advertisement


It's Taylor DAC v2.0

Kit Taylor

Well-Known Member
UPDATE 27/3/05

I've know collated choice info from various threads and other posters on other forums, and posted it in the FAQ Room:
http://pinkfishmedia.net/forum/showthread.php?s=&threadid=9784

UPDATE 12/11/04

Thread continues here:

http://pinkfishmedia.net/forum/showthread.php?s=&threadid=8287

General Construction
- Built on high quality Roth Electronik board with round solder pads
- Four 1/2" PCB pillars as feet, plus one under the local resevoir cap to stop the dac assuming the launch position :)
- Support is two pieces of foamed butyl rubber topped with a light stiff piece of cork tile, in an antistat bag
- All grounds connect to same plane
- P2P wiring and signal earths are SCSI ribbon cable
- All resistors are Multicomp 1206 1% 0.125W

Power Supply
- DC source wallwart with schottky diodes and 3300uf 35V Panasonic FC (original LM7812 removed)
- Local resevoir cap 3300uf 35V Panasonic FC
- LM317 supplying each pin, with and 4700uH/20R inductor before and after
- Chip pins decoupled with 1uF 1206 X7R and 10nF 0805 X7R
- Voltage drop has been compensated for like so:

CS8412 VD+ / 5.18V / 754R ADJ
CS8412 VA+ / 5.5V / 816R ADJ
TDA1543 / 6.25V / 960R

NB Kept these values, but they are calaculated for 25R not 20R

S/PDIF
- Interconnect is a cheapo RCA plug connected to two strands of twisted SCSI ribbon cable, hardwired at the dac end

CS8412
- 10nF NPO 1206 coupling pins 9 and 10
- Loopfilter is 100R 1206 & 1uF X7R 1206
- 100R pn digital output pins
- 3300uF 16v Rubycon ZL on pin 22
- Foil groundplane wraps around both sides of CS8412

TDA1543
- Supply +5V
- SCSI ribbon cable as hookup wire.
- 869R I/V and 910R Vref

Output
- Coupling cap is 1uF LCR polypropylene
- Interconnect is hardwired 0.5mm silver in 0.91mm PTFE tubeing

Transport
- Cheap Wharfdale DVD player, SMPS upgraded with the biggest Rubycon ZLs that physically fit

Some related LM317 questions
http://pinkfishmedia.net/forum/showthread.php?s=&threadid=7099

Some related opamp and I/V questions
http://pinkfishmedia.net/forum/showthread.php?s=&threadid=7385

---------------------------------------
ORIGINAL FIRST POST
---------------------------------------

This is a super basic non-oversampling passive output dac based around the CS8412 reciever and TDA1543 convertor, in essence a tweaked clone of this Peter Daniel 3D dac. It's not flawless, it did cost less than £100 to build after all, but fed from my Sony 7700 DVD player it's perhaps my favourite out of all the digital sources I've tried at home.

It sounds very "live," busy and loud with instruments overwhelming one another in a way that makes it impossible fit all the parts neatly and clearly together. It doesn't give an overview of the grand scheme of things, so to speak, and there always seems to be more going on than at once than is possible to pay attention to.

I think this is a natural and compellingly atmospheric kind of presentation. With respect to fox and the other Manatroids, the "at last the music made sense and I finally understood it" sound is an artificially simplified reinterpretation. Music is inherently chaotic and mysterious.

It also has a ferocious bass kick, is very good at picking out the little pauses and wavers that make things sound human and has Naim-like sense of pace.

Faults? There's a bit of glassyness in the upper mid that I suspect can be tweaked away, and it's a little soft, dark, airless and slightly honeyed. Timing is spot on in terms of an individual musician's phrasing, but there's not that sensation of an umblical linking the different instruments.
 
Taylor DAC 2.0 is built onto a piece of Farnell's pad protoyping board (breadboard with solder pads) supported by foamed butyl rubber squares and perched on a John Lewis pine chopping block.

Output caps are my favourite Sanyo OSCON 10v 220uf G-type, which seem to have the best available balance of everything for reasonable money. I/V is via 2k2 Audio Note tantalum resistors, which have a lush, chunky, somewhat airless sound. The filter on pin 20 is a 470R AN tant followed by a 4.7nf SMD ceramic cap.

For conveniences sake all grounds connect to the same plane, though some have recommended separate planes for digital and analogue. I've hooked it all up with silver wire. You could just bend the chip pins and solder them down, though they do look a bit fragile.

Each powersupply pin is fed by an LM78XX fixed regulator, each connected via a 1uf inductor to the same Rubycon ZL 16v 3300uf cap. This is fed 12v DC via a linear wallwart fitted with schottky diodes and a Panasonic FC 35v 3300uf resevoir cap.

Due to a parts shortage decoupling is via Panasonic FC 10v 120uf bypass with a 100nf SMD ceramic rather than my preferred quadruplets. I think the sound is abit unrefined because of this. Inductors between the reg and chip pin might also help.

The output wiring is hardwired, 0.5mm silver wire in 1.91mm PTFE tubing for the signal and plain multicore mains cable for the earth.

The copper foil is 3M self adhesive embosed and tinned tape from Farnell.
 
Due to a parts shortage decoupling is via Panasonic FC 10v 120uf bypass with a 100nf SMD ceramic rather than my preferred quadruplets. I think the sound is abit unrefined because of this

A quadruplet is a Panasonic FC 25v 120uf bypassed with these SMD ceramic caps:

100nF X7R 0612
220nF X7R 0805
Murata 4.7uF 16V Y5V 1206

I got the idea here:

http://www.diyaudio.com/forums/showthread.php?threadid=4562&perpage=15&pagenumber=1

Note that when I originaly recieved the caps from Farnell there was a bit of a mix up with the labelling. It seems the 4.7uf caps are larger, dark brown types whilst the 220nf are smaller yellow caps.
 
I've put a 10nf polyprop across pin 20 & 21, recommended by dint of pin 19 spraying 11MHz all over the place. I've also added 1uH inductors to on each chip supply pin.

A worthwhile improvement. Tougher and cleaner, and even more stygian bass.
 
Great stuff Kit.
Are you using a CS8412 for the interface? If so, I have some things for you to try if you wish....IME the 8412's performance is *very* dependant upon implementation.

M.
 
Nice work Kit:cool:

Is there anybody in the UK who supplies the CS8412?
I can't find them anywhere, I want to add one to my TDA1543 dac so I can use it with my dvd player, currently if I want to listen to it I have to stick it inside my modded Arcam
 
Martin Clark - I'm using the CS8412. Spill the beanz!

Ron - Oliver (coredump) might be able to help you.

Thanks for the kind words chaps, but you might retract them if you saw my hilarious/dreadfully dangerous Heath Robinson construction job. I want to build another one, as I've some ideas for a more stable layout. SOme pictures will be a must for that one.

The dac seems to be a blank slate, with the sound determined overwhelmingly by component choice and the rest of the system. The defining characteristic are super muscular bass and a soft, textured top end that makes my other digital sources sound unaturally polished. Perhaps this is a little euphonic distortion or compression, or is it just the truth of what's on the disc?

I'm using it with an Olson mains block fitted with an encapsulated silver filter from Schaffner, and Eupen Adusa mains leads on the pre and power amp. I think the sound lacks a bit of colour and shape without them.
 
When the DVD player transport isn't engaged there's a high pitched tone from the dac. I assume this is something to do with the earthing.

When I use the analogue outputs of the DVD there's no noise even if I disconnect the signal earth, whereas if I don't connec the dac's analogue signal earth it buzzes furiously.

It doesn't seem to be having a negative impact on the sound, but I'd like to squish any bugs for elegance's sake. Any ideas?

TIA
 
Sorry for the delayed reply Kit - had to go hunting in the 'old-projects' box

The CS8412: it has two voltage supply rails - +5V to Pin 7 for the digital content (pin8 is digital ground), +5v on Pin 22 for the analogue (pin 22 is analogue ground).

What is not made explicit is that the analogue supply is for the Phase locked loop, therefore noise on this pin will directly impact the jitter performance of the receiver - any noise here gets rolled into the PLL loop, and increases jitter. The CS8412 is specified for 200ps of jitter, though I have a suspicion this is actually its contribution over-and-above whatever is on the transports output already rolled into the inbound SPDIF stream.

The answer is to get paranoid about supply noise to pin 21. Use a dedicated local regulator; a well-sorted LM317 is better in this application than either a 7805 or a shunt regulator (no matter whether an amplified zener or TL431-based, shunt-types have lousy noise performance - esp 1/f noise - which has an adverse audible effect in my experience). Preceed the reg. with a small inductor or RF choke, say 1-2mH, and maybe add-in a ferrite bead or two between the reg and pin 21. Put the reg. output caps right at the body of the CS812 - your 'composite quad' sounds a good idea, connect it directly between pins 21 and 22 bypassing any socket.

Apply the similar care to pin 20 if you have space, but it's only the supply for the schmitt trigger and other digital gubbins- much less significant. The inline choke is a good idea however, whatever you do. I'd always recommend the small type that look like a resistor - they usually have a small effective series R of 10-13ohms which helps damp things nicely.

That takes care of the PSU side of things.

The other major implementation detail which is not obvious from the datasheet is the job of the components on Pin 20. It's labelled 'FILT' and the datasheet shows a series network to ground of 1000 ohm/0.047uF for CS8412 and 470 ohm/0.068uF for CS8414. In fact this controls the cutoff frequency for the PLL. The quoted values give a cutoff frequency above the audio band and so there is not really any reduction in data-correlated jitter from the transport and cable interface. This is necessary because the input receiver is generally designed to lock very quickly to a wide range of input data rates.

For the DIYer, this is not necessarily a concern ;-)

To improve the jitter reduction the resistors can be reduced substantially or shorted and the capacitor can be increased right up to the point where the receiver won't lock reliably to the incoming datastream. You shouldn't have any problems with 100 ohm/2.2uF here; you can try increasing the capacitance even further. I found 2.2uF on its own OK - it would still lock consistently to my marantz CD63SE. Use a good quality, film-type cap whatever you try.

The final thing - I know in your implementation the CS8412 is as close to the DAC as physically possible, but do try adding 50-100ohm resistors in series with the MCK, SCK, FSYNC and DATA lines. The CS8412 is pretty intolerant of driving *any* capacitance - I think these outputs are specified for 10pF max(!) so some series-R will reduce ringing and other nasties from interacting with the 1543 input capacitances and so on.

Hope this is of some interest - it's pretty much the sum total of what I learnt 4 years ago playing with a Midiman Flying Calf. Then I modded my CD63 further, which left the Midiman in a steaming crater - while it sounded good it was fundamentally incapable of carrying a tune. It's now a paperweight...

Oh, and sorry - no idea where the DAC idle tone is coming from.

M.
 
Thanks a bunch Martin. I'll be trying your suggestions soon.

What should determine the ratio of cap to resistor value on pin 20?

Are separate analogue and digital grounds a good idea? Reading around this seems a contnetious point, so I've gone with a single common ground plane as I find the simplicity philosophically appealing.
 
Are separate analogue and digital grounds a good idea? Reading around this seems a contentious point
Indeed. I think the real aim of separate grounds is that it provides control of where the currents flow in the planes and so minimise 'ground bounce' interference.

If you can arrange things such that the return currents from the respective digital and analogue sections do not cross, then there should be no difference, because interaction is minimised. This can be hard to arrange. It might be helpful to remember that the return current in a plane flows in a mirror image of the supplied current over it - so you can cut slots and tracks into the ground plane to force currents from certain pins to flow where you want them. Its also one more reason to keep HF decoupling as close to the item served as possible (at the pins), because keeping HF current out of the plane will minimise local ground modulation due to inductance in the plane.

I'll have a hunt for the pin20 info...
 
Suddenly had the urge to do a bit of midnight modding, and then stayed up until 3.30AM watching Sam Peckinpah's Cross of Iron on the telly.

I've swapped the 220nF ceramic cap on the loop filter for a 1uf Wima polyester. The TDA1543 is now bypassed with a 120uF FC/4.7uF cer/100nF cer combo, and has a 1uH between pin 5 and the LM7808.

Topology means that the digital feed to pin 3 of the dac has to be sent via a 5cm or so silver link, so I've soldered a 75R Welwyn RC55 to pin 28 on the CS8412. These have always sounded quite stark and brutal when I've used them in the analogue stage, but somehow that makes me imagine they'd sound pretty good in the digital domain.

These tweaks add up to quite a difference in sound. Bass is heavier and smoother and spatial positioning is much more precise. Images are smaller, and the tone is drier and not obviously warm or cool. There's more acoustic ambience too.

I'd say the sound was more precise and evenly proportioned (it's clear which musical lines are upstage and downstage), and less billowy and romantic. It doesn't really sound like any other source I've heard, so I'm very curious to see if I can push things further in the same direction.
 
Just thought of something.

Solder the decoupling cap to the TDA1543 before any other components. This allows you to flatten the cap leads flat against the body of the dac and keep them as short as possible.

The way I did it was to solder the I/V and Vref resistors first and put the cap on top. With the big Audio Note buggers I use this means the cap leads are about twice as long as they need to be.
 
It's a black art getting these things to sound right :)

Too dry and lean today, so I've moved my Royd RR2s right up to the side walls. A bit lusher and fuller, but the basic character is the same. Very pleased with how stable the imaging is, considering the speakers are almost 5m apart!
 
To improve the jitter reduction the resistors can be reduced substantially or shorted and the capacitor can be increased right up to the point where the receiver won't lock reliably to the incoming datastream. You shouldn't have any problems with 100 ohm/2.2uF here; you can try increasing the capacitance even further. I found 2.2uF on its own OK - it would still lock consistently to my marantz CD63SE.

Martin

Is the resistor on pin 20 actually part of the filter, or is it just to help the CS8412 interface with the cap?
 
It is part of the filter, but the relationship/formula is not in the public domain AFAICT. I think it constitutes damping, so values 100ohm- 1K are probably appropriate. The cap value has far greater effect on how long the dac takes to lock to the transport, i.e how tight the PLL 'window' is.

M.
 
Interesting stuff!

I had planned some late night tweaking but have decided to spend my evening doing other things.

I've got 12 cans of 'Harp' and I plan to get absolutely wasted listening to 12 Gold Bars by the mighty QUO.
Whilst doing this I shall be perfecting my stomping...

For those of you who are not familiar with the 'Quo Stomp', simply think of Oliver Reed on that chat show a few years ago.
 
I had planned some late night tweaking but have decided to spend my evening doing other things. I've got 12 cans of 'Harp' and I plan to get absolutely wasted listening to 12 Gold Bars by the mighty QUO.
Whilst doing this I shall be perfecting my stomping...
Do NOT be tempted to go through the cans while tweaking...been there, done that, have the shrapnel to prove it.

A great way to fault-find in kit you despise though ;)
 
I spent some time with an M-Audio Superdac a while back. It used a CS8414, IIRC.

For a clue as to the maths of the loop filter, I found this: -

Techy maths stuff

Take a look at table 1, page 3. The equations aren't valid for the 8412, but the mathematical relationship between the resistor value and the cap value holds.

The basic rule of thumb is easy, half the resistor value, 4x the existing cap value. You can iterate this process (i.e. 1/4 the original R value, 16x the cap value) until it won't lock, but there does seem to be a sweet spot with the 8412.

The next thing you can do, is change the filter response to a third order, this reduces the jitter peaking that ALL PLL receivers suffer and gives greater jitter attenuation above the knee of the PLL response.

I recommend trying 470R + 0.22u, then adding an additional 3300pF directly between the loop filter pin and ground, via the shortest path possible.

The 8412 data sheet is available here. The 3300pF should go directly between pin 20 (FILT) and pin 21 (AGND).

Look at figure 3, page 8 for the PLL response; below 10^4 Hz (10 kHz) it has NO jitter attenuation, and actually increases it, due to jitter peaking, before the 'knee' of the reponse.

Andy.
 
You're pin numbers in your original long post seem out-of-kilter with the data sheet. You seemed to be stating pin 21 was a +ve supply pin, where the data sheet says it's analogue ground.

Were you on the harp at the time ;)

Andy.
 


advertisement


Back
Top