For a USB 2.0 Device you can use the CDC class for no-loss communications at the rate of one SOF (start of USB frame) every millisecond, or 64,000 bytes (64 data per SOF) per second. However, USB is half duplex, with an ACK needed from the host, so the theoretical max to transmit is about half (32,000 bytes/sec). In practice you won't see rates that high due to latency on device and host ends.
Regular CAN has a total packet size of roughly 8 to 16 bytes (8 to 72 for FD-CAN), if you include the header and timestamp (that's important for a sniffer). Average case your thruput would be around 2,000 CAN packets/second delivered through USB. In practice some packets will be only 8 bytes, with no data field, so it might be a little better. So the question is, how many packets/sec can your CAN network support? Collisions increase with active node count and the CAN protocol used, so the more nodes running, the lower the theoretical maximum per second. You can calculate the absolute worst case with only two nodes present, and only one sending as fast as possible. That will tell you if you can sustain a CAN rate of 500Kbaud.
Alternatively, you can go with a controller that supports USB 480M/bit HS mode. Plenty of USB bandwidth then, but an order more complex for software.
I would not recommend HID mode unless you intend to interpret the CAN messages into something meaningful for an HID device. Converting manifold pressure transducer measurements from CAN to a mouse operation results in the cursor bouncing all over the display.
Jack Peacock
