Noise Factor measurement

[trh01uk]

Folks,

just reading the thread about the TF1106 NF meter raised another question in my mind. My impression is that NF meters are a thing of the past. No-one seems to be measuring noise factor these days, and I am not aware of any modern equivalents of the CT82 or TF1106.

So what are modern RF engineers doing instead?

Richard

[Radio Wrangler]

Try looking for 'Noise Figure' instead. It's the same thing, but expressed in dB not linear power ratio. You'll find tons of stuff.

Noise figure is the 0-60 time for small RF transistors, amplifiers etc.

The long-term standard box for measuring noise figure was the HP 8970 family, which was replaced in 2000 by the N897x family of Noise Figure analysers which are synthesised, DSP based and a bit more accurate. HP had changed name to Agilent the year before. Along with it is a family of noise sources with built in calibration tables and an electronic thermometer. Various firms are now often including noise figure options in spectrum analysers, but they lack some of the optimisation of the dedicated boxes and accuracy suffers.

Noise figure is a terrible measurement. The probable errors are greater than the noise figures of many devices that could be measured.

There was a big conference at Churchill College, Cambridge in August 1012 "EME 2012" and the first talk of the weekend was on errors in noise figure measurement by Ian White, G(M)3SEK and yours truly. It caused a bit of a stir, there were people in the theatre who'd bought preamps with claimed noise figures a fraction of the uncertainties that even NPL could offer. I was there because I was the originator of the Agilent NFA and its new noise sources... it began as a 1-engineer unofficial investigation, I got a breadboard working spread across a couple of benches and we managed to justify staffing up a project on the back of the working breadboard.

There's details of the measurement in Agilent AN57-1 and furter details on errors and how to minimise them in Agilent AN57-2

David

[SteveCG]

David,

Re: Churchill College, Cambridge in August 1012 "EME 2012"

You should have gone just up the road to the Radio-Astronomy Group of the Cavendish Laboratory - they have had a few years of measuring distant objects in the Universe where a knowledge of noise floors and levels is rather important.

[G0HZU_JMR]

Originally Posted by trh01uk

So what are modern RF engineers doing instead?

At work I've used the classic old school HP346 noise source + HP8970B NF meter many times over the last 25 years or so and we still have several of them kicking about. However, the HP8970B does have a very wide IF bandwidth of several MHz which can cause problems for some applications.

Nowadays the top choice at work is to grab an Agilent PSA spectrum analyser with the noise figure personality (option) fitted and use it with an external HP346 noise source.

At home I have a homemade noise source (NoiseCOM diode + precision low VSWR attenuator) that is very flat and stable to about 2.5GHz and I use this and a precision low noise preamp along with a decent spectrum analyser (eg TR4172) and a simple Windows program I wrote that gives the corrected noise figure using the Y factor method.

I get good results with it and I have a suite of gold standard LNAs and precision attenuators with known gain/loss and noise figure vs frequency that I use as a crude confidence checking system to make sure it stays reasonably accurate

[G0HZU_JMR]

Richard
Here's a link to a decent app note from Rohde and Schwarz about the Y factor method when used with a hot/cold noise source and a modern spectrum analyser.

I used the equations in their 1MA718 app note in my software and they give some worked examples that help prove my software is correct.

It also covers the theory behind the overall measurement uncertainty of the whole process and shows the many limitations of this approach especially wrt mismatch uncertainty and the noise source's own uncertainty.

http://cdn.rohde-schwarz.com/pws/dl_..._technique.pdf

The classic AN57 Agilent app notes on noise figure measurement suggested by David are also well worth a read

[Radio Wrangler]

Oh the Agilent NFA has done the rounds of all the usual low noise people. NIST, NPL, JPL, Deep Space group etc. Hot/cold loads are the usual backstop for noise source calibration, and the National Physical Laboratory (Teddington) were found to offer the lowest uncertainty figures. The American standards bureau, NIST ran into some trouble, they hinted that funding cuts were to blame, but a badly repaired roof let water in and flooded their ENR cal rig and they are taking years to get their uncertainties back down and proven. Noise/Com have done a great job of making nolise diodes available to ordinary people... and their noise sources were calibrated against an Agilent calibration rig, last I knew. The rig was developed for a certain super-power's navy, but became a product on sale because of demand.

The problem is that the uncertainties in the results from the best commercial gear, or at the nest national standards labs, are several times larger than what is claimed for some preamps. Amplifiers have certainly got ahead of even the best test gear.

Radio astronomers tweak things up for best difference in use between a cold piece of sky and a piece with something warmer in it and so get what they need.

The trouble with y-factor methods is that the Z of noise sources change from on to off and the amp insertion gain changes in sympathy. Anyway, all Ian and I have to say on the matter is in the conference proceedings

We do need something better to measure current technology amplifiers.

Apologies... this isn't very vintage, but it's interesting that only little progress has been made since the vintage noise fig instruments were made.

David

[trh01uk]

Folks,

thanks for all the updates. I was aware that noise figure is still being measured, but not being a professional any longer, and thus no longer falling over this sort of test kit on a daily basis, I noted that we see plenty of the vintage noise sources on the 2nd hand market - but very little that is relatively recent.

If there has been only a little progress in NF measurement techniques then it sounds like the older kit is still usable.

And I might add, that despite 27 years in the profession on various RF projects, I never once had to measure NF. At least not directly. I guess it got included by default when checking receiver sensitivity.

Clearly saving the odd dB in such systems as satellite links is pretty important, but all the systems I have worked on have been terrestrial, non-line-of-sight, with fade margins of 20 - 30dB typically. In that context shaving the ultimate fraction of a dB off the NF is of little interest.

Richard

[Radio Wrangler]

The 346 family are still fully useable noise sources. But go for the low ENR models. They have more attenuation between the diode and the output connector, so changes in impedance are diluted. The change in source Z applied to the amplifier being tested causes the amplifier's insertion gain to change (and also its noise figure) Unfortunately this correlates perfectly with the noise on/off switching for the Y-factor method, and so the noise figure result is erroneous. Depending on where the input match of an amplifier is set, this can make the displayed noise figure result read higher or lower than reality.

So, if you take a preamp, and put it on a noise figure instrument with a noise source, you can tweak the amplifier to give a commendably low displayed noise figure. Unfortunately what's been done is to set the preamp to maximise the instrumentation errors in the negative direction. Even more unfortunately there are no clues to what's gooing on, the user just sees a really nice result and he isn't going to be inclined to argue with that.

There's a measurement uncertainty calculator somewhere on the Agilent website for the brave individual.

There are noise figure measurementfests at most Microwave Roundtable meetings and EME events. Some of the amplifiers tested have claimed noise figures of about 0.06dB, but the measurement gear will have a statistical uncertainty of 0.15dB or higher. Usually there isn't even a thermometer in use, and an error of 6 degrees C in the ambient temperature guessed, is equivalent to 0.1dB error in the result.

High ENR noise sources (about 16db) are still valuable. Add a good 10dB attenuator and keep it fitted. Calibrate the whole thing as a low ENR source. I use an HP346 (APC-7 connector option) noise source with a 10dB APC-7 precision attenuator fastened to it. 33 year old kit, and with a thermometer and good calibration, it's as good as the latest.

David

[G0HZU_JMR]

Clearly saving the odd dB in such systems as satellite links is pretty important, but all the systems I have worked on have been terrestrial, non-line-of-sight, with fade margins of 20 - 30dB typically. In that context shaving the ultimate fraction of a dB off the NF is of little interest.

Agreed. In my case I often had to design to achieve less than a maximum noise figure across a very wide (terrestrial) receiver frequency range. A typical design for me back in the 1990s would be a high performance RF up/downconverter ahead of a DSP rack.

The converter would tune across from a few MHz up to a few GHz with a final IF BW of a few MHz feeding to the DSP.

So I would usually have to be able to measure system noise figures of typically 8 to 12dB across this range. Often it was sufficient to use a trusted and decent spectrum analyser (HP8568B or TR4172) and a decent sig gen. Simply measure the converter gain and then measure the noise floor coming out of the converter in dBm/Hz (if it has enough noise level to dominate the analyser noise floor)

From these measurements the system NF can be calculated.

This could be automated and was sufficient to produce a quick and dirty measure of system noise figure across the full range of the converter albeit with various (significant) contributors to overall measurement uncertainty.

For design and verification of LNAs the system above would be inadequate and for stuff like this I either use the HP346+HP8970B or the PSA spectrum analyser with the NF option fitted.

I've not really ventured into the ultra low noise LNA field though. The lowest designs I've worked on were nothing special and the accuracy of measurement was not critical eg 1dB NF max with 0.5dB noise figure 'typical'.

[Radio Wrangler]

Aye, ultra low noise figure only becomes worthwhile for things with their antenna pointing at the sky, clear of the ground. The definition of noise figure and noise factor involves comparison with the noise of a matched load at 290 Kelvin, so all things at comfortable temperatures make comparable noise to things with 3dB NF.

David

[trh01uk]

The other thing this thread highlights in my mind is the gulf between the high and low end RF design work. To do NF measurements, I presume you must have 50 ohm (or at least defined impedance) ports at input and output of your amplifier (or sub-system you are measuring. Indeed much RF theory assumes such.

I was in the commercial field where every penny counted, and thus 50 ohm ports were non-existent. Clearly we could have transformed - on a temporary basis - to 50 ohms, but there is always the suspicion that such text fixtures are going to mess up the measurement and add further losses which can't be accounted for.

I wouldn't want to give the impression that we designed "on a shoestring". Indeed, I managed to justify the purchase - in the 1980s - of one of the first series of HP8753 VNAs for our development work. At a cost of some £25k - which was a lot back then. My boss was extremely miffed - as it stopped him getting his own pet project through - a first network for all the labs pcs!

Richard

[trh01uk]

And a personal note to you David, I surmise from your comments in this thread that you work at Agilent at South Queensferry. At the time I am referring to in my post above, in the 1980s, I was at Racal, then near Ratho (Lochend Industrial Estate). On that site, we had the two extremes of types of RF design on the same site. We had Racal MESL, which was doing high-end microwave stuff, and we had Racal Security (later Racal Guardall, then Guardall, then ??), where I worked and we did the very cost sensitive, high volume stuff for the alarms market.

I always wanted to work at (then) HP at South Queensferry. Indeed I landed a job there - only to have the offer withdrawn due to some economic downturn. I recall the bosses at HP elected to take a 10% pay cut then, to avoid making anyone redundant. That's what I call a good company to work for!

Richard

P.S. Mods - you may think this off-topic, but I couldn't see quite where else to post while keeping continuity with the thread.

[Radio Wrangler]

To give the ending to the tale of Noise Figure measurement R&D in the UK; Agilent South Queensferry is no more. Only bulldozers roam across the site. There are a few people left doing sales etc and a bit of software localisation in rented office space at the Gyle, but the rest of us are long dispersed.

The NFA and its noise sources were developed there with a good many other things. I kicked-off the NFA investigation then led the hardware/metrology design. I did the noise source as a little solo project. It was rather profitable and kept the place going for a few years until that product line (and its income) were shifted elsewhere.

The Ferry was a place everyone in the area wanted to move to if they got a chance, but things got unpleasant towards the end. 10% voluntary pay cuts by everyone were the usual way of handling downturns. It meant everyone was in place and ready to go again on the upturn.

David