Full duplex LVDS

Straight from a FPGA with LVDS IO ? I don't know whether they provide tristateable LVDS IO though. I remember having seen multipoint LVDS drivers with HCMOS inputs. They didn't operate at 600MBit though, a limitation of the HCMOS. They did 400MBit only.

Rene

--
Ing.Buero R.Tschaggelar - http://www.ibrtses.com
& commercial newsgroups - http://www.talkto.net

If you want to do 600Mbps in *both* directions simultaneously (over one path) then you'll have to do something along the lines of the telephone hybrid transformer.

I've never attempted this type of rate using a hybrid (sometimes called transhybrid) transformer technique, but it should be possible. You'll need this type of transformer at each end of the twisted pair.

The main reason why it might be possible is because destination path impedance is known and fairly stable. This is what allows the transformer circuit to subtract-out each side's own transmit signal from full-duplex path.

You could also (theoretically) use an active technique (e.g., opamps), but getting this to work at this speed might be tricky, although easier to put together.

Do some searching and you'll find the transformer and opamp topology. You'll probably have to design and wind your own transformer, if you go this route. If you use the transformer technique, it will also be necessary to encode (and decode) your data to ensure a minimum number of edges per group of bits. This will keep any DC component out of the signal (which can't be passed by a transformer). 8b10b or 64b66b encoding/decoding techniques are two examples of this.

Bob

It's certainly possible. Whether it can be done simply or economically or in such a way that it qualifies for FCC or other agency approvals is another question.

Gigabit Ethernet uses simultaneous bi-directional signaling, But not at such a high data rate per pair. In fact, I think each pair is operating at

25 Mega-symbols per second.GigE doesn't use transformers for subtraction. It uses either signal processing or perhaps simple analog subtraction to remove the near-end contribution.

At your data rate I think it would be hard to go analog. But then again, maybe not. Maybe you could use a high-speed differential amplifier to buffer the incoming signal, then subtract a replica of the local signal. With a truly differential amplifier, this could work, but you need lots of bandwidth.

Best of luck!

--Mac

Thanks for everyone who has replied. From your answer I understand that this type of solution (for this data rate) is far from being common, so I'll probably try another approach for my problem.

Thanks again, Avishay