Re: [SI-LIST] : PCB track length and ringing

Larry Smith ([email protected])
Wed, 24 Sep 1997 09:04:08 -0700

We are missing big picture here...

The ultimate mismatch for a transmission line is an open circuit. But,
this is exactly what we have for many (most?) of our transmission lines
on PCBs. The receiver or receiver/driver on the far end of most
address or data lines is a capacitor (3 to 10 pF). A 50 ohm transmission
line into that load will have a rise time of less than 50 ohms x 10 pF =
500 pSec (about the rise time of our signal). In other words, it is an
open circuit and the transmitted voltage is going to double. This causes
far more reflection than any via, changing planes, stubs, etc.

The ringing commonly seen on unterminated transmission lines and
reflected noise are 'one and the same'. Measure the period of the
ringing as seen on the lab scope and you will find that it is at the
frequency associated with the quarter wave length of the transmission
line. This is most apparent in an unterminated, point-to-point trace.
The driven end of the line is commonly tied to AC ground with a low
impedance source (5 to 25 ohms). This closely approximates the
quarter wavelength resonator stub (shorted on one end, open on the other)
that we learned about in transmission line or antennas class so many
years ago. The ringing is the resonance of the stub.

If it were not for source resistance, there would be very little
damping, and the system would ring for a long time (more than 100 nSec).
PCI drivers are great for this because they are strong (close to 5 Ohms
output impedance). A series resistor at the source that is about 2/3 of
transmission line Z0 cleans up the ringing very nicely. (A good target
is to have the driver Ro + Rseries = 2/3 Z0. A slightly under damped
system makes crisper edges.) A wavefront of slightly more than half
the power supply voltage proceeds down the transmission line and doubles
at the open circuit far end, giving just the voltage you wanted to see
at the receiver. The reflected wavefront is almost completely absorbed
in the series resistor and driver. Sophisticated drivers will attempt
to match the transmission line impedance so that no external components
are necessary.

Daisy chain nets behave similarly. Clustered loads are somewhat the
same. Star topologies introduce multiple Eigen values, so there
will be multiple resonant frequencies on the net, associated with the
several lengths of the star stubs.

Larry Smith
Sun Microsystems.

> Simply due to the reflections caused by impedance mismatch somewhere
> in the trace between source and destination. Mismatch can occur for a
> multitude of reasons such as vias, changing planes, stubs, tees,
> capacitive inputs, inductive traces etc., etc. That is based on
> transmission line theory found in elementary electromagnetics
> textbooks.
> Hans
> ______________________________ Reply Separator _________________________________
> Subject: [SI-LIST] : PCB track length and ringing
> Author: Non-HP-gkrishna ([email protected]) at HP-ColSprings,mimegw5
> Date: 9/22/97 8:32 PM
> Hi,
> Can someone explain how the PCB (printed circuit board) track (signal
> trace) length causes ringing in the High Speed Digital Signal systems.
> Regards,
> Gopalkrishna