April 10, 2009

Sensor interface in 373

by Andrey Filippov

I’m considering using Xilinx high speed Serial I/O – 4 of the GTP in XC6SLX45T in CSG324 package (15mm x 15mm) to interface the sensor boards – it can be very attractive to be able to migrate to higher performance sensors from the current Aptina 5MPix ones.¬† It would also enable higher bandwidth to the next version of the 10359 board – up to four of 3Gb/s channels, reduce EMI and enable longer cables.

The problem is that normally you need a large chip on the other (sensor) end capable of encoding data in a way compatible with the GTP – it would be an overkill for the simple CMOS sensor module. But it seems there is a way to use a tiny (5mm x 5mm) 3 – channel (plus clock – total 4) Texas Instruments SN65LVDS301 FlatLink(tm)3G that encodes 24 bit parallel data plus 3 sync signals into serial data. Maximal “pixel” rate – 65 MHz (not enough for MT9P031) but with additional register or better CPLD (like XC2C64A in CP56 6mm x 6mm package) it would be possible to double the pixel rate and add some scrambling to reduce disparity – not that critical as the data rate per pair would be only 480Mb/s. It seems that it would be possible to make GTP to accept that slow data rate, bond 4 channels (each still has to be able to recover it’s own clock) and use SN65LVDS301’s clock as a master in this bonding with a comma mask ox 10’b1111111111 and comma pattern of 10’b1111100000.

Most likely I’ll still use smaller XC6SLX25 (w/o “T”) first and use the bank that has GTP-s in 45T to connect the 30-pin parallel flex connector (compatible with the 10353 and the existent sensor boards) first, upgrading to high speed serial I/O in the next revision/model. that will be rather simple task as it will not require major PCB redesign – FPGA is side-by-side with the sensor connector, so those traces do not intersect with other ones.

Leave a Reply

Your email address will not be published. Required fields are marked *

1 + = ten