BECKHOFF ET1100 - Manual (Page 16)

General Issues
12 Slave Controller – Application Note FAQ
3.15 Distributed Clocks: Resolution, Precision, Accuracy
The Distributed Clocks system is operating internally with a 100 MHz clock source, i.e. 10 ns cycle
time. Nevertheless, most DC values are represented by multiples of 1 ns (only the pulse length of DC
SyncSignals is counted in multiples of 10 ns).
Question: How good is DC, 10 ns or 1 ns?
To answer this question, different terms have to be considered. The short answer is that the resolution
of most values is 1 ns, but the precision is basically 10 ns. By averaging, the accuracy can be
increased so that it can come close the 1 ns resolution. This is a more detailed answer:
Resolution
The resolution is the smallest value you can read or set in a DC register. The resolution of the DC
values is 1 ns, only the pulse length of DC SyncSignals has a resolution of 10 ns.
Precision
The precision is somehow the “quality” of a single event or measurement, i.e. the deviation between
actual time and the ideal time. The precision of the DCs is mainly caused by jitter, e.g. due to the 10
ns / 100 MHz operating clock of the DCs.
Accuracy
The accuracy is like the “average of the precision”, i.e. the average deviation between a couple of
measurements and the ideal value. The accuracy of most DC values gets better and better the more
measurements are made. This is a statistical effect, with better accuracy for an increased number of
measurements.
Ideal reference
For most DC functions, the precision and jitter reference is the ideal and continuous time value of the
reference clock, the global system time.
The system time value read from the reference clock registers is already subject to a 10 ns jitter
because of the 100 MHz operating clock for DC. So the precision of the system time read from the
reference clock is 10 ns, the accuracy of the system time at the reference clock is 0 ns.
Synchronization
This system time value is spread over the network into the slaves, which requires knowledge of the
propagation delays. These delays are calculated from the DC receive times. The precision of the
receive times at the ports and the core depends on the physical layer. For EBUS, the precision is 10
ns, for MII it is 40 ns (because of the 25 MHz MII receive clock). By averaging the calculated delays,
their accuracy can get close to the 1 ns resolution.
The system time of a synchronized slave is generated by averaging system time differences. The
accuracy of a synchronized slav
e’s system time depends on the accuracy of the delay measurement
(down to 1 ns), but it can only be read with a precision of 10 ns due to the 100 MHz clock source. This
is nearly the same for all synchronized DC slaves.
SyncSignal start time and LatchSignal time stamps
The SyncSignal generation and the LatchSignal time stamps depend on the local representation of the
system time. The start time of a SyncSignal is subject to a jitter of 10 ns. This adds to the jitter of the
local system time, so the total precision of the SyncSignal start time is about 20 ns. Averaging many
SyncSignal start times will lead to accuracy near the resolution limit. The precision of LatchSignal time
stamps is also about 20 ns, because of the input synchronization stage with 10 ns jitter and the local
system time with another 10 ns jitter. The accuracy of multiple latch events gets down to the resolution
limit.
NOTE: The precision/jitter and accuracy values are based on the algorithms; real hardware and real clock
sources further increase the jitter by a small fraction, and inaccuracies can sum up.