Proposal of Mechanism for Source Clock Recovery through Asynchronous Network
Original Publication Date: 1994-Jul-01
Included in the Prior Art Database: 2005-Mar-27
A real-time communication (voice, video, multi-media...) between 2 points of a network needs a synchronization between them (usually at 8 kHz).
Proposal of Mechanism for Source Clock Recovery through
communication (voice, video, multi-media...)
between 2 points of a network needs a synchronization between them
(usually at 8 kHz).
the network itself is not synchronized because it
is not required for packet transmission, a method to achieve the
clock recovery, known as "adaptive clock recovery", is proposed in
the Standard. Based upon the monitoring of a play-out buffer, this
method has been criticized because it was not supposed to be able to
filter the low frequency jitter (wander) which may be introduced by a
a system which is able to recover a source clock
in such a way that the wander can be filtered out.
filter for very low frequency means long time
constants, which are incompatible with short delay required on
real-time data connections. For that reason, the proposal is based
upon the idea of dissociating the clock transport and recovery from
the communication itself.
Two options are possible :
1. Use a dedicated
connection for the clock transport. This
brings the following advantages :
o The initial setting up of the output clock
frequency can be
as long as needed for the required quality, without drawback.
o Only one connection per clock is needed, and
can be used by
all the data connections which need to be synchronized with
the same source clock.
o The clock monitoring is going on, whatever
can be the dynamic
re-allocations of the data connections.
2. Use a data
This alternate option is to use a data connection (it is obvious
that this connection is a real-time connection like AAL1, which
grants that packets are sent from the source at a constant rate),
but without using its own play-out buffer, taking the packets at
the input of this play-out buffer, that is to say, as soon as
they arrive in the node.
option does not change anything to the mechanism
described in this proposal, but with the drawback that there is
no more clock monitoring when the connection is off.
At the source
node, it is assumed that packets are generated at
a constant rate related to the source clock, with a nominal period
known by the destination node. A short period would make possible to
have a better precision or a shorter setting time of the output
clock, but would require a higher bandwidth.
are ATM cells, a period of 6 ms would require about
64 kbps bandwidth. Variable length packets would allow to carry the
same clock with a much smaller bandwidth requirement.
It is assumed
that there is no packet loss between the source
and the system input. That means that in case of packet loss in the
network, one supposes tha...