I know this thread is a bit older, but it still comes up when you search for technical details of isolating USB.
I found one more isolation product: A USB 3.1 Gen 1 (SS with 5 GBit) Hub with isolation.
I found it sold under these names:
coolgear CG-U31iS4PH,
https://www.coolgear.com/product/4-port-usb-3-gen1-isolated-hub-15kv-esd-surge-protectionExsys EX-1182VIS,
https://www.exsys.de/index.php?page=product&info=2062The OEM developing & manufacturing these is most probably:
Comwise Tech MG-14346IS,
http://35.222.228.198/usb-hubs/When opening it, you see a quite interesting concept: isolating USB 2 (from Low Speed to High Speed / 480 MBit) is done differently than USB 3 Superspeed. USB 2 isolation is done with the Silanna ICE08USBC, similar as in the HifiMeDIY described above. USB 3 SS is isolated with ceramic coupling capacitors.
There is a regular GL3522 hub IC at the input. One of it's downstream ports goes to the isolation barrier, splitted by protocol as described above. This is most probably being done to buffer the USB 3 SS signals, to create a clean and known signal across the coupling capacitors. On the other side of the isolation barrier is another GL3522 hub, which connects the 4 downstream ports.
There is a Minmax MA01-05S05HI DC/DC voltage regulator on board. It provides 200 mA, just enough to power the hub itself, it is not designed for onnecting bus powered peripherals. You can disconnect it with the jumper on the pcb. The DC/DC isolates 5.2 kV for 60 seconds. That translates to about 400 VDC permanent isolation, which is more than is offered on the data path, see below.
For bus powered peripherals you are supposed to connect an external 5 V power supply through a 5.5 / 2.1 mm barrel connector.
Finding details about the Silanna ICE08USBC is a bit difficult, as Silanna doesn't publish any datasheets. The best I could find is this presentation about their portfolio:
http://spacewire.esa.int/WG/SpaceWire/SpW-WG-Mtg17-Proceedings/Documents/Session 7 - SpaceWire Link Isolation (G.Baterina).pdfOn page 7 you can see how the IC is internally constructed. It seems to use capacitive coupling internally. So all claims about "optical isolation" you see in the Exsys datasheet are misguided.
On page 8 they claim isolation for 2.5 kVrms. Comwise Tech claims 3 kVrms in their datasheet. It could be that Silanna improved the isolation in the ICE08USBC, the presentation is from 2011 and they show the SIL11USBx which is probably a predecessor of the ICE08USBC. It could also be that Comwise Tech confused DC and AC rms.
Also they clearly state "Working voltage >100V". That means the 2.5 kVrms are just specified for a short period, usually 60 seconds. After that the isolation slowly begins to break down. The working voltage is what they consider safe for permanent usage. 100 V permanent matches what is commonly specified for DC/DC converters rated 2.5 kVrms @ 60 s.
The ceramic capacitors crossing the isolation barrier are 1812 size with 2 mm height. I measured them at about 1 nF. Several manufacturers offer such caps in X7R rated for 3 kVDC. I didn't find any for higher voltages in 2 mm height. So one can assume that they are rated for 3 kVDC. In the Datasheets they allowed for short-term overload to 120%, so no 3 kVrms. Also they warned not to use them for AC applications with continuously changing polarity.
The problem with capacitive coupling is that if the downstream side is floated to AC, a current starts to flow across the isolation barrier. With the voltages safe for permanent usage and a sinusoidial 50 Hz the current is in the µA range and doesn't matter much. But when using rectangular AC this can quickly become an issue and for example destroy the hub ICs because the data lines go outside their allowed common mode voltage.
There are no bulk capacitors anywhere on the pcb. The USB spec requires a minimum of 120 µF per downstream port, connected to the same net as the socket, after any protection measures. This is clearly missing. When plugging in a new device or when a device creates a current spike, the power for the 3 other ports or the hub IC can droop. This could be fixed by modding the board and soldering some polymer caps in the right places.
It seems like there are ESD protection ICs, the MSOP-8s labeled "118 6yUHG". I couldn't find a product name or datasheet with this marking. The SOT23-5 near the downstream ports are most probably short circuit protection ICs.
I did some USB data transfer measurements: transfer with USB 2 HS was a little bit slower (336 MBit/s) than directly connected to the PC (352 MBit/s). But this is very similar to what I measure on all my other hubs, so probably more related to having a hub in between at all than this specific hub. I couldn't measure a difference in latency, as USB latency tends to be hard to measure precisely. These measurements were done with the "benchmark" applet of Glasgow revC1 (Cypress FX2 + FPGA)
https://github.com/GlasgowEmbedded/Glasgow.
I do not have a device to constantly saturate a USB 3 SS link. So for USB 3 SS I just did stability and data integrity tests, as I've also had USB 3 hubs which had connection problems under load. I wrote and read back 1 TB without disconnect or a data error.
Final Verdict:
Good:
- Isolated USB up to 5 GBit/s, no shortfalls with USB LS, FS and HS
- Ready-to-buy solution
- Convenient to use as it directly provides 4 ports and a power plug
- Solid construction
- ESD protection
Bad:
- Permanent voltage allowed across the isolation just about 100 VDC, so not suitable for connecting USB instruments floating on mains levels
- Capacitive coupling precludes some usage cases with AC
- No bulk capacitance on downstream ports as mandated by USB spec
- Costs about 200 EUR