Used Spectrum Analyzers for under $1k

[TSL]

Hey joeqsmith,

Well that's impressive - whats that written in ?

thanks

Tim

[Wuerstchenhund]

Here are some screenshots of the 28MHz 0dBm sine from a Rohde & Schwarz CMU200.

I used a different cable (standard RG-58 BNC) with H&S BNC-to-N adapter and another 10dB attenuator.

The input attenuator of the CMU200 has three operation modes (Normal, Low Noise, Low Distortion) so I captured the results in all three modes.

Let me know if you want to see any other tests. Unfortunately I can't do things like the two tone test suggested by G0HZU as I don't have a suitable power combiner at the moment (and not the time to build one).

[G0HZU]

Hopefully you can now see how noisy the basic LO synthesiser is in the E7495A as it performs very poorly in all the tests.

The plots from the R&S test set look a lot better but I think that if you measured the carrier noise at 100kHz, 500kHz and 1MHz it would not be in the same class as an old school lab analyser. There will be compromises made here because this is a test set and not a lab grade spectrum analyser.

The same applies for spurious performance and signal handling. If I had both test sets here I think I could find aspects of the RF converter performance that are severely compromised when compared to a dedicated analyser. Obviously, the test set has a modern DSP back end and a host of other tools and features that don't even exist on an old school analyser but if (like me) you want to do fairly serious RF design then test sets are not the tool of choice as they are too compromised in terms of the raw RF performance they offer in the converter section.

[Wuerstchenhund]

There will be compromises made here because this is a test set and not a lab grade spectrum analyser.

Yes, but I never said that they are. But reality is that not everyone needs the performance of a high end SA, and both sets are still better than what I've seen from a Rigol DSA830.

All I wanted to do is to show that there are alternatives between these old high end SAs and the Rigol, and if you can live with their performance then they might be a reasonable option, and without the size, power consumption and reliability issues of the old kit.

I can see the E7495 as a great tool for someone starting with RF stuff, as it's a not just a VSA but also a RF generator/AWG and a Power Meter (although you need a measurement head), is compact and can often be found for little money.

The other thing is that the E7495 is running Linux, and that there's a huge potential to hack it into a nice unversal portable test set for other stuff than cell phone testing.

[Rupunzell]

For serious RF-Microwave folks, performance of the RF sections matter most. Extra features offered by DSP are never going to make up for a RF section with limited performance.

It might be worth noting that a good spectrum analyzer is a absolute requirement for communications RF and microwave work as they can provide information and test data that a time domain instrument can not. Different tool for a different job.

And yes, that hp 8568 looks as good as ever. BTW, one good old hp 8568 was used to develop one of the many cameras that is currently living on the Mars rover Curiosity.


Bernice

Hopefully you can now see how noisy the basic LO synthesiser is in the E7495A as it performs very poorly in all the tests.

Obviously, the test set has a modern DSP back end and a host of other tools and features that don't even exist on an old school analyser but if (like me) you want to do fairly serious RF design then test sets are not the tool of choice as they are too compromised in terms of the raw RF performance they offer in the converter section.

[G0HZU]

Hey joeqsmith,

Well that's impressive - whats that written in ?

thanks

Tim

Thanks.  That was done with Labview 5.

Yes, very impressive!

You should do a youtube demo

[alex.forencich]

You can also do higher resolution screenshots with these old analysers.

The display has 1000 x 1000 datapoints inside the grid area so each box in the grid has 100 x 100 data points in system memory.

The plot below is the highest resolution that the plotter program can go to but I think the analyser is better than this... The analyser can do 1000x1000 inside the grid area and the image below is only 1024x768 for the whole image so the plot still looks a bit grainy.

(making screenshots for example )

Can you show a screenshot from your modern E7495A analyser with finer screen resolution than this 35 year old technology?

I suppose I could try writing my own plotter capture program. The TR4172 allows full access to all system memory via GPIB so I should be able to grab the 1000x1000 grid data and produce a nicer image still...

Is it actually stored as a 1000x1000 grid or is it stored as a vector?  My guess is a vector, not a grid (1000 points at 1000 possible levels per point) - this is how the 856x series analyzers work, and why the display is so much crisper than the 859x series.  Anyway, if you want "native res" for vector, what you need to do is store it as a vector file like an svg.  I have been working on some Python code that translates HPGL read out of an analyzer to XML and writes it out as an SVG file.  It works with the trace from my 8593E, and it should work for other instruments as well, possibly with some bug fixes.  You can then open the svg in a vector graphics editor like Illustrator or Inkscape and play around with it, or export it as a raster file like PNG.  (It's actually not terribly difficult to convert HPGL to SVG, the existing plotting software is horribly complex because it actually renders the HPGL to a raster format instead of just converting it)

[G0HZU]

Hey joeqsmith,

Well that's impressive - whats that written in ?

thanks

Tim

Thanks.  That was done with Labview 5.

Yes, very impressive!

You should do a youtube demo

Thanks.   The start of this video shows my home made discriminator that I use with that particular Labview program.    If you would like, I could show how to make a measurement with it. 

I'd be very interested to see that if you have the time to do it.  The discriminator looks very interesting and you have built it very well!

[G0HZU]

Is it actually stored as a 1000x1000 grid or is it stored as a vector?  My guess is a vector, not a grid (1000 points at 1000 possible levels per point)

I'm afraid that I can't answer your question very reliably as I've never tried to capture screen images with my own code but I think it is possible to access any part of the system memory via GPIB.

So I think I could dump a string from the analyser containing the trace memory.
I think the start address for the trace memory is 0xC018.

I'd expect to be able to dump 1000 memory points from here where each point takes up two bytes.

In the byte pairs would be a data value from 0000 to 03E8 in hex where 3E8 = 1000 in decimal. If it was 3E8 then it would be at the top of the grid and zero would be at the bottom. But all this is just guesswork looking at the user manual.

[G0HZU]

It would be interesting to see it measure and plot phase noise and see how far down it can measure using this method. But please only do this if it is convenient for you and doesn't take up too much time...

[alex.forencich]

Is it actually stored as a 1000x1000 grid or is it stored as a vector?  My guess is a vector, not a grid (1000 points at 1000 possible levels per point)

I'm afraid that I can't answer your question very reliably as I've never tried to capture screen images with my own code but I think it is possible to access any part of the system memory via GPIB.

So I think I could dump a string from the analyser containing the trace memory.
I think the start address for the trace memory is 0xC018.

I'd expect to be able to dump 1000 memory points from here where each point takes up two bytes.

In the byte pairs would be a data value from 0000 to 03E8 in hex where 3E8 = 1000 in decimal. If it was 3E8 then it would be at the top of the grid and zero would be at the bottom. But all this is just guesswork looking at the user manual.

There isn't a command to just read the raw trace data directly?  That would seem like a very odd omission. 

[G0HZU]

I had a go at dumping raw trace data from the TR4172 memory.

It does seem to be arranged as 1000 pairs of bytes each holding a value up to about 03E8.

The analyser allows the data to be dumped out very efficiently so the plot below takes a fraction of a second to appear. Obviously I haven't bothered to add the scaling or other onscreen text but it shows the raw trace data against a basic grid background.

I think I might take this further because I think I can get about 5 screen dumps a second with my code and this is in contrast to the KE5FX toolkit plotter that takes maybe 20 seconds to grab just one screen plot

There isn't a command to just read the raw trace data directly?  That would seem like a very odd omission

I'm not really sure if I'm dumping the info correctly or not. But there is a command that allows the trace to be dumped out as a huge string or a set of strings. I think I can dump out any memory address range in any size chunk I choose. However, the manual appears to have been produced by a dyslexic japanese to english  translator and there are quite a few typos in the manual wrt the relevant commands to control the analyser via GPIB

[alex.forencich]

With my 8593E, you can just send TRA? and it will spit out the data points.  There are a couple of other options to change the data format to binary or ASCII as well.  It's rather the whole point of the remote interface, I would think.  I presume there is a similar command for your analyzer as well, though it may be hard to find. 

[joeqsmith]

You may want to have a look at CERN's program.   I have used this for several years to convert HPGL files and it works great.   

http://service-hpglview.web.cern.ch/service-hpglview/download_index.html

[joeqsmith]

The corrected video is up. 



The video is about 25 minutes long.   The attached picture shows the end results.  Blue trace is the measured phase noise and Green is the noise.    Also shown are the calculated values from the simulation allowing you to compare the results.

[Electro Fan]

raising the bar above $1k...

Anyone have experience with both the 8560E (or other same series models) and the Agilent E4400 Series (E4403, E4402B, etc.), or have any insights/thoughts on how the 856XE compares to the E44XXB?  The specs might be better on the 856XE but the user interface might be better on the E44XXB?  (I noticed in a couple of videos it looks like w2aew uses a E4411B.)

PS, joeqsmith - very cool posts, I'd be happy and honored to share this thread with you - please keep posting your excellent work!

[G0HZU]

The corrected video is up. 



The video is about 25 minutes long.   The attached picture shows the end results.  Blue trace is the measured phase noise and Green is the noise.    Also shown are the calculated values from the simulation allowing you to compare the results.

That looks very good, thanks for doing the video. The noise floor limit does look to be quite good

The only phase noise measuring capability I have here at home is via my TR4172 and HP856x spectrum analysers but I'm lucky to have access to a couple of Agilent E5052A SSAs at work. I keep meaning to make something for home use but never get around to it.
In the meantime I'm hoping that one of the works E5052A analysers will fail and be declared BER and I'll have a chance to buy it cheaply

[G0HZU]

Anyone have experience with both the 8560E (or other same series models) and the Agilent E4400 Series (E4403, E4402B, etc.), or have any insights/thoughts on how the 856XE compares to the E44XXB?  The specs might be better on the 856XE but the user interface might be better on the E44XXB?  (I noticed in a couple of videos it looks like w2aew uses a E4411B.)

At work we have dozens of lab grade analysers and we also have some of the E4407 type analysers. These don't offer the same RF performance as the HP856xE analysers but they do have a more modern user interface and a fresher looking display.

The E4407 analysers we have at work have an optional self cal routine that makes an annoying cheepy/chirpy noise each time it runs. A bit like there is a small hatchling bird inside singing for food when it gets hungry every few minutes. You can't hear it in a big noisy lab but in a small room it is very noticeable and I find it very distracting 

It can be turned off in the menus after it has warmed up but this is at the risk of losing some accuracy over time.

The E4407 is also a very good looking and nicely built analyser and it is quite easy to move around (not that heavy) and has very chunky rubber support blocks around its extremities giving it a reasurringly robust feel.

They tend to get used for field trial work or for basic verification tests. Especially by people doing basic system testing or faultfinding. But for serious RF design work the HP856xE analysers are much better in terms of the raw RF performance. eg noise and spurious performance and overall measurement uncertainty.

But top dog at work are the E444x PSA analysers and we also have a few PXA analysers although some of these may be rented. I'm assuming that you aren't including the E444x in your list as these are very expensive. I'd like to have one here at home but can't justify the cost of a used one 

[joeqsmith]

Thanks for the comments.   I realized after making the video that I never showed the final jitter number and I am sure I missed some other details.   Hopefully there were enough details that people will have some idea of how this measurement is made.   

The only phase noise measuring capability I have here at home is via my TR4172 and HP856x spectrum analysers ..

The Advantest looks to have a fair amount of features.   The 3589A can also display complex data using an external coupler.   The problem with it (besides the limited BW) is that their internal software is crippled without having a real test set connected. 

[LabSpokane]

On the Agilent E7495B, does any one know if the option 510 CE signal generator is a tracking generator or just a single frequency? 

[alex.forencich]

On the Agilent E7495B, does any one know if the option 510 CE signal generator is a tracking generator or just a single frequency?

Product brief calls it a "reverse link signal generator" for several cell network standards.  So it seems to be a modulated CW source of some sort, not a tracking generator. 

[LabSpokane]

On the Agilent E7495B, does any one know if the option 510 CE signal generator is a tracking generator or just a single frequency?

Product brief calls it a "reverse link signal generator" for several cell network standards.  So it seems to be a modulated CW source of some sort, not a tracking generator.

Yup. That's what I saw on the data sheet as well, but it's a different wording on the installed option screen of the device. So, thought I would ask.

[Wuerstchenhund]

On the Agilent E7495B, does any one know if the option 510 CE signal generator is a tracking generator or just a single frequency?

The E7495 is a VSA (Vector Spectrum Analyzer) which works by FFT and not by sweeping what essentially is a superhet receiver, and since there is no frequency sweep there can't be a tracking generator.

The E7495 has three signal generators. One is standard (a swept generator used for the Insertion/Return Loss measurements), and option 510 enables the other two, a CW generator (option 500) and an Arbitrary Waveform Generator.

[LabSpokane]

D'oh!  I knew that for earlier in the thread. I think I can pick one up for less than $1000 US. Any issues to look out for? 

What features will be missing from the spectrum analyzer versus say an 8593E?  I know that the latter will be more sensitive and more accurate, but will the handheld be missing essential analysis tools?

[Wuerstchenhund]

D'oh!  I knew that for earlier in the thread. I think I can pick one up for less than $1000 US. Any issues to look out for? 

What features will be missing from the spectrum analyzer versus say an 8593E?  I know that the latter will be more sensitive and more accurate, but will the handheld be missing essential analysis tools?

Depends on what you need. The E7495's SA frequency coverage starts at 500kHz but below 375MHz the noise floor is higher than above that due to the input amplifiers (can also be seen on the screen shots I posted a while back which show signals below 375MHz; still had no time to do some more thorough tests with mine). Functionality-wise it's the same as any SA. You get 5 markers (although only one can be read out at any given time), manual and automatic scaling, manual and automatic RBW settings. There are other goodies like a Spectrogram or measurements for Occupied Bandwidth (OBW) which can be useful.

The frequency accuracy of the E7495 is also pretty good, especially when GPS stabilized. And if you wanted a TG for checking cables and filters then the E7495 can do that as well. And having all that in a relative compact format that doesn't take much desk space and is ruggedized is a bonus. As is a display that can even be read in full daylight.

One of the drawbacks is the limited remote control capability. There's no SCPI or stuff like that, there's only a JAVA GUI (actually the same GUI as on the device itself) which requires an antique and insecure JRE (1.4) to work.

Generally I'd say if a Rigol DSA830 would do the job for you then the E7495 should be fine. Just be aware that the E7495A only goes to 2.5GHz while the E7495B goes to 2.7GHz.