problem with high speed data transfer

[mingyuexin]

I'm trying to work on communication between two board A and B equipped with FPGA,440 bits data must be transfered from A to B in 3 us,
there are three lines for communication.In addition,there are 8 A boards must communicated with B simutaneously, I use DS coding to transmit,
which works at 200MHZ for communication of one A and B,but when it comes to communication of 8 A and B,it doesn't work even at 100MHZ,probably
it's because crosstalk.I don't know how to solve this problem?Anybody has solution?Perhaps I can change the way of communication,but what else
communication protocol can I use? I don't have many experiences about design of FPGA,can anybody help me to work this problem out,any suggestion
is appreciated.
PS,board A contains virtex2,board B contains virtex4,so SERDES is not suitable here.
board A has only virtex2,nothing else processor.
board B has only virtex4,nothing else processor.

 

[JohnDuq]

We need more infomation to start helping on this.

What are your three lines for communication? If you don't have a dedicated ground on each channel you have an EMI/crosstalk nightmare. Differential can help here if done correctly (many engineers mess this up).

Do the 16 boards share common power supplies? That can be a potential crosstalk path.

Are all 8 channels running at the same time? Can you stagger the communication (time-division-multiplexing)?

What is the physical medium (cable?) between board A and Board B?

One channel works and eight channels fail. What about 2 channels? What about 4 channels?

 

[mingyuexin]

Thank you for the reply.I'm sorry I didn't describe it clearly.
>>What are your three lines for communication? If you don't have a dedicated ground on each channel you have an crosstalk EMI/crosstalk nightmare. Differential can help here if done correctly (many engineers mess this up).
Actually the cable is network cable,which have 4 twisted pair,three of them can be used for this communication,I use two of them with DS coding and LVDS is adopted here
>>Do the 16 boards share common power supplies? That can be a potential crosstalk path.
In total there are 8 A boards and one B board,they should have different power supply.
>>Are all 8 channels running at the same time? Can you stagger the communication (time-division-multiplexing)?
yes,8 channels must running at the same time.I can't use time-division-multiplexing here,because each board have almost 440 bits to transfer,they must all arrive there in 3 us.

>>One channel works and eight channels fail. What about 2 channels? What about 4 channels?
I just did the test in only B board,which means B sends via RJ45 and B receives it back via anothere RJ45,the two RJ45 are connected with 5 meters calbe.
For one link,it workes reliably.
But for more than one link,it doesn't work,the received data is wrong.when I do post simulation in questasim,more than one links receiver have unknown state.even if I reduce the speed to 100 MHZ,I don't know why.

 

[JohnDuq]

"But for more than one link,it doesn't work,the received data is wrong.when I do post simulation in questasim,more than one links receiver have unknown state.even if I reduce the speed to 100 MHZ,I don't know why."

If your simulation doesn't work there is an extremely high probability that the real world will never work. Simulators don't have crosstalk. Track down the root of that unknown level and you are probably getting close to solving this.

Have you tried putting the same data into all eight channels and see if they all get the same result? That is another way to eliminate any crosstalk issues.

John