The claim that TV RF noise is un-attackable doesn’t convince; it can be no stronger than AES’s resistance to a chosen plaintext attack, and could be much weaker. The video signal and tuner is AM not FM, so there is no capture effect (good) but if the transmitted signal is enough stronger than the noise power, the number of bits of noise remaining at the A/D output after Automatic Gain Control (AGC) can be arbitrarily limited.

Please run your capture and statistical test with the tuner set to a strong nearby TV channel (analog not HD-TV, although Analog mode tuner with HD-TV signal would be interesting too) .

What effect would using a non-random, non-secret key and IV for the AES step have on entropy dispersion? On attack by jamming? If I beam NTSC equivalent of all-ones on your channel strongly enough, and I know your key and IV and above procedure, your only entropy comes from any quantization errors in the A/D, if any, and dispersion liimiting ability to detect which point capture was initiated?

If I broadcast a pre-recorded snow-patttern, with proper framing, would you notice?
If I broadcast deterministic PRNG output, without framing, could i the attacker determine where you started capture except by brute force? With that, how much does the dispersion of quantization errors save you?