Just a simple little test amp I'm currently boxing up to add to my "lab". Uses ultra-low voltage noise transistors in a differential stage. The intended use case is that the output goes to a sound card with differential microphone input, and spectra are recorded and plotted from this.
The ZTX851's are 0.275nV/√Hz or so at 10mA, and a differential pair like this with 3.3 ohm emitter resistors gives about 0.5nV/√Hz floor for measurements. The supply is two 9V batteries, the 1.8V is actually a red LED as a voltage source.
I have a Scarlett Solo 3rd gen USB sound card which is 1.5nV/√Hz on the microphone input at max gain. This amp gains 9.5dB of extra noise sensitivity to this and retains the differential mode. Measured gain is about 35dB into the 3k input impedance of the Scarlett Solo.
In the circuit I've shown some ways to measure simple transistors and opamps for voltage noise (you can ignore V3 as it was solely for LTSpice's simulation purpose to allow noise runs).
The large 470uF caps are essential for handling low frequencies down to 30 Hz or so.
Some example measured spectra:
(I'm very intrigued by the Signetics NE5532 being sensitive to hum, I think this is due to lower PSRR/CMRR, the current setup is not boxed up and is close to mains wiring!)
[The bottom "preamp_0R.wav" trace is the noise-floor of the amp. The gradual roll-off at 10k or so is an artifact of the Scarlett Solo's microphone input frequency response.]
Axes are in dB(V^2/Hz) - so -180dB = 1nV/√Hz for instance.
To calibrate the setup I used a range of resistors across the inputs, plotting the results, then adjusting the Y-axis scaling to match expected voltage noise levels:
A more smoothed version allows better measurement by eye:
For 1k source resistance (984ohms in reality) the current noise of the ZTX851's is starting to appear I think, especially the 1/f part of the current noise spectrum. The 99 and 50 ohm traces seem to be the most reliable guides - theory says they should measure as 1.375 and 1.038 nV, or -177.2dB and -179.7dB (assuming my amp is 0.5nV (-186.0dB))
The ZTX851's are 0.275nV/√Hz or so at 10mA, and a differential pair like this with 3.3 ohm emitter resistors gives about 0.5nV/√Hz floor for measurements. The supply is two 9V batteries, the 1.8V is actually a red LED as a voltage source.
I have a Scarlett Solo 3rd gen USB sound card which is 1.5nV/√Hz on the microphone input at max gain. This amp gains 9.5dB of extra noise sensitivity to this and retains the differential mode. Measured gain is about 35dB into the 3k input impedance of the Scarlett Solo.
In the circuit I've shown some ways to measure simple transistors and opamps for voltage noise (you can ignore V3 as it was solely for LTSpice's simulation purpose to allow noise runs).
The large 470uF caps are essential for handling low frequencies down to 30 Hz or so.
Some example measured spectra:
(I'm very intrigued by the Signetics NE5532 being sensitive to hum, I think this is due to lower PSRR/CMRR, the current setup is not boxed up and is close to mains wiring!)
[The bottom "preamp_0R.wav" trace is the noise-floor of the amp. The gradual roll-off at 10k or so is an artifact of the Scarlett Solo's microphone input frequency response.]
Axes are in dB(V^2/Hz) - so -180dB = 1nV/√Hz for instance.
To calibrate the setup I used a range of resistors across the inputs, plotting the results, then adjusting the Y-axis scaling to match expected voltage noise levels:
A more smoothed version allows better measurement by eye:
For 1k source resistance (984ohms in reality) the current noise of the ZTX851's is starting to appear I think, especially the 1/f part of the current noise spectrum. The 99 and 50 ohm traces seem to be the most reliable guides - theory says they should measure as 1.375 and 1.038 nV, or -177.2dB and -179.7dB (assuming my amp is 0.5nV (-186.0dB))
Last edited:
interesting how good could be there a Low-noise matched dual monolithic transistor MAT02-903H if at 1mA it has <1nV/Hz^.5.
The differential configuration is convenient and advantageous, but is it worth the 3dB penalty on the measurement floor?
Its 9.5dB better than what I had before, and capable of measuring something like an AD797's voltage noise, and allows measurements of microphones which I'm also doing.
Why not connect N transistors in parallel and get sqrt(N)-fold improvement in noise? The existing emitter resistors can be used for ballasting .
Also why not use the Zetex PNPs? According the "The Art of Electronics" they have a bit lower noise than the NPNs.
Also why not use the Zetex PNPs? According the "The Art of Electronics" they have a bit lower noise than the NPNs.
On the input side, both BE junctions are in series effectively. Since their noise voltages add geometrically, you get 1.414 times the voltage noise or 3 dB more noise power.
Rbb and RE are the most significant noise sources of an BJT.
Last edited:
That is quite exactly the circuit I used in #126 of the slightly derailed TI new low noise FET thread.
< https://www.diyaudio.com/forums/ven...se-jfets-texas-instruments-5.html#post6752643 >
Paralleling complete op amps is simpler in the end.
cheers Gerhard
Last edited:
A SE LNA is perfectly workable, there are a number of examples on the forum.
Mine only manages 0.9nV/sqrt Hz, but it uses a single fet, and it does not even require a battery supply. The mains supply is connected via an umbilical.
Sanity-check + end-result: a x 1000 measurement preamplifier
A single ended bipolar amplifier can achieve a very low noise level, even with just one input transistor.
Some time ago I made three different amplifiers inspired by the design in this thread:
My version of the G = 1000 low noise measurement amp (for Ikoflexer).
One was a dual amplifier, with the LSK389 as the FET. Actually the FET was placed in a socket and the purpose of that design was not to measure noise from external sources, but to sort the LSK389's, which had huge variations in noise level 😱
The next one was a single amplifier with a BF862 as the input FET. It has a decent performance with an input noise level of around 0.84 nV/rtHz.
The third one was a single amplifier with a bipolar input transistor (ZTX851). This one achieved an input noise level of around 0.27 nV/rtHz, so almost 10 dB better than the BF862 based.
One of the nice things about the design is that the emitter resistor is 1 ohm, keeping the noise level very low.
The bipolar amplifier is supplied by a single 9 V battery.
0.27 nV/rtHz is roughly equivalent to the thermal noise of a 4.4 ohm resistor!
Some time ago I made three different amplifiers inspired by the design in this thread:
My version of the G = 1000 low noise measurement amp (for Ikoflexer).
One was a dual amplifier, with the LSK389 as the FET. Actually the FET was placed in a socket and the purpose of that design was not to measure noise from external sources, but to sort the LSK389's, which had huge variations in noise level 😱
The next one was a single amplifier with a BF862 as the input FET. It has a decent performance with an input noise level of around 0.84 nV/rtHz.
The third one was a single amplifier with a bipolar input transistor (ZTX851). This one achieved an input noise level of around 0.27 nV/rtHz, so almost 10 dB better than the BF862 based.
One of the nice things about the design is that the emitter resistor is 1 ohm, keeping the noise level very low.
The bipolar amplifier is supplied by a single 9 V battery.
0.27 nV/rtHz is roughly equivalent to the thermal noise of a 4.4 ohm resistor!
Wow Gerhard, JFET input measurement amplifiers with feedback all over, are they stable or not?
Last edited:
This is an amazing coincidence; I posted almost the exact same circuit (schematic, Gerbers, board photo), as the MC input stage in my HPS 6.1 in the “HPS 6.1” thread on January 24th 2019, 12 days after yours: https://www.diyaudio.com/forums/analogue-source/333000-hps-6-1-a.html#post5675186, which eventually evolved in a measuring amplifier months later, with a JFET cascode instead of bipolar. Meaning that the same circuit was developed at the same time, 5000 miles away, what are the chances? Great minds think alike, that’s all I can say 😀.
Last edited:
In a differential stage, the two input (low noise) halves are uncorrelated noise sources; uncorrelated noise sources add up geometrically, so overall a 3dB noise penalty is unavoidable.
Did you measure the current noise?
If you need really low input noise its hard to beat a transformer.
No, I didn't. I have mainly used the amplifier to measure noise from voltage regulators, so current noise was not a major concern. But I will try to measure it.
That has been done in Art Of Electronics, ed. 3.
BTW I have tested their ribbon preamplifier, but only single ended 🙂
That did cost some ugly capacitors but only 16 and not 64 ZTX851.
Remember the capacitor must be much larger than for the -3dB frequency.
It must short the base resistors noise through the low impedance DUT.
That leads to unpleasantly long time constants when you attach a DUT.
The 16 ZTX deliver 70 pV/rtHz, as promised in AOE3.
Cheers, Gerhard
BTW I have tested their ribbon preamplifier, but only single ended 🙂
That did cost some ugly capacitors but only 16 and not 64 ZTX851.
Remember the capacitor must be much larger than for the -3dB frequency.
It must short the base resistors noise through the low impedance DUT.
That leads to unpleasantly long time constants when you attach a DUT.
The 16 ZTX deliver 70 pV/rtHz, as promised in AOE3.
Cheers, Gerhard
noo, to me SE LNA is almost practically useless. Ok, if well isolated i.e. battery-powered or <30pf trafo in its power supply, I can consider that as a make sense config.
The ZTX851 is quite interesting. The noise is not specified on the datasheet BUT 20A peak and 5A continuous in a TO92? My concern is that usually low noise transistors have some process parameters closely watched that are less important for power devices or so I have been told.