TPC DAC-1 redesign

[Pete Goudreau]

This is meant as a follow up to some older posts of mine covering
modifications to the TPC DAC-1. I have attempted to summarize the
total mod sequence here so this post may get a bit long...thanks.

--------------------------------------------------------------------

If anyone out there remembers a series of posts on modifying the TPC
DAC-1 which I wrote a while back, this is a followup with more mods
discovered along the way. The great delay has been due to trying out
various pulse xfmrs at the coaxial input, along with a few other
strange things. I'll summarize the mods to date as follows and then
go into the newer mods afterward.

All local bypass caps were replaced with Panasonic HFQ 120uF/25V
'lytics paralleled with AVX 06123E274M and 06125C103M ceramic chip
capacitors. The ceramics are special ultra low ESL (<400pH) SMT caps
that are basically 1206s with the long side metallized. The E and C
mean Z5U and X7R respectively and the values are 270nF and 10nF,
respectively...just for completeness. The 0612s are stacked and
located directly on each power trace immediately adjacent to every
active component with the exception of the digital side of the '1702
DAC ICs where a pair of AVX 10uF/10V tantalums were installed with
zero lead length directly between the +/-5V rail pins and the digital
ground pin on the back of the board. All 7805 and 7905 regulators
were bypassed locally with Panasonic HFQ 12mF/6.3V 'lytics and the
7815 and 7915 regs were bypassed locally with Panasonic HFQ 8.8mF/16V
'lytics at their outputs. The main filter caps were replaced with
Panasonic FA 3.3mF/35V 'lytics and the xfmr was replaced with the 25VA
version. The filter cap after the 75 ohm power resistor and ahead of
the input receiver 7805 was replaced with a Marcon 82uF/35V 'lytic.

Both sets of op-amps were removed from their sockets and soldered
directly to the board with the output op-amps being replaced with
AD811s and their feedback resistors changed to 750 ohms. All
resistors in the signal path were replaced with Vishay VMTA55/0.1%
parts, the caps in the reconstruction filter were replaced with RelCap
RT series polystyrene parts, the feedback filter cap in the I-V stage
was replaced with an AVX glass cap, and the I-V topology was altered
to increase gain and modified to improve noise characteristics. The
series resistors at the output were replaced and mounted on the back
side of the board directly from the output op-amps to the mute relay
outputs which were rewired to act as a shunting mute which eliminated
the contacts from being in series with the signal path. The 10uF
'lytics around the 1702 DAC ICs were replaced with the Pan HFQ 120uF
parts. The digital input select relay was removed along with the
associated R and D, the clock test header was removed, and an Amphenol
EMI filtered BNC connector installed and wired direct with a Vishay
termination resistor.

At the local hifihut, this modified DAC-1 was so close to the
performance of the Levinson No.36 that it was very, very difficult to
tell them apart. The main differences being that the bass impact and
extension and clarity of the upper octaves wasn't quite as good. The
imaging was very close but that last little bit of definition and
transparency just wasn't there. Overall, the sound of the unit was
worth far more than the cost invested.

New modifications since these include installing series termination
resistors at the four digital output lines from the oversampling
filter as 51R1 1206 SMT parts bridging a cut in each trace immediately
at the IC pins. This mod really cleaned up the upper octaves'
digitalitis pretty much more than any other single mod for the simple
reason that a) this value critically damps the LC network of the long,
thin trace and b) slows the edge somewhat so that less noise current
is coupled between the digital and analog side bond wires internal to
the DAC IC package. Definitely worth the grief to install, one of the
most important fixes to date. There is also a pretty nasty layout
error on this board where the ground plane on the topside, over these
four traces, is broken to pass the +/-15V feeds to the analog output
stage. This creates a slot antenna which is driven by the interrupted
image currents induced by the digital signals in the traces
immediately underneath on the solder side of the board. A future fix
for this is to cut and peel the two power traces and bridge the break
with copper tape, solder it at each end, and bridge over the tape with
a pair of jumpers, carrying ferrite beads, to carry the power. This
should dramatically reduce the high freq noise induced on the power
rails and carried out on the analog outputs due to the poor high freq
PSRR of the op-amps and DACs.

The output op-amp ('811) was then replaced with an AD/PMI BUF04 after
this was suggested to me by Avery Montz in a series of private e-mail
exchanges. The improvement was amazing with a clearly audible
increase in transient verisimilitude across the band and a significant
improvement in tonality...thanks, Avery! It seems that the current
limit circuitry in the '847 and the '811 are acting to slightly
compress the signal and since the '04 has no active output current
limit circuitry, this simply doesn't occur. By incrementally removing
the 182 ohm dead load and 1nF output filter cap at the output of the
'811, it became clear that these loads were compressing dynamics and
limiting the perceived power of the bass octaves, this lead to the
conclusion that the peak current at the op-amp output was likely
causing the internal output stage current limiting transistors to
slightly compress the drive to the output devices. The '04 is a drop
in replacement for the '847, etc. but the feedback resistor can then
be removed as it is no longer needed.

Next, an attempt to improve the connection of the RC compensation
network at the FLTR pin of the CS8412 indicated that simply removing
the parts from the board and hairpinning them off the FLTR and AGND
pins actually made things worse, supposedly by acting as a nice little
antenna to inject noise current into the high impedance FLTR node.
Removed the parts and replaced them with a pair of 1206 SMT parts of
the same values but mounted on the back of the board in a shallow "V"
configuration with the open end of the "V" at these two adjacent pins.
This made another very clear improvement in the overall smoothness of
the sound of the unit, better than the original configuration. After
some more thought, it seemed logical that the FLTR pin, which is the
output of a transconductance amp, should be directly bypassed with a
small valued cap to reduce noise on the control signal for the
internal VCO and thus make an additional attempt to reduce overall
jitter in the recovered clock. After some calculation based on the
closed loop response curve in the databook and talking with the
factory, experimentation showed that a 1nF 1206 X7R chip cap directly
across the pins made almost as big a difference as the series
resistors did, pretty amazing...

After some more discussion with several other regulars here, the NPC
SM5842AP oversampling filter was suggested as a replacement for the
DF1700. Got the data sheets (the local NPC office is just down the
street) and figured out the conversion, bought one of the ICs ($72),
and installed it. Since it has digital deemphasis and mute, the mute
relay and driver, the JFET and deemphasis circuitry were all removed
and their associated traces peeled from the board. The deemphsis
control pins on the '5842 were wired to an extra gate. Details
available if you want 'em, just e-mail me. At the same time, found
that the dither function is atrocious...don't use it. The improvement
over the '1700 was well worth the cost and effort with an ease evident
that simply wasn't remotely there before, finally a true sense of
microdynamics that had been so sorely lacking in the original
configuration, and a smoothness that verged on what I remember the
sound of LP being like. Not to say that it sounded like an LP but
simply that sense of smoothness and ease was there along with a
crystalline treble and virtually inaudible edginess in the upper mids.
Highly recommended...

After listening to this state of modification for a few weeks, the
bypass of the digital section was changed from tantalums and 1210 chip
caps to the same configuration as that used in the analog section
which resulted in another improvement in loss of grain and edginess,
much to my surprise...each little change kept showing just how much
the previous result, which at the time was thought to be nearly
perfect, was just a tiny bit further away from perfection than was
thought at the time. BTW, the 7805 was bypassed with a 12mF 'lytic at
the same time.

After another conversation with Avery, he suggested that the bridge
rectifiers be replaced with hyperfast/soft recovery diodes. I didn't
think it would make any difference but he had heard an improvement so
samples were ordered and then installed. Well, I was wrong...and with
a little thought it was fairly clear why they would improve the sound.
The capacitive input filter of the power supply really causes a
terrible reverse recovery snap on the diodes when the AC waveform
breaks over the fully charged cap potential. This kind of slow snappy
recovery of the regular diodes in the circuit is going to generate
some serious RF in the 20-50MHz band, typically, and it will basically
crawl all over the circuit and really can't be gotten rid of easily.
The Harris RHRD440 diodes generate orders of magnitude less hash in
this application and thus lowered the power rail noise somewhat. This
of course lowers the overall noise floor and seems to make the sound
of the unit much clearer, I was really astonished at the level of
improvement even though it was subtle...and I sure wasn't expecting
it. Another recommended, but severe pain in the ass to complete
tweak...

The last mod so far is the testing oft a couple of digital input pulse
xfmrs, with dissapointing results. Apparently the open E-E core and
loosely coupled toroids in some products generates enough radiated
hash to really destroy a lot of the past gains in clarity and
tonality. With either of two units in circuit, and one of them
configured in two different installation configurations, vocals that
had been so clear and so easily understood were now basically so bad
that making out the lyrics was at times difficult. Transients were
blurred and the tonal balance of the system seemed more like an AM
radio. Well, out went the xfmrs. Instead, the chassis connection of
the shield side of the coax feed was lifted making the feed to the
'8412 essentially full differential with capacitive coupling and at
the same time, the four protect diodes at the RXP and RXN inputs of
the chip were removed as they act as small coupling caps from the
signal to the rail and common. That made yet another audible
improvement beyond what had gone before...again. Apparently the
simple chassis connection was allowing a ground loop from the
UltraJitterbug to contaminate the local analog common through the
chassis to common bypass caps in the DAC-1 and its removal acted to
remove a source of noise floor contamination.

Approach this mod with caution in that if the digital source is not
referenced galvanically to earth then there exists the possibility
that the inputs of the '8412 can be taken outside of its common mode
range and the part will latch up, I know, I've had it happen a few
times. The 75 ohm power resistor ahead of the 7805 limits the current
to a value low enough to avoid completely crispy crittering the part
but a "G" version part is on its way and due to replace the existing
part soon.

In terms of optimizing the digital link, it appears that the AES/EBU
coupling xfmrs in the UltraJitterbug are Scientific Conversions
SC916-01s and are currently not being used in this sytem
configuration. They are EP core type xfmrs so they will not radiate
noise anywhere near as much as the other xfmrs would so the next thing
to do is to pull one and install it in the DAC-1. If it at least
doesn't make the sound quality worse, it will be left there to protect
the new '8412. Another option being investigated is the installation
of an ST optical receiver as the xport (EAD T-1000) has an ST output.
It is likely that this won't be as good as the coaxial feed as holding
the LED transition jitter to <10ps is quite a feat and I have no data
from EAD indicating that the optical jitter is as low as the coaxial
output jitter. When the HP eval kit arrives, it will be used to make
a temporary link for evaluation, if it works out, then a metal
packaged part will be ordered and a small board built to adapt it into
the input circuit of the DAC-1 and the UltraJitterbug will be removed.
Will post results of this testing as time permits and more testing is
to be completed at the local hifihut as soon as one of these two links
demonstrates acceptable performance and adequate reliability...more to
come.

Overall, the improvement in performance for this series of mods is
easily of the order of magnitude that the original sequence was. All
of the deficiencies heard in comparison with the No.36 in previous
testing are now gone, or nearly so. It seems like it will really give
the 36 a run for the money when compared again, maybe even better
it...certainly sounds that way to me. But then I've been wrong
before...

Thanks for your time,
Pete