Consoles 979Where this approach really shines is main mix bus
monitoring—one can mess about with the monitoring
bus as much as one likes without affecting the main mix
at all—nondestructive soloing becomes a reality, simply
implemented at that.
Talkback is simply treated as one of the sources to
the mixer; it can get routed into any of the mix outputs
with no necessity of creating a separate subsystem, with
IFB (Interruptible Foldback) talkover ducking or
muting deriving from modified coefficients.
Note there is no distinction made between the group
outputs as to what their ultimate purpose will be, group,
aux, cleanfeed, etc. All that distinction is done at the
control surface and the interpretation of its requirements
by the host microcontroller—in other words the differ-
ences are all in the controlling software and not in the
hardware that implements the mixes.
25.23.2.6 Coefficient Compounding
This is a rather fearsome title for a rather nice concept.
This is how master fadering and group fadering are
achieved. Rather than have a separate downstream gain
stage after a mix has been achieved to effect overall
level control, a convenient approach with a soft matrix
mixer such as has been described here is to take the
sensed level of the real, physical, group fader and then
multiply each of the coefficients feeding that particular
bus by its value. This is a direct analogy of VCA
grouping, where one fader actually modifies the level
contribution of each source to the mix bus, rather than
gain changing the mix after the fact. Since all of these
numbers (source contribution coefficients and group
fader) exist in the host microcontroller, the arithmetic
manipulation is quite straightforward. The database
management aspect of this on a large console can get
quite interesting, but this pseudo-VCA grouping
approach is widespread and very powerful.
25.23.2.7 Coefficient Slewing
Rapidly altering coefficient data in a DSP runs into
exactly the same tone click problem as do MDACs in
analog; even small transitions made when the audio data
sample is nonzero stand a very good chance of being
heard as a click. A fader swipe can generate the
all-famous zipper noise, and just as with MDACs,
without care and attention the effect in EQs is little short
of comical.
Sensing zero-crosses in digital is practically impos-
sible, since particularly at high levels of high frequen-
cies there may well not be any samples anywhere near
zero—remember this is not a continuum like analog, the
samples are just a regular set of stabs in the dark. A
wide enough window to capture enough zero-crossings
would probably be wide enough to still allow some
transitions to be audible. Never mind the fact that the
processing overhead for doing a window compare and
decision on each and every coefficient would be over-
whelming; it would probably cut the potential number
of crosspoints in a mix stage down by an order.
A good solution is to allow the DSP to ramp rela-
tively slowly between its present value and the new
desired value, creating its own interpolating steps on a
sample-by-sample basis small enough that each is inau-
dible. (This, by the way, is one of the necessary
processing elements that eats up a chunk of mix-DSP
cycles, limiting the maximum number of crosspoints
available to significantly less than the raw cycles avail-
ability of the device would suggest.) A slightly different
approach is to “pre-slew” the coefficients in an interme-
diate processor (often also a DSP) to offload the effort
from both the host and the target DSPs. The inter-DSP
communications can start to get a bit fierce, however.
It is a nerve-wracking moment when first trying
on-DSP slewing. After all, the coefficients for IIR filters
such as in EQs can be very, very touchy and have little
tolerance for error before doing very odd things most
unlike the filters they were intended to be. Amazingly
though, it seems as though provided the filter set is
stable where it starts, and stable where it ends up, it
stays stable in between as the coefficients are slewed; it
might get just a little wonky, but not enough to cause
any serious sonic issues and certainly not enough to
explode into what has been charmingly called
“screeching cats from hell” (DSP audio guys and gals
hear lots of them).25.23.2.8 ClockingA major subsystem within a digital mixer is
clocking—making sure that each of the various circuit
elements get the necessary hard, clean clocks required
to operate properly. In this design alone there are six
clocks for processing: 12.288 MHz master clock (actu-
ally divided down from 24.596 MHz to ensure
symmetry), 6.144 MHz used as a master clock by
AES/EBU transmitters, 3.072 MHz as the main serial
bit clock for the standardized native serial data format,
an inverse of that used by some A/D or D/A converters
of less serial format flexibility than others, then 48 kHz,
which of course is the data sample rate and houses
left/right clock.