It would be cool but I don't see it happening any time soon. Plenty of demos have been created, but the ASICs that the SDR would be competing with are much more efficient at doing their particular job, and they're very cheap.

I vaguely understand that modern high performance rf chipsets are largely software defined anyway, which is why they require binary firmware blobs.

The modulations have gotten so complex that there's really no chance of a CPU or even a general purpose FPGA keeping up in real-time. Someone unencumbered by an NDA would need to be smart enough to design a specialty FPGA with the right sort of accelerators in silicon. They would also need to avoid any active patents.

Like pretty much anything, I imagine someone will figure it out for any given technology a decade or two after it's mostly obsolete.

Are patents a problem for software distributed in source code form? I don not think so. I remember something related to font hinting and apple patents that could be circumvented by simply uncommenting part of the code recompiling.

I specifically mean patents for the "generalized" accelerator hardware implemented in silicon that would presumably be needed to make a specialty FPGA that's fast enough to handle modern RF modulations, even in a vaguely software-defined way. That's presumably companies like Qualcomm's secret sauce, and you can bet they have it locked down with patents and NDAs. If someone came up with a general purpose CPU powerful enough to actually do everything fast enough purely in software, then that might be different (but would likely have plenty of its own patent issues!)

The latter already exists, at least for LTE.

Well, SDR-based clients that you could package in a smartphone form factor. They're fairly power-hungry, though.