Tektronix Tek RSA306 6GHz signal/spectrum analyser

[G0HZU]

Anyone else got one of these?

I've got one here to have a play with and I wondered if there were any common issues to look out for when first installing everything. Are there any patches or updates or known bugs/issues?

I'm running Win 8.1 64bit but my test PC is a fairly old quadcore and the USB 3 card in it is fairly basic so I'm not expecting to get much in the way of optimal (real time)performance from it. I've run it up and done a few RF tests on it. Some results are quite good and some things not so good but I guess you can't complain on something that only costs about £2400.

It doesn't belong to me so I'm just having a play to see how useful it could be for various future tasks/projects compared to some alternatives.

[dadler]

Shahriar played with one:

http://thesignalpath.com/blogs/2014/11/04/tektronix-rsa306-usb-real-time-spectrum-analyzer-review-and-experiments/

[G0HZU]

Yes, that's a good video. I watched it when he first released it. I don't know if Shahriar still has the 306 but he hinted he may do more videos about it in the future.

I'm currently tasked with trying to characterise the frequency plan and the spurious limitations. However, my PC doesn't meet the recommended system requirements so I'm seeing a few niggles in operation. I'm hoping to have a better PC available next week.

I've done a bit of reverse engineering by sniffing the RF ports of the 306 and I've now created a computer 'model' of the frequency plan and this model mimics the places the 306 hops between it's (two) first IFs as it sweeps up to 6.2GHz. It seems to swap back and forth in quite a few places in order to minimise spurious issues etc. My computer model now predicts where all the major spurious tones will be plus it predicts where there may be issues with IF breakthrough and some of the more significant mixer spurious terms.

The model also predicts the internal spurious tones that appear in Shahriar's video and why they sometimes appear and disappear.

So far the model has predicted quite a few problem areas where the instrument may not meet the -50dBc (typical) spurious spec and my initial tests on the real hardware confirm this. The worst places appear to be the little pockets of frequencies in the upper ranges where it swaps back and forth between either the 1190MHz IF or the 2440MHz IF.

[G0HZU]

Just to put this in perspective this particular example of the RSA306 can be set to certain frequencies and spans where there are -9dBc spurious or -15dBc mixer spurious as predicted by my model for a -10dBm test tone into the RSA306. i.e. inject -10dBm and see the wanted tone at -10dBm but also see spurious terms also appear with it just 9dB and 15dB below this.

These aren't internal spurious or 'birdies' these are mixer spurious terms caused by the frequency planning in certain parts of its range.

There's nothing wrong with my signal sources (all HP or Agilent or Marconi) and they are all checked for spurious on a HP8566B analyser up to 22GHz anyway. All are clean.

The computer model can find these special places instantly but it is much harder to find these places with a manual (blind) search because you have to know what precise RSA306 centre frequency to use and what span to use and what test tone frequency to inject to get it to misbehave. With the model I can find other spurious mechanisms at -15dBc in other places and they are also proven to be there when the hardware is tested.

Maybe my RSA306 has duff firmware or maybe it has a fault? Or maybe my setup is causing it to misbehave?

I'm hoping someone else on here can replicate some of my tests and compare numbers for board version, FPGA version and firmware version. Can anyone advise?

[G0HZU]

Here's a few screenshots to show some of the issues.

I used the computer model to predict a place where both the -9dBc term and the -15dB spurious term can be seen together on a single (big) span. Then I replicated this with a clean lab grade sig gen and the real hardware in the screenshots below.

In the first plot there is no signal in and the spectrum is acceptably clean with minimal internal spurious.

In the second plot there is a clean -7dBm signal fed in and the spectrum is acceptably spurious free.

In the third image I program the RSA306 to shift down in frequency just a few MHz and you get to see one of the spurious terms suddenly appear onscreen as predicted by the model. Note that I haven't changed the sig gen settings.

In the final screenshot I've arranged the RSA306 CF and span and the sig gen to show an additional spurious term. So there is the wanted signal and one term at -9dBc and one at -15dBc as per the model. The sig gen is not causing these two extra terms. They are not there on a HP8566B analyser when set to the same CF and span.

Note: They are not caused by overloading with the test signal. I can turn down the sig gen 10dB (or 20dB) and the spurious terms track to still be at -9dBc and -15dBc.

The reason they are there is easy to calculate once you know the frequency planning the RSA306 is using when it tunes across the span range and stitches this this span plot together in various segments.

On a conventional spectrum analyser this level of mixer spurious performance would be a bit of a disaster but I guess the RSA306 is really a signal analyser (and not a conventional spectrum analyser) and some sympathy needs to be given to it under certain test conditions. It doesn't prevent the RSA306 from being an exciting and new type of portable real time analyser but if this example is typical of others then it can't always be relied on to maintain a reasonable spurious free display. Its real strength is in the 40 MHz real time capture bandwidth it offers and the post processing tools in the free bundled Tek software in the PC.

I can post up why these terms appear if anyone is interested? I can post up other scenarios where similar thing happen and also a few places where some very odd things happen because of the way it manages its frequency plan and display.

I think this behaviour could be different with different firmware versions but I only have one RSA306 here to play with

[G0HZU]

Here's couple more examples:

There's only one -10dBm tone being fed into the analyser in both cases below so each plot should just show one test tone at -10dBm. But this particular RSA306 shows extra mixer terms at only a few dB below the wanted term. I can imagine that this could cause some confusion for some users. When the sig gen is turned off all the tones disappear. The sig gen is clean and verified on a HP8566B analyser. Also, I only have to tune the RSA306 a few MHz and the phantom tones suddenly disappear. They also change frequency spacing relative to the wanted tone when I tune the RSA306. So that proves again that they are generated internally in the RSA306 as I'm not altering the sig gen during the test. They can also be predicted to be there with a few (mixer spurious) sums based on the RSA306 frequency plan.

[Psi]

Just a crazy thought, (ive not watched the video)

Are you sure it's not something to do with the CPU clock of the PC your using being picked up?
Core i3/5/7 cpus and i think Core2 as well will boost the cpu clock speed between two points depending on load, temp and number of cores being used.
So your CPU clock can jump around depending on workload.
2.4ghz is a common cpu frequency as well.

[w2aew]

One thing I noticed in your screenshots (lower left corner in the status bar) says that the unit is "Not Aligned". Running the  alignment may help a little (Tools>Alignment). I also know that there have been some mid-production changes which improve the internal shielding, and an improved vector calibration. All that being said, there will still be a few Spurs that will be worse than the typical spec. - par for the course for a 40MHz R/T BW 6GHz SA that had to be built with a 4.5W power budget.

[hendorog]

It would be interesting to see a comparison between the Tek and the Signalhound BB60C. Perhaps they would send you one to evaluate?

Signalhound have a comparison chart here:
http://signalhound.com/download/signal-hound-bb60c-and-tektronix-rsa306-comparison-chart/

[Hugoneus]

It would be interesting to see a comparison between the Tek and the Signalhound BB60C. Perhaps they would send you one to evaluate?

Signalhound have a comparison chart here:
http://signalhound.com/download/signal-hound-bb60c-and-tektronix-rsa306-comparison-chart/

That review/comparison is coming up soon. I'll try to speed it up.

[w2aew]

It would be interesting to see a comparison between the Tek and the Signalhound BB60C. Perhaps they would send you one to evaluate?

Signalhound have a comparison chart here:
http://signalhound.com/download/signal-hound-bb60c-and-tektronix-rsa306-comparison-chart/

There are a few inaccuracies in this chart. The latest version of firmware/API delivers streaming corrected IQ over the full 40 MHz BW, and delivers swept spectrums.  Lots of other RSA exclusive features are not mentioned.

[Hugoneus]

It would be interesting to see a comparison between the Tek and the Signalhound BB60C. Perhaps they would send you one to evaluate?

Signalhound have a comparison chart here:
http://signalhound.com/download/signal-hound-bb60c-and-tektronix-rsa306-comparison-chart/

There are a few inaccuracies in this chart. The latest version of firmware/API delivers streaming corrected IQ over the full 40 MHz BW, and delivers swept spectrums.  Lots of other RSA exclusive features are not mentioned.

w2aew,

I am aware of all the new features of the RSA and will mention them in my next review. Not to worry.

[Rupunzell]

This effort could be a good instructional item for many who are curious about or want to learn more about spectrum analyzers in general, why spurious images on the display matter and where they come from.

Good teachable moment here.


Bernice

I can post up why these terms appear if anyone is interested? I can post up other scenarios where similar thing happen and also a few places where some very odd things happen because of the way it manages its frequency plan and display.

[G0HZU]

One thing I noticed in your screenshots (lower left corner in the status bar) says that the unit is "Not Aligned". Running the  alignment may help a little (Tools>Alignment). I also know that there have been some mid-production changes which improve the internal shielding, and an improved vector calibration. All that being said, there will still be a few Spurs that will be worse than the typical spec. - par for the course for a 40MHz R/T BW 6GHz SA that had to be built with a 4.5W power budget.

Hi Alan
I get the same results even after a 30 minute warmup and an alignment. I'm guessing that the alignment helps with amplitude calibration using an internal test tone.

I did try and model a case where I could display three spurious tones in one sweep that are above -50dBc. I can try for four cases but it gets harder for me to configure the search model.

But see below for three large mixer spurious terms generated from a single clean signal into the RSA306. All of these are (RF , nLO1) terms where n <>1.

I'm thinking that this RSA306 is misbehaving or maybe it has the wrong firmware because I wouldn't expect to see a frequency plan that would allow terms like this to appear.

In the plot below tone the biggest tone at about -4dBm is the expected tone and everything else is a mixer spurious term that appears with it.

In case anyone thinks this is caused by stray interference or my sig gen is faulty I just have to tune the RSA306 a few MHz to one side to see all its mixer spurious terms disappear.

See the second plot below to see what I mean. I've used the same sig gen settings for the 'clean' plot below. My model also predicts a clean plot here with no major mixer spurious terms.

[Melt-O-Tronic]

This effort could be a good instructional item for many who are curious about or want to learn more about spectrum analyzers in general, why spurious images on the display matter and where they come from.

Good teachable moment here.


Bernice

I second that!!!   

[G0HZU]

A decent tutorial on downconverter design would take a lot of time to prepare and (sadly) I simply don't have the spare time to do this

But I can show you why the frequency planning of this RSA306 is so flawed in one place (of many).

For example, in the tuning range 3.7GHz to about 4.56GHz this particular RSA306 analyser has a first IF of 2440MHz and it reverts to a 'low side' LO signal across this sub range.

So to convert a centre frequency of 3.7GHz to the 2.44GHz first IF it uses an LO frequency of 3.7-2.44 = 1.26GHz. This is not a good frequency plan. It is such a poor plan that I can't believe it is true. There must be something wrong because it is such a poor frequency plan for something that is marketed as a spectrum analyser.

The reason it is such a poor frequency plan is that the mixer type used in the analyser has fairly decent mixer conversion efficiency at odd order multiples of the LO frequency.

In the above case the 5th harmonic of 1.26GHz is 6.30GHz.

So if you tuned the analyser to 3.7GHz and injected a signal at 6.30-2.44 = 3.86GHz you will see the analyser display this 3.86GHz signal as if it were at 3.7GHz but maybe 15dB lower than expected.

In other words, when tuned to 3.7GHz centre frequency, this analyser will only have 15dB rejection of a signal at 3.86GHz.

There are worse examples than the above in other sub ranges. eg there is a case where the rejection is only 9dB where the third harmonic of the LO provides the unwanted mixing.

My overall impression of this analyser is that it is capable of surprisingly good performance over parts of its tuning range but the frequency planning above about 2.3GHz is not good. This results in lots of very significant spurious responses like the example above. So the mixer spurious performance is extremely woeful in various sub bands across approx 2.3GHz-5.8GHz. I suspect that there could be a bug in the firmware in this particular RSA306 that sometimes configures a very poor frequency plan in terms of IF choice and LO management (in terms of the choice of low or high side injection)

There are also lots of internal spurious responses or 'birdies' caused by harmonic mixing of the local oscillators. The behaviour of these can be predicted (and subsequently managed) if the design were developed and improved to be a bit more flexible.

I no longer have it here but I could probably have it back here at some point in the future. In the few hours I had access to it I only looked at the downconverter performance because that was my task. I did test the downconverter for lots of other things. I also tested it for input match, phase noise performance, various classic SFDR tests and image/IF/Alias tests and DANL.

Overall, it did quite well considering its size, power and cost limitations. However, I do feel it is better described as a signal or 'transmitter' analyser rather than a spectrum analyser because of the poor spurious performance in parts of the tuning range. I'm still clinging to the hope that the example I had was a dev version with experimental firmware

[w2aew]

I had a conversation with some of the designers - and learned that there have been recent changes in the frequency plan, unit firmware, RF path vector calibration and operating software.  They sent me a screenshot of the same result you showed (CF=3GHz, Span=2GHz, RBW=1MHz, with a 3.78GHz input signal) that showed that all three spurs were gone.  FYI - I duplicated the results you showed (I haven't had my unit updated yet).

[xygor]

Tektronix (seller "tektronixused") has refurbs for half-price on ebay. 2 remaining.