Clock Distribution

[BenM]

I can sort of understand how a clock is distributed across the circuits in a CPU and how the separate injection points can be made to run in synch however on the scale of a chip the distances and therefore propagation times are tiny.

On a machine as big as the LHC complex, just how do they make sure the clock works properly? every accelerating magnet must push the beam at exactly the right time within a nanosecond or two and when you are talking about a device 23km around propagation delays would preclude a single clock injection point. even if you have multiple clock sources they all have to be synchronised...and you have the same problem the distances introduce significant delays on the scale of the machine as a whole.

My colleague who used to work at CERN (indeed some of his displays, developed when he worked on the LEP, are still used in the control room today and can be seen in the portal ) tried to explain how the clock worked back in those days but I couldn't quite grasp how it actually worked.

so two questions...
1. Have I explained my puzzle clearly?
2. Can anyone come up with an easy to understand description of how the machine keeps itself in synch?



B.

[tomey36]

http://atlas-service-enews.web.cern.ch/ ... clocks.php

I rembered reading somthing about that a while ago so i found the link for you. I hope this answers your question.

[BenM]

Thanks for that link... it goes some way towards sorting it out in my head...

I guess what I am looking for is how the trigger processors get the clock signal to the experiements before the beam arrives...

thinking aloud - if the clock signal is just used to discipline the experiment clocks propagation isn't a problem because the trigger will be used by the experiment equipment in a manner somewhat similar to the ntp protocol so it won't matter when the trigger arrives (for given values of when), just as long as it does and that, in a round about fashion, does answer the question

B.

[pcatom]

For the machine there are 2 timing systems, slow and fast ... the terms are relative

The slow system allows synchronization of events at the millisecond level. Here the Signals are broadcast - with tof delays built in (but not really so significant), the whole system is kept in synchronization with GPS receivers.

The fast timing is in the range of a fraction of a nanosecond. This uses the distribution of the RF frequency signals (and derivations from it). Here the tof is important and taken into account in all signals. In addition the effect of temperature and aging in the fibre optic links is also compensated for.

[BenM]

Thanks pcatom... millisecond timing I can cope with - after all ntp manages that just fine so it must be a relativley simple problem

My brain melts just thinking about how to account for propagation losses at a sub nanosecond level; I am glad there are many more clever people than I on the problem

B.