Level translator has fast edges. termination needed?

• Level translator has fast edges. termination needed?
  Posted by JPortici on 09 Feb, 2017 21:11
• designing a pic32 board..
  there are seven signals that will go offboard and will control 5V logic.
  
  - a full duplex spi, this will go on a motherboard (this is a submodule) to shift registers, as IO expanders. Not expanding onboard because less connections.
  - three (slow) analog mux selectors, this will go on the motherboard as well as off board, on a ribbon cable, along other signals (i think 20 cm total cable lenght)
  
  to accomplish this i was thinking to use a TXB0108 by TI, which is a bidirectional level translator with push-pull outputs (all the others i've seen are open drain types)
  the datasheet specifies for 3V3 <-> 5V operation 10ns risetime and output impedances of 40 ohm on the 3V side, 50 ohm on the 5V side.
  
  i think i can be safe with the 3V line, the driver will be as near the mcu as possible. the SPI bus too because the shift registers on the motherboard will be near the connector.
  problem is, the three mux lines.
  I'm wondering if i should add termination to the line because of the edges, if so how to do it. i was thinking to terminate at the source, adding a resistor to match the other side impedance.
  
  so, if for example i put 60 ohm to have 100 ohm output impedance...
  assuming i had three mux connected to the line i should have 300 ohm impedance at each input to avoud reflections. correct?
  or, another way to slow down the edge, the switching rate will be far lower than the SPI line (400 kHz against 10 MHz)
  
  http://www.ti.com/lit/ds/symlink/txb0108.pdf
• #1 Reply
  Posted by danadak on 10 Feb, 2017 14:22
• Looking at datasheet I cannot seem to find a current drive spec. So i am thinking
  termination to 50 ohms will destroy noise margin ?
  
  One approach to de-Q-ing the parasitic LC problem is serial R, range 10 - 50 ohms
  depending on layout. I have used that on fast clocks with fairly good results. 50
  Mhz kind of speeds. Not a well controlled timing environment, in my cases almost
  all applications asynch. But simple to eval and use.
  
  
  Regards, Dana.
• #2 Reply
  Posted by vealmike on 10 Feb, 2017 14:40
• If your propagation delay is equal to or longer than three times your rise time then you need to treat the system as a transmission line (think about it!)
  
  If you determine that you do have a transmission line, then you need to consider when you will be sampling the waveform. It could be that any reflections have died away by the time you sample.