|Glass Audio Article nr. 6/98|
A 24 Bit DAC
by Stefano Perugini
article appeared originally in Glass Audio
Vacuum Tubes Renaissance appears more and more as an
characteristic of every sector audio; as well as in realizations
you can see this fascinating and bright glassy bulbs in
symbiosis with the youngest products of the solid-state
presence of vacuum-tubes into solid state amplifiers (hybrid
and in the output stage of the dac converters or in the musical
of the pro-audio sector (where in
few cm2 of PCB you can
gathered DSP IC's and electron tubes) is
symptomatic, in my
of the convicement that attributes to valves the ability to
the asperities produced by the solid state devices
the reproduction of a sound less straining for the human ear.
musicians for instance seem to have a preference for the presence of
in power amplifiers
as well as in the analog signal
and pre-amplification unities, while in the High-End sector
Audiophile Community reserve a
great attention to
with valves in output.
this article I'm going to introduce the project of a DAC converter
uses the recent Crystal's CS4390. This component that you can see as a
up-grade is a complete 24 bit stereo digital-to analog converter,
in addition to the traditional D/A
function, include a digital
filter and a 128X oversampled delta-sigma modulator.
sigma modulation 
come also to a technological
and can quietly rivalry both
sonically and technically with the
traditional multibit modality.
this project I have reserved great importance to the designing of
power supplies even featuring the eccentric use, for this context, of a
output stage, thanks to the versatility of the
can lend to the most varied
designing interpretations. In this specific
you will see the implementation of a passive unity realized with a
full block-diagram of the converter. The signal
from the transport unity of CD-Player is submitted to an initial
that consists in an amplification and slope front
, Block 2. The Block 2 output is the input of the Interface
Block 3. I have used here another Crystal's device, the
CS8412. The CS8412 receive an decode audio and digital data
a digital audio transmission line according to the AES/EBU,
S/PDIF and EIAJ CP-340 standards. The (low jitters) clocks
= 256 X Fs
= 64 X
= Fs or 2 X Fs
Fs is the sampling frequency.
the fourth produced signal, is correlated directly with the audio
data flow is input to the DAC,
Block 4, realized with
the CS4390. The 90's full Block-Diagram is shown in Fig. 2.
factor permets the selection of an output filter based on out-
noise attenuation requirement rather than anti/image filtering .
delta-sigma modulator, Block 2, convert the interpolation filter
into 1 bit data @128X Fs. This data flow is input to the DAC
Block3, where the digital world bridge the analog world and
digital word translated into analog signal. Block 4 perform a low-pass
and allows two analog output with a phase difference of 180
Fig 3. Each output produces 1 Vrms for a full scale digital
In differential mode, where you can exalt the cancellation of
mode errors, noise, distortions and offsets, you can get up to 2
noise spectrum in output from the 4390 presented in Fig.4, shows
the uselessness to apply a strong
low-pass filtering. In fact for this
the same Crystal recommends the implementation of a 2nd
filter, however some experimental
tests show that good results can
be got with an a simple 1st order filter. Nevertheless the
good characteristics don't have to let think that the output filter is
because some measures, that have found full confirmation in
tests, have put clearly in evidence the presence of small levels of
when the signal in output of the DAC is input to the
of the audio-chain directly. In this project the filtering process has
realized resorting to the natural low-pass characteristic that a real
transformer exhibits in the region of the high frequencies. This has
the realization of an output stage, Block 5 for the L-channel
Block 6 for the R-Channel in Fig. 1, with excellent sonic characteristics.
The output stage schematic is in Fig. 5.
you can see this circuit as:
a differential mode-to single ended mode converter;
a low-pass filter;
very simple to profit some correlated advantages with the architecture
the output stage internal to the
the simplicity of this stage is only apparent since a lot of
can be dissipated for the search or the realization of a
transformer that answers to our specifics and for the circuit optimization;
load seen by the transformer is in
fact very complex because you
have here a shielded cable that connects the output of the
unity with the input of the preamplifier. The knowledge of the
between the parasitic elements of the transformer,
cable and the preamplifier input
circuit , it's of primary
in order to establish the position and the entity of all the
certainly out audio band, that would be able to move by
noise and distortion into the high region of the
have conducted with profit this investigation entrusting me, as
to the circuit simulation . The circuit used for the
is shown in Fig. 6. The components designated with an
represent the parasitic elements of the real transformer.
Rti2*, Rti3*, Rti4* are
the windings DC resistances;
L_leak_pr1*, L_leak_sc2*, L_leak_pr2* are
the equivalent lumped capacitances between windings.
connection cable (the
Signal Cable black-box) has
schematized as transmission line , Fig 7.
have extracted the
sample of the High-End and Consumer production; by measurements
have extrapolated a real model of
Cp, Rp) = (3uH, 300pF, 0.1ohm)
I have used for the simulations. Fig.8 is a Monte Carlo
related to the frequency response
of the circuit in Fig. 6 when R1
C_pre have 5% and 100% tolerance rispectively. Fig.9
a histogram evaluated with respect to Bandwidth("node", 1db) Goal
, extrapolated from Fig.8. You can see that the
@1dB always falls between 21.6kHz and 23.5kHz, therefore
output stage well "defend" the correct frequency response from most
last results are not accidental or causal
since with a transformer of
quality the results won't be so good. From these simulations
peculiar behavior of the used transformer emerges:
low-pass characteristic that you
can observe in
typical of every audio transformer, has been wanted
during the transformer building, on the values of
have shown that this parameter has to lie between 10mH and
Lower values of 10mH produce an excessive extension of the
worsening the DAC's noise levels. Greater values of 20mH
a premature cut of the frequency response reducing the
content of the audio high frequencies. Simulations and
have shown an optimal behavior when the leakage inductance
a value of 16mH.
a lot of equally valid variations to the scheme of Fig 5
If in this context you don't desire to renounce to vacuum tubes
take in consideration the simplified scheme of Fig. 10; as electron
I recommend the followings:
3A167M, EC8010, E810F in pseudotriode, 417A, E55L
is small valves with high mu, high gm and a comforting plate
in order to make easier the search or the construction of a
transformer, to get a better impedance matching and to produce,
case of necessity, also a meaningful output power . If you love to listen
music with headphone this can be
the road to follow for achieving
sonics results. Moreover since a
lot of Sound Card have a
output, you can use this circuit to improve the sound reproduced
by the speakers of your Computer dramatically.
renuncing to the elegant
of a transformer output stage (active
or passive), you can
yourself to the natural tendency of an op-amp to the differential-to
single_ended mode conversion.
Fig. 11 the scheme of a 2-pole
response. This filter has been designed with
50kHz, a 40dB/Dec slope and 6dB
gain. Obviously the circuit of Fig.
can be implemented with vacuum tubes technology getting good results
however, Fig. 13.
can find the grounds of this
circuit in GA 1/95
Nevertheless in comparison to the original version this schematic use only
in order to simplify the realization I have chosen a SRPP as output stage.
limit instability phenomenae in
sub-sonic band that this modifications can produce.
fact Fig.14 shows what happens in the frequency response
with C3=4uF, C10 is varying between .2uF and 10uF with .2uF
. When C10 <C3 overshoots are
present in subsonic band. As
steps you can for instance choose C10=3.3uF and C3 = .22uF and
with a following tuning on the real circuit. Unfortunately the
of a low value for C3 introduces some disadvantage; Fig.15
the simulated frequency response when Rl varies between
and 20Kohms with step of 500ohms. A minimal value of
is necessary to avoid an excessive limitation of the low
In this context, since the realization of a vacuum-tubes op-
is less critical than that of a
wide-band transformer , the influence of
parasitic elements can consider some more negligible and therefore the
results more satisfactory.
The full schematic of the 24 bit DAC is show in Fig. 16.
it features a good ability
You don't have to worry about the elevated capacitive value of C35 and C36
the repetitive peak current, even in the most serious condition of
operation is lower of the maximum value, as Fig 17 show.
act in the decoupling of the high frequencies residuals; nevertheless their
values doesn't engrave heavily on the resonances of the power supplies filters,
shunt-type regulator for the voltage of the ' 90 have been chosen because it
small degrade of the sonic
performances in comparison to the series
whereas audio analog stage are present , the load is not onerous and,
in this case, it is not possible to
renounce to the power supply
, I prefere it to the serie-type regulator.
hands "in pasta"
have realized a first prototype of this converter with a point-to-
wiring, using small Teflon
clippings as support, Photo 1.
this solution is sonically effective , lately I have opted for
PCB realization in order to furnish in briefer times copies of this
to the friends of the audiophile cenacle remained
intrinsic sonic characteristics, Photos
2, 3, 4, 5.
this approach determines an increase in the difficulties. In
as every High-Resolution Mixed-Signal PCB Layout you will have to
the followings constraints:
Separate analog and digital circuits and segment by functionality and speed;
Distribute power supplies and grounds taking care to minimize loop area and
return current paths;
Isolate noisy return current paths from more sensitive analog circuits;
Minimize interference from clocks;
Decouple ALL IC power supply pin;
Reducing the effects of capacitive and inductive couplings;
Minimize return current path impedances to reduce ground bounce
instance with reference to c), the splitting of the ground into separate
and digital grounds is the best way to guarantee that noisy digital
will not flow in the sensitive analog area.
can see the adopted solutions in the Figgs. 19..22 that shows
artworks of the DAC.
ambitious objective to make the digital source timbre similar to
of an analogic source, or more simply to make harshless the resultant
is pursued generally acting entirely on the output stage side.
of the high-end converters builders realize
this circuit using vacuum
and the results they get are good.
Nevertheless, also using an output
with vacuum tubes, I think that a margin of improvement still exists
we also re-consider the constructive philosophy of the power supplies.
this light you can see the presence of a vacuum rectifier
and a shunt-
in the circuits of this converter.
the presence of a vacuum rectifier into a low-voltage
supply can appear a bit " freak ", nevertheless I invite you to
a similar solution, not necessarily in this same context, since,
am sure, the results will impress you positively.
project was conceived initially two years ago thinking at the
(the predecessor of 4390, with the same pin-out). During the
continuous improvements have been effected in order to get both
minimalist and well sounding object . A possible evolution of
object could foresee power supplies with
choke input filter and
T. Tanaka, T. Sugimoto, C. Kubomura
D/A converter with integrated digital and analog filters
AES convention, October 1991, New York, Pre-Print
Crystal Semiconductor Corp.
Stereo D/A converter for Digital Audio
DesignLab Eval. 8, 1997;
Radiotron Designer's Handbook, Fourth Ed.
World, Vol. 103 No. 1738, October 1997;
Crystal Semiconductor Corp.,
Board for the CS4329
Vacuum Tube Op Amp
Audio, Vol. 7 No. 1, 1995;
Vacuum-Tube Models for PSpice Simulations
Audio, Vol. 8 No. 5, 1996;
What did you
think of this article?