Specifications

6
Upsampling
Upsampling has been in and out of vogue for many years. Its
benet has been obvious in one DAC and nonexistent in another.
We do not do upsampling as such in the SELECT model. We have
the accuracy and precision which is compromised with upsampling.
Custom MSB Digital Filter
The DAC IV has a custom DSP based Digital lter. The sine x
function is the ideal shape to apply to the audio lter task, but
unfortunately to work perfectly it must sample an innite number
of samples. Our 16x lter contained 3200 taps, a very large sample
and worked well. The 32x Filter contains and amazing 6000 taps,
and the increased size of the lter more closely approximates the
ideal lter. Immediately you can hear the increased clarity of the
music.
MSB has applied noise shaping technology, not directly to the
audio as SACD and Delta Sigma DACs do but to the actual digital
lter. This novel approach reduces that digital harshness without
loss of detail or focus.
Just as glasses bring an image into focus for eyes that are not
perfect, we have applied a corrective algorithm to the slight
imperfections specically found in our DAC architecture. This
new innovation focuses the instruments more accurately on the
soundstage.
About the Clock / Jitter Control
Jitter control devices (and inputs on most DACs) normally reclock
the input signal in attempt to lessen the jitter of that incoming
signal. The DAC IV does no such re-clocking. We actually pay no
attention to the clock on the input signal. All internal clocks are
generated by an extremely accurate plug-in clock. Since the input
clock is no longer related to the clock of the DAC IV, an intelligent
½ second buffer is used to maintain data synchronization.
A clock header is a part of the DAC IV that will allow even more
accurate clocks to be installed as an upgrade as they become
available. Clock jitter seems to be the single biggest contributor to
digital harshness. MSB clocks are the lowest jitter audio clocks in
the world. The DAC IV plus models are supplied with the Femto
140 Clock with about 140 femtoseconds of jitter. The SELECT
comes standard with the Galaxy Femtosecond Clock which offers
about 77 picoseconds of jitter.
MSB Digital Filter History and Detail
One of our primary goals at MSB is to provide the music lover with the
most accurate musical experience possible. During years of careful
design and improvement of our custom discrete DACs, which form the
heart of your Diamond DAC, we realized that the Diamonds sound quality
was no longer limited by them. We next narrowed the problem to the
Digital Filter which was feeding our DACs. While the excellent Burr-Brown
(Now owned by Texas Instruments) DF1704 Digital Filter had served
us well in the past, it had became the bottleneck once we started using
our new Second Generation DAC modules. After a thorough search of
all the available off the shelf and custom DSP based Digital Filters we
realized that little improvement could be had from any of them. With no
other option in sight we decided to build our own solution.
Converting the ones and zeros of Digital Audio into music is an
enormously delicate and critical process. Each individual sample that
makes up the audio stream must be converted into the high resolution,
continuous analog voltage that can be transformed into the sound that
you hear. Any misstep can corrupt the nal result ending with audio that
does not sound anything like the original recording. Errors in translation
can make a harsh, veiled, muddy, and/or tonally colored result. Minimizing
each potential problem allows the original recording to shine through.
Audio reproduction starts when the DAC receives the binary coded
information from the source. The rst step requires recovering the audio
samples, which represent the nal output voltages, and the timing, which
tells the DAC when to output those voltages. Next the sample rate is
raised and the data is digitally ltered. While it is possible to feed the
DAC with the original audio samples thereby avoiding the use of a digital
lter skipping this step has many unintended consequences. After being
digitally ltered the digital stream is feed to the DAC. The DAC receives
the digital audio samples and converts them into a continuous analog
voltage. Our DACs instantly convert the data into a precise continuous
voltage waveform with timing determined by the DACs conversion clock.
The digital lter is necessary because mirrored image frequencies
created during the conversion process must be removed. If the DAC
did not have a digital lter, an analog lter with an aggressive response
must remove these image frequencies. These brick wall analog lters
seriously damage the signal by corrupting the original phase of the sound
and cannot fully remove the high frequency images. This results in harsh
or rolled off high frequencies and poor soundstage focus.
Traditional digital lter designs consist of cascaded FIR (Finite Impulse
Response) lters, each of which raise the sample rate by two. The
intermediate data between the lters is usually stored at less than 40
bit resolution. Since the next lter works with previously computed data
the resolution decreases with each lter pass. This limits higher quality
digital lters to a low oversampling rate (usually 8x) before the output
starts to deteriorate. The loss in resolution is typically not apparent when
using the best conventional digital lters with standard DAC chips, but in
combination with our high resolution DACs the problem is very apparent.
The sound becomes muddy, veiled and un-involving when using any
off the shelf digital lter. To counter this problem the MSB Digital Filter
does it’s ltering in one lter stage that raises the sampling rate by 32.
FIR lters operate by multiplying each sample in the data by a set of
lter coefcients and then summing the result. Most digital lters round
the result of each addition before the adding next sample. This repeated
roundoff results in a similar problem to the cascaded 2x lter approach,
muddy sound. MSBs digital lter uses bit perfect accumulation in an 80
bit accumulator completely eliminating these debilitating roundoff errors.
Only as the last step do we carefully convert the audio to the 24 or 26
bits our DACs require. The high sampling rate of the output allows us to
include advanced ultrasonic dither and noise shaping techniques in this
step to achieve greater than 24 bit effective resolution.
Through extensive listening tests we have found that the choice of lter
coefcients has a great impact on the tone of the music. We have found
that steep, phase perfect “Brick Wall” lters tend to sound the most
neutral but are also the most difcult to implement without problems.
Improvements we have made in our digital lter, with its single stage
design and 80 bit computation, allow us to use very steep lters with
no compromises.