From "The ISDN Studio" by Dave Immer
Audio Engineering Society 99th Convention
Oct. 8, 1995 , New York City
"CASCADING" Using Frequency-Domain Coding
This applies to the MPEG and Dolby AC-2 algorithms. If you think
about the digital audio bit-reduction process, you will begin to appreciate
the hazards and penalties of multiple encode/decode cycles. It can be more severe
than the problems of multiple A/D - D/A conversions. At the lower bit-rates
of 112/128 Kbps the risk of audio artifacts (noise) accumulating due to sending
a stereo waveform through multiple CODEC systems is very high. For this reason
it's best to keep cascading to a minimum and avoid doing things like boosting
mid & hi-end EQ at the receiving end. But at data bandwidths of 256 Kbps
or higher, you can get away with more cascading and post processing of the received
stereo audio before noise is perceived.
There is a larger issue here, however, and it deserves careful attention. As
the use of these bit-reduction algorithms increases throughout all steps of
the audio (and video) production and broadcasting process, the odds of multiple
encode/decode cycles also increases. Plus there is a greater chance of "transcoding"
from one algorithm to another. To give you an idea of this, here are some of
the audio coding applications now, or soon to be, in use.
1. Talent import. This includes voiceovers and multi tracking.
2. Sound Effects, Sample distribution/delivery.
3. Sub mix delivery for editing/post-processing.
4. Final Mix delivery.
5. Surround-Sound systems for movies, laser discs and HDTV.
6. Satellite distribution of Radio and TV programming.
7. Radio STL (studio - transmitter link). In use now by many radio stations.
8. Digital Radio Broadcast. Available in certain test markets now.
9. MiniDisk Recorders. 384kbps or higher.
10. Storage and editing of digital audio on personal computers.
It is quite conceivable that a piece of audio could pass through five or six
(or more!) bit-reduction cycles before it arrives at the intended consumers
ears. If most of the CODEC cycles were at the relatively low bit rates of 112/128
Kbps stereo, the resulting waveform would quickly accumulate noticeable noise
& artifacts. Generally speaking, the way to avoid this waveform deterioration
is to use the CODECs at the highest bit-rates possible. For example, extensive
testing of the MPEG1 layer 2 algorithm by ISO in the early 90's in Stockholm
revealed that a stereo program could withstand 1, perhaps 2, CODEC cycles at
128 Kbps without noticeable artifacts appearing. But at 256 Kbps, listeners
didn't hear any deterioration until about 4-5 pass-throughs; and at 384 Kbps,
material could withstand 10 or more cycles depending on the waveform complexity.
(By contrast, a time-domain algorithm can withstand more CODEC cycles at 384
Kbps before the waveform noticeably deteriorates.) The point here is: Use frequency-domain
CODECs carefully. Consider transferring unreduced audio files (SoundDesigner
II, etc.) for critical mastering.
"Bridging" or "Hybrid Sessions"
Due the existence of at least 4 popular incompatible algorithms,
DIGIFON, EDnet and GlobeCast offer bridging services whereby incompatible users
- either independents or competing network subscribers - can both call in to
a central station which has both systems on hand, and be simply "patched
through" from one to the other. This is a less than elegant solution to
system incompatibility, since it requires the waveform to be decoded and "re-coded",
thereby adding another layer of bit-reduction to the transmission. (See "CASCADING".)
At low bit rates this can be very detrimental to the audio, but if done with
care, bridging will yield excellent results. Hybrid session fees add to the
cost a transmission. But for all users it greatly expands the number of CODEC
systems you can connect to, opening up new markets and talent pools as well
as allowing for greater flexibility in recording touring talent and in program