When I started with VHF and UHF amateur radio back in 1982, it was somewhat difficult to even get on the 1296 MHz band. My friend who was an RF designer used to joke that the L-band power transistors at the time had more reverse gain than forward gain.

Yeah that. Particularly when it came to test equipment which was pretty much down to building it yourself (or being rich). I saw a lot of things that looked like plumbing in the 80s rather than electronics. Not until the 90s did I understand this stuff and it was actually pretty amazing what people were building with no gear and no tools. One of the finest things I saw was a Lecher line or movable short. And some weird stuff on 10GHz that used micrometers for tuning!

As for plumbing goes: relatively recent 25GHz+ breadboard single transistor sine wave oscillator[0]

[0]: https://www.youtube.com/watch?v=yR9TweWf3w8

Ah yeah I watched that thread in EEVblog.

For an alternative and a similar system, TI has an RF DAC available now but the RF ADC is probably in the pipeline [1].

Imagine a flipper zero version on steroid based on this RF ADC/DAC platform but the price will probably going through the roof [2].

[1] 16-bit, 2-channel, 20.48-GSPS 12-GHz RF digital-to-analog converter:

https://www.ti.com/product/DAC39RF10

[2] Flipper Zero: Multi-tool Device for Geeks:

https://flipperzero.one/

It is amazing that we can do A/D and D/A at anything faster than audio... let alone 28 GHz D/A and 20 GHz A/D. The hardware required to anything more just process the samples at those speeds and do anything more than frequency shifting is one of the compute intense things I can imagine.

It's very rare to actually utilize the full bandwidth of the ADC to do any processing in these. The data gets immediately decimated down.

Correct. If you look at the block diagram of the AD9084 (from the article link), it has on-chip DUC (Digital Up Converter) and DDC (Digital Down Converter) blocks.

https://www.analog.com/en/products/ad9084.html

That's not really true, there's plenty of applications that use these sort of bandwidths. A big one is optical communications (28 GHz bandwidth is actually not super high there), but also some specialised microwave comms and radar/military. They alm do processing at 10s to 100s of GS/s (obviously highly parallel).

While full bandwidth applications certainly exist, it is rare. Most direct sampling schemes are used for having a purely digital RF front end within the Nyquist range of the ADC/DACs. It's cheaper and easier to develop firmware/gateware than debug analog RF issues and respin hardware.

That's not even crazy considering you can do tens of Teraflops/s on modern GPUs.

Get your data into the system via PCI-E, do some RDMA magic to get the data onto your GPU and put the cores to work.

That's an incredible level of integration; having so many ADC's and DAC's with 20Gbps sampling rate all in a single package.

What would someone do with one of these?

Applications:

  · Radar and communications, L/S/C/X/Ku band radar and electronic warfare
  · Phase array system,
  · Broadband communications systems
  · Electronic test and measurement systems
  · Satellite communications
  · Microwave point-to-point, X-band and 5G mmWave

https://www.analog.com/en/products/ad9084.html

So most probably subject to export controls.

There are things that you can build with a dozen $30 rtlsdr dongles that would be export controlled.

The bar is pretty low these days. It reminds me of the time Phil Zimmermans PGP was considered a munition for USA export purposes

The KrakenSDR is certainly toeing the line.

I think that it should be possible to distribute clock lines to at least a 16 channel engine using Tayloe polyphase detectors, which have really good noise characteristics, image rejection, and incredible dynamic range, they also make efficient use of the received signal. One 8 phase clock to feed it all, and coherent downconversion is guaranteed.

Parts cost would be $100 or less via Digikey.

I was thinking about:

> Multiple loopback (ADC to DAC) supported ~45 ns without DSP path

If you can get a tiny bit of DSP in there without boosting latency out of the water, then I bet this could be used to pretty thoroughly defeat the supposed jamming immunity of frequency hopping techniques.