TUSB551: Problems with TUSB551

how long is the failed cable? can you draw a block diagram as well as the schematic?

The failing cables are from 40 cm to 2 m long. The cable from camera manufacturer, which works fine, is 3 m long. The other working cable is 30 cm long, but frame losses happen sometimes with it.

The block diagram is:

CAMERA -- (mini-usb B connector) -- usb cable -- (usb A connector) -- REDRIVER BOARD -- (mini-usb B connector) -- usb cable -- (usb A connector) -- COMPUTER

Our redriver board is designed according to the TUSB551 datasheet, the relevant part of the schematic is attached below.

Please note that the TUSB522 EVM is behaving virtually the same way as our TUSB551 board: cables that work when connecting the camera directly to the computer often fail with the TUSB522 redriver in the middle.

can you take the waveform to see where it was stuck?

Unfortunately, we do not have access to any oscilloscope with bandwidth sufficient for USB3 capturing waveforms.

It is worth noting that the camera was tested with USB3 ports on an add-on PCIe card equipped with Renesas uPD720202 controller (which was suggested by camera manufacturer). The camera itself has Cypress CYUSB3014 and Littelfuse SP3010 surge protector onboard. The USB3 traces on the camera board are short (about 1 inch), but there is a visible asymmetry between + and – lines, at least in the top layer.

Our board is carefully designed. The USB3 lines have no stubs, no vias, void under TX capacitors, and its impedance is controlled during manufacturing. They are put at an angle to the prepreg fibers. The total length of the line is about 8 inches, and the redriver is put near the middle.

In the meantime, we did many additional experiments and achieved some improvement.

First we have found that our USB plugs and sockets were contaminated with soldering flux. One of the sockets must have been left unwashed well enough and, through the frequent changing of cables, all sockets and plugs finally got dirty. After every contact in every plug and socket was cleaned thoroughly (including the socket in the PC), the problems with hard disk drives and USB memory sticks mostly disappeared. However, the camera still did not work, neither with TUSB522 EVM, nor with our TUSB551 board. Depending on the OS/DE/EQ settings, the device was not recognized at all, or constantly recognized and disappearing. The only exception was the ‘good’ long cable. It worked with TUSB522 at DE1=F, OS1=H, EQ1=F, DE2=L, OS2=H, EQ2=F. Particularly, increasing the DE2 causes problems. None of the cables worked well with TUSB551 (at any DE/OS/EQ settings).

Second, we have tried another PC: a notebook with Intel Kaby Lake chipset onboard. It performed much better. For each camera-redriver cable, except one (very short, but apparently poor quality), we were able to find such settings for TUSB522 EVM that the camera performs reliably. These settings differ from cable to cable, but generally low DE is preferred. However, for our TUSB551 board, no reliable settings could be found, with the exception of one cable (the long cable provided by the camera manufacturer).

Third, we got rid of the surge protectors (TPD4E02B04) from our board. No improvement.

We have drawn a conclusion that, despite of careful design of our board, there is some compatibility problem between TUSB551 and the camera. We cannot change the camera, so we had to try different redriver, preferably without making a new board. There are at least two pin-compatible redrivers, but only from other vendors. We have chosen NB7VPQ701M due to its Gen2 compatibility. The results were surprising: the camera worked well with both computers (Intel and Renesas USB3 controllers) using most combinations of cables and either the default redriver settings or lowered de-emphasis and output swing.

This seems to support the above conclusion. There is apparently something that makes TUSB551 incompatible with the rest of our setup. Maybe it works “too strongly” (it has the largest output swings of the three tested chips)? Anyway, does it make sense to try another TI’s redriver? Can a linear redriver, as TUSB1002, work significantly better? Or older chips, like TUSB501?

NB7VPQ701M looks same with TUSB551, are you using TI EVM? can you TUSB1002 as well to see if it works?

Indeed, NB7VPQ701M looks very similar to TUSB551, except smaller delays and jitter (related to its 10 Gbit/s capabilities). However, surprisingly, it works much better in our application. We have assembled another piece of our board using NB7VPQ701M, and it also works well.
We are using TI EVM for TUSB522, as there is no EVM for TUSB551 (at least we have not found it).
At the present, we have exhausted the research budget for the project, so we are going to stay with the chip that works without any re-design of the board, i.e., NB7VPQ701M.