Maintaining signal integrity is a matter of degree. Any discontinuity in 
impedance will produce a reflections and changes in the received 
signal.  But, if you are using ordinary pecl, 1-2 ns rise times, what you 
want to do is easy. With high speed pecl, 200-300 ps rise times, you 
will  have to be more careful, but extreme measures are usually not 
required. Discontinuities in the controlled impedance signal path that are 
short compared to the signal rise time will have very small effects on the 

First, keep the pull down resistors on the pecl output close to the driving 
chip.  Do not place a Thevenin equivalent termination on the other board 
(you don't want dc power supply currents between the two boards).  The best 
is a simple parallel termination (no connection to ground) at the receiving 

If you can't start the twisted pair at the signal source (close to the 
chip) then run two 50 ohm controlled impedance lines to where the twisted 
pair will start. For ordinary fr4, a line (on an outside layer) about twice 
as wide as the layer spacing (outside layer to next layer down, which 
should be either a ground (best) or power plane, not a  wiring layer) will 
give you about 50 ohms.  The two lines spaced 3-4 line widths apart 
(uniform spacing) will give you an approximately 100 ohm differential line. 
Connect to the next board with a short piece of twisted pair. Typical 
twisted pair has an impedance about 100-120 ohms. One inch is only about 
130 ps delay.

There must be a ground connection between the boards, but this does not 
need to be extremely close to the twisted pair signal path.  You can also 
use two pieces of 50 ohm coax, which will connect the grounds. Unless you 
are very sensitive to emi, twisted pair will be fine.

Continue on the next board with controlled impedance if necessary. 
Terminate with a 100 ohm resistor between the 2 lines of the differential 
pair. This should be very close (or a little beyond) the receiver.  Do not 
terminate with resistors to ground or to a power supply.

If you can launch the twisted pair close to the driver, and land close to 
the receiver, you can dispense with the controlled impedance lines on the 
pcb, and use a longer piece of twisted pair. This will improve the received 
signal, by eliminating the discontinuities in the total path (where the 
path changes between pcb and twisted pair). You can transmit differential 
pecl (or ecl) over several feet of twisted pair with very little signal 

I hope this is clear!  (and helpful)


>A couple of questions.
>Does anyone have a good way of doing board  to board connection (only 
>about an inch) for a differential 50 ohm connection?
>Can I make two parallel connections with 100 ohm twisted pair?
>We have done some calcs for a FR-4, 1 oz copper, double sided, 1.6mm 
>thick, and we get traces 140thou, 8thou spacing over a gound plane for a 
>50 ohm differential impedance. Can anyone confirm our result? 140 thou is 
>getting up there!
>The reason for the questions is that we need to take a differential PECL 
>signal from one board to another and maintain the signal integrity.
>Scott Ellis
>Novatex Research - Excellence in Electronic Research & Development
>41 Yule Road, Merewether, Newcastle, NSW 2291, Australia
>Ph 0412 988408   Fax 02 49636058

Cheesecote Mountain CAMAC
24 Halley Drive; Pomona, NY 10970
845 364 0211,

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* To post a message: mailto:[EMAIL PROTECTED]
* To leave this list visit:
* Contact the list manager:
* Forum Guidelines Rules:
* Browse or Search previous postings:
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Reply via email to