Advisable PCB solutions for a DAC

[Sandor]

I am going to design a PCB for a CS8412 + TDA1541 DAC. I can only manage a double sided PCB for a two layer routing (no internal layers).
I have two ideas:

First idea
Bottom layer is used for all connections. Analog ground connections and digital ground connections adopt a bus schema (two separate buses).
Top layer is used as a ground plane, insulated from all components. The ground plane on top is connected, in a single point, to the analog and digital ground buses on the bottom layer.

Second idea
Top layer is used as a ground plane, analog and digital ground connections are directly soldered at the component pins; so ground connections spread all over the plane and there is no separation between analog and digital ground.
Bottom layer is used for all the remaining connections.

Are these "good" ideas? And if yes, which is better?
What about swapping the side role (i.e. ground plane on the bottom layer) ?

I'd greatly appreciate suggestions from you skilled friends (I am an absolute beginner...)
Best regards.
Paul

[Guido Tent]

Originally posted by Sandor

I'd greatly appreciate suggestions from you skilled friends (I am an absolute beginner...)
Best regards.
Paul

Hi Paul

This may be of help

http://www.tentlabs.com/Info/Article...decoupling.pdf

best

[Sandor]

Hello Guido,
thank you for the valuable link.
I have read your paper about decoupling and now I guess an answer to my own question:
The first idea (ground connections routed as buses and then connected in a single point to the ground plane) leads to long return loops. So the second idea is more advisable. Correct? But, is it the best I can do with a two layer PCB?

I have now a few more questions:
My circuit is very similar to the ones found in this ng. For each IC supply I use local TL431-based regulators. A T-filter (LCL) is inserted between power supplies and the local regulators input . Is a ferrite bead necessary between the TL431 output and the IC supply pin? Is a VK200 bead well suited for this job?

Best regards.

Paul

[Extreme_Boky]

Better use the bead before the 431.

Otherwise, you can end up with unwanted ringing which is function of soooo many things: track length/width between 431 and Vcc pin, bead+wire inductance, Vcc pin impedance, layout, capacitors... It is hard to have a clear picture and understanding of the complexity – I’ve been trying, though.

I always try to keep everything after the regulation as simple and as short as possible, with only one local decoupling cap located as close to Vcc pin as possible, with shortest possible lead length to ground. You have to have very clean power supply rail to start with + very low noise ground plane of large area! And think current charge/ discharge between regulator and decoupling cap + current flow from the cap to Vcc pin. I was under impression for a long time that small currents do not need this approach, but was wrong! Have implemented this on already very good 1541A NOS design with astonishing results.

I got carried away here, so ease to get emotional

Extreme_Boky

[Guido Tent]

Originally posted by Extreme_Boky

I always try to keep everything after the regulation as simple and as short as possible, with only one local decoupling cap located as close to Vcc pin as possible, with shortest possible lead length to ground. You have to have very clean power supply rail to start with + very low noise ground plane of large area! And think current charge/ discharge between regulator and decoupling cap + current flow from the cap to Vcc pin. I was under impression for a long time that small currents do not need this approach, but was wrong! Have implemented this on already very good 1541A NOS design with astonishing results.

I got carried away here, so ease to get emotional

Extreme_Boky

The advantage of a shunt is that you can place the sunting part (ie the 431) very close to the load, so no need for ferrites

Best to feed the shunt with a current source

best

Guido

[Extreme_Boky]

Yes I knew that will come

I have wrote CCS for each shuntleg. A CCS will set the current for each TL431 for about 50mA + Chip current

I can put in resistors-> because each CCS can be a serie reg instead-> up to people

Thank you both

Kim

Hi

Good !. A small resistor + bead in series with the CCS will maintain high impedance beyond the 1st pole of the CCS.

best


__________________
Guido Tent

The advantage of a shunt is that you can place the sunting part (ie the 431) very close to the load, so no need for ferrites

What the ...

Extreme_Boky

[Guido Tent]

Originally posted by Extreme_Boky





What the ...

Extreme_Boky

Oh

No need for ferrites at the output of the shunt, assumed placed close to load. Please use ferrites at the input of the shunt.........

[Sandor]

The outputs of the shunt regs are just a few mm from the supply pins; the decoupling caps are directly soldered on those pins on the bottom layer of the PCB and connected to the ground plane on the top layer. The shunt regs are feeded by LM317/LM337-based preregulators.
At the input of the TL431 I use a t-filter from RS as the one showed in the picture.
Quesions:
1) Should The ferrite bead substitute the t-filter or should I use both?
2) Can you suggest a suitable single IC as a CCS? (the LM334 seen on the ng cannot supply enough current for the TDA1541 and CS8412).

Best regards.

Paul

[Extreme_Boky]

1) Should The ferrite bead substitute the t-filter or should I use both?

T-Filter is a bead in series - capacitor to ground - bead in series configuration. I don't like them because they have steel legs + I do not have a full control over those 3 components contained within a T-Filter!. You can use solid core copper wire with a bead. Or pure silver solid core wire, annealed. Stay away from silver plated copper wire!! The type of ferrite bead and its length can be determined by measuring the noise level and spectrum at voltage regulator pins, and at Vcc pin. You can also put 2 beads in series to achieve close to double the noise attenuation.

I always try to achieve lowest noise with the least amount of components. Beads are usually needed around noise "generators": uPC, DAC's and clock oscillators, but they can and should be avoided if you have a very low noise design (read: NOS!!!) + good ground topology.

Can you suggest a suitable single IC as a CCS? (the LM334 seen on the ng cannot supply enough current for the TDA1541 and CS8412).

I think Pedja Rogic used 431 (more current!), and connected + to common!!! See Pedja Rogic 1541A DAC. 431 will give you enough current for all your 1541 / 8412 needs.

Extreme_Boky

[Giorgio]

Very interesting project and discussion indeed!

Now a have question: it's mainly for Guido, as author of the very clear and specific document referred to in a previous mail, but I would appreciate to have the point of view of anyone else...

I see that Guido suggest to have return currents to flow through the ground plane, after filtering out high frequency components; BB uses this technique too, but AD does not use it and I remember that at the time of CS4929 Crystal did not use it either.

So I must assume that the question is controverial.

In the past I used both solutions, but I have not been able to understand which is the best.

Clearly in this discussion the fact of working on a mixed signal environment (very high frequency and very low power digital signals, and very low frequency and sometimes very high power analog signals) can be an answer.

The questions for Guido are:

1) would you suggest to use the technique of the ground plane as return path also for
- a low current, discrete components, purely analog audio stage?
- the PSU and regulation stages of a low current, discrete components, purely analog audio stage?
- a high current, discrete components, purely analog power audio stage (obviously assuming the currents do not destroy the plane... by the way, the skin effect on ground plane should be very limited at audio frequencies)?
- the PSU of a high current, discrete components, purely analog power audio stage
or in any of these cases you would better use the old star ground concept?

2) Should you prefer the ground plane, it is necessary to filter out any audio frequency current in the return paths, and how? or, from your experience, we can assume that, even though perhaps not perfect, the ground plane as a return path even at low frequencies is superior to star ground? for all and any aspect?

3) The main problem with split ground planes is that often there a thin ground connection in which an high return current flows. In case it is possible to decouple the DAC and the following audio stages for example with a transformer, would you still suggest a ground plane approach for the audio stage?

mmmhh... better to be very clear: I am just trying to learn and understand, not to discuss or be polemic....

Thanks in advance for your help

Giorgio

giorgio@tnt-audio.com