USB network analyzers -- as bad as USB oscilloscopes, or not?

[rs20]

I agree entirely with the statement that USB-based oscilloscopes are a fairly bad choice compared to, say, a DS1054z. Performance, waveform/sec, OS compatibility and responsiveness issues aside, the knobs and buttons on a "real" oscilloscope are so lovely and tactile that you can find and use them without even looking. I see that this is a good thing, compared to messing around with a mouse.

However, I saw a recent (excellent) video by Shariah discussing the Signal Hound BB60C USB signal analyser, and it got me thinking. I was considering the Rigol DSA815-TG 1.5 GHz spectrum analyzer, but the specs of these PC-connected units appear (to my untrained eye) to be vastly superior. Am I missing something; does the DSA815-TG have any advantages to justify the equal cost for 1.5 GHz (compared to the 4.4 GHz Signal Hound USB-SA44B + USB-TG44A)?

                         DSA815-TG                     SA44B+TG44A
Price                      $1495                          $1475
Frequencies            9 kHz to 1.5 GHz               1 Hz to 4.4 GHz
Dynamic range               ??                    -151 dBm to +10 dBm
Amplitude error          "<1.5 dB"                    "+/- 1.5dB"
DANL               -135dBm ("typical")      -135dBm at worst (from 10kHz to 3.3 GHz)
Phase noise     -80 dBc/Hz @10 kHz offset                  ??

Now I know that "which X is better" type question are often dumb and often best answered with "depends on your usecase" -- so to be more clear, can someone please contrive a usecase for which the DSA815-TG would be better?*

And what about the PC vs standalone device question? With oscilloscopes, the interactivity and speed of custom-layed out physical knobs is considered by many to be critical for usability. Is the spectrum analyzer workflow typically so different that this isn't true? (Seems plausible to me that there's less "hunting for a signal" or deep zooming/scrolling with an SA compared to an oscilloscope, and that with an SA it's more common to set up a fairly intricate configuration, and then you hit "go" and watch the result roll in -- more computer-suited, less interactivity-suited).

* For the record, my usecase is "random/general use" -- e.g., testing cables, exploring the RF in my neighbourhood, analysing control loops, messing about with time-domain reflectometry, ultrasonic transducers, etc, etc.... just random hobbyist stuff.

[tggzzz]

* For the record, my usecase is "random/general use" -- e.g., testing cables, exploring the RF in my neighbourhood, analysing control loops, messing about with time-domain reflectometry, ultrasonic transducers, etc, etc.... just random hobbyist stuff.

If that's "all" you want to do, you may like to consider using an RTL SDR dongle plus noise source. There are significant limitations, of course, especially dynamic range scanning time, 25MHz-1.5GHz, but they only cost £40 or so! Think of it as a reason to use thought and imagination

[pascal_sweden]

The general issue with using a computer is that the operating system could start to exhibit stability issues, when you install too many applications on it.

This can be of course solved, by reserving a dedicated computer, which you only use for connecting the USB signal analyzer, and on which you don't install any extra software.

But still you will always have to boot your computer, and boot times can be longer with a computer compared to with embedded test equipment.

Last but not least, a real Spectrum Analyzer is just more impressive and show off on your test bench. The Rigol DSA815-TG is very affordable and a nice show off equipment. Much better than an old laptop and small USB box

Most electronics hobbyists here use computers for reading email, browsing the internet, and playing some games in a while. When it comes up to test equipment, they prefer the real deal.
That includes also analog test gear from several centuries, covering the 70's, 80's, 90's and beyond

[pickle9000]

* For the record, my usecase is "random/general use" -- e.g., testing cables, exploring the RF in my neighbourhood, analysing control loops, messing about with time-domain reflectometry, ultrasonic transducers, etc, etc.... just random hobbyist stuff.

If that's "all" you want to do, you may like to consider using an RTL SDR dongle plus noise source. There are significant limitations, of course, especially dynamic range scanning time, 25MHz-1.5GHz, but they only cost £40 or so! Think of it as a reason to use thought and imagination

And just fun to play with!

[rs20]

If that's "all" you want to do, you may like to consider using an RTL SDR dongle plus noise source. There are significant limitations, of course, especially dynamic range scanning time, 25MHz-1.5GHz, but they only cost £40 or so! Think of it as a reason to use thought and imagination

A very interesting concept, and I thank you for suggesting it. And I already have an RTL SDR in fact! But I don't much enjoy the prospect of writing the code to convert it for all the reasons above, especially when the chip in question is not designed for anything other than video and the manufacturer does not support this usecase at all. It also doesn't cover the control loop, ultrasonic and TDR cases particularly convincingly either?

I'm mostly interested to hear from people that have used both USB and standalone network analyzers, and have opinions comparing the two.

[G0HZU]

Sadly, few people have the knowledge and experience to do a decent technical review of the RF converter section of a spectrum analyser. I've not seen anybody on the various blogs who really knows what to look for. From a technical point of view the reviews are pretty lame and don't even do any basic/classic tests of a spectrum analyser.

But then I think the rules for what makes a good spectrum analyser have changed in the last 5-10 years or so because there appears to be a new type of user. i.e. someone who doesn't have a typical RF background but who wants access to modern analysis technology (eg RF>digital IF and maybe also SDR) and is prepared to sacrifice RF performance and design integrity in order to get these modern features at an affordable price.

So you get disruptive products like the DSA 815 and the SH44 and the Tek RSA306 appearing on the market and these are tailored towards this type of user rather than someone like me who wants/expects some classic design integrity from a spectrum analyser. But this integrity normally comes at huge expense in terms of size, power, weight and cost.

I'd like to review both of the 815 and the SH44  but I doubt the manufacturers would want someone like me to do this as I would do a classic analysis of the RF converter design with a view to predicting (and measuring) the spurious performance and also do the various classic tests of dynamic range.

I did get a few hours with a Tek RSA306 demo unit and after some basic reverse engineering I managed to model/predict how awful the spurious performance was in some places. It's claimed to achieve about 50dB spurious free range but I was quickly able to model it and measure it and prove it was only about 9dB in places due to some pretty awful frequency planning in the RF converter. I only had one evening or so with it before it had to be returned.

I would expect to find similar holes in the performance of the Rigol and SH analysers but I doubt I will ever see one of these.

[G0HZU]

I'm mostly interested to hear from people that have used both USB and standalone network analyzers, and have opinions comparing the two.

I've not used the USB VNAs but some of them do look to be quite useful/accurate up into the VHF region. I'm a classic user of a modern lab VNA with access to various modern xxGHz VNAs at work and my typical use of a VNA is to take 2 port (or n port) models of components/circuits for subsequent modelling on an RF simulator on a PC. i.e. use the VNA to create/export  .n2p s parameter files for various components etc.

I've not seen much evidence of people using the USB VNAs like this but I'm pretty sure it could be done successfully if the bundled SW allowed it. I'd be interested to see just how successful this could be with a little USB VNA because I'd consider buying one for LF/HF use even though I have a reasonably good HP VNA here at home.
My VNA is fairly old (1990s 3GHz model) and it would be nice to have something smaller and more efficient when it comes to taking accurate 1 or 2 port models of various components up into the VHF region.

[w2aew]

Sadly, few people have the knowledge and experience to do a decent technical review of the RF converter section of a spectrum analyser. I've not seen anybody on the various blogs who really knows what to look for. From a technical point of view the reviews are pretty lame and don't even do any basic/classic tests of a spectrum analyser.

But then I think the rules for what makes a good spectrum analyser have changed in the last 5-10 years or so because there appears to be a new type of user. i.e. someone who doesn't have a typical RF background but who wants access to modern analysis technology (eg RF>digital IF and maybe also SDR) and is prepared to sacrifice RF performance and design integrity in order to get these modern features at an affordable price.

So you get disruptive products like the DSA 815 and the SH44 and the Tek RSA306 appearing on the market and these are tailored towards this type of user rather than someone like me who wants/expects some classic design integrity from a spectrum analyser. But this integrity normally comes at huge expense in terms of size, power, weight and cost.

I'd like to review both of the 815 and the SH44  but I doubt the manufacturers would want someone like me to do this as I would do a classic analysis of the RF converter design with a view to predicting (and measuring) the spurious performance and also do the various classic tests of dynamic range.

I did get a few hours with a Tek RSA306 demo unit and after some basic reverse engineering I managed to model/predict how awful the spurious performance was in some places. It's claimed to achieve about 50dB spurious free range but I was quickly able to model it and measure it and prove it was only about 9dB in places due to some pretty awful frequency planning in the RF converter. I only had one evening or so with it before it had to be returned.

I would expect to find similar holes in the performance of the Rigol and SH analysers but I doubt I will ever see one of these.

For what it's worth, I tested an RSA306 that was updated with the latest firmware, which includes updated frequency plan and vector calibration, and the SFDR under those test conditions you showed was dramatically improved.

[tggzzz]

If that's "all" you want to do, you may like to consider using an RTL SDR dongle plus noise source. There are significant limitations, of course, especially dynamic range scanning time, 25MHz-1.5GHz, but they only cost £40 or so! Think of it as a reason to use thought and imagination

A very interesting concept, and I thank you for suggesting it. And I already have an RTL SDR in fact! But I don't much enjoy the prospect of writing the code to convert it for all the reasons above, especially when the chip in question is not designed for anything other than video and the manufacturer does not support this usecase at all. It also doesn't cover the control loop, ultrasonic and TDR cases particularly convincingly either?

It depends on what you mean by "convincingly". For the TDR example I can already clearly measure the lengths of 19 cm and 29cm open/short circuit stubs located 3.1m along cable. In theory 1.5GHz => "distance quantisation" of 3.3cm. I'm still improving my technique for measuring the impedance/magnitude of discontinuities.

As and when I'm satisfied with my techniques, I'll publish my code.

People like G0HZU will, quite correctly, point out all the imperfections and limitations of the tools - but I merely regard that as a challenge (After all, how do you test the fastest scope or highest frequency SA in the world? )

[Howardlong]

In some ways it depends how often you use it and the convenience that a purpose made device offers. Personally, while I can see the initial excitement, I'd find it immensely boring using an SA to search out random signals across a great swathe of spectrum. SAs also tend to be significantly less sensitive than SDRs, so I'd suggest that using an SA on its own to search out elusive signals from the ether isn't necessarily the best tool, at least not on its own. Equally, SDRs tend to have limited spans compared to SAs. But as you increase the span on your SA, typically the sensitivity decreases, or the sweep takes a lot longer.

So what do I use? For the occasional field work for testing things like satellite uplinks I use a TTi PSA hand held SA and an Anritsu Sitemaster for a VNA. However most of my work is at my home office/lab, so I have a number of SAs of varying ages and frequency ranges. The usual go-to SA is the DSA815-TG, but I also have SAs to 3GHz and 26GHz but they are of the boat anchor variety, and although reasonably trustworthy, they don't have all of the modern features of the DSA815. For probing receiver signal paths, I often use an SDR with a home brew probe, and signal generator: you don't have to look up all the time, because you can hear the S/N by ear and identify where gain sections aren't working. While an SDR is little use for absolute measurement, they are reasonably handy for relative measurements, especially with weaker signals.

For VNA in the lab, I use my trusty old 8753A, it still works well and since replacing the CRT with a Newscope colour LCD it looks like a relatively modern instrument. I don't use it every day. Like the SAs I have, their use tends to be in sporadic bursts, unlike the oscilloscope which is on most days.

[pascal_sweden]

I did get a few hours with a Tek RSA306 demo unit and after some basic reverse engineering I managed to model/predict how awful the spurious performance was in some places. It's claimed to achieve about 50dB spurious free range but I was quickly able to model it and measure it and prove it was only about 9dB in places due to some pretty awful frequency planning in the RF converter. I only had one evening or so with it before it had to be returned.

I would expect to find similar holes in the performance of the Rigol and SH analysers but I doubt I will ever see one of these.

So you are saying that the R&D department at Tektronix, with many skilled engineers, having years of knowledge and experience, have missed things that one person can spot in a few hours?

How is this even possible? Hard to believe for me. Imagine they read your comment here.

I am sure that the R&D department is very proud on their art, and would never tolerate such basic sacrifices. If they have made design decisions to reduce the cost, they probably have thought about them in a clever way, so that these design decisions don't affect performance and that the product is still within spec and margins allowed in the RF industry.

Or to put it in another way: if you are that smart, and are the only person who knows how to do it right, how come you don't work for Tektronix? =)

[Wuerstchenhund]

So you are saying that the R&D department at Tektronix, with many skilled engineers, having years of knowledge and experience, have missed things that one person can spot in a few hours?

How is this even possible? Hard to believe for me. Imagine they read your comment here.

Based on the information available it wasn't really that hard to see where the flaws of the Tektronix device will be and to predict how it will perform. G0HZU also wasn't the only one who came up with that assessment.

I am sure that the R&D department is very proud on their art, and would never tolerate such basic sacrifices. If they have made design decisions to reduce the cost, they probably have thought about them in a clever way, so that these design decisions don't affect performance and that the product is still within spec and margins allowed in the RF industry.

That paragraph is a load of bollocks. "Art"? "Spec and margins allowed in the RF industry" (what "RF industry" exactly, as there isn't just a single one)? Seriously?

The point is that the Tek device was made for a price point, which meant they had to cut some corners. The corners Tek cut are different than the corners Signal Hound has cut to get the price point for their USB SAs right. If you believe that cutting corners won't affect performance then you're deluded. There are laws of physics which can't be ignored, even not by Tek engineers.

Or to put it in another way: if you are that smart, and are the only person who knows how to do it right, how come you don't work for Tektronix? =)

One reason could be that people with certain skills rarely have problems finding good jobs, and that means not having to work for a company that has been in decline in its core markets for over a decade and has seen more lay-offs than any other big T&M brand. Tek has been the to-go brand for scopes in the analog scopes days, but these days they are merely following while others (predominantly LeCroy) are the ones pushing scope technology forward. In addition, Tek was never renowned for their spectrum analyzers, which even in the good old days were mediocre at best (the old saying was "the scope from Tek and everything else from HP"). The companies that really understand spectrum analyzers are Keysight and R&S.

I often find myself disagreeing with G0HZU, but in this case I have to defend him, as I find your attack highly unfair, even more so when it comes from the corner of "no clue".

[pascal_sweden]

I was just wondering, as the spots that Tektronix seemed to have missed, seem to be fundamental issues, affecting basic functionality. Or did I misinterpret?

So what are the weaknesses in the Rigol DSA815-TG?

In which areas doesn't it perform well in the 1,5 Ghz operating range?

I thought it is limited up to 1,5 GHz compared to the A-brands that go above 7,5 GHz, but that it performs well within its 1,5 GHz operating range.

[KJDS]

I was just wondering, as the spots that Tektronix seemed to have missed, seem to be fundamental issues, affecting basic functionality. Or did I misinterpret?

It was indeed a basic miss from Tek. It didn't take much to make a significant improvement, only a few tweaks of the software, which tells you a lot about how far Tek have fallen from the greatness of their analog scopes.

[G0HZU]

I did get a few hours with a Tek RSA306 demo unit and after some basic reverse engineering I managed to model/predict how awful the spurious performance was in some places. It's claimed to achieve about 50dB spurious free range but I was quickly able to model it and measure it and prove it was only about 9dB in places due to some pretty awful frequency planning in the RF converter. I only had one evening or so with it before it had to be returned.

I would expect to find similar holes in the performance of the Rigol and SH analysers but I doubt I will ever see one of these.

So you are saying that the R&D department at Tektronix, with many skilled engineers, having years of knowledge and experience, have missed things that one person can spot in a few hours?

How is this even possible? Hard to believe for me. Imagine they read your comment here.

I am sure that the R&D department is very proud on their art, and would never tolerate such basic sacrifices. If they have made design decisions to reduce the cost, they probably have thought about them in a clever way, so that these design decisions don't affect performance and that the product is still within spec and margins allowed in the RF industry.

Or to put it in another way: if you are that smart, and are the only person who knows how to do it right, how come you don't work for Tektronix? =)

Yes, they made various basic errors in the design of the frequency plan of the RF converter in the RSA306. I would suggest that the person who chose the plan had limited design experience and the thing that amazes me is that this thing got through a basic design review of the frequency plan let alone a critical design review involving many engineers and other types of employee.

how come you don't work for Tektronix?

To give you some background, I was designing high performance and very expensive RF downconverters for wideband SDR type radios for gov/military use as far back as 1990. So I've had plenty of time to gain experience in what is good/bad and how to evaluate a frequency plan for spurious performance etc. I doubt Tektronix were doing anything like this back then. The RSA306 that I was given to try out had a frequency plan that was laughably bad in places. We are talking unforgivable errors (as in errors that break even the most basic rules of frequency planning) that only an inexperienced engineer could make. Therefore, I can't believe they have any proper design review process in place for products like this.

It was indeed a basic miss from Tek. It didn't take much to make a significant improvement, only a few tweaks of the software, which tells you a lot about how far Tek have fallen from the greatness of their analog scopes.

Agreed. I think it's the fact that this basic error wasn't spotted that is the most amazing thing. I found it by reverse engineering their frequency plan using another spectrum analyser but I was laughing even before I entered the frequency plan data into a spurious prediction program because it was so obviously flawed.

[Hugoneus]

As someone who regularly uses instruments ranging from $50 to $500k on a daily basis, I can tell you that I have to tailor my reviews to the instrument's intended use and audience. This is inevitable. For example I put the Keysight S-Scope through much tougher tests than I did the USB-based spectrum analyzers (although I specifically showed and complained about the spurious response of the RSA). I specifically design tests to show the instrument's capabilities and limitations and that already takes more than one hour of video and more ten times longer to prepare. It is not matter of expertise necessarily, it is a matter of practicality and time.

[G0HZU]

The other thing that surprised me was they Tek's frequency plan was so basic that they didn't include even the simplest of spurious shift/avoidance algorithms. These are things you can get for free in firmware as long as the HW can support this (and I think it can).

I would hope/expect they have improved this as well. My first thought was that someone had given me an early beta version of firmware but apparently it was a general release version.

I doubt I'll get to see a revised RSA306 because I don't think the company is going to buy one even though I thought it was still a useful tool for parts of the frequency range that were reasonably spurious free. But hopefully someone else will review one of the updated versions.

[G0HZU]

Based on the information available it wasn't really that hard to see where the flaws of the Tektronix device will be and to predict how it will perform. G0HZU also wasn't the only one who came up with that assessment.

To predict and find the spurs that I found required a degree of reverse engineering which I didn't see others doing. I was able to predict and find places where you put one signal in and it displayed it twice with the ghost version only 9dB lower.

This applied to all signal levels so if you put in a fairly large signal at -10dBm it would show the -10dBm signal and another copy of it (not far away) at just 9dB lower! This is very different to the internal low level 'birdie' spurious signals seen in other reviews. To find these special places where this happens at its worst you ideally have to know the frequency plan and know how to model the generation of the spurious signal so you know how to set the centre frequency and span that opens the door for this spurious mechanism to rear itself.

Of course, you could also find them by pure chance but I didn't see any review with this particularly nasty type of spurious mechanism present

I did also find other issues (bugs?) with the way it builds up a wide span on the display but these weren't as significant. But still surprising to see stuff like this from a big name like Tek.

If I had been allowed to keep it for a week I would have probably found more issues but I only had a few hours to reverse engineer the plan and then evaluate the RF downconverter section with a few classic tests.

To give an example see the image below. There is only supposed to be one test signal on the screen. But I was able to predict a particular centre frequency and span that displayed the same test signal several times at various levels. eg it displays a copy at -9dBc and another at -15dBc and another copy lower down still. There's only supposed to be one tone displayed!

So the spurious signals in the image below are all ghost copies of the main test signal and they suddenly appear along with the main test tone as it is injected into the analyser. This isn't very impressive for something that is marketed as a spectrum analyser

[hendorog]

Isn't it (more) interesting that the performance of the Tek was able to be radically changed with a firmware upgrade?

Rapid development (with known issues), and reacting quickly to resolve issues is in line with the software world these days.

Thats one advantage of a USB instrument - more of the instrument is software, and so can be upgraded.
Signalhound did this recently with a rewrite of their UI which was a massive improvement on the old UI.

[G0HZU]

To clarify things a bit the issues I found were to do with how they were managing/programming the hardware inside the box wrt the frequency planning for each tuned frequency (rather than this being something fixed in the PC/DSP/SDR part of the system)

i.e. these were fundamental mistakes in the management of the hardware in the RF converter and they also failed to deploy even the most simple of spur avoidance algorithms anywhere within this hardware.

[MagicSmoker]

I bought the Signal Hound SA44B a couple years ago just to do "pre-compliance" EMC testing (along with a homemade log-periodic antenna) and it performed reasonably well, especially given the price and ease of use. This is the type of test equipment that I might use once a year, so a more capable device was neither justified on price, nor on the inevitable steepness of the learning curve.

I did not buy the tracking generator because I had no intention of doing immunity testing and a scalar network analyzer (ie - characterizing frequency response but not phase) isn't much use to me.

My opinion - based on very limited usage - is that the Signal Hound is a capable instrument and a good value.

[w2aew]

Based on the information available it wasn't really that hard to see where the flaws of the Tektronix device will be and to predict how it will perform. G0HZU also wasn't the only one who came up with that assessment.

To predict and find the spurs that I found required a degree of reverse engineering which I didn't see others doing. I was able to predict and find places where you put one signal in and it displayed it twice with the ghost version only 9dB lower.

This applied to all signal levels so if you put in a fairly large signal at -10dBm it would show the -10dBm signal and another copy of it (not far away) at just 9dB lower! This is very different to the internal low level 'birdie' spurious signals seen in other reviews. To find these special places where this happens at its worst you ideally have to know the frequency plan and know how to model the generation of the spurious signal so you know how to set the centre frequency and span that opens the door for this spurious mechanism to rear itself.

Of course, you could also find them by pure chance but I didn't see any review with this particularly nasty type of spurious mechanism present

I did also find other issues (bugs?) with the way it builds up a wide span on the display but these weren't as significant. But still surprising to see stuff like this from a big name like Tek.

If I had been allowed to keep it for a week I would have probably found more issues but I only had a few hours to reverse engineer the plan and then evaluate the RF downconverter section with a few classic tests.

To give an example see the image below. There is only supposed to be one test signal on the screen. But I was able to predict a particular centre frequency and span that displayed the same test signal several times at various levels. eg it displays a copy at -9dBc and another at -15dBc and another copy lower down still. There's only supposed to be one tone displayed!

So the spurious signals in the image below are all ghost copies of the main test signal and they suddenly appear along with the main test tone as it is injected into the analyser. This isn't very impressive for something that is marketed as a spectrum analyser

I would say that the current firmware made a substantial improvement:

[G0HZU]

That does look cleaner, but I'd have to have one here to see if the new frequency plan is truly better across all tuned frequencies. The only way I could do this is to reverse engineer it again to map out the complete frequency plan for every frequency step in the whole range and then analyse it for issues.

Or maybe Tek could release the revised plan in paper format for people to review. The first plan was so flawed and basic it will take more than one screenshot to convince me that the analyser is now 'fixed' in this respect

[pascal_sweden]

Agreed. I think it's the fact that this basic error wasn't spotted that is the most amazing thing. I found it by reverse engineering their frequency plan using another spectrum analyser but I was laughing even before I entered the frequency plan data into a spurious prediction program because it was so obviously flawed.

Nice job about the reverse engineering. Hats of for smart people like you!

[rs20]

Thanks everyone for your input; there's a bunch of part numbers and model listed that I shall take a look at in the coming days! It's also interesting to note that no-one seems particularly opinionated on the standalone-vs-USB thing. I also really appreciate the other discussion in this thread.

Just briefly, does anyone know what the Signal Hound actually sends over the USB link? I/Q samples, frequency/magnitude pairs, or something else? I'm trying to get a sense of what proportion of the cleverness is in driver/UI software vs actually on the device.

It depends on what you mean by "convincingly". For the TDR example I can already clearly measure the lengths of 19 cm and 29cm open/short circuit stubs located 3.1m along cable. In theory 1.5GHz => "distance quantisation" of 3.3cm. I'm still improving my technique for measuring the impedance/magnitude of discontinuities.

As and when I'm satisfied with my techniques, I'll publish my code.

Cool! What software did you use to interface to the SDR? I.e., did you write your own code to talk to the USB directly, or did you use some software to take a frequency sweep or something for you? Did your code start with a list of frequency/magnitude pairs, or a pile of I/Q samples?

Thats one advantage of a USB instrument - more of the instrument is software, and so can be upgraded.
Signalhound did this recently with a rewrite of their UI which was a massive improvement on the old UI.

Is that pre-or-post Shahriar's review? Speaking of which:

I specifically design tests to show the instrument's capabilities and limitations and that already takes more than one hour of video and more ten times longer to prepare. It is not matter of expertise necessarily, it is a matter of practicality and time.

Your efforts are very very much appreciated!