992 Chapter 25
countless varieties, all proprietary and utterly incompat-
ible, of course, since there is a strong commercial
impulse to keep everything in house. A standard,
AES-50, attempts to bring sanity and compatibility into
the gigahertz realm, with the myriad 100 MHz schemes
already considered a lost cause.
At its simplest, a big packet consisting of a header
and then however many audio words of whatever width
is constructed, then sent down a dedicated Ethernet
hardware circuit, every sample period; at the receive
end the simple format is readily decoded and the
constituent samples recovered. It is a reasonable
assumption that on a dedicated line the packet will be
unimpeded and arrive intact, and such links typically
run raw with no mechanism for error trapping. Of
course, it is entirely possible to build in error detection
and correction in case bits get hurt somewhere. It would
have to be a fairly weak link for this to get exercised
much, and such mechanisms raise the bugaboo of
TCP/IP systems—building out a fixed latency to allow
for the randomness of the errors. In short, these systems
run just fine without and typically do.
Audio is only part of the whole picture. Metadata
accompanying it, logic switching contingent on control,
control data and metering data all have to be considered
and accommodated within the link for it to be a fully
usable system in any meaningful context.
Such UDP links are typically bidirectional (but
sometimes unidirectional) end-to-end closed links. In
and of themselves, they don’t constitute a network,
which can loosely be described as anything to
anywhere—any source connected to the network may
be picked up by any destination. There are two general
schemes for turning these one-to-one links into
networks: cascading them node-to-node, with a modi-
fied signal passing along each link (Serial), or arranging
them all to radiate from a central hub (Star).
25.25.3.3.1 Serial or Loop Networking.
In this methodology a single unidirectional line is run
passing through each area that needs access to the
network; access is achieved by nodes or breakout boxes
of varying complexity depending on the requirement,
each of which has a unique address for programmability
purposes. At the simplest, a small fixed number of
inputs to the network and outputs from the network may
be offered at the node, along with unique control data.
These may be analog ins/outs or digital ins/outs or a
combination, and each may either look at (in the case of
outputs) any of the (say) 64 program slots, or select a
slot into which to place their audio (in the case of
inputs).
More advanced nodes could, by way of example,
look at many slots, mix them, mix that with local input
material, and even place the composite mix into a slot(s)
in the network. A frequent application is to retrieve
audio from a slot and replace it with local input mate-
rial. Signal processing specific to a local need (e.g.,
crossover/EQ for a speaker cluster) can be done within
such a node; indeed, such products can be thought of
primarily as a processor that just happens to have wide
connectivity through the network and is marketed as
such. Third-party or multiple vendors can be interoper-
able, providing they’re all licensees of the same
networking protocol.
This modified stream is then sent downstream to the
next node, and so forth. The stream can be unidirec-
tional (serial) or looped back upon itself (surprise,
loop), whereupon the originating node seemingly
perversely sees as its input the stream after it has passed
through all the other nodes.
Such networks tend to be quite efficient, since they
are able to reuse slots along the way. Disadvantages are:- That serial anythings tend to be badly affected by
single-point failures—in other words, one node
failing makes orphans of all the others downstream of
it, bisecting the net. - It takes an appreciable time to receive the packet of
slots, disassemble it, modify slots to whatever degree,
reassemble it, and send it on its way, and such
processing latency is obviously cumulative with that
from previous and successive nodes. That said, laten-
cies can be low, in the handful-of-sample-period
range, trivial compared to those of 100 MHz TCP/IP
systems. - The network cabling routing has to be carefully
thought through and follow a logical progression of
where the audio needs to go next. This sometimes
isn’t easy.
25.25.3.3.2 Star Network Topology.In contrast to relying on the packet addressability of an
IP-style Ethernet network—which belies the need for
centralized command and control—UDP-style networks
which dumbly if faithfully and with low fixed latency
propel a fixed amount of audio from one end to the
other of a straight pipe, requires a central switch: this
unpacks each incoming stream, decides which elements
within them need to go where, and assembles outgoing
streams appropriately. Such routing systems were long