This is not Ethernet, it is not only about cable losses and their amplification, the times in which information passes through the cable are also important.
Time restrictions are much stricter than in the ETH.
USB is a symmetrical serial bus, therefore for D + D-, as in ETH, we use twisted pair.
USB is quite time strict,
the total delay caused by the cable connecting two USB devices should not be greater than 30ns.
The theory is about the use of cables
For low speed:
two untwisted 28 AWG wire pairs
For full and high speed:
shielded twisted pair D + D- 28 AWG
28AWG unshielded power pair
So for cables 6-10ns / m we get distances of 5-3m.
In practice, the 10m cable works too, 15m not with every device.
Additionally, if we used USB hubs, their number also cannot be arbitrary.
The total unidirectional delay in the transmission path should not exceed 350ns.
On the other hand, the hub + cable should not introduce a delay greater than 70ns.
So if we have 30ns cables and properly functioning hubs, they cannot form a chain longer than 5 separating devices.
In practice, it is different, you can often bend the standards, but some devices do not work at all, or work unstably (they disconnect and reconnect).
There are also problems with USB load capacity, and with larger hub configurations, one device that constantly starts the USB enumeration process may affect the others.
Also remember about the time restrictions in USB transmission,
often placing the hub in the middle of the transmission channel can help,
if the hub is bad it can often hurt.
Also, beyond the limits of the USB specification, experiments remain.
By the way, I wonder how USB ETH USB devices work,
they must have cleverly made data buffers.
