The following document was intended as
introductory guide to understanding and selling the dCS DACs and Upsamplers. While
intended specifically for store sales staff, we thought it contained enough general
information to make it worthy of posting for all to read. This was written before
dCS began "upsampling" to DSD. So do keep in mind, that while conversion to DSD
(the format for SACD) isn't mentioned in this paper, many of the advantages of upsampling
to 24/192 also apply to DSD. In most cases we've found conversion to DSD to be
superior to upsampling to 24/192.
dCS DACs and Upsamplers
A Suggested Approach to Selling dCS Components – compiled by
Audiophile Systems, Ltd.
Contents
Introduction
The Numbers Game
There’s Digital and then there’s Digital
Loose Lips Sink Ships
More On Wavelets
Additional dCS Resources
Upsampling FAQ
An Introduction to Wavelets
Summary
Introduction
By far, the most effective way to sell any dCS product is to
demonstrate it. Our experience has shown that, in actual demonstrations, dCS Upsamplers
and DACs will:
- Significantly improve the performance of any quality audio system.
- Consistently outperform all competition.
While that is a rather bold claim (and suspect, coming from the
distributor of the product), it can be easily verified through your own listening tests.
And, we strongly encourage you to do so.
That said, we do understand the inherent limitations imposed by taking
a strict "listen to it – if it sounds better, it is better" approach. It is
simply impractical to have every competing DAC and Upsampler readily available for use in
A/B comparisons. Besides, most people make their buying decisions on two levels. On an
emotional level the demonstration and the involvement with the music are key. But, there
is an intellectual component as well. Buying any dCS product requires a major investment.
People need enough factual information to be able to justify that investment.
The Numbers Game
Our first thought was to take the easy way out and stick to the
numbers. dCS has always led the "spec" race. In 1999 it seemed that every DAC
manufacturer was coming out with new 24/96 models or upgrades, five years after dCS
launched the world’s first 24/96 DAC in August of 1994. Three years ago, when
24/96 was still on the wish list of most manufacturers, dCS was already shipping a 24/192
DAC! [1]
While all this sounds very impressive, evaluating performance based
solely on published word lengths and sampling rates does present some concerns. A growing
number of off-the-shelf 24-bit DAC chips are currently finding their way into consumer
DACs and CD players. Most manufacturers using these chip sets freely admit that these DACs
are 24-bit in name only. Yes, they will do 24-bit math, but their true resolution is
typically on the order of only 20 bits or so.
To the best of our knowledge, the only consumer DACs currently
available with true 24-bit resolution are the dCS Elgar and Delius Ring DACs. That is to
say that a change in the state of the LSB (least significant bit) actually results in a
corresponding change in the output of the DAC. [2] [3]
One of the problems with all "numbers games" is that there is
rarely a 100% correlation between the numbers and the real world performance. In fact, it
is frequently possible to come up with impressive numbers through means that virtually
guarantee poor performance. Case in point – amplifiers in which massive feedback
produces phenomenally low distortion measurements on sine waves and utterly disastrous
sonic performance with music.
So, while the numbers are interesting to mention in passing, attaching
too much significance to them actually understates the dCS advantage and leaves one open
to attacks based on hyped-up specs rather than achievable performance.
[1] To view an impressive list of dCS "firsts",
visit www.dcsltd.co.uk.
[2] For additional information on the
unique performance advantages of the dCS Ring DAC may we suggest the Arcam Alpha 9
Technical Paper available at both the Audiophile Systems and Arcam web sites –
www.aslgroup.com and www.arcam.co.uk. This paper provides an overview of the chip version
of the Ring DAC used in the Alpha 9 and FMJ CD23 CD Players. While the discrete version
employed in the Elgar and Delius does offer additional performance advantages, the basic
principles are applicable to both the IC and discrete versions.
[3] For test measurements confirming the
24-bit performance of the dCS Elgar, see the Mike Story’s paper Resolution, Bits,
SNR and Linearity available at the dCS web site – www.dcsltd.co.uk.
There’s Digital and then
there’s Digital
The real dCS advantage comes from their unique approach to digital.
While most audio companies rely on commonly available chip sets, dCS
designs their DACs from the ground up using discrete components. This allows dCS to
achieve a level of performance and precision that simply can not be matched by companies
that have chosen "off-the-shelf" solutions.
dCS also likes to say that they have "the softest hardware in
town." The choice to make their hardware reconfigurable through the use of software
allows easy (and affordable) upgrades that protect an owner’s long-term investment.
This approach also facilitates the quick implementation of the latest digital processing
techniques and goes a long way towards explaining the time-to-market lead (as much as five
years!) that dCS has consistently maintained over the competition.
While both of the above dCS "advantages" seem compelling, we
feel that there is a third, more significant and more fundamental, dCS advantage. Unlike
traditional audio companies, dCS is engaged in front-line digital research (and holds a
number of significant patents in this area). Their background in both scientific research
and military [4] & industrial consultation provides them with an understanding of the
digital process that is far beyond that of traditional audio companies.
Thus the performance advantages offered by dCS components have less to
do with factors that the audio industry typically associates with high-performance
digital, and far more to do with the fundamental soundness (and uniqueness) of the dCS
developed algorithms that carry out the upsampling and digital to analogue conversions.
[4] While you may know dCS for their
work in both consumer and professional audio, outside our industry they are better known
for their development of ultra-high performance A to D converters for use in military
aircraft radar systems. This is an area in which one extra bit of resolution can make the
difference between life and death. So it should come as no surprise that they are able to
maintain such a substantial lead over conventional audio manufacturers.
Loose Lips Sink Ships
Coming from a background that includes the development of ADCs for
fighter nosecone radar, it’s understandable that dCS isn’t predisposed to talk
openly about the digital techniques they employ. [5] However there were two
questions that came up so often that we were finally able to persuade Mike Story at dCS,
to provide at least partial answers.
Question 1: How does upsampling work?
Question 2: What’s the difference between upsampling and
oversampling.
The first question was particularly intriguing, since, in theory,
upsampling should not result in a performance benefit. Upsampling (as well as
downsampling) has been around for years and had never been thought of as a way to improve
performance. These types of digital to digital conversions were used primarily in
professional applications where you simply had data in one format and needed it in
another.
It wasn’t until the dCS 972 Digital to Digital Converter (a
professional version of the dCS Purcell) was found to provide a substantial improvement in
performance when upsampling conventional 16/44.1 CD data to 24/96 that
"upsampling" became the new digital buzzword in high-end audio circles.
This unexpected result naturally led to a great deal of speculation as
to the mechanisms in play. Explanations have ranged from video line-doubler analogies in
which digital "gaps" are filled, to the relaxed filtering requirements at higher
sample rates, to improved management of jitter.
While some of these explanations may indeed account for the subtle
benefits provided by the plethora of upsamplers currently appearing on the market that use
conventional conversion techniques and off-the-shelf chips, they didn’t seem
sufficient to explain the level of improvement in spatial perspective and apparent detail
provided by the dCS Purcell.
Additional confusion was contributed to upsampling speculations by the
fact that CD players and DACs had long been using "oversampling," a technique
that is, in some ways similar.
To help sort out the differences between upsampling and oversampling,
and to shed some light on the possible mechanisms behind the sonic improvements that
result from upsampling, we now refer you to the attached Upsampling FAQ kindly
provided by Mike Story. We should point out that, while we asked Mike for a
"simple" explanation, there seems to be some divergence between his definition
of simple and ours. So, if you start getting a little bogged down on the technical
details, don’t worry. It all comes together at the end with the introduction of
"wavelets."
[5] We are not joking about the sensitivity of this issue.
We actually had to get clearance before we could publish this document. (And yes, we were
required to delete some information.)
[6] Unlike upsampling, oversampling was always intended to
improve performance – for example to increase the S/N ratio of one-bit DACs. However
it should be noted that this usually involved achieving "acceptable" performance
from otherwise flawed DACs, not in stretching the performance of the 16/44.1 Red-Book CD.
More on Wavelets
For those of you that end up wanting additional information on wavelets
after reading the Upsampling FAQ, we have also attached the introductory section
from a paper by Amara Graps entitled "An Introduction to Wavelets" that appeared
in the IEEE Computational Science and Engineering Volume 2, Number 2, Summer 1995.
We particularly enjoyed the reference to being able to "see the
forest AND the trees" at the end of the second paragraph. What a great line!
Also of interest is the list in the last paragraph of other fields
making use of wavelets, which includes Radar. Coincidence? We think not!
Additional dCS Resources
For those wanting to explore the dCS advantage on a more technical level we suggest the
following papers, available on the dCS web site – www.dcsltd.co.uk.
A Suggested Explanation for (Some Of ) the Audible Differences Between High Sample
Rate and Conventional Sample Rate Audio Material
Resolution, Bits, SNR, and Linearity
Effects in High Sample Rate Audio Material
Timing Errors and Jitter
Upsampling FAQ
M.J.Story, dCS Ltd.
What’s the difference between Upsampling and Oversampling?
Upsampling, or interpolation, is an activity carried out
in the digital domain. The sample rate of a data stream is increased (interpolated or
upsampled) whilst keeping the same (baseband) spectrum. It is a synchronous process.
Whilst the two terms, interpolation and upsampling, can mean the same thing, interpolation
is often used for small changes in sample rate - 44.1 kS/s to 48 kS/s, for example.
Upsampling is a convenient term where larger changes are involved – 44.1 kS/s to 96
kS/s, for example.
Oversampling is generally applied to conversion between
the digital and analogue domains, and is so called because the sample rate is generally
significantly in excess of that needed to represent the signal of interest. Oversampling
can also be applied to a transfer between two (asynchronous) digital domains, as in some
asynchronous sample rate converters. In dCS’ view, oversampling is not the correct
term to use for synchronous digital sample rate changes.
More Background:
Upsampling – An input digital signal at a sample rate Fs
will have a spectrum that occupies the frequency range 0 to Fs/2. The spectrum repeats as
the frequency goes up – for a simple signal, it is reflected about Fs/2. For example,
a signal at Fs/25 will be reflected at Fs/2 and also now appear at 24.Fs/25. The whole
spectrum from 0 to Fs repeats from Fs to 2.Fs, 2.Fs to 3.Fs, etc. If the sample rate is to
be increased to a new rate n.Fs then generally one of the jobs of the upsampling or
interpolation filter is to clear out (filter out) the region between Fs/2 and n.Fs/2 by
digital filtering. For example, a signal might be upsampled 4 52/147
times (from 44.1 kS/s to 192 kS/s, say. In the course of the upsampling process, the
spectrum between 22.05 kHz (44.1/2) and 96 kHz (192/2)
would be digitally filtered and cleared. The upsampling ratio does not have to be an
integer, but it is desirable to make it a rational fraction (m/n, where m and n are
integers).
Oversampling is the term used to describe an A/D or D/A process
where the sample rate used is (usually substantially) higher than the minimum to reproduce
the spectrum of the wanted signal. An oversampling ratio of 64 is not uncommon in audio.
Generally, oversampling is used with a second process – noise shaping. In ADCs it
allows the use of low bit, inaccurate ADCs to produce digital signals that are extremely
accurate over a small part of their spectrum. In DACs it allows the use of short word
length (5 bit, say, or even down to 1 bit) DACs operating much faster than the input
signal requires. Again, this can produce a very accurate representation over a small part
of the output spectrum - the small part is enough to cover the spectrum of the input
signal. Oversampling does not have to be used with noise shaping – it can just be
used to give small signal/noise improvements, or to allow partly digital anti alias
filtering. Digital ‘scopes and some video codecs operate in this way. However, from
the audio point of view, this is not an interesting approach.
Why does Upsampling improve sound quality?
From the explanation above, there is apparently no extra information in
the upsampled signal that was not present in the initial signal. With a 44.1 kS/s input,
both the input data stream and the upsampled data stream will only contain a spectrum that
must be between 0 and 22.05 kHz and is probably only between 0 and 20 kHz.
This conventional analysis starts from the viewpoint that the behavior
of the ear can be described in mathematical terms using Fourier analysis. This assumption
is probably pretty good – it means we are interested in frequency responses, for
example, and these do provide good guides to the performance of equipment and to
descriptions of what we hear. The analysis was right at the heart of the definition of the
audio coding used on CDs.
For those working with audio, it is also apparent that theories based
on these descriptions are not completely adequate, and that there can be significant
differences in the performances of pieces of equipment with similar
"conventional" specifications. It seems that two things are going on here –
the ear may have more than one mechanism at work, and sine waves may not be the best
function to use as the basis for analysis. On the mechanism front, it seems highly likely
that the ear has a sound localization mechanism ("where is it") that is fast,
and independent of the mechanism that says "it’s a violin", and that is
related to transient response. There may also be a third mechanism at work. On the
analysis front, it may be that some form of wavelet is the best basis for mathematical
modeling. The problem here is that sine wave theory is relatively simple, and has been
fully worked out by generations of mathematicians, following on from Fourier. Wavelet math
is just plain hard work, and does not yet have anything like such a solid core of
mathematical results to call upon. Our ears, however, are not waiting.
If one gets the frequency response of some equipment right, but the
provision of transient information wrong, one or more of the ear’s mechanisms cannot
work properly, and so we are unable to separate out echoes and cues about where a sound is
coming from the rest of the "what is it anyway" signal. dCS’ upsampling
filters are designed to help sort this problem out. They are best analyzed not in sine
wave terms, but using wavelets.
An Introduction to Wavelets
[7]
Amara Graps, Stanford University
The fundamental idea behind wavelets is to analyze according to scale.
Indeed, some researchers in the wavelet field feel that, by using wavelets, one is
adopting a whole new mindset or perspective in processing data.
Wavelets are functions that satisfy certain mathematical requirements and
are used in representing data or other functions. This idea is not new. Approximation
using superposition of functions has existed since the early 1800's, when Joseph Fourier
discovered that he could superpose sines and cosines to represent other functions.
However, in wavelet analysis, the scale that one uses in looking at data plays a special
role. Wavelet algorithms process data at different scales or resolutions. If we look at a
signal with a large "window," we would notice gross features. Similarly, if we
look at a signal with a small "window," we would notice small discontinuities.
The result in wavelet analysis is to "see the forest AND the trees."
Can you see why these features make wavelets interesting and useful?
For many decades, scientists have wanted more appropriate functions than the sines and
cosines which comprise the bases of Fourier analysis, to approximate choppy signals. By
their definition, these functions are non-local (and stretch out to infinity), and
therefore do a very poor job in approximating sharp spikes. But with wavelet analysis, we
can use approximating functions that are contained neatly in finite domains. Wavelets are
well-suited for approximating data with sharp discontinuities.
The wavelet analysis procedure is to adopt a wavelet prototype
function, called an "analyzing wavelet" or "mother wavelet." Temporal
analysis is performed with a contracted, high-frequency version of the prototype wavelet,
while frequency analysis is performed with a dilated, low-frequency version of the
prototype wavelet. Because the original signal or function can be represented in terms of
a wavelet expansion (using coefficients in a linear combination of the wavelet functions),
data operations can be performed using just the corresponding wavelet coefficients. And if
you further choose the best wavelets adapted to your data, or truncate the coefficients
below a threshold, your data is sparsely represented. This "sparse coding" makes
wavelets an excellent tool in the field of data compression.
Other applied fields that are making use of wavelets are: astronomy,
acoustics, nuclear engineering, sub-band coding, signal and image processing,
neurophysiology, music, magnetic resonance imaging, speech discrimination, optics,
fractals, turbulence, earthquake-prediction, radar, human vision, and pure mathematics
applications such as solving partial differential equations.
[7] This introductory section from "An
Introduction to Wavelets" by Amara Graps was published in the IEEE Computational
Sciences and Engineering, Volume 2, Number 2, Summer 1995, pp 50-61.
There are clearly many different approaches that could be taken towards
"telling the dCS story". Information going far beyond that offered in this paper
can be found in the dCS literature and on their web site. This would include a look at
their involvement in the pro side of the business where their pioneering work on the Sony
DSD (SACD) and an impressive list of recording studios employing dCS converters can be
used as additional evidence of their expertise.
A review of the manuals for the Elgar and Purcell (both available for
download from their site) may also prove useful.
With so much information available there is a tendency for any
retelling of the story to run on, and on, and on…
So, we would suggest that the focus in selling dCS always be on the
demonstration with the "story" serving in a support function only. In our view
the key parts of the story are:
Audiophile Systems, Ltd. • 8709 Castle Park Drive •
Indianapolis, Indiana 46256