And to be clear, any change in configuration can interrupt the interaction -- dropping hub power might not kill communications directly/immediately (the hub chip may still be powered by the host cable for example), but the peripherals might not be receiving full power, so they don't respond to messages correctly, or at all, and the host driver may show a dropped connection (device disconnected). Basically equivalent to Windows producing the ba-dump sound when this state changes (operating/disabled).
Which, YMMV, I've seen embedded Linux platforms where interrupting USB causes the driver to fail silently and it has to be manually cycled to restore function, it doesn't auto-retry. No idea if there's a setting for that or what. Not that that should be a primary issue, but leakage works both ways and maybe the transmitter (if applicable) can knock it out, too.
Does the power go through the same conduit, and are there any ground connections -- is this in a vehicle perhaps, so things may end up grounded by virtue of mounting hardware?
Could you make a drawing showing everything in the conduit, and what power/ground connections are made at each end?
Also, have you tested the radio on transmit, does that knock out the USB?
It's entirely possible that the USB connectors or cables have poor shielding, and noise is leaking out, at pretty much any connector. It's not obvious if this is leaking into the antenna cable (or whatever it is?) directly, or being radiated and then picked up by the antenna. RG-58 is notoriously leaky, enough that it might be a problem when paired with a noisy cable like that. But there are plenty of other ways for noise to get into the radio system, it's just one of many possibilities.
Also, do you have a rough idea of how much signal is being received? If it's only like +6dB from baseline, yeah, ferrite beads may prove useful. If it's just screaming, there's not much you can do with filtering alone, it'll be leaky shields or something and the only meaningful change is improving them (or adding more).
To that end, you can add cable grounding clips. This works by first stripping back some of the outer jacket, exposing the ground braid. (So, this is NOT a sealed environment anymore; that will take more work to achieve!) Clip the braid into a finger contact, much like a fuseholder; there are sources for these (cable grounds) specifically, though I don't have any links offhand unfortunately. The clip must be placed somewhere that grounding will be effective; for example, at the ends of the conduit, assuming it's metallic. The low ground impedance also makes ferrite beads more effective: FB adds series impedance to the cable, on the order of 30 to 300 ohms (depending on size, or more with multiple passes/turns through the core), and noise, from the source towards the grounding point, acts as an impedance divider. The ground point may have an impedance of 1-10 ohms at this frequency. So a reduction of 20-40dB is reasonable in this way, more with multiple clips (and FBs between) or careful construction.
Tim