User Manual

User's Guide Digiface AVB © RME
61
20.2 AVB Network Latency
All devices in an AVB network share the same time. This allows the sending device (talker) to
specify the precise point of time when its audio samples should be played out at the receiver
side (listener). This is achieved by adding an offset to the current time and sending the resulting
timestamp with each sample transmitted. The timestamp is called "presentation time" and has
nanosecond precision. For comparison, a single sample at 48 kHz has a duration of over 20800
ns.
The receiver compares the incoming presentation time of each sample to the current time and
buffers the sample until the presentation time is has come. The offset (maximum transit time) is
specified by the AVB standard as 2 ms for class A traffic, which is enough time for the signal to
pass through a very large network under full load with up to seven 100 MBit/s switches along
the way. By default, most AVB products will use this offset. In smaller networks with less hops
or 1 GBit/s link speed, the offset can be adjusted to lower values, such as 0.3 ms, 0.6 ms or 1
ms. In the event that the chosen offset is too low, the listener detects that the requested presen-
tation time has already passed and the audio data is discarded.
The Digiface AVB acts both as a talker with an adjustable offset of 2 ms down to 0.3 ms, and as
a listener - where the latency depends on the talker.
In AVB networks, the latency is always specified by the talker and guaranteed by the listener -
with nanosecond accuracy. This behavior is plug & play and does not require any user interac-
tion or monitoring.
Due to the large number of possible listeners and talkers in a network, it is usually not possible
under AVB to speak of typical network latency. Rather, depending on the configuration of the
individual devices, their current presentation time, and the usually not included AD/DA
conversion delay, a large number of possible latencies occur, depending on which device
communicates with which one and where the data is fed in and out.
Due to the specification of a maximum presentation time of 2 ms, however, including AD/DA
conversion, a maximum value of 5 ms can be assumed. With a setting of 0.625 ms and current
low latency converters 1.5 ms can be reached, for the complete path Analog In -> Network ->
Analog Out.