My first attempting to design a RF noise generator

[G0HZU]

Hi Joel. That's impressive and humbling.

My design is nowhere near as complicated as I imagine yours is, because like you mentioned, we now have access to nicer components. It's a two stage amplifier (two GaAs MMIC gain blocks), that already provide very good performance. The difficulty was in preserving that performance

Yes, it's much easier these days with modern RFICs. A few years ago I had to quickly design a 'sacrificial' preamp that protected a spectrum analyser from damage.

The noise floor spec for the RF box under test required the analyser attenuator to be 0dB and the frequency range for the noise floor test was HF through to 10GHz. Sadly, the RF box could also emit high levels of RF power so the preamp was designed as a sacrificial buffer. It would be cheaper to swap the external preamp chip than the front end of a microwave spectrum analyser.

I achieved it using a couple of evaluation boards from Atlanta Microwave. I think I used this part here although I used their bare eval boards and fitted them in a metal enclosure. I didn't use their boxed version.

https://www.atlantamicro.com/wp-content/uploads/2018/05/AM1063-1-Datasheet-Latest.pdf

I managed to get a very flat response up to 10GHz and this included the losses for the inner (precision) RF adaptors used to connect it all together inside the metal box. I recall it was short RF cable>>amp>> pad>> amp>> pad>> to the analyser. I did it without needing a soldering iron or any CAD tools. It just shows how things have changed due to modern technology. RF design is becoming more and more like building a system in discrete blocks like you would build something with LEGO.

[G0HZU]

Would be interested to hear some part numbers for the diodes - the NoiseCom ones are pricey!

I've still got a few more diodes to try but none of them are expensive although some are obsolete sadly. I'll have a proper play over the weekend if I'm able. To get the nice 3GHz plot I literally cut a small rectangle of PCB material and made the tracks and pads with a scalpel and drilled the vias by hand. I robbed some R and C parts from another board and just lashed it together. I then tried several diode types.

I think it may be possible to reach 6GHz with a better diode and a better PCB design. I've got one of these T-Tech machines here so I can make a much nicer PCB if I spend a bit of time on this.

https://t-tech.com/store/systems-and-upgrades/quick-circuit-systems/quick-circuit-qc7000/

[G0HZU]

The 1dB dip at 2.3GHz was due to a lack of via holes along one edge of the PCB. I soldered a strip of copper tape across the PCB sides to make this go away.

It's looking quite good now up to 3GHz. I have some even tinier and even faster diodes to try next...

The plot below is scaled at 1dB/div so this looks quite flat and this has the beginnings of a decent little project if I can get a good response up to 6GHz or so.

[Joel_Dunsmore]

show a plot with the noise source off.  Also I think you can add trace averaging to that trace to remove the jitter.

[G0HZU]

Hi Joel. With the noise source off it just shows the noise floor response of the PSA analyser with the internal preamp in and 0dB attenuation.

This shows a linear increase in noise from -169.8dBm/Hz at 100MHz up to -166.7dBm/Hz by 3GHz. This means there's about 10dB difference between hot and cold at 3GHz so there will be maybe 0.5dB contribution from the analyser at 3GHz with the noise source on. However, I don't know how flat the PSA analyser response is with the preamp on and there could be mismatch uncertainty as well. I've only spent a few minutes tinkering with it so far.

I think I can try smaller and faster diodes and maybe get a reasonable response to 6GHz. However, I think the capacitance of the diode will dictate things here. I might be able to change the PCB design to compensate for this a bit. I haven't tried optimising/shrinking PCB pad shapes or anything. I've never really tried to make a noise source up at these frequencies before. I have made a few diode detectors  that work well to several GHz but not a noise source. Many, many years ago I made a precision high level noise source (-85dBc/Hz) but this was band limited to 180MHz. It is very flat though with a decent crest factor. I still use this a lot.

[Gerhard_dk4xp]

Zener or avalanche of some common low power cheap RF BJT BE / BE / BC can be used for RF noise also.

I tested several hi-freq BJTs: BFR93A, AT42035, BFP420. They do generate noise and quite wideband, but the level is relatively low, so it's difficult to get a good ENR after an attenuator of at least 15 dB, which is necessary to obtain a nice RL.
My impression is that hi-freq BJTs, having a very low BE reverse breakdown voltage (4 V or less), don't enter avalanche breakdown, but remain in zener mode, which is only moderately noisy. Besides, their base is fragile and current must be kept low.

I have made a source based on BFR93 some 25 years ago, mechanically not exactly excellent
and had it measured at the boot of a univ at the Weinheim VHF/UHF ham radio meeting and it
yielded 22 dB up to 2.5 GHz. It was pretty much like the W1GHz design mentioned above.

Now I have a 346C + Rosenberger 10 dB PC3.5 attenuator but I don't have the calibration table.
I wonder if I should send it in for calibration to Keysight but I wonder if they would accept the
Rosenberger attenuator and if it would cost more than an arm & a leg.

From my measurements at DC-1MHz with direct comparison to 50 Ohm thermal noise,
Schottkys are hopeless as noise generators.
They are half-thermal in relation to the slope resistance.
Really 60 Ohms instead of 50 because that's 1 nV/rt Hz at room temperature.

cheers, Gerhard

[srce]

Now I have a 346C + Rosenberger 10 dB PC3.5 attenuator but I don't have the calibration table.
I wonder if I should send it in for calibration to Keysight but I wonder if they would accept the
Rosenberger attenuator and if it would cost more than an arm & a leg.

Keysight recently calibrated a NoiseCom 346B for me so I presume happy to do other vendors stuff.

Incidentally, the calibration data had barely changed since it was previously done ~10 years ago.

[jadew]

Keysight recently calibrated a NoiseCom 346B for me so I presume happy to do other vendors stuff.

When was that? I tried to get a 26 GHz Eaton noise source calibrated with them and they said they no longer calibrate non-Keysight noise sources, unless it comes with a bigger batch of TE.

Would be nice to have access to a calibration lab that is equally good, in the EU.

[srce]

Keysight recently calibrated a NoiseCom 346B for me so I presume happy to do other vendors stuff.

When was that? I tried to get a 26 GHz Eaton noise source calibrated with them and they said they no longer calibrate non-Keysight noise sources, unless it comes with a bigger batch of TE.

Would be nice to have access to a calibration lab that is equally good, in the EU.

~6 months ago in the UK. Was that item on its own.

[G0HZU]

One alternative to a formal calibration (not as good as sending it to Keysight I know) is to find someone with a healthy Agilent PSA analyser with the noise figure option fitted.

These analysers are old and starting to become available at lower and lower prices now. If you carefully measure a Minicircuits gain block on a VNA for gain at a very low drive level (to prevent tiny amounts of compression)
it is possible to then measure the same gain block on the PSA analyser in 'noise Figure' mode to display the noise figure and gain. Ideally the gain should agree with the VNA after the VNA has had a full 2 port calibration.

To give an example, if I do this with my 346A and the PSA analyser and check gain and noise figure at the spot frequencies stamped on the noise source then I always get very good agreement for gain with the VNA. It's usually within about 0.15dB for gain (often within 0.1dB) and the noise figure consistently agrees with the Minicircuits datasheet within about 0.2dB.

https://www.minicircuits.com/pages/s-params/GALI-51+_VIEW.pdf

If the ENR was wrong then the gain and noise figure would be wrong on the PSA analyser.

The uncertainty on the analyser should be very low because it has a digital IF (which means the log accuracy will be close to perfect meaning it can measure Y factor very well) and it is designed to measure noise in a fast and repeatable way. It also has remarkably good port match even with the preamp fitted. This method should be very good because the PSA analyser doesn't have to be able to measure absolute noise power accurately. It just has to measure relative noise for the Y factor.

[G0HZU]

The other neat trick the PSA can do is that it lets you tweak the ENR cal table on the fly and this speeds up the tweaking of the ENR table on a homebrew noise source.

In other words, you don't have to keep recalibrating because it appears the PSA stores the raw uncorrected calibration data when in calibration mode. This lets you continuously change/optimise the ENR table with just a few button pushes and then you can look at the effect on any improvement in accuracy for the noise figure and gain of a test amplifier.

[G0HZU]

Here's an example. I just measured my GALI51 test amp again this afternoon on the PSA+346A and then on the the VNA and I then checked the results against the GALI51 datasheet at 65mA bias.

There is good agreement on the noise figure. The gain is pretty good too. There will be some spread in the GALI51 from device to device but probably not much and the gain can always be verified with the VNA. This can be repeated across several types of gain block although it's best to stick with devices that have low VSWR at the input and output.

[G0HZU]

From my measurements at DC-1MHz with direct comparison to 50 Ohm thermal noise,
Schottkys are hopeless as noise generators.
They are half-thermal in relation to the slope resistance.
Really 60 Ohms instead of 50 because that's 1 nV/rt Hz at room temperature.
cheers, Gerhard

Interesting... I'm pretty certain the microwave diode I'm currently using is a Schottky diode. I've salvaged these diodes from old dev boards and these were used as ALC detectors.

I managed to improve the response of the noise source slightly by adding an SMD resistor pad at the output. This has added a tiny bit of pre-emphasis and the response is slightly flatter to 3GHz.

I also cross compared (across 50MHz to 3000MHz) against the 346A for a GALI51, ERA2 and ERA3 test amplifier and the agreement was really close in every case for noise figure and gain on the PSA analyser with the ENR table below for the new noise source.

100MHz  9.02dB
1000MHz  8.93dB
2000MHz 8.80dB
3000MHz  8.60dB

Check it out... only 0.42dB droop to 3GHz!

The noise source diode has a shunt resistor and a total of 16dB attenuation after it to get this ENR of just under 9dB. The return loss needs to be improved. It is 36dB worst case to 2GHz across hot and cold but I haven't tried to improve this yet. I also need to improve the thermal stability. the NoiseCOM diodes are rated at 0.01dB per degC and I'm nowhere near that with this schottky diode. I think it can be improved but the other unknowns are repeatability over time (12 months?) and also reliability in avalanche mode. I'm running this diode at 4mA via a current source at present. I think I'll have to fudge a temperature correction onto the current source to get good thermal stability across a change of (say) 20degC.

[Gerhard_dk4xp]

Hm. Do you operate the Schottky diode with a huge backward voltage,
so that it breaks down and is really used as an avalanche diode? That's very un-Schottky-ish :-)
That would have a lot of dissipation for the small junction, even if the small junction would
make it wideband.

Schottky diodes when used as a Schottky ring mixer for example do not generate much noise;
the noise figure is not much worse than the mixing loss, maybe 7 dB, most of which is loss.
I did measure my diodes in forward direction to create bias voltages.

You need an output attenuator anyway to get an acceptable return loss. That is much more
important than a flat spectrum. You can calibrate away the varying spectrum, but a bad return
loss makes the source unusable even as an indicator, For sub-dB noise figures you won't
get far with a 15 dB source. That's why I want to have my Agilent 15 dB source calibrated
with that Rosenberger PC3.5 precision attenuator.

DJ9BV RIP has written an excellent article on that in DUBUS some 20 years ago.

Found one on the net, there are more: 
<  http://f1chf.free.fr/hyper/article%20DJ9BV%20from%20F6DRO.pdf    >

[G0HZU]

Yes this diode seems to go into avalanche at about 8V. There seems to be a sweet spot at about 4mA where is is quite stable in terms of bias versus output noise level.

I have to confess I'm out of my comfort zone with this diode. I may be stressing it a lot, but then again I might not be. I left it running for several hours and it seems consistent. I also gave it the hot hairdryer test of reliability and it survived. I'm currently getting 9dB ENR and about 36-40dB return loss hot/cold to 2GHz.