low cost open hardware VNA (vector network analyzer)

[xaxaxa]

EDIT: The kickstarter has now been launched:
https://www.kickstarter.com/projects/1759352588/xavna-a-full-featured-low-cost-two-port-vna

--------------------------------

I've been working on a low cost 140MHz - 2.5GHz 2-port VNA and got reasonable results with my first iteration:



Measured vs simulated graphs (simulations done using RFSim99):




Measuring a 900MHz power amplifier:




More details (block diagrams, schematics, pcb designs, software, etc) are on github:
https://github.com/xaxaxa/vna

Some inspirations I got from the hforsten and loxodes VNA; Mine is designed to be cheap and accurate, and forgoes 6GHz support because parts get extremely expensive above about 3GHz. It covers the major useful ISM bands, and I was more interested in the lower bands because they propagate much better.

It uses 3 receivers, 20MSPS ADCs, and all DSP is done on a spartan 6 FPGA. The usb interface is implemented using a usb PHY and a usb virtual serial port core; it can transfer up to 20MB/s but that's not utilized here since we are only transferring a few numbers (already averaged by the fpga logic).

I'm thinking about doing a kickstarter (after redesign for manufacturability) and wanted to know if people would be interested in a ~$200 140MHz to 2.5GHz T/R VNA.

[razberik]

Perfect. Watching your topic.

[lukier]

Good work!

Recently I've noticed this project as well, maybe I'll try to build one at some point:
http://hforsten.com/cheap-homemade-30-mhz-6-ghz-vector-network-analyzer.html
http://hforsten.com/improved-homemade-vna.html

[loxodes]

Good work! The software looks nice.

I like the coax cable board to board connections, that looks like a great way to cut SMA connector costs when prototyping 

I'm thinking about doing a kickstarter (after redesign for manufacturability) and wanted to know if people would be interested in a ~$200 140MHz to 2.5GHz T/R VNA.

You've probably already noticed, but if you are doing a manufacturability respin it looks like the PE4302 and CMY210 are both EOL.

I'm don't know what kind of a market there is for a VHF/UHF kickstarter VNA. The chazwazza (https://www.kickstarter.com/projects/1913839854/chazwazza-a-low-cost-2-port-400mhz-27ghz-vna?ref=category_newest) recently failed, but it cost twice as much.
I haven't done any research, but I'd speculate that most of the demand for hobbyist VNAs would come from hams who might care about HF (~kHz to 30 MHz).

[Wirehead]

Any specific reason not to include the HF band? Could make a great tool for Ham radio operators as well 

[bjcuizon]

Subscribed to this topic...
This looks like a nice and neat project.

[henrikf]

Nice to see more open source VNA projects. This one looks really good.

I would make a full two port VNA instead of one direction only. It has advantage of supporting more advanced calibrations. Eight term calibrations like TRL, unknown thru are hard to implement on one direction VNA. Especially the TRL calibration is very useful for calibrating out the connectors on the test boards. Hardware shouldn't need that many changes, just switch after the source and fourth receiver or switch before the reference receiver. Sharing reference receiver is not a bad idea since during normal operation the non-active port reference is not needed. Some commercial VNAs also use this method to reduce costs. Make sure that the source switch has >100 dB isolation if you go this route, but it shouldn't be too hard at these frequencies.

MAX2871 might be a better choice for the source if you want to extend the frequency range downwards and it wouldn't even increase cost. HF range could be implemented for example with a switch and DDS chip.

[Zebryk]

I sure would be interested even for $499 or $999!
My application is for tuning the antenna matching network in 433MHz-915MHz-2.45GHz products.
A must have is calibration stability over temperature.
(Apparently the miniVNA is poor in this regard?)
The PocketVNA looked good but apparently now has business issues.
(Liked the aluminum case; also a must have.)
Have you seen this:
http://www.analog.com/en/products/rf-microwave/rf-power-detectors/rms-responding-detector/adl5920.html
The specification says "zero insertion loss"!
Having researched the market, there is definitely nothing much between the low end <$500 and the high end >$3K
A possible business strategy is offer a basic $499 plus options.
Also, I suggest not waste any R&D on a display, go straight USB and let it's processor deal with the graphics.

[xaxaxa]

Any specific reason not to include the HF band? Could make a great tool for Ham radio operators as well 

The main issue is that there are no single chip PLL synthesizers that go this low; I've looked at the designs of all the VNAs on the market and this is typically solved by first upconverting the signal; this would add quite a bit of cost and complexity, and possibly worsen the signal quality;

The other way is to use an additional prescaler (and switches) to divide the output frequency; this can cover down to 10MHz with the prescalers I can find. The other consideration is that the HPF in front of the mixer needs to pass rf but block IF; the IF being 1MHz means probably 5MHz is the lowest rf frequency doable; lowering the IF frequency may lead to 1/f and other noise issues. The LPF after the mixer also has to pass IF but block rf.

You've probably already noticed, but if you are doing a manufacturability respin it looks like the PE4302 and CMY210 are both EOL.

The pe4312 is a drop in replacement for the pe4302, but cmy210 is where the main trouble lies; there is an alternative, just much more costlier.

I would make a full two port VNA instead of one direction only. It has advantage of supporting more advanced calibrations.

I will consider that; it may take several iterations to get right though since it means the port 1 and port 2 routing have to come much closer on the PCB, and even when two traces are separated by a solid metal plane i wouldn't hope for more than 40dB of isolation from experience. I may have difficulty fitting it onto a 10cm x 10cm board though if I want to keep the separation to a good amount.

Make sure that the source switch has >100 dB isolation if you go this route, but it shouldn't be too hard at these frequencies.

Easier said than done   

MAX2871 might be a better choice for the source if you want to extend the frequency range downwards and it wouldn't even increase cost. HF range could be implemented for example with a switch and DDS chip.

it does increase cost quite a bit if you consider that I can source adf4350 at about $1 each  (and QA them myself of course)


As for stability over temperature, I haven't noticed any issues but I'll do a persistence plot later and post it here.
I don't wait for it to warm up before using it; just plug it in, calibrate, and go.

With the current prototype the main issue is mechanical drift; if you nudge the cables around a bit the readings will change slightly. Probably because i'm using cheap ass cables 

[xaxaxa]

Have you seen this:
http://www.analog.com/en/products/rf-microwave/rf-power-detectors/rms-responding-detector/adl5920.html
The specification says "zero insertion loss"!

Interesting chip but it's only a magnitude detector so makes for a SNA; it would be nicer if they do away with the magnitude detectors altogether and include some mixers instead; analog magnitude/phase detectors will always have the problem of linearity, dc offset, 1/f noise, etc, which is why all the higher end VNAs use the mixer/adc architecture and then do the magnitude/phase comparison digitally. This might also why the lower end VNAs using a "vna on a chip" solution tend to have a reputation for drift.

[TheSteve]

Great project, I hope you can get them into production.

[xaxaxa]

Here are the temperature drift persistence plots of the 3 calibration references;

frequency: 200MHz - 2180MHz, 100 points
accumulation time: 10 minutes for each plot

The device was powered off to cool down before each persistence plot; the persistence plot is started less than one minute from powering on.
The scanning speed is about 23 points per second, meaning a 4 second full update time for a 100 point plot. This is adjustable in software, and is a tradeoff between precision and speed.

Open circuit reference:



Short circuit reference:



Load reference:



just for fun I decided to blast one of the receiver sections with a 100*C heat gun:



Since the vna has a reference path coming from the directional coupler it is not as sensitive to phase and amplitude variations from the signal generator or LO, but if you heat up the reflection receiver while keeping the reference receiver cold, it will show much more drift. However in normal operation the receivers don't generate enough heat for this to be a problem.

[xaxaxa]

Here's the persistence plot measuring a resistor plus transmission line;
200MHz - 2180MHz, 5 hours accumulation time.
procedure is:
1. calibrate
2. unplug and let cool
3. power on, load calibration, and start persistence

[Theboel]

Unfortunately the frequency range is not suit to my need.
1 Mhz to 1000Mhz is all I need of course up to 6 Ghz is very nice thing.
Maybe You can offer 2 type of VNA a low and high band range

[exe]

I'd love to use your VNA as impedance analyzer, but that would require a much lower minimum frequency  At least 100KHz or below.

Still, if it goes into production, I'll donate just to support open hardware initiative.

[xaxaxa]

Update:
The kickstarter is now live:
https://www.kickstarter.com/projects/1759352588/xavna-a-full-featured-low-cost-two-port-vna

Thanks for all the suggestions.

[hcglitte]

Will it feature port extension, i.e. electrical delay? This is handy so you can do cal off board (the DUT) and then tune the electrical delay to the Open on the PCB (the matching PI placeholders).

[xaxaxa]

Will it feature port extension, i.e. electrical delay? This is handy so you can do cal off board (the DUT) and then tune the electrical delay to the Open on the PCB (the matching PI placeholders).

Yes; this is a fairly simple feature to implement.

[hcglitte]

That's great. The miniVNA tiny only has this feature by manually adding the coax length and velocity factor.

Now I calibrate normally, then do a measurement with nothing at the PI placeholders (open), then I divide the s11_open_dut with the cal s11 dut (short to antenna) or something (don't remember exactly).  Anyway this gives the correct result. I did the math and also tested in rfsim99. Adding this as a step procedure in your GUI would be awesome.

[richnormand]

Very nice project. Nice price-point.
Bookmarked your kickstarter as I want one!

Any ETA on the version 2. From my experience going to a working prototype to manufacturable version 2 as your are doing always seem to have hidden surprises.
Other surprises also seem to be getting the software (that works well in Linux) to play well in various versions of Windoze ...
Best of luck with it.


EDIT: Done, backer! Again, cheers and best of luck with the project.

[xaxaxa]

Update: these are now in stock and available for order on the website (xaxaxa-dev.com)

[rhb]

I did not find any information about ordering.

[markus_jlrb]

@xaxaxa,

great job and engineering!

Some words concerning the lower bands < 150MHz for HAM porpose.
If you take two of ADF4350/55/56 or the new one ADF535X you could
mix frequency from DC to 1/2-fmax of the chip (avoiding to use internal
divider - no more sinusoidal output). Makes your design a bit expensive
but offering a fulle rang of frequency for all kind of hobbysists.

If you compare low/mid cost competitors like the new SVA1015X VNA
from Siglent (500kHz-1.5GHz) for 1300€ in Europa your design is round
1k€ cheaper, but you heve to use an additional computer to control the
VNA. 

Good luck and good ideas for your future OH projects.

Markus

[xaxaxa]

I did not find any information about ordering.

If you go to the website you can order directly from there

If you take two of ADF4350/55/56 or the new one ADF535X you could
mix frequency from DC to 1/2-fmax of the chip (avoiding to use internal
divider - no more sinusoidal output). Makes your design a bit expensive
but offering a fulle rang of frequency for all kind of hobbysists.

That's an interesting idea, I'll have to see about fitting 4 x adf4350s on the same board; though I do plan to switch to the adf4351 for the T/R version which would support down to 35MHz. Still not ideal, but I think to cover <10MHz a completely different receiver architecture will be necessary, since the RF frequency gets too close to the IF. Most of the VNAs I have seen all upconvert the signal first, just like a spectrum analyzer (the hforsten design being the exception).

If you compare low/mid cost competitors like the new SVA1015X VNA
from Siglent (500kHz-1.5GHz) for 1300€ in Europa your design is round
1k€ cheaper, but you heve to use an additional computer to control the
VNA. 

Unfortunately the VNA add-on is another $500 

[markus_jlrb]

@xaxaxa,

The effort that you have with two ADF's will bring you the
advantage not to make measurements with square signals
but with proper sinus signals.
The ADF chip family provides a clean vco signal from 0.5fmax
to fmax when not using the internal divider.
So you are able with the ADF4350/51 to generate 2.2-4.4GHz
in good spectral purity.
With two of these chips you could fix one to 2.2GHz and do a
range tuning from 2.2 to 4.4 GHz with the second one.
Mixing both sigs you get depending on your LPF or HPF
you use (or even make them switchable) DC- <2.2GHz
or/and 4.4 - 6.6GHz and if disabling the fix freq. ADF
you will get 2.2-4.4 GHz.

So it's up to you which effort you would like to do.

Btw not knowing if you know the link (German doc concerning
the scalar Analyzer with ADF4350/51 from BG7TBL found
on Ebay) :

https://www.agaf-ev.org/agaf-de/NWT4000-AGAF.pdf

This doc covers the problem using rectangular sigs for
measurements. And for vactorial measuremts this issue
is much more important to care about concerning reliable
results.

What concerns the SVA1015X - I think they will reduce
the price in the future and the Siglent device have the
advantage that you could enable all options by hacks you
find in this forum easily.


Markus