Why is there no merger of the TinySA project with the NanoVNA V2? Since the hardware is very similar, it would be possible to combine the functions in a single product?

A SA has a very different architecture from a VNA.

In a SA you must upconvert the input signal to a IF frequency higher
than twice the highest input frequency to avoid spurs and image signals.
A 3GHz SA requires a 6GHz IF. A VNA doesn't need to do that and can go
directly to a low 10kHz IF.

So a modification allowing to combine the functions of the 2 products would be too complex?

VNA + SA combo instruments usually are based on a SA architecture. There
are no low cost 3GHz SAs on the market because of the 6GHz IF required.

Well, I've been thinking about this a bit. A half-usable 3GHz SA in a similar price range as the S-A-A-2 would sureley sell like hot cakes... For a classic SA design, I think you would put the first IF somewhat above 3GHz, not at 6GHz. But you would still need a >6GHz LO (and first mixer). There are a couple of ways around this, but they come with cost/complexity/performance trade-offs.

With a two band design, frequencies much higher than 3 GHz would not be needed. You could have a <1.5GHz low band and mix that up to maybe 1.7GHz IF1 with a 1.7-3.2GHz LO. For the 1.5-3GHz high band, you could mix down to 0.7GHz IF2 with 0.8-2.3GHz LO. The high band still needs a preselection filter for image (and second harmonic) rejection, but I think two different switchable bandpass filters should be enough in this configuration. Still, now you have (at least):
- 4 mixers
- 3 LOs
- 2 RF switches
- 1 low pass filter (for low band)
- 2 switchable bandpass filters (for high band)

The "modern" approach is to basically ignore image rejection on the RF side. In principle a single mixer and LO is enough. Image rejection is then done with DSP. Essentially you record your spectrum (at least) twice, and offset the LO for the second pass. The images will shift in the wrong direction when the LO is changed (and second harmonic will shift at twice the rate of the fundamental etc.) so you can filter them out:
http://pxisa.org/files/Achieving%20Excellent%20Spectrum%20Analysis%20Results_final.pdf (Agilent)
The latest Keysight PNA and PNA-X network analyzers have an SA option that works like this as well.
The downside is, there are cases where this will go wrong and also you need a relatively fast IF ADC. I think at least twice the bandwidth of the signal you want to look at, otherwise the image will overlap. But then again you need a couple of MHz IF bandwidth anyway if you don't want analog RBW filters and still have a reasonable sweep speed.

(btw, the Signal Hound SA44B does something similar, but with two mixers and two slow ADCs; here is the patent: https://patents.google.com/patent/US8977519B2/en)

So is this economically viable right now? Maybe not, and I suspect Gabriel has already considered these options as well. In the end, some things you can get away with in a VNA will not give good performance in a SA. You will also very likely need to calibrate against a levelled signal generator at the factory. But I think the second option eventually will be.

Another niche that I hope will be filled sooner rather than later is a cheap 3 or 4GHz signal generator with sine wave(-ish) output. At least I think this does not exist yet. But it should be possible with two ADF4350, a mixer and some switches and filters. The idea (roughly) is to use the output (low pass filtered) directly for 2-4 GHz and mix with 2 GHz (also filtered) from the second one for 0-2GHz.

I have tested the tinySA in harmonic mode with DSP mirror elimination till 3GHz. It works but not good enough for the uninformed user

all those things sure can be solved ... but not with developing a simple
circuit with a price well below 100 bucks (saa2 is half of that roughly)

nice thing on the saa2 ... it is cheap ... and still accurate (more or less)

my idea would be ... a bit bigger unit (say double size) ... bigger
screen with better resolution ... fast sweep rate ... change clocks to
units up to 6 gig (and in software add harmonic measure also even if
with questionable dynamic range and accuracy) ... n sockets ... 2-4x
18650 in parallel (yes 10 to 12 amp hours for really long stand alone
work) ... a bigger cpu with more pins (why not add parallel and spi
display?!?) ... sd card (if we have that on the display why not use
it??) to store sweeps and maybe even update firmware that way?? ...
additional dc in with a switcher/charger (5 to 30v in without cooking
the lipo cells) ... bluetooth optional?? ... a smartphone app like we
have for sark100/mini60 ???

ok ok it is still not an r and s ... hp ... or similar ... but can be
used for same things (together with the clever pc software)

on startup the firmware should ask what cable length is used (say 4
preprogrammed/stored setups) ... one with no cable (only double male
connectors tor measure direct on antenna) ... one short (say 2-4 inch)
one mid (say 10 inch) and one long cable (20 inch?? ... longer??)

thats also a reason why more memory is needed ... the user should have
the option to preprogramm 4 setups (that he regularly uses) ... with a
LOOOOOONNNNG sweeeeeeeep (lots of calibration points stored "somewhere")
... so that when you wanna use same setup again (same cable length)
there is no need to do a calibration run ... also not when you change
sweep range or centre frequency ... means after the user made the 4
longer sweeps he can recall them on startup and on every settings change
no need to recalibrate

ok that would mean you have to do that long long sweep for each setup
(cable length used) ... but only ONCE!! and that would in the end make
the usage a lot faster ... no need to recalibrate after each change of a
setting cause those hundreds/thousands of calibration points already
exist ... (in between points could be interpolated maybe??)

if that unit could be a good sa ... and nwa and would then need a bit
different clock layout with mixers, filters, better baluns etc etc) and
it would cost more ... it would still be supercheap compared to
"professional gear" ...

say double or even tripple the price we have now ... it would still be
cheap ... and would sell like sliced bread (or hot cakes or whatever you
prefer hi hi)

sa and nwa combined ... bigger case and bigger screen!!! (we are all
getting older) ... n sockets ... up to 6 gig (with harmonic 18 or even
30 gig maybe??) ...sd card ... bluetooth etc etc ...  199 dollars

hot cakes??? yes ... hot cakes with free coffee :-)

most tricky thing sure is to develope the rf frontend to be accurate
also on the high end ... and if you redo the frontend you  could also
think about using real rx tx chips (ad936x or similar) ... and an fpga
... to speed up things a bit

dg9bfc sigi


Am 05.08.2020 um 19:11 schrieb switchabl@mailbox.org:

> Well, I've been thinking about this a bit. A half-usable 3GHz SA in a similar price range as the S-A-A-2 would sureley sell like hot cakes... For a classic SA design, I think you would put the first IF somewhat above 3GHz, not at 6GHz. But you would still need a >6GHz LO (and first mixer). There are a couple of ways around this, but they come with cost/complexity/performance trade-offs.
>
> With a two band design, frequencies much higher than 3 GHz would not be needed. You could have a <1.5GHz low band and mix that up to maybe 1.7GHz IF1 with a 1.7-3.2GHz LO. For the 1.5-3GHz high band, you could mix down to 0.7GHz IF2 with 0.8-2.3GHz LO. The high band still needs a preselection filter for image (and second harmonic) rejection, but I think two different switchable bandpass filters should be enough in this configuration. Still, now you have (at least):
> - 4 mixers
> - 3 LOs
> - 2 RF switches
> - 1 low pass filter (for low band)
> - 2 switchable bandpass filters (for high band)
>
> The "modern" approach is to basically ignore image rejection on the RF side. In principle a single mixer and LO is enough. Image rejection is then done with DSP. Essentially you record your spectrum (at least) twice, and offset the LO for the second pass. The images will shift in the wrong direction when the LO is changed (and second harmonic will shift at twice the rate of the fundamental etc.) so you can filter them out:
> http://pxisa.org/files/Achieving%20Excellent%20Spectrum%20Analysis%20Results_final.pdf (Agilent)
> The latest Keysight PNA and PNA-X network analyzers have an SA option that works like this as well.
> The downside is, there are cases where this will go wrong and also you need a relatively fast IF ADC. I think at least twice the bandwidth of the signal you want to look at, otherwise the image will overlap. But then again you need a couple of MHz IF bandwidth anyway if you don't want analog RBW filters and still have a reasonable sweep speed.
>
> (btw, the Signal Hound SA44B does something similar, but with two mixers and two slow ADCs; here is the patent: https://patents.google.com/patent/US8977519B2/en)
>
> So is this economically viable right now? Maybe not, and I suspect Gabriel has already considered these options as well. In the end, some things you can get away with in a VNA will not give good performance in a SA. You will also very likely need to calibrate against a levelled signal generator at the factory. But I think the second option eventually will be.
>
> Another niche that I hope will be filled sooner rather than later is a cheap 3 or 4GHz signal generator with sine wave(-ish) output. At least I think this does not exist yet. But it should be possible with two ADF4350, a mixer and some switches and filters. The idea (roughly) is to use the output (low pass filtered) directly for 2-4 GHz and mix with 2 GHz (also filtered) from the second one for 0-2GHz.
>
>
>

Interesting, I took a look at the tinySA and it looks quite neat. I might pick one up just to play around with. Presumably the Si4432 doing second LO, second mixer, RBW filter and log detector in one cheap IC is what makes this possible in this form factor and price point. It also looks like you did a really nice job with the firmware.

Digital image rejection might work better with more than just low/high LO. Keysight is using up to 8 different offsets on their VNAs:
http://na.support.keysight.com/vna/help/WebHelp12_70/Applications/Spectrum_Analyzer.htm#AdvancedSettings
Even then, this is what the datasheet says: "Image response: Mostly eliminated. Intermittent image response may be seen when making multi-tone or modulated signal measurements."
But in any case, this will not work with the Si4432. And the separation of just 2x870kHz with a 600kHz RBW might be marginal as well.

An interesting option for a higher end, full-digital IF might be AD9629. LCSC is selling the 40MSPS 12-bit version for <7$ and the specs look impressive. If a Cortex-M7 is enough to do the processing, this might even be viable.

I am also curious how well a AD963x-based solution would work, although it would of course be way too expensive. I've been meaning to do some experiments with a PLUTO SDR, but I haven't gotten around to it. Unfortunately to use the full 56MHz bandwidth, some processing or buffering has to be done directly on the Zynq, USB2 is just too slow.

@dg9bfc I think the calibration feature you want (or at least something similar) you could easily implement on the current platform. All that is needed is calibration interpolation and automatic port extensions. Then you do one high-resolution calibration at the test ports and save that. You don't need to go crazy with the number of points. As long as you don't miss any phase-wraps and take care of the discontinuity at the band-switch, you should be fine. Then just leave the cable open (or connect an open/short standard) so the VNA can measure the cable length using the reflection and correct for it.
Of course it is more accurate to calibrate at the test cables and on the correct frequency grid, but for non-critical measurements it will be fine. And if you really care, you have to re-calibrate anyway because of drift.