Building Real Time Spectrum Analyzer

[Mechatrommer]

last time i checked in google, there very little info mostly on theoritical level from rtsa manufacturer such as signalhound.. from the way i understand it.. rtsa is similar to how rt dso work in sampling method, ie there is adc at some GSps sampling the input and go to some fpga, memory,mcu and then processed in fft and go out to display. The difference is the front end attenuator i guess. Where dso can accept 300V, rtsa is classical 0-10db max input or 20V max before damage. Questions...
1) is this mentality correct?
2) anyone with greater detailed circuit of how the rtsa is built? may be a chance for diy or hobby community?...

ps: Bandwidth isnt relevant, topology is. It can be 1MHz for learning purpose or it can be GHz range for more practical purpose nowadays.

[Someone]

Google is your friend:

http://cdn.teledynelecroy.com/files/whitepapers/interleaving_process_in_dbi_scopes.pdf
http://download.tek.com/document/55W-29371-2%20ATI%20Whitepaper.pdf
http://www.analog.com/en/analog-dialogue/articles/whats-up-with-digital-downconverters-part-1.html

[BravoV]

From one of my bookmark entries -> http://lea.hamradio.si/~s53mv/spectana/sa.html

[hagster]

An RTSA is basically a software radio like HackRF. It digitises a chunk of bandwidth (usually 20 or 40MHz) at a given frequency. They either operate in a sweep mode where the centre frequency is moved in steps to cover the entine span, or in staring mode when the span is less than the instantaneous bandwidth. In staring mode you have the ability to have 100% chance of seeing any short transient signals and can normally set a trigger to capture these.

The big difference between an SDR is calibration. An RTSA should provide accurate measurements and automatically compensates for internal gain or attenuation settings.

You can also have some cool extra software features that are not possible with a standard SA.

[jbb]

I believe the ‘Real Time’ part means that the FFTs are done fast enough to keep up with the ADX samples (instead of capturing a buffer and FFTinf it now and again).

I also understand that to get 100% probability of detection you need to overlap several FFTs in time (think like overlapping brocks).

Finally, I hear that if you manage to make a fierce enough instrument you may run into military export restrictions.

I’m conclusion, FEAR THE MULTIPLIERS!

[TheUnnamedNewbie]

A real-time spectrum analyzer, on a basic level, is just a RF receiver that digitizes the IF. However, it does so while having massive range of frequencies you can "tune" to, as well as compensating for internal errors due to non-flat frequency response of the internals. It can then use a swept FFT to display the spectrum.

The reason they cost so much and are massively complex units is because the do so at incredible sensitivity, linearity, wide range, very low phase-noise, and often have a big stack of software on top of that to allow you to do measurements. (Kinda similar how at the surface a scope is just a real fancy ADC with a trigger)

[Mechatrommer]

Google is your friend:
http://cdn.teledynelecroy.com/files/whitepapers/interleaving_process_in_dbi_scopes.pdf
http://download.tek.com/document/55W-29371-2%20ATI%20Whitepaper.pdf
http://www.analog.com/en/analog-dialogue/articles/whats-up-with-digital-downconverters-part-1.html

you are refering mostly to how modern DSO architecture sampling works at higher speed, either by interleaved/up sampling or downconverting. are you saying i'm correct with the idea that RTSA work in similar way as modern DSO?

They either operate in a sweep mode where the centre frequency is moved in steps to cover the entine span, or in staring mode when the span is less than the instantaneous bandwidth. In staring mode you have the ability to have 100% chance of seeing any short transient signals and can normally set a trigger to capture these.

i'm interested in how this so called "staring" mode (ie 100% probability capture chance) works. thats why i ask, since sweep mode may lose some intermittent signal (usually in frequency hopping radio transmitter). my best guess so far after the RTSA is triggered, it will sample as fast as it designed to be, in sequential manner similar to DSO...

I believe the ‘Real Time’ part means that the FFTs are done fast enough to keep up with the ADX samples (instead of capturing a buffer and FFTinf it now and again).
I also understand that to get 100% probability of detection you need to overlap several FFTs in time (think like overlapping brocks).

from my understanding reading signalhound literature, 100% probability of detection is due to the trigger mechanism tuned for a particular spectrum, once the spectrum is detected, the RTSA will capture the rest of the signal for display, but how exactly done, is not described. i have a strong belief that multiple overlapping FFTs method is used to enable the small chunk of FFTs are calculated in parallel in quick time so "real time" experience will be realized virtually, the result of these multiple FFTs will be combined in some (proprietary or paid IEEE?) way to achieve greater FFT resolution, it has been mentioned by some members here time ago.

Finally, I hear that if you manage to make a fierce enough instrument you may run into military export restrictions.
I’m conclusion, FEAR THE MULTIPLIERS!

this thread, or this forum for this matter is not about complying or fear to this restriction. yes its been discussed in signalhound thread and link to their reply how they crippled their SW to comply with this restriction. but i guess this rule is only worth concerning in high greater than 10's of GHz of BW, which i believe still beyond the reach of hobby purpose in components cost alone, let alone the expertise to design the PCB at those RF BW.

i'm thinking this thread more as preliminary education or insight lets say if we want to design a smaller scale RTSA at 10-100's MHz of BW or 1GHz BW tops. i saw several links to PIC RTSA at audio BW, but its too simplistic imo and impose less challenge. we can even make our DSO as RTSA by downloading data to PC and do FFT processing, but the experience will be far from "real time" and no spectrum trigger, and DSO will be alot less sensitive to RF due to their input attenuator (resistor divider) design. i need something more realistic and closer to practical application like implementation of SA attenuators, spectrum trigger, and then interleaved sampling (to lower buying cost of ADCs), fpga to save to memory maybe?, and then FFT processed by mcu or some dsp (multicore for parallel processing preferably).

[rhb]

The Nov/Dec 2017 issue of QEX has an article by DC9ST describing an SDR built from an AD 9467  eval board and a zedBoard that digitizes 100 MHz of BW at 16 bit resolution which pretty much puts it at the top end of SDR performance until you get to the "government only"  stuff.   The main hardware costs about $1000 US.

http://www.panoradio-sdr.de/

[cncjerry]

Looking at that Panoradio, I'm thinking that in the not to distant future, we will be able to build high-function spectrum analyzers from off the shelf parts (setting aside input conditioning which is a large part of any analyzer).  I was looking at Analog Devices product brochures that they send out occasionally and I continue to be amazed at the quadrature mod/demod chips they have.  They are packing more and more function into a single device these days.  They have phase detectors, filters, log amps, log detectors, all kinds of neat little devices that can be thrown together in a million or so ways.  I've been working on Dual Mixer Time Difference (DMTD) setups and Tight PLL configurations and used an 8302 phase detector with a filter on it and was able to do some very repeatable Allan Deviation plots with the little AD8302 and an HP 3457a voltmeter.  Anyway, I guess technology is just moving ahead so quickly because of all the SDR related work.

[rhb]

It's a lot more than just spectrum analyzers.  With Gs/S ADC, fast processors and large memory all manner of things become just a "matter of software".  The latter not being trivial, but a lot of stuff is reaching the level of you just need to know the math.

I picked up an NOS Instek GDS-2072E from Amazon in one of their discount sprees for $222 plus tax.  I've put one of Leo Bodnar's 40 pS rise time pulsers on it and it is about 150 MHz out of the box.  No hacking or other funny business.  It will collect 10 Mpts on each of the two channels at 500 Ms/S or 1 Gs/S on one channel.  It's built on a Zynq chipset. using the Xilinx Linux distro.  I'm hoping to pop it open so it can be used as a basis for custom instruments.  Most instruments today are being built using open source firmware from chip OEMs.  So it's a lot lower barrier than posed by custom ASICs make by Keysight and Tektronix.  I'd love to see the West beat the Chinese using Western IP and Chinese hardware.  Not that they won't catch up in time.

Rigol and Siglent are happy to let hobbyists hack the licenses on their low end kit, but what I really want is documented hardware that can be programmed for bespoke applications.  My goal is to demonstrate with the Instek  that OEMs can make money selling well documented hardware that can be custom programmed by the end user.  The production volume associated with easy customization should make it very profitable.  People like me are not cheap, but the cost of my delivering  a bespoke product improves  dramatically if I can buy a reliable piece of hardware that I can customize with software.  This was the great virtue of the Unix workstation during the course of my career.  Everyone sold one and they were sufficiently similar that I could make my programs work on all of them with no changes.

[cncjerry]

I picked up an NOS Instek GDS-2072E from Amazon in one of their discount sprees for $222 plus tax. 

I have an older Instek function generator/counter.  It's worked fine since day one which was at least 15yrs ago.  No complaints, no bugs, decent signal, etc. That deal you got was pretty damn good. 

I have to start unloading some of my boat anchors and get updated equipment but the stuff I can afford is all so riddled with bugs I seem to be stuck.  I was looking at one spectrum analyzer, don't know the number but it goes to 3.2G and it would drive me crazy to have spurs only -70db down even if it meets the specs.  And then the channel imbalances I've seen with some of the scopes would also be tough to deal with.  I'll have to take another look at Instek because there have been some great deals lately.

Jerry

[rhb]

I briefly had a Siglent SSA3021X which I returned after experiencing a lot of weird behavior.  Interestingly I notice that Rohde & Scwartz sell an SA with a 1 GHz upper limit for $1900 US which can be field upgraded to higher BW.  The highest BW version costs only slightly more than the Instek GPS9300B.  One thing I've noticed about GW is that they are pretty active updating their FW.

There was a big kerfuffle with the Siglent SSA3032X recently with spurs as high as -50 dB.  That seems to be a work in progress situation.