411. The LF411 works with a rather high impedance signal and one thus need an OP-amp with low current noise. So more like an OPA134 (could be available in DIP), OPA1641 or OPA205.  With well matched FETs there may still be little effect from the LF411.

Yes, I notice that, which JFET amp is the best to replacing LF411, that’s question I didn’t have idea to consider now.
I am RF designer, not much experience on amp selection in low frequency application.
Thanks FYI.   

I have SMD to DIP adaptive board on hand, JFET amps of sop8 can be work too.
I will have glance on those amps and try it later on.

Stay tune.

[Kleinstein]

The OPA145, OPA1641, OPA140 or OPA141  would be my choice, with not much difference between the last 3. They would need an adapter.

[egonotto]

Hello,

I would be interested to know what the noise looks like in the time domain. Is this preamplifier still useful when using a modern oscilloscope?

Could you please short-circuit your input (or with extern 50 Ohm termination) and show what it looks like at gain 100 with an oscilloscope at 1 ms/div?

Best regards
egonotto

[sky2city]

Hello,

I would be interested to know what the noise looks like in the time domain. Is this preamplifier still useful when using a modern oscilloscope?

Could you please short-circuit your input (or with extern 50 Ohm termination) and show what it looks like at gain 100 with an oscilloscope at 1 ms/div?

Best regards
egonotto

Post the measured data here.
Common configuration:
1. input JFET = JFE2140
2. OP37 change to AD797
3. LF411 change to AD797
4. SR560 in low noise mode, first stage amp  Gain = x110,~ 40.8dB, Total gain = ~x1100
5. SR560 in high dyn. mode, first stage amp  Gain = ~x10, Total gain = ~x1000
6. For square wave test, amplitude of square wave of SG ( Agilent 1102G ) =~2mV.


===========================================
Following Test is base on configuration: SR560 in Low noise mode

1. short input




2. Square wave test.



3. bandwidth test


=============================================
As comparation data.
Following Test is base on configuration: SR560 in dyn. mode.

1. Square wave test.



2. bandwidth test


:-）

[sky2city]

update measured data.
Configuration:
1. input JFET = JFE2140
2. OP37 change to ADA4898-1
3. LF411 change to OPA140
4. R113，R114 change to 110 //10 ohms
5. SR560 in low noise mode, first stage amp  Gain = x110,~ 40.8dB, Total gain = ~x1100