Spectrum analyzer design?

I don't know where a question like this should go in the forum so I'll stick it in the brewery. Mods, feel free to move it to a more appropriate place if needed.

Does anyone have access to schematics for modern spec-ans? I've been tasked with studying spec-an designs and learning possible ways to make them better.

I know the basic structure and a lot of the pieces to modern spec-ans but I feel that there is likely a lot of tried-and-true technology that each company must use to get the performance they do.

I'm especially interested in elegant and novel solutions to issues that might arise during a design cycle.

Anyone?

What kind, what range ?

You could try the HP / Agilent site for articles & info, and since they have lots of service-docs as free pdfs, I guess they'll have the service docs for a few of the classics as well (say for instance the HP3585A). Might be more detail than you'd like though...

Bye,

Peter

Isn't this sort of thing done with DSPs and Fast Fourier Transform these days. You can get FFT algorithms as samples when you buy the DSP trial boards and software, I think. Here is a company that implements a 31 band spectrum analyzer with a DSP and uses it to control a 31 band compressor/limiter, also implemented in the DSP.

http://www.vorsis.com/audioprocessors.html#ap1000

Let's get this clear first, is this about a spectrum analyzer or about an RTA ?

5mhz-1Ghz

Direct sampling for the bandwidth and depth needed is definitely in the design cycle for a future product but for now we are investigating a "standard" DDS driven mix-up/mix-down front end like a lot of the mid level spec-ans use these days.

I have looked at all of the Agilent docs, Agilent is one of our main suppliers for test equipment here but of course they don't give us the schematics.. :roll:

This product will be cost efficient and direct sampling is not cost efficient right now due to the costs of A/Ds and FPGAs(for hardware DSP) that are fast enough for what we need to do direct sampling. To get what we want for the target price we have to mix up and filter to get away from close harmonics and/or Nyquist problems and then mix back down to get to an IF that we can sample with more cost effective A/D and FPGA parts. The LOs are driven via DDS/PLL/VCO combinations, something that a lot of spec-ans are using these days.

I guess we are looking to see if we are on the right track and if we can see any issues with our designs based on seeing what others have done. Just basic research.

We can decode the IF into I/Q and/or X/Y in FPGA/DSP no problem so we can do spectrum, vector, network and o-scope plots as long as we get a clean digitized signal.

Might check out a fairly old book by Ulrich Rohde, Digital Pll Frequency Synthesizers: Theory and Design, ISBN 0132142392. I don't have it, but his ones on communications receivers are pretty good. In fact they might be apropos too.

Of course you want an analyzer, not a synthesizer, but there's a lot in common.

Well we can use the synthesizers to drive the LO ports of the mixers. We are planning on using a DDS(direct digital synthesis for those that are unfamiliar) IC to drive PLLs. The DDS ICs can barely go above 400Mhz so we need to drive a PLL with the DDS to get to a usable frequency range, 2-3Ghz for the first LO. VCOs may be used as additional control loops but the main frequency control for the LO will be the DDS itself. This allows us to sweep the LO and thus sweep the spectrum without having to "lock". Our current architecture forces us to lock to a frequency and wait for the tuning elements to settle before moving to the next frequency. Entirely too slow for any real spectrum usage, but then again it was never meant to do spectrum.. Marketing now demands, so now we have to deliver.

I've been working with Analog Devices and some of their DDS parts for about a month now and I can say that these parts are simply amazing in what they can do. Its great when you can write a register and within a couple microseconds you are MHz away and steady as a rock.


On a similar note and one that really sparks my interest in all of this, besides the obvious paycheck necessity:

We hacked a DDS IC and forced it to do BPSK to transmit data to field units. Management and Principal RF engineers completely ignored us lowly associates and even tried to get our little project canned stating that we were wasting our time, that we needed to build a discrete analog(varactor, PLL/VCO, etc.) FM encoding system in order to make the most of our time. One senior guy told our group that it would never work. Another stated that it was impossible to get RF from an IC and such things don't exist..(where has he been for the last 15 years?)

It took us 3 man days to get the system up and sending random data. It took another 2 man days to have it transmitting data over RG6 and less than 1 man day to debug the FPGA code.

It took 3 guys a little less than 2 full days to do what we were told would take 4 weeks to do in discrete analog. It also cost us 15$ in parts and layout whereas the estimated BOM for the discrete board was closer to 70$ after layout and tuning. One of the engineers actually congratulated us, the others simply ignore us now. We've heard that we "were just lucky".

Again we've run up against the "old school" guys and their nay-saying. I'm determined to prove them wrong again. :green:

Man when we told them that we were going to use DSP on FPGA you should have seen their faces. :thumb:

DDS parts in wideband mode (ie without a tight band filter) traditionally suffer from spurs, which is obviously Not Good for any decent SA. Rumor has it that newer parts (particularly the AD9912) are better but not quite perfect.

In a traditional VCO upconversion->filter->fixfreq downconversion SA the ultra wideband sweeps are hardest (and, IME & IMHO least useful). A DDS-PLL combo (such as documented in http://www.thegleam.com/ke5fx/synth.html ) can work well, but still you have a trade-off to make between sweep speed and close-in noise when picking your PLL loop filter width. Making this switchable might help.

An interesting technique I've seen (and toyed with) to get fast accurate wideband sweeps is to have the upconversion VCO sweep over the band of interest without aid of a PLL, but by simply applying a voltage ramp to its control input. Now keep a frequency counter running in parallel with your measurement, and stretch the horizontal (frequency) axis on display to correct for the non-linear nature of the VCO CV curve. For bonus points, use a PLL and an ADC to precompute a rough CV-> frequency table, and then disconnect the PLL from the VCO and directly apply a DAC to the CV in to step your VCO to the desired frequency. Won't work all too well with a 100Hz video BW, but you don't want to use that on a DC-daylight sweep anyway.

Good luck,

JDB
[built a custom VCO+DDS+SDR network analyzer for filter/antenna tuning in a previous project. Surprising how far you can get with a few AD eval boards plus the Mini-Circuits catalog]

:shock:

That link is very close to the idea that we came up with..

I hope nobody thinks we copied it.. :roll:

Very good information Jdbakker, I'll keep it in mind as we research.

The product doesn't need to have the extreme quality of a *real* spectrum analyzer, just enough to accurately test docsis equipment and coax conditions of various kinds. Our build specs are still being developed by marketing so we honestly don't know exactly how *good* we need to make this device yet.

EDIT:

I use the AD9913 parts. They are really pretty good. SFDR is really good too.

Congrats.. you make my efforts to perform low audio frequency FFT seem puny by comparison.

If an engineer says something can't be done, it only means with any certainty that he won't do it. I was told drums couldn't be tuned electronically.
------
I have an old (old when I bought it used 25+ years ago) Singer SA that uses a hetrodyne (beat or alias frequency?) approach, but this was to get response all the way up to 35 kHz...

I gather you're doing some modern variant on that. I've seen some tricks with sub sampling where you grab an alias of the higher frequency, but you kind of need to know what you're looking for. I suspect you could delta the sample rate between crunches to parse aliases from real frequency spikes.

JR

[quote author="JohnRoberts"]If an engineer says something can't be done, it only means with any certainty that he won't do it.
JR[/quote]

So true. Especially engineers in a given company whose turf is encroached.

ADI is pretty good at avoiding that culture, and the results speak for themselves.

I'm reminded of some fairly recent help I was giving to an engineer that inherited some designs I did for a low-frequency magnetic sensor system for use primarily by plumbers. Because he had not been following some of the developments in undersampling techniques, but only remembered Nyquist as more-or-less of an admonition, he threw out aspects of the existing working system when addressing a new instrument design. I guess he managed eventually, but it was silly to have this attitude that "If I don't understand it, it can't work"---despite hundreds of instruments working in the field demonstrating the contrary.

EDIT: It's interesting to go back and look at some eminent scientists' and engineers' comments about feedback as well, to the effect that it wouldn't be beneficial at best, or even that it couldn't work.

"If I don't understand it, it can't work"

Click to expand...

If I had a dollar for every manager or engineer that I have worked for/with that LIVED that philosophy, I'd be able to pay for a nice steak dinner.

I may have ranted about this before but at my previous job, I was designing a relatively simple motor controller. One direction, variable speed, IGBT switch, PWM IC and redundant protection for the motor and power supply. My boss, whom spent his life working with relays, rheostats and power resistors told me "there is no way in hell that little doodad right there(pointing at a TO247 100A/900V IGBT) could ever handle that motor! What are you crazy? You waste any more time on that and I'll run you right out the door!"

My little credit card sized controller not only handled that motor, it also handled it at full rotor lock too. For about 15$ compared to the 600$ in the current scheme.

I was poised to replace a box full of relays and variable AC transformers about the size of a computer case with a board the size of a credit card...

The project was nixed and I quit soon after. :green:

I still have my prototype too! That guy was so against using my design that he let me keep it, stating that he would never use anything like that as long as he owned his portion of the company and to "get it out of his sight".

:shock:

No wonder that place never made any money.

I once had an engineering manager not let me incorporate a circuit approach into a design because it wasn't proven in mass production... Precisely how do you prove a design besides using it? (It still is a good idea IMO). I am well aware of such risks and was actually designing it into a lower volume niche product precisely to production test it, but noooooo.

OTOH that street goes both ways, I have also worked as an engineering manger and even heaven forbid in marketing. You would be surprised how many engineers tried to BS me about what could or couldn't be done.

Of course they can work very hard to make something not work, at which point you have to take them off the project. I once had to kick an engineer out of meeting because he became argumentative to the point of being disruptive. I'm not a "my way or the highway" kind of manager. I want engineers to be passionate about their work, but there's a place for debate and it's not in meetings in front of the suits. They don't understand the difference between refining a technical concept via heated discussion and someone being wrong. I want my people to question me, just at the right time and place.

Corporate politics can be influenced by nonsense like that.

JR

[quote author="JohnRoberts"]I once had an engineering manager not let me incorporate a circuit approach into a design because it wasn't proven in mass production... Precisely how do you prove a design besides using it? (It still is a good idea IMO). I am well aware of such risks and was actually designing it into a lower volume niche product precisely to production test it, but noooooo.
JR[/quote]

I had the generally good fortune to have a free hand and little vetting from managers in some early multimedia designs. I was hoist by some shoddy parts at one point, but a particular invention (never filed on) worked fine (and I still have a working system on my desk from those days circa 1995, that's been in more-or-less continuous use).

When I described the "invention on the critical path" in the product, another far-more-cautious consulting engineer said Hmmm surprised you don't get some effects of half-wave rectification in the transformer that would be deleterious. I conjectured that there would be a certain chaotic behavior that would move things off of such a situation, thus I doubted it was a real concern.

He said So did you ever build one of these to try it? I said "David, it's in the product---we've made and shipped thousands of them."

He was visibly horrified.

Working in thousands of products is usually good proof of concept, but a little late for patent applications. :grin:

JR

[quote author="JohnRoberts"]Working in thousands of products is usually good proof of concept, but a little late for patent applications. :grin:
JR[/quote]

I figured out a mod to it since then to address the other consultant's concerns, but I doubt I'll bother to file. Patents are expensive to get and maintain.

And expensive to defend..