Post by David Pickett
This whole business of low noise microphones and preamps is in my
experience a non-issue in the vast majority of cases.
That said, I was upset to discover that MOTU publish no details on their
website of the 4Pre that can be construed as truly technical. [...]
Obviously they should publish their figures. In fact I don't think we
should really implicitly trust any figure given out by any equipment
purveyor. Instead we should, how was it now, "trust, but verify".
My personal waking up experience there was when I as a teenager upgraded
from a Gravis Ultrasound soundcard to a Gravis Ultrasound MAX. At the
time those were the most cost efficient means of producing sampled sound
from a PC expansion card, and I believe it's settled the original Gravis
card, at least in its later iterations, can still be called a small
marvel of sound, low-cost engineering.
The MAX on the other hand immediately sounded noisy and irritating to
me. Of course I still just had to have it, because it was the first
credible consumer range card on the market to not only support 44/16
playback, but to have it done at low central processor load. Gravis
achieved that by tacking on a separate Crystal Logic two channel
converter chip (CS4231), while leaving their existing custom GF1 ASIC
chip to do offloaded sample playback synthesis (32 voices!!!). The
efficiency gain in direct playback was because you could do DMA over the
ISA bus, without the tortuous peeking and poking and unreliable
DMA-to-device-memory, and chunking to fit register constraints, and
without any hardware synchronization on board the GF1, and whatnot.
Programming that chip was sheer hell, which showed in any application
not doing precisely what it was meant to do, and everybody liked the
stupider, more direct access of the MAX's new chip.
Except that the card sounded like shit. No user accessible feature of it
made it silent. It not only hissed, it let through power supply
fluctuations. Those emanating from hard drive seek servoes, in
I eventually got incensed enough to actually contact Crystal Logic, and
procure the new chip's databook. (Then they still used to send them for
free, over mail.) My first analysis of the problem was that they had
this new, unknown to me thingy called "dither" in there, and after duly
disassembling half of Gravis's driver code, I determined it was turned
on. So I dialed, a number of times over successive days or perhaps even
weeks, Gravis's Finnish helpline. (They had to have that, because the
Finnish Demo Scene was a considerable driver of Gravis sales then. Demos
can't waste any cycle, now..? ;) ) I pestered the hell out of a couple
of service reps and even a couple of engineers, to no eventual avail. I
resigned to the reality of a now-shitty soundcard, and eventually just
switched it out as part of technological development.
A few years later, after I'd finally taken a slightly more involved look
into mixed analogue-digital engineering as well -- and still nothing
spectacular, just the barest of basics -- I suddenly thought to look
back at my noise problem with the MAX. As it happened, I still had the
databook of the new converter chip with me, with its reference PCB
design, not to mention both the physical card *and* a '486 carrying
motherboard in which to plug it.
If I remember correctly, I ended up cutting a couple of line level
connects on the board and rerouting them through a haphazard out of
board capacitive traps. Did something similar to separate a couple of
digital signals, including the the clock signal to the new converter
chip, from the analogue earth. Put a bit of actual tin foil encased
in saran wrap over the digital section, strategically grounded.
...and then most crucially completely reworked how the analogue
reference of the new converter chip was fed. Because that was the true
root cause of the problem: for some reason unfathomable to me, Gravis's
engineers had decided to mostly adapt Cirrus's reference layout for the
new section, *except* for the *explicit* warnings against in *any* way
directly coupling the pin to the workings of the digital section. Even
touching its (in itself well designed, wide) ground willy-nilly, because
of what ground resistance and reactance do to all of them funny-funny
digital switching transients (remember, at that time the voltages and
currents we worked with on the digital side were at least a decade and
sometimes more beyond today's figures). The circuit board also relied on
external regulation alone, with typically *very* substandard performance
as far as audiophiles would have it; while the first iteration with only
the GF1 ASIC and its analogue output circuitry had been well designed to
withstand that, the added-on Cirrus chip couldn't cope with that at all.
The final version of my modifications brought down the total noise
somewhere in excess of 30dB. The end result just fell silent, as it
should have been from the start. I'm also rather certain the same result
could have been achieved at negligible cost had the PCB been designed
right from the start; even going with Cirrus's reference design and the
ample commentary which came with it, the board could have been as first
rate as the first iteration was. Combining those two separate teams'
efforts fully, I'm pretty sure they could have exceeded my eventual
achieved performance by at least a couple of decibels in inherent noise
and power supply rejection -- my modifications having costed something
in the vicinity of 15€ in current money, while being total overkill, and
achievable by sane, careful circuit design at a tenth or even a hundreth
of the total cost.
Long story short, openness and the sane engineering discipline which
goes along with it can achieve a lot. Even in the lowest realms of
engineering. In this case, with one half-clueless teenager rectifying an
ex post *obvious* engineering fault of a well-performing, already mostly
well-engineered mass-market product.
Sampo Syreeni, aka decoy - ***@iki.fi, http://decoy.iki.fi/front
+358-40-3255353, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2