Yes, this does sound like a very interesting, and useful idea.
One problem with using "the Internet" for truly anonymous communications is that everybody has an IP address. We may try to hide it, by means of VPN's, or TOR, or both, but such communication is still a bit risky. Have a big-enough opponent, say the NSA or GCHQ, and somebody may trace it.
Imagine a person with a computer on a canonical hill, with an hyper-wideband SDR radio, spreading out over perhaps 100 MHz in bandwidth. (unused UHF station bands? Military frequency allocations?) I'm thinking of a transmitter of 10-100 watts, which when spread over 100 MHz, is not particularly loud.
Conceivably, depending on his transmit footprint, 1 million potential listener-computers can hear his signal, also by SDR. He receives data, presumably through VPN or TOR, etc, by means of a packet which is doubly-encrypted: The operator decrypts by using his private key, revealing a still-encrypted packet that he then transmits by ultra-broadband, using a secret provided by the decrypted packet.. Anyone who can 'hear' his signal, and who knows the encoding secret, can decrypt the data, but it can only be further decrypted by some further key. So, nobody knows who actually was the intended recipient.
This amounts to anonymizing "the last 10 miles", making it essentially impossible to learn who is actually receiving this information.
This process could be reversed, with possibly a different person's setup 'listening' to a similar ultra-wideband signal, with a code provided by an encrypted packet. If the receiver hears something, it could be sent by prearrangement to a VPN or TOR, or perhaps re-transmitted to a different ultra-wideband facility.
It should be possible to 'listen' for a large number of simultaneous transmissions, at one site, since each will be encrypted by a different 'secret' transmission encoding.
It may not be obvious, but it will probably be necessary to charge for these services, even if the providers would want to do so for free. Remember "denial of service attacks"? If this were provided as a fairly well-known service, chances are good that somebody would want to gum up the system with overloaded, useless requests for transmissions or receptions. Charging for this system, at least when the service is heavily occupied, would ensure that users would continue to have access to it, or at least the operator is getting rich, or both.
Jim Bell