Gain Staging on the K2000


First I will discuss the basic voice gain structure, then I will
discuss the "output" gain setting.

The gain path for one voice is primarily controlled by the
algorithm for that voice's layer. The order is pretty well made
clear by the diagram, but a little explanation is needed.

First of all, the samples in ROM or RAM have at least 6dB of
headroom built in due to the way Hobbes processes samples. If a
sample is not normalized, there will be more than 6dB of
headroom. Typically in our programs this is all eaten up by the
adjust setting on the F4-AMP page, which is often set to be +6dB
or more.

But if we assume that we are using Algorithm 1 and the F4-AMP
page is set to all zero, and say we select the parametric EQ
function with no pad, then we can always boost the EQ gain
(F3-AMT) at least 6dB before that voice will exhibit clipping.
Padding the input to the EQ on the F1-FRQ page will give even
more... 18dB pad will give a total of 24dB headroom available for
boosting the EQ. Of course different frequency ranges of
different sounds have more or less energy, so with EQ you
actually can often apply more dB's of boost than you have
headroom (the frequency range you are boosting may be many dB
below the overall level of the sample) but if you are boosting
near the fundamental it will work a lot like simple gain in terms
of headroom.

Due to the ordering, if you boost the parametric EQ until it's
clipping, no amount of turning it down at the F4-AMP page will
correct it. Each box in the Alg that applies gain must be aware
of/provide it's own headroom.

Hobbes generated waveforms do not have any headroom built in,
they are full scale, except that the "saw+" and others that mix
with the sample have 6dB headroom built in (so it won't clip when
it adds 'em, but once they're added, not much headroom is left).

The output of the distortion, wrap, shape, etc. tends to be full
scale or near it when the distortion is audible. Padding down the
input to these doesn't help, it just subtracts from your adjust
setting.

I believe the mixers (+amp) pad both inputs down 6dB before
mixing to prevent clipping, but the output of the mixer will then
tend to be near full scale.

All of this implies that you usually cannot set F4-AMP higher
than 0dB unless the signal is filtered, cut, or padded somewhere
along the line, or unless you apply no DSP, in which case you can
always boost at least 6dB there. Don't forget that keytrk can
cause clipping, as can source 1 and 2. Velocity tracking on the
F4-AMP page will never add clipping though, it actually subtracts
gain. Velocity tracking on other functions often does add gain.

Most layers created by soundware have been tweaked so that there
is little or no headroom left, as soon as the user goes to add
more boost anywhere, it will clip.

Low pass filters, high pass filters, bandpass filters, and EQ cut
all reduce gain, providing any box following them with more
headroom.

Now, about output gain.

The output gain setting on the MIDI Channels page simply replaces
the one on each layer of that channel's program. A setting of
Prog makes it use the output gain programmed for each layer. We
probably should have designed it as an offset instead of an
absolute value, but that's what we are stuck with.

The output gain specifies how much gain to add when summing a
voice onto the DAC with the other voices. If we assume that each
voice's individual gain structure has been tweaked to be just
below clipping, then an output gain setting of 6dB will tend to
prevent "group" clipping when 24 voices are playing. If you were
only playing a single voice, you could set the output gain as
high as 30dB without clipping, but then when you played a second
voice, both would clip.

On method I have used to maximize programming flexibility is to
set the first pad in the Algorithm to 18dB, set the F4-AMP adjust
to 0dB, and set the output gain to 30dB. Initally this is similar
to (prog 199) no pad, F4-AMP set to 6dB and output gain set to
6dB, but I've turned it down at the beginning and middle of the
signal path, and turned it up at the very end (output). The
overall gain has not changed. Now I can boost the parametric with
no clipping, then I just turn down F4-AMP to compensate for the
increased volume. If I am making a kick drum and I want it to be
a bit louder than other layers, I can just turn up F4-AMP,
instead of having to go to the output page.

The disadvantages to this method are: more quantization noise
from the EQ due to shifting the input down 18dB (I've never
noticed this though), more likelyhood of getting "group" clipping
(although little or no chance of indiv. channel clipping), and
finally it doesn't work well for many algs/functions, esp. those
that use distortion type functions or Hobbes waveforms.

Finally, MIDI volume adjusts the F4-AMP page, and only turns it
down. You may have noticed that some patches sound different
and/or better when you adjust MIDI volume, this occurs when the
patch clips the F4-AMP page.