SFP+ optical transceivers are data-agnostic and they’re actually very simple devices, the only function they perform is optical to electrical conversion. As long as your signal is a CML-like differential signal and has the correct swing level, SFP+ modules will happily transmit it – even if low-frequency operation is not specified, in practice it works. I heard some people have even successfully used it for 10 MHz clock distribution.

Speaking of receiver detection, the solution is also simple. The quick and dirty solution is simply connecting two AC-coupling capacitors and 50-ohm resistors to the SSRX pair. This hack is surprisingly effective, it simultaneously serves two purposes at once: First each line has a common-mode termination of 50 ohms so it can be detected by the transmitter. Second, it also level shifts the USB signal. USB 3.0′s signal swing is up to 1200 mVpp, but many SFP+ modules only accept 600-850 mVpp. With an 50-ohm resistor it’s now 50/25 voltage divider, reducing the swing by 0.66x, to 800 mVpp. Third, one may ask: doesn’t the resistor destroy the impedance control of the transmission line? A quick simulation tells you that the return loss from a 90-ohm differential transmitter (USB 3 PHY) to a 100-ohm differential receiver (SFP+) with two 50-ohm resistors to ground is still better than 10 dB, which is good enough.

Of course, unconditionally connecting the resistor basically fools the transmitter into thinking that a receiver is always there regardless of the readiness of the far end, it will attempt to establish the link over and over again via LFPS. It still works in real life! Nevertheless, if you want to be a good citizen and follow the specs, I think you can simply put a high-speed RF switch in the data line or in the termination, and only close the switch when the far-end is ready, using the TX_DISABLE and RX_LOS for side-band coordination.

Finally, there are two limitations.

1. Loss budget. USB 3 allows up to 20 dB loss from the host to the device. If the device is not plugged directly into the USB port via USB-A (you can do this, but I’m really not a fan of a huge device hanging on the USB port), but through a long cable, like USB-C, I think SFP+ modules may have difficulty transmitting a clear signal. Fortunately, you can use a USB Hub as a repeater to decouple the SFP+ and the cable.

2. The USB 2 is indeed a problem, no ideal solution exists, and it’s not possible to solve the problem without violating the USB spec. But a good enough workaround is acceptable, the solution would be the VL670/VL671 ASIC by VIA Labs, Inc. It basically transparently upgrade an existing USB 2 device to USB 3 by emulation.

I’m currently developing a prototype reference design to do exactly what I said. It’ll be published as a free and open source hardware design. Watch my blog for update ;-)

The USB 3.0 spec is explicit about that all wiring, both USB 2.0 and USB 3.0 wires are connected between each device and host / hub, so you will definitely be violating the spec by letting only the SSTX / SSRX wires through for one of the devices.

Having said that, I can tell from personal experience that if you connect only the SSTX / SSRX wires, it works perfectly well. You should also have the GND connected to prevent excessive DC voltages.

However since you’re breaking the rules, and is not in control of the device, I can see a possibility that the device refuses to play ball with the 3.0-only wiring. Not very expected, but yet possible.

If it’s very specific hardware involved, I would get myself a USB 3.0 extension cable, and cut off the non-3.0 wires, and see what happens when the device is behind it. Just be sure to test that it connects as USB 3.0 before doing the cuts. Some extension cables are junk.