 |
Networked Audio Transport
Looking at the methods and factors
 Pick your audio network transport highway: good ol’ CAT5 and up-and-coming fiber optic. |
In the (still) emerging world of networked audio transport, there
are two major categories in which a system may fall: a fully standards-based
network, or a proprietary network that may or may not use standards-based
transport.
These two different methods for routing digitized audio around a venue.
Both have their advantages and disadvantages and, of course, are subject
in varying degrees to the problems associated with transforming an analog
signal into a digital stream of ones and zeros and then back again.
Let’s first explore the biggest question that a system designer
should ask when someone shows them a new piece of digital gear: “What
is the latency as it relates to audio being transported in large networks?”
As you’ve already learned if you’ve read the related articles
earlier in this issue of Live Sound, latency, or what some manufacturers
call “propagation delay,” is the amount of time that an
audio signal is delayed due to digital processes including analog to
digital conversions (A/D), digital to analog conversions (D/A), and
digital signal processing (DSP).
For live sound applications, excessive amounts of latency can wreak
havoc on the audience as well as the performers by creating listener
fatigue and poorly reconstructed audio.
Generally speaking, the extent to which latency will cause a problem
is a function of the ratio between the direct sound and the sound that
is delayed. In a large-scale sound system, where there is little to
no direct sound, the delay does not become a problem until there is
a noticeable delay between sight and sound.
However, when the direct sound is within 8 dB or so of the delayed signal,
20 milliseconds (ms) to 30 ms difference will be audible. As for performers
on stage, the acceptable time is generally shorter, especially when
it comes to their monitors.
Echo from a system - whether it is acoustic or electronic - contributes
to performer and audience fatigue. A performer using an in-ear personal
monitoring system will be conscious of any latency in the system to
as little as 5 ms to 10 ms. Small latencies on the order of 100 microseconds
(µs) can cause phasing problems that can result in high frequency
roll-off when they are added back into the mix with an un-delayed source.
A COMMON CLOCK
It should be noted that all digital network components should be synched
to a common clock. The best way to achieve this is with the use of a
separate word clock. This will allow all devices in the chain to be
locked to each other and thus eliminate any phase shifts that may occur
between devices from un-synched sampling rates.
With advances in digital technology, latencies for audio gear have
dropped dramatically, to the point that some digital mixers are below
3 ms for analog in to analog out with no added time for DSP. The problem
lies in cascading multiple digital devices together if not digitally
linked. When having to make A/D and D/A conversions for every piece
of digital gear in a signal chain, it’s easy to see how the latencies
from conversion alone can add up to an unacceptable total.
Two key factors in comparing digital audio networks are the sampling
rate and the bit resolution that the network supports. The sampling
rate is the ”rate” at which a digital device ”samples”
the composite analog sound waveform over time.
The sampling rate is important for the way digital audio can describe
the frequencies in a sound. Bit depth, or resolution, describes the
potential accuracy of a particular piece of hardware or software that
processes audio data. In general, the more bits that are available,
the more accurate the resulting output from the data being processed.
For example, audio recorded with a 48 kHz sampling rate and 24 bits
of resolution will have 48,000 measurements of which there are 16.7
million different values that each measurement can be per second. Sampling
rates and bit resolutions are important to digital audio networks because
they need to remain consistent to keep the latency to a minimum. A sampling
rate conversion typically takes the same amount of time as an A/D conversion.
The major limitation on the number of channels a network can support
is bandwidth, which is the amount of information that can be sent down
the chain at one time. The bandwidth required to pass one channel of
audio varies with the sampling rate and the bit depth selected. As both
increase, so too does the bandwidth demand reducing the maximum number
of channels which can be transported over a particular network’s
architecture.
Well, what if you could have just one A/D conversion at the start of
the signal chain and one D/A at the end with all the other gear still
in the chain? Now we’re talking about the digital audio networks.
Again, there are two groups of these digital audio networks, so let’s
take a closer look at both.
PUBLISHED RULES
A fully standards-based network abides by a published set of rules,
which specifies a recommended interface for serial digital transmissions.
This includes the description of data format for transport and the method
of transport.
The best part of a standards-based audio network is that any manufacturer
can implement the format into their products without having to pay the
often-expensive licensing fees of a proprietary solution. This allows
the use of “manufacturer A’s” reverb unit with “manufacturer
B’s” digital console without the conversion to analog, resulting
in a total system latency that is much lower than if a D/A and A/D conversion
was necessary.
 A look at digital transport for Eric Clapton on tour. Spectrum Sound implemented Soundweb DSP and networking at FOH, with audio routed via CAT5 to QSC RAVE A/D converters with CobraNet technology. Then it’s a quick flight for the signal on fiber optic to RAVE D/A converters at the amplifier racks, with another short , CAT5 hop to the system’s four-channel CyberLogic amps. |
The most common standard is AES3, more often called the AES/EBU (Audio
Engineering Society/European Broadcasting Union) standard. AES3 uses
a 110-ohm shielded twisted pair cable to send two channels up to 300
feet. The standard allows up to a 24-bit resolution with no maximum
sample rate.
Another standard developed by the AES is AES10, more commonly called
MADI (Multichannel Audio Digital Interface). MADI offers 64 channels
at a 48 kHz sample rate and 32 channels at a 96 kHz sample rate with
a resolution of up to 24 bits per channel. Transmission over a single
75-ohm coaxial cable has a limitation of 150 feet, but the use of fiber-optic
cables can extend the length to two miles.
Other multichannel standards include ADAT Optical (ADI), Sony/Phillips
Digital Interface (S/PDIF), and Tascam Digital Interface (TDIF). See
Table 1.
The only latency added into standards-based digital audio distribution
is the act of making the A/D and D/A conversions and any subsequent
DSP that occurs. This is unlike some proprietary solutions that also
require additional time to transcode and transport the data.
VIABLE & AFFORDABLE
The two most common proprietary systems, which use a standards-based
transport, are CobraNet from Peak Audio, and EtherSound from Digigram.
These systems transport audio data using standard IEEE 802.3 Ethernet
protocols. Ethernet is used because it’s relatively inexpensive,
reliable, and the technology has been, and will continue to be, upgraded
by the computer industry.
Just a few years ago, 10baseT was the best thing going. Now, gigabit
(100baseT) Ethernet has become viable and affordable. In order to integrate
either of these technologies, a manufacturer must license CobraNet or
EtherSound technologies from the developer in order to use them in their
product(s).
CobraNet is capable of up to 64 bi-directional channels at a 48 kHz
sample rate with a resolution of up to 20 bits per channel over a single
100Mbit link. This could mean that all you have to pull out to the front
of house is a single CAT5 cable.
However, CobraNet comes with a healthy dose of latency, 5.33 ms just
to transport the signal from one CobraNet device to another. The reason
for this delay is due to the 256 samples required to buffer the audio
data into Ethernet packets. (The latency due to A/D and D/A conversions
must still be calculated into the equation to get to total delay.)
 Table 1: A point of format comparison. |
CobraNet is still in its growing stages, and is beginning to be implemented
by more manufacturers into different product types including amplifiers,
speakers and DSP boxes. However, currently no mixing consoles, digital
or otherwise, offer integrated CobraNet. Therefore, a run out to the
front of house mix position is going to require two networks, adding
more than 10 ms of latency before even thinking about any other digital
processing.
This is one example of why cascading digital equipment can lead to unacceptable
latency totals. However, note that consoles with integrated CobraNet
may be in the not-too-distant future with both Midas’ parent company
Telex, as well as Mackie, becoming recent licensees of the technology.
OFF THE SHELF
EtherSound is different from CobraNet in that audio only flows in one
direction. It offers 64 channels of 48 kHz sample rate at 24-bit resolution.
The same off-the-shelf Ethernet switches and cabling can be used for
distribution, but any audio that is added into the signal chain is only
available to be peeled back off down stream of where it was added.
At the 2003 New York AES convention, Digigram announced that it will
be offering a bi-directional EtherSound architecture. Nevertheless,
there is still a trade-off. The system can only operate using the daisy-chained
“ring” network topology or in direct transmitter to receiver
applications. The latency in an EtherSound network is six samples, with
.00122 ms added for each EtherSound device in the chain. EtherSound’s
licensees include InnovaSON, Fostex, and Nexo.
Another lesser-known network is the Media-accelerated Global Information
Carrier (MaGIC) developed by Gibson Labs, the technology division of
Gibson Guitar Corp. MaGIC grew from the development of the digital electric
guitar. Like CobraNet and EtherSound, MaGIC conforms to IEEE 802.3 physical
layer and offers up to 32 bi-directional 32-bit channels with up to
a 192 kHz sample rate.
The problem with Ethernet is that it is a non-deterministic system,
meaning that the data will arrive when it feels like it. CobraNet and
EtherSound have developed systems that make the arrival times very predictable
and allow for the network to be synchronized with only a small margin
of error.
For instance, CobraNet’s master unit regularly broadcasts beat
packets onto the network either from its internal clock or an external
master clock. Other devices on the network lock onto the arrival time
of this packet and regenerate the clock locally. The error in clock
delivery is ±1/4 sample period; this translates to about .005
ms at a 48 kHz sampling rate.
While digital seems to be the future of audio networks, it still requires
attention to detail and proper setup. Monitoring overall system latency
and keeping a consistent sample rate and bit resolution are both new
requirements for the digital age. The answer to the question of what
system is best for your situation is still the one that best meets your
needs, which might be good old-fashioned analog.
Digital is just another hammer in the sound designer’s tool belt
which can either drive the nail home if used correctly or smash a thumb
if its requirements are ignored.
David A.McNell holds a Bachelor of Science degree from Purdue University
and is an audio-visual engineer in the Special Technologies Group of
Newcomb Newcomb and Boyd, a multidiscipline firm based in Atlanta.
