tggzzz wrote: ↑Sun Mar 10, 2024 9:23 am
mansaxel wrote: ↑Sun Mar 10, 2024 6:37 am
Given the right hardware, and a lot of legwork, we can lock two clocks into fractions of a µs over long distances. Which is a requirement we actually have.
Locking clocks like that I can understand.
But what might it mean w.r.t. agreeing the time? (.... refers to Leslie Lamport's partial ordering)
Basically, we have the requirement to be able to reconstruct -- at a later time -- the exact relation in time between two events, so as to be able to compensate for it.
I'll use audio streams for the example, because I know the intricate parts better on that side of the media business. We have, in a venue, a couple of microphones, that both are connected to A/D converters, producing an AES3 sample stream each. This stream is chunked into time slots of one ms, so 48 samples go in a slot, which then goes into a RTP packet, which is then sent over a network using IP multicast. Now, the two microphones can have different converter hosts, in effect entering the network from two different network nodes. But since it is distinctly possible that they both will hear the same sound (being in the same venue) the phase relationship that was present at their diaphragms needs to be preserved. This is achieved by time-stamping the RTP packets using the host clock. Phase being something we humans are very sensitive to, synchronisation of said clock is critical. The relevant standards require 1µs, which is sufficient. At the receiving end, the device responsible for observing the phase relationship of these two signals (the audio mixer, typically) will then time-shift the reconstructed bit streams into sync again, using the time stamps as guide. (Sadly, the digital earlierfication device has not yet been invented; which means that the first arrived stream will have to wait like the rabbit for the tortoise.)
This is achieved by having a PTP clocking system present on the network; and locking all sources and sinks to this clock. Consequently, we do care that a clock in Eugene, Oregon (at some sports event) is locked to our GM in Stockholm. With some care and elbow grease, good transmission systems can be predictable enough to be able to carry the clock across. For many applications, though, we are satisfied with being able to achieve a frequency synchronisation. This is true when we have different venues that need to transmit media between them, as long as all devices in each venue are in phase locally.
