Need LNA for low-frequency low-level preamp

[dnessett]

I'm not sure what is the proper forum board to ask this question, but as it relates to signal conditioning of the output from a phase noise detector, I imagine Metrology is as good a place as any.

I have run up against a problem analyzing the output of a phase noise detector using a low-frequency spectrum analyzer (PicoScope 4262). The project I am working on is documented in this EEVblog thread, which in total is quite long. For those unfamiliar with the project, the 3 posts beginning with this message summarizes the approach using a HP11729C in Frequency Discriminator mode. While I am now trying to use the HP11729C in Phase Detector mode, the problem I am attempting to solve is applicable to both configurations.

Specifically, this problem is the noise floor of the PicoScope 4262 is around -92 dBm, whereas the phase noise information in the output of the HP11729C is below that. For example, I was trying to analyze the phase noise of a Bliley NV47A1282, which has phase noise specs: (10Hz: -120 dBc/Hz ; 100 Hz: -130 dBc/Hz; 1KHz: -140 dBc/Hz). For the given oscillator carrier signal level, all of these values are below -92 dBm. (It is a bit more complicated than this, but the power spectrum information in the HP11729C output, which requires correction to get the phase noise spectrum, is also below the PicoScope 4262 noise floor.)

This suggests using an LNA with suitable bandwidth to amplify the HP11729C signal before processing it with the PicoScope. The problem is I am having trouble finding an LNA suitable for this purpose. Specifically, I need an LNA with bandwidth 1Hz - 100KHz and gain around 30-40 dB. Actually, I need one with 60-80 dB gain. However, I could gang two LNAs with the lower gain together to obtain what I need. Furthermore, the LNA must amplify very low signal levels, as low as 10 nV_RMS.

I have searched extensively for such an amplifier. Most LNAs I have found have bandwidth in the 100 KHz to GHz range. An exception are audio LNAs, but the ones I have found generally have bandwidth 20Hz-20KHz. This would eliminate the low offset frequencies that are of significant interest in phase noise analysis. Furthermore, the audio LNAs I have found are not intended for the low signal levels that the application must handle.

I did find one possible device, the AlphaLab Oscilloscope Preamplifier LNA10, that might work. It is claimed that the gain of this device is up to 1000X (60dB - clarification on 8-3-2019: 60 dB voltage gain; 30 dB power gain Nope - dumb error. As long as input and output loads are equal, voltage and power gain are equal), but the device's noise spec is 4 nV/sqrt(Hz), which means it is probably useful in this application at low offset frequencies (say 1Hz-100Hz), but above that range may introduce significant noise that will mask the phase noise in the signal.

If anyone has experience with this device or has suggestions for other devices that might work, I would appreciate their input.

[awallin]

Hoffmann "lono" could be a starting point:
http://www.hoffmann-hochfrequenz.de/downloads/lono.pdf

it uses many (N) ADA4898 op-amps in parallel to get 1/sqrt(N) improved input-referred noise.

Johnson noise from 50 Ohm at room-temp (~1 nV/sqrt(Hz)) should be doable, but 1/f noise in the op-amps is increasingly problematic if you want <<10Hz frequencies...

[jhenderson0107]

Perhaps the Standford Research SR560 https://www.thinksrs.com/downloads/pdfs/catalog/SR560c.pdf? 

[splin]

What is  your source impedance?

[chuckb]

The older HP 461A preamps are nice but they roll off the gain below 1khz.

The AlphaLab Preamp amp should definitely help your setup.

Attached is a simple amplifier designed by Charles Wenzel with less than 1nV / rt Hz noise density. This is 4-6x less noise than the Alpha Amp.

Another simple step to improve your sensitivity by 6dB is to remove the 600 ohm load on the output of the HP11729C test fixture.
Good Luck!

[David Hess]

Like awallin suggests, there are some operational amplifiers with low enough noise to be used directly but 1/f noise is really going to hurt performance.  See below for how to handle this.  Paralleling several of these parts for even lower noise will help.  None of them include current feedback amplifiers but these can be cascaded later for gain stages.

An alternative I would consider for the first stage is a discrete amplifier.  Medium power bipolar transistors like the BC327/BC337 (and BC639/BC640?) are good as low impedance low noise amplifiers.  FETs have higher 1/f noise but may be feasible as well if that is removed.  Again, see below.  Many integrated transistors like the MAT03 have well defined noise characteristics and lower base spreading resistance than discrete parts.  I have never tested it, but ring emitter type transistors might be a discrete part with low base spreading resistance.

Somewhere below 10 Hz, 1/f noise will become a real problem.  Chopper stabilized amplifiers have flat 1/f noise because the chopping process removes it.  So a parallel path amplifier with the chopper removing the 1/f noise of the main amplifier is feasible.  Because of the low bandwidth involved, the frequency breakpoint for lowest noise can be found with just a DC voltmeter; adjust it for lowest low frequency noise measured using the standard deviation of the output.  I am a little fuzzy if choppers can be used in parallel for better performance in this application but I would at least try it.  Just make sure that they are synchronized which probably means using a dual or quad part.  Some single parts allow synchronization however via a separate clock pin.

With enough bipolar inputs in parallel, maybe 1/f noise will not be a problem above 1 Hz.

Watch out for the construction of the gain cascade.  It needs to be well constructed to avoid oscillation caused by feedback from the output back to the input.  I would not count on more than 40dB in a single enclosure.  With as much gain as you are asking for, I might convert the signal to differential before most of the amplification.

Long ago I did something similar to what you are suggesting but without the need for such high performance although it was a differential amplifier complicating things.  I used a pair of LT1028s in a differential x1000 configuration with a dual LTC1151 chopper stabilized amplifier to remove 1/f noise and drift by driving the adjustment pins of the LT1028s.  Noise was low enough to measure low values of resistance via Johnson noise alone.  Adjustment of the frequency breakpoint was done exactly as I described above and the results matched the datasheet noise specifications exactly.

[David Hess]

Attached is a simple amplifier designed by Charles Wenzel with less than 1nV / rt Hz noise density. This is 4-6x less noise than the Alpha Amp.

Getting below 1nV/SqrtHz was pretty easy even 40 years ago.  But getting that below 10 Hz or even 1000s of Hz if you try to use JFETs or especially MOSFETs is very difficult because of flicker noise.  Modern CMOS RF processes are incredibly bad for this and RF bipolar transistors are not great.

Usually it would not be a problem however because the close in phase noise of the oscillator would be greater than the flicker noise of the amplifier (1) so I am not sure Dnessett's requirements (2) are realistic.  So I just described how I would tackle the flicker noise problem anyway.

I suspect a precision low noise bipolar amplifier will handle it just fine without anything fancy.  At most, parallel a bunch of precision low noise operational amplifiers at a relatively low gain for the first stage.  Take advantage of that 25 ohm source impedance.

(1) Low noise oscillators use low flicker noise transistors for this very reason which suggests medium frequency amplifiers using bipolar transistors will always be sufficient for this application.

(2) "I need an LNA with bandwidth 1Hz - 100KHz ..." - Dnessett

[Kleinstein]

The source impedance is an important parameter. With still some 600 Ohms output impedance, resulting from a series resistor, there is no more need to get < 1 nV/sqrt(Hz).  A 600 Ohms resistor is already a little more than 3 nV/sqrt(Hz).
With BJT based OPs, one has to keep an eye on current noise too. The 1 nV/sqrt(Hz) of the LT1028 and similar BJT based OPs is combined with a high current noise that can give more noise if the source is more than some 50 Ohms. In the low frequency range the current noise tends to be even more important.
So paralleling BJT based amplifiers is tricky and may not be a good idea, unless the source is extremely low impedance.

Paralleling JFET based amplifiers is possible and allows to reach 1 nV noise level with less current noise, at least for some 10-100 Hz and up.

Oscillators tend to have more noise in the low frequency band - so some 1/f noise may be acceptable.

[dnessett]

What is  your source impedance?

600 ohms.

[dnessett]

Thanks to awallin, David Hess and Kleinstein on the complication of 1/f flicker noise in the amp at frequencies below 10 Hz. I hadn't considered that.

[dnessett]

Hoffmann "lono" could be a starting point:
http://www.hoffmann-hochfrequenz.de/downloads/lono.pdf

it uses many (N) ADA4898 op-amps in parallel to get 1/sqrt(N) improved input-referred noise.

Johnson noise from 50 Ohm at room-temp (~1 nV/sqrt(Hz)) should be doable, but 1/f noise in the op-amps is increasingly problematic if you want <<10Hz frequencies...

This design seems promising. Is it available as a finished product or, perhaps, a kit?

[dnessett]

I am not an experienced amplifier designer and that is especially true of low-noise amplifiers with the requirements I am seeking. So, while I found the comments about how one would design such an amplifier interesting, I really am looking for something already designed, i.e., a product or, perhaps, a kit I can purchase and assemble. That isn't to say these comments are not appreciated, since, for example, they made it clear that I failed to consider 1/f flicker noise in the preamp as a design issue. But, actually, I am looking for something I can purchase, rather than something I might design.

[chuckb]

The HP11729C has a bipolar (2N5943) preamp built in. Does this built in amplifier need more gain to drive the Picoscope or do you want a lower noise amplifier connected to the 600 ohm output?

The AlphaLab Preamp should provide low noise and enough gain for either situation.

[dnessett]

The HP11729C has a bipolar (2N5943) preamp built in. Does this built in amplifier need more gain to drive the Picoscope or do you want a lower noise amplifier connected to the 600 ohm output?

The AlphaLab Preamp should provide low noise and enough gain for either situation.

The HP11729C has a 40 dB LNA built in. However, when measuring 10 MHz, because of the way the instrument is architected, the LNA is driven into compression. John Miles pointed this out on a different forum (see this post on the HP/Agilent/Keysight test equipment forum). Consequently, I was forced to use a phase noise output (the 1Hz-1Mhz output) that bypasses the LNA.

Added 8-04-2019: It occurred to me after writing this post that I had actually addressed the LNA compression issue on the EEVblog. The post is located here.

[Kleinstein]

Hoffmann "lono" could be a starting point:
http://www.hoffmann-hochfrequenz.de/downloads/lono.pdf

it uses many (N) ADA4898 op-amps in parallel to get 1/sqrt(N) improved input-referred noise.

Johnson noise from 50 Ohm at room-temp (~1 nV/sqrt(Hz)) should be doable, but 1/f noise in the op-amps is increasingly problematic if you want <<10Hz frequencies...

This design seems promising. Is it available as a finished product or, perhaps, a kit?

The design with paralleled ADA4898 is only suitable for really low impedance source - that is something like 40 Ohms / N.
So with 4 amplifiers this would be something like  < 10 Ohms and less if more OPs are in parallel.

AFAIK there is a newer JFET based version, discussed somewhere in the forum.

For just a 600 Ohms source a single BJT based OP like OPA209 would be more suitable. It's higher voltage noise from the OP, but due to less current noise lower overall noise.
As someone noted already much of the 600 Ohms is due to a resistor in the instrument (to get the standard 600 Ohms values and maybe for protection too). For lowest noise the signal should be taped off before that resistor.
The other Forum points to the idea of adding some filtering in front of the internal LNA, so that this could be used.

[ap]

You may e.g. wish to check with Femto if you need something finished. Not cheap though.

[Gerhard_dk4xp]

This design seems promising. Is it available as a finished product or, perhaps, a kit?

Not as a kit, but there are some leftover boards.

Note that the LONO's noise rises faster than 1/f below 50 Hz or so. That comes from
the  fact that the input capacitor at 100 or 200 uF foil is not large enough to short the
10K bias resistor through the source impedance.

That amplifier makes only sense for low impedance signal sources. A 60 Ohm resistor
delivers 1 nV/rt Hz ; the LONO is 15 or 16 dB better ( 200 pV really). But in a 50 Ohm
system you could not take advantage of that.

I have a  95% completed follow-up version that uses a wet slug tantal 4700u/25 in the
input. That removes the steep noise rise at the low frequency corner but requires a lot
of discipline when connecting DUTs. I managed to kill a LED "reference" with the 
precharged 4700u cap or when connecting a 12V source suddenly could put the
op amp inputs into jeopardy. But then there are 20 of them. Shared grief is divided grief...
The current version has protection relays against that.
Best handled by computer over the opto-coupled SPI interface. Gain & bandwith
can also be set with analog switches.

I'm also working on a FET version with a dozen BF862 or a IF3602 pair. Most amplifiers
you find on the internet are not really stable, even if they claim to.
The eye opener is plotting the real part of (Vsource / Isource). For most amplifiers that
goes negative at 50 KHz and/or up. Not acceptable for a measurement amplifier.
May add "personality" for audio enthusiasts.

The problem is the feedback loop back into the source. Stability or bandwidth,
pick one.

cheers, Gerhard

[TimFox]

What is  your source impedance?

600 ohms.

Is the actual impedance from the source (seen into the source terminals) actually 600 ohms, or is the source designed to work into a 600 ohm load?  This makes a difference for the low-noise input stage.

[guenthert]

Thanks for sharing notes and design.

[..]
The problem is the feedback loop back into the source. Stability or bandwidth,
pick one.
[..]

  The otherwise not-so-remarkable (and certainly not-so-low-noise at lowest frequencies) schematic of the SR560 addresses that by adjusting the current fed into the input FETs depending on the sum of their outputs.  I thought that was quite nifty, even though it's more interesting for high-impedance sources.

[splin]

How about this 700 ohm noise equivalent amplifier:

http://www.emelectronics.co.uk/a29.html

Ok, not a lot of gain at 100kHz but enough to allow a much cheaper low noise amplifer for the rest of the gain. Not especially low noise at higher frequencies but no flicker noise and only around 9nVpp .1 to .33Hz - you'd be hard pushed to get much better from an off the shelf product for less than $200.

How about using different amplifiers for different frequency bands? You could use the even lower noise, 25ohm equivalent, A23 < $500. LF noise is only 1.3nV pp but unity gain bandwidth is only 10kHz. Use it for say < 100Hz and an LT1028 for 10Hz to 100kHz, (.85nV/rtHz).

If you want to go the route of paralleled opamps, bipolar amps have too much current noise but the JFET ADA4625-1 is quite impressive; 150nV pp 0.1 to 10Hz, 3.3nV/rt(Hz) @ 1kHz but only 4.9fA/rt(Hz) current noise so you can parallel as many as you can afford. Expensive at $7.8, @ 10off; the dual channel version is cheaper (per amp) but only seems to be available direct from Analog @ $5.99, MOQ 100.

If you need to make the measurement in one step, you could use both amps with a crossover set at some frequency where any gain and phase errors aren't important.

TiN has played with some A23s - take a look on his xdevs site.

[EDIT] You could parallel lots of A29 amps - current noise doesn't dominate below about 10k ohms source impedance. A bit pricey mind. 

[dnessett]

You may e.g. wish to check with Femto if you need something finished. Not cheap though.

Thanks. There are Femto products that would nicely meet the requirements. Perhaps the best is the DLVPA-100-BLN-S, the relevant specs of which are: 1) DC-100 KHz, 2) voltage noise 700 pV/rt(Hz), 3) variable gain of 40dB-100dB in 20 dB steps, 4) input impedance 1 Mohm. I contacted the North American distributer and was quoted a price of $2,170, which as you mention is not cheap; in fact too expensive for a hobbyist such as I. Maybe I can find a used one on ebay.

[dnessett]

This design seems promising. Is it available as a finished product or, perhaps, a kit?

Not as a kit, but there are some leftover boards.

Note that the LONO's noise rises faster than 1/f below 50 Hz or so. That comes from
the  fact that the input capacitor at 100 or 200 uF foil is not large enough to short the
10K bias resistor through the source impedance.

That amplifier makes only sense for low impedance signal sources. A 60 Ohm resistor
delivers 1 nV/rt Hz ; the LONO is 15 or 16 dB better ( 200 pV really). But in a 50 Ohm
system you could not take advantage of that.

I have a  95% completed follow-up version that uses a wet slug tantal 4700u/25 in the
input. That removes the steep noise rise at the low frequency corner but requires a lot
of discipline when connecting DUTs. I managed to kill a LED "reference" with the 
precharged 4700u cap or when connecting a 12V source suddenly could put the
op amp inputs into jeopardy. But then there are 20 of them. Shared grief is divided grief...
The current version has protection relays against that.
Best handled by computer over the opto-coupled SPI interface. Gain & bandwith
can also be set with analog switches.

I'm also working on a FET version with a dozen BF862 or a IF3602 pair. Most amplifiers
you find on the internet are not really stable, even if they claim to.
The eye opener is plotting the real part of (Vsource / Isource). For most amplifiers that
goes negative at 50 KHz and/or up. Not acceptable for a measurement amplifier.
May add "personality" for audio enthusiasts.

The problem is the feedback loop back into the source. Stability or bandwidth,
pick one.

cheers, Gerhard

Thanks for your kind offer of making a surplus board available for purchase. That is very generous.

However, after reading you comments, I think, perhaps, this is a project still under development. So, I probably will make do with a less capable product that is more mature.

[dnessett]

Is the actual impedance from the source (seen into the source terminals) actually 600 ohms, or is the source designed to work into a 600 ohm load?  This makes a difference for the low-noise input stage.

The HP11729C operators manual says the output impedance of the 1Hz-1Mhz noise output is 600 Ohms.

[Gerhard_dk4xp]

600 Ohms is 3.1nV/rt Hz alone; a single AD797 would be adequate then; it's voltage noise (0.9 nV/rtHz)
would simply go under. It adds geometrically. Even an OPA140 could be justified if you want to go JFET.

At $5 for 10 boards from PCBway, there is not much purchasing.  :-)

I have solved my phasenoise problem with a Timepod. No more delay lines, mixers
and calibration guesswork.

My own interpretation of of the Wenzel / the NIST design is this here:
It has a 12 year old precursor.

[RolandK]

Just look at the works of Bernd Kaa DG4RBF:
- Low Noise amplifier with 0,4 nVeff/Hz^-0.5 from 10Hz to 50 kHz.
- He uses it to design low noise powersupply with LM723
- Phase Noise Test set
His articles are in german, but google is your translater-friend. The first 2 are available if you search him in google. For the third article are only build descriptions available, but the original article explains best the principles and the calculations. You can get them from "UKW Berichte" or perhaps from the author.  Scan them to pdf, ocr them with pdf xchange viewer and translate with google.

Good luck.