[dsavitsk]

Quote:

Originally Posted by Nisbeth I'd replace the traces with two long jumpers. It may not be neat but it keeps the ground plane intact

Interesting suggestion. It has a side benefit as well that one can use a seperate offboard PS for the DAC if one is so inclined. To that end, I added an extra ground connection.

 

[dsavitsk]

A few more changes, a little cleaning to the ground plane, a large picture for seeing details.

 

[Francis_Vaughan]

A few thougths.

Is there any chance you can find a smaller transformer? There are plenty about. That one is just huge for your needs.

You should delete the fuse. Use an IEC mains cable connector with an internal captive fuse holder. These are far nicer and safer as well.

Get the mains traces away from the edge of the board. If it goes in a Hammond case the traces will be right next to the board guides. In fact get the mains traces away from everything else you can.

Have you read the PCM2706 DAC thread? We discussed much there, lots about grounding. Also, you may care to check on diyaudio in the digital section. There has been much discussion there too.

But in brief. You need to look very carefully at the ground return currents. For everything. A single ground plane is almost always the right idea. Dividing the groung between digital and analog is falacious. Especially with the highly integrated systems on chip we have here. At any moment current can be flowing from any source through just about any path to ground through any ground pin. Your job is to protect the integrity of the signals in the traces from the receiver to the DAC. These signals will tend to confine their ground returns to the ground-plane directly under them. But since the trace continues past the pad and into the chip - and still continues out fo the chip again to any ground pin, you need to be especially careful not to cut up the ground plane under the chips. If you do you confine the return current to flow through whatever gap you have left. This may force it to wander about too much. Currently there is quite a bit of cutting up under the reciever. The cut on the left hand side is almost perfectly badly placed. You must close that side up totally.

In the 2706 DAC it was all one chip, so attention was put onto a ground-plane as close to the chip as possible. With a two chip solution this is a bit less critical, so a ground-plane on the other side is fine. But keep the vias to it as close to the corresponding pins as you can.

 

[dsavitsk]

Quote:

Originally Posted by Francis_Vaughan Is there any chance you can find a smaller transformer? There are plenty about. That one is just huge for your needs.

There is a step smaller in this line, though it does not use a whole lot less board space. I'll see how much difference it would make.

Quote:

Originally Posted by Francis_Vaughan You should delete the fuse. Use an IEC mains cable connector with an internal captive fuse holder. These are far nicer and safer as well.

Unfortunatly, there is no room on the case for a large IEC -- see below.

Quote:

Originally Posted by Francis_Vaughan Get the mains traces away from the edge of the board. If it goes in a Hammond case the traces will be right next to the board guides.

I would never use something so gauche.

It is going in a mahogany box the same size as a Grado RA-1 (see the middle box on my avatar). A regular IEC just fits, so the large one with a fuse won't, and there is no room for a panel mount fuse holder, so either it has to be on the board, or floating inside. You may be right that a smaller transformer would help here, though.

Quote:

Originally Posted by Francis_Vaughan Have you read the PCM2706 DAC thread? We discussed much there, lots about grounding.

Do you have a link? I couldn't find it.

How different are the 2706 and 2707? Am i using the wrong one here?


Quote:

Originally Posted by Francis_Vaughan But in brief. You need to look very carefully at the ground return currents. For everything. A single ground plane is almost always the right idea. Dividing the groung between digital and analog is falacious. Especially with the highly integrated systems on chip we have here. At any moment current can be flowing from any source through just about any path to ground through any ground pin. Your job is to protect the integrity of the signals in the traces from the receiver to the DAC. These signals will tend to confine their ground returns to the ground-plane directly under them. But since the trace continues past the pad and into the chip - and still continues out fo the chip again to any ground pin, you need to be especially careful not to cut up the ground plane under the chips. If you do you confine the return current to flow through whatever gap you have left. This may force it to wander about too much. Currently there is quite a bit of cutting up under the reciever. The cut on the left hand side is almost perfectly badly placed. You must close that side up totally.

I have beeen trying to get things off of the ground plane, but I am really stumped on the left. Do you have any suggestion as to where they can go? Everytime I move them up, someting else moves down that is worse than what's there. A 3rd layer would help, but since that is not an option I am unsure how else to get the 3.3v over to that side of the board. I'll keep moving stuff about and see what happens.

Thanks for the suggestions.

 

[bigmike216]

How about using a wall wart or line lump stype transformer? Or, alternatively, making your own line lump in a small project box? It can just sit on the floor.

It'll get the transformer right outside of your box..

 

[dsavitsk]

Quote:

Originally Posted by bigmike216 How about using a wall wart or line lump stype transformer? Or, alternatively, making your own line lump in a small project box? It can just sit on the floor. It'll get the transformer right outside of your box..

The thing is, you are correct. It makes total sense and is the "correct" solution, except that I am allergic to wall warts -- won't have them in my house.

Back to the smaller transformer -- it is 5mm smaller per side and is actually getting low on the amount of current it can supply. If I were to ever, say, cram a buffer in the box, the smaller transformer looks like it might be stressed. Probably it would still be enough, but the overhead seems like a good thing. My experince is that overspec'd transformers can do some sonic good.

Here is today's update. I moved around the I/V to get the traces all to the top. I also moved a couple of other traces around the chip to the top.

 

[tangent]

I would remove one of those 1000uF power supply filter caps, allowing you to move the entire regulator assembly back away from the big SMT chip. As it is, you have unregulated power too near the SMT pins. And, 2000uF of filter capacitance is more than plenty. I'd even consider removing two of those caps, if you can find a way to use the space.

16V might not be sufficient for those PSU caps. Amveco says the unloaded voltage for that transformer is 13V, and rectification will push that to over 18V. Now, in your design, the transformer is always loaded, so the question is whether the load will bring the transformer voltage down enough before the rectifier. I wouldn't change anything right now, but if the caps explode on startup or you measure the peaks with an oscilloscope and find them over 16V, swap some 25V caps in there. (You do have access to an oscilloscope, I hope...)

Since space is tight, I'd definitely use a monolithic bridge. Not only are they smaller, but the diode matching is likely to be better than a collection of random discretes, so there will be less switching noise.

Speaking of switching noise, I don't see any snubber caps on that bridge...

I don't like the circuitous trace from the N pad to the cap. Perhaps the easiest fix is to move the L pad so it uses the left pad of the fuse holder instead, and then put the N pad where the L one was and try to squeeze the trace between the fuse and the mounting pad. If there's not enough room for that, it looks like you can make some room. The filter section is rather loosely packed. Also, you could try switching to an 0.047uF cap in the line filter.

I'm not sure it makes any difference, but I would prefer to put the oscillator's load caps between the osc and the IC, simply for inductance reasons.

Finally, I like halman's idea of using the second winding separately instead. If you go with the 70001 transformer, fewer filtering caps, and the monolithic bridge, you may have enough room for two filter and regulator sections. You'll probably need to use an LDO on the 8V side. You should be using high-performance regulators anyway, due to the high speeds this circuit will be running at. Ye olde 3-terminal regulator is generally only good out to 10 kHz. Beyond that, you're depending 100% on the bypass caps. A good regulator can help out a lot.

 

[dsavitsk]

Quote:

Originally Posted by tangent 16V might not be sufficient for those PSU caps.

I caught that too. 25V will fit fine, so I think it is not worth the risk.


Quote:

Originally Posted by tangent (You do have access to an oscilloscope, I hope...)

What's that then? I do, though I am not sure I'd know what to do with it.

Quote:

Originally Posted by tangent Since space is tight, I'd definitely use a monolithic bridge.

It didn't actually save any space here. It is not obvious, but the diodes are meant to be tombstone'd, so the bridge is pretty small.

Quote:

Originally Posted by tangent Speaking of switching noise, I don't see any snubber caps on that bridge...

SMD under the diodes.

Quote:

Originally Posted by tangent Finally, I like halman's idea of using the second winding separately instead. If you go with the 70001 transformer, fewer filtering caps, and the monolithic bridge, you may have enough room for two filter and regulator sections. You'll probably need to use an LDO on the 8V side. You should be using high-performance regulators anyway, due to the high speeds this circuit will be running at. Ye olde 3-terminal regulator is generally only good out to 10 kHz. Beyond that, you're depending 100% on the bypass caps. A good regulator can help out a lot.

For regulators I am doing dual regulation using a to-220 LM7809/5 to do the first round, then a TO-92 low dropout for the final bit. Peter Daniels swears by these, so I thought I give them a try.

Anyway, I did switch to dual supplies and a smaller transformer.

 

[tangent]

Quote:

Originally Posted by dsavitsk Quote: Originally Posted by tangent (You do have access to an oscilloscope, I hope...) I do, though I am not sure I'd know what to do with it.

There are several places in this circuit where it's easy to know in advance what sort of waveforms you should see. You should check all of them, even if the circuit appears to be working. It's a high-speed circuit.

Quote:

[A monolithic bridge] didn't actually save any space here. It is not obvious, but the diodes are meant to be tombstone'd, so the bridge is pretty small.

Eck... I'd much rather use the monolithic than resort to tombstoning diodes.

One more comment: consider switching to the 290 series of ELF chokes. They're better at high frequency, it's very nearly the same size as the 16Ms, and they're available in slightly higher inductances.

 

[Francis_Vaughan]

OK, a few more thoughts. Back to the left hand side of the 2706.

You need to look at the function of the various pins and the traces connecting them. Some are critical to the success of the design, others are very tolerant to how they are treated. Currently the priorities are inverted.

The most crucial trace you need to protect is the bit clock. But it isn't just the trace, just as important is the ground return for that trace. Remember, as much as possible the return current will attempt to stay as close to the signal trace as it can. But it can't because of the +V supply to some of the other pins on the 2706. The worst offender is the supply to pin 14 - the DT pin. But this pin is only used for the external ROM, and just needs a pull up. You can get that to it by whatever circuitous route you can find after the fact. It won't matter. But at the moment the ground return for the bit clock is forced to go around the cut in the ground plane.

Worse, the cut is about equally long either side of the bit clock - so current will probably flow either side - if any return current flows on the lower side of the cut it will be unavoidable mixed with the ground return currents for common mode noise on the USB input. Effectively placing USB noise onto your bit clock - which is the reconstruction clock for the DAC. This will substantially damage the S/N performance if nothing else.

At least remove the cut in the ground plane that feeds DT, and move the feed to DT to wherever.

But I think we can do better. Strangely at the moment the input signals from the USB are far better treated although they are vastly less important. I'm going to suggest that you rotate the 2706 90 degrees right. This is going to make a mess, but we can sort it out. Doing so allows you to get the critical bit clock trace right next to the DAC. Don't worry to much about the USB inputs, they can run along the left side of the board and back under the connector. You will probably need to push the USB connector a bit further down. There is also some scope for tightening up the crystal oscillator layout. But it is not too bad now.

Now turn attention to the PS and de-coupling. The layout violates one little rule. The power goes to the cap first, then to the pin. Otherwise there is a little tiny bit of trace that adds some inductance. If you rotate the 2706 and get the USB connections out of the way you have a lot more scope to carefully route the power.

You do need to get rid of the cut in the ground-plane under the USB input pins. The USB signal is differential - so the actual USB signal does not need any ground return, but any external crud that comes from outside and, in particular, any common mode noise on the USB lines will need a solid ground return, otherwise it will spray about the circuit and get into other signals. At the moment it is a great conduit for noise inside your computer's case to get into the circuit.

If you do find it unavoidable to pop a trace down onto the ground-plane side, keep that part of the trace as short as you can, and then pop back up. There is no need to cut the ground-plane up gratuitously. Don't feel shy about using the gap under a component to run a less critical trace.

 

[dsavitsk]

Quote:

Originally Posted by tangent Eck... I'd much rather use the monolithic than resort to tombstoning diodes.

Allright, you talked me into it. Do you happen to have a decent datasheet for the W01G, or at least know the pin measurements off hand? The datasheets I found had quite a range for the pin spacing. Is it best to just split the difference and bend what needs to be bent?

 

[dsavitsk]

Quote:

Originally Posted by Francis_Vaughan OK, a few more thoughts. Back to the left hand side of the 2706.

I just want to make sure I am taking this in, and that I can implement your suggestions: To be clear and so we're on the same page, this is a PCM2707, not the 2706.

 

[Francis_Vaughan]

Yes, sorry, they are on the same datasheet. I wasn't watching. There are minimal differences, but indeed my comments about the DT pin only make sense for the 2707. On a 2706 the pin is left open. Which raises the question - you could just swap to a 2706 and forget the pain of that pin. Since you are not using the serial programming interface there is no additional functionality in the 2707 that you need. They are otherwise identical chips.

 

[dsavitsk]

That certainly makes things a little easier.

I'll see how the 90 degree turn looks. So far, it is not too bad other than the traces from the usb go all over.

 

[Francis_Vaughan]

Rats, I had another look at the datasheet - and it is ambigous. Or rather poorly worded. I suspect you need to retain the pullup for DT even on the 2706. The pinout for the PJT package was mute about the pullup, but the pinout for the DB package explicitly says it is needed on the '06. So I think we can assume that it was an ommision on the PJT page, and you do need it. But as I said, don't worry about this pin until last, it isn't critical how it is fed. You could always jump the WS line to the DAC (pin 2 on DAC) with the pullup resistor. Then send the trace around the other side.