I suppose it depends on what attack vectors you're trying to protect against and how vulnerable you'd be if a successful attack were performed, however, the two concerns that immediately come to mind are:
1. If the radio stops working for some reason, you'll possibly be fed a repeating sequence that represents pure silence on all frequencies?
2. An attacker that is aware of your algorithm could potentially transmit a strong enough signal from close enough to your equipment to effectively overpower any "noise", resulting in a predictable set of input data.
First of, thank you for the valid and pertinent answer.
I hadn't think of #1, but I can add a response validation algorithm, either at the MCU or computer checking for patterns or repeated bytes.
As for #2, it has to be potent enough, has to "guess" when the print key would be pressed, as that the only time the random bytes are actually used, and it has to be a FM jammer, as the attacker has also to "guess" which frequency is being listen to and if or not shifted. Thus a jammer would probably render a pattern, throwing an error with the fix applied to #1 and having to running it in continuum, people around would start to complaint of bad radio reception.
Your attacker could simply broadcast on all potential channels for an extended period of time. You could potentially counter this by personally listening to the channel to confirm there is no interference, however an attacker could counter this counter by learning which channel you are listening to, and when, and using this information to learn the input to your "randomness".
seed in a way that is not affected by any external variable. Any external variable you use is going to open up the potential that someone will observe when you generate your seed and can look for information generated around that time.