Would it make sense to mate a Duet rcvr w external I2S DAC?

[Pete Fowler]

I will be more than happy to pass on my knowledge, as my time allows. You came here to share and learn. I can help with the latter.

Thank you! I just hope I can keep up...

OK......jitter.......

In a nutshell, the PSRR of a simple CMOS gate is -6 dB at the midpoint. Both halves are going from the full on to full off state. So, each are on (or off) by the same amount.

Translation: the output is an open spigot to whatever noise is on the supply rail. And you can bet there is a lot of noise on the Xilinx supply. I don't know diddly about that chip, but in general, the less ground pins, the more crud you will have to deal with. With only one supply or ground pin, everything has to go through it. With its intrinsic inductance. Ergo, curd on the rail. And possibly ground bounce, as well.

For a 256 x Fs clock, those of us who design this stuff all seem to agree that the jitter has to be in the single pSec or so range. There is no way that you can get jitter that low out of any large chip. It is hard enough using single gate devices.

You've brought up several items to ponder. In order to keep it manageable (for me) how about we work on possible jitter reduction first (clock, chip or PS induced), and assume the onboard DAC will be used for simplicity? Similar to when my physics professors would "assume zero friction" to simplify a homework problem? ;-)

And since you want both 44.1 kHz and 48 kHz operation......if you think about what a nightmare it would be to have 2 clocks after the Xilinz, and then find some way to switch them........well, now you know why Sean has the Xilinx in the first place. Catch 22.

OK, your turn to digest and ponder.

Pat

And away we go...in an earlier post I asked if the Xilinx was the weak link in the system (or perhaps limiting factor might be better) and I think you've just confirmed that. On the other hand, it is a nice way to allow toggling between CD and streaming FM. A reasonable tradeoff even if it limits the ultimate jitter performance of the system.

Pat - you mention above that most designers agree that a very low jitter (single pSec) is the goal. Is it an all or nothing goal? In other words, is there any benefit to say, cutting the jitter in half? If not then we can stop here on jitter. If there is merit to reducing jitter by some amount then it's worth pursuing this (to me, at least).

It's pretty clear I'm stuck with the Xilinx since adding a way to toggle a downstream reclocker is prohibitively complex (and completely beyond a noob). Fair enough?

Fortunately, the Xilinx may not be as bad as you fear (based on the specs - no hands on experience). I read the data sheet (XC9536XL) and the chip has 3 separate power and 3 separate gnd pads, so it's possible it may be a well-behaved design (if we're not aiming for $5,000 DAC performance).

That's the chip limitation, so what about power?

Unfortunately it appears the Duet's PCB routes the 3.3V supply to the Xilinx in the middle layer. I see no way to isolate the Xilinx and feed it clean 3.3V power (or at least I haven't found it yet). Thus I think cleaning up the power supply for the Xilinx in not in the cards unless you have some tricks up your sleeve that a noob would not consider. There are decoupling caps on the back of the PCB mounted opposite the Vcc pins - a perfect place to tap in a clean supply. But there's no way to isolate the dirty power. Strike two.

That leaves adding a low jitter clock upstream of the Xilinx and hoping for the best. The cost and effort involved in swapping the stock oscillator for an XO and clean power supply is not bad from my perspective. Would it yield any benefit? I can't answer that (dang ignorance, again).

My first thought is to ask whether the jitter inherent in the Xilinx is affected by the incoming jitter from the oscillator - would a cleaner clock reduce the jitter from the Xilinx? Is jitter in this situation additive, multiplicative or absolute? If less jitter in equals less jitter out, then it makes sense (to me). If not, strike three.

And that's it w/respect to the Xilinx and jitter. Did I miss anything?

If not, then could we look at the possibilities on the other side of the I2S bus (both internal and external)?

Regards,

Pete

[ar-t]

Wow.......a lot to deal with at one time.

OK, 3 power and ground pins. Divide the number of functioning gates by 3, and see what you get.

Clock jitter: no matter what you feed the Xilinx, it will come out noisier. No way around that.

That single digit goal: best to call it what it is. Not all jitter is alike, so let's just concentrate on Gaussian type. The goal is to reduce it to that range, as that is the area where most designers feel that you can no longer hear its effects. You don't actually hear the jitter (common misconception), you just hear how it mucks up stuff. Usually loose bass and strident highs. It simply gets better as the absolute number goes down. Below a certain point, you can not hear any improvements.

OK, my turn to query you.

What are you hoping to gain by this? The DAC chip in the Duet is not a bad sounding one, on its own. Ever think of working on improving it, and then moving "up" to a "better" DAC? The are things that can be done to it that are well within your grasp.

Pat

[Pete Fowler]

OK, my turn to query you.

What are you hoping to gain by this? The DAC chip in the Duet is not a bad sounding one, on its own. Ever think of working on improving it, and then moving "up" to a "better" DAC? The are things that can be done to it that are well within your grasp.

Pat

Uh...no I didn't think of that - at least not as the first place to focus. I was (good or bad) thinking in terms of cleaning up the data as much as possible before it got to the DAC (internal or external). Once garbage gets in the signal, it's in forever, at least in analog playback. Then (as you mentioned earlier) there's the problem of getting said data from point A to point B intact. Lots to learn.

I think many DIYer's lack a system perspective (for whatever reason). Where will mods actually make a worthwhile difference vs. just throwing things blindly at the wall to see what sticks. If you think the internal DAC would be the best place to focus I'm all ears (or eyes as the case may be)...

Pete

P.S. Thank you for the jitter threshold explanation.

[ar-t]

You are not only listening, but comprehending. Both are good, especially together.

Ok, here is what I see as the most common DIY mistake: making a change for change sake. Call it lack of system perspective, or whatever. Just swapping out parts (usually over-sized and over-priced caps) is the usual first step. Without regard as to why they are even there in the first place.

OK, let's stick with the internal DAC. What changes could be made, and why? Let's not bring up the coupling caps.......

This one may be off the wall, especially since it can not easily be changed. But since we are discussing supply noise, here is something that has lots of room for learning.

All circuits have to have a return path to ground. That piece of coax that you want to run to an external DAC chip..........the supply bypass cap on it (or any chip) has to return to ground. What is the right way?

Well, lots of wrong ways, and lots of so-so ways, but not many really good ways. You have to view the return path in a 3 dimensional manner. The bypass cap couples energy from the supply pin back to the ground point of the chip.

So, look at how the bypass caps are arranged on the DAC chip. A small cap in close, and a larger one further out. Both go to the ground plane. Now follow how that energy goes from that point to the ground pin of the chip.

Not a straight line, is it? Or the shortest. (A pet peeve of lots of knowledgeable designers is the positioning of the supply and ground pins on the chips. They are almost never side-by-side.) OK, not much we can do about that, but we can look for ways to make it shorter.

I have something that I want you to read. Go to Tentlabs (yeah, the guy who makes expensive clocks) and rummage around for a technical paper on proper grounding. Trying to find the URL so that you don't have to read his sales stuff.

OK...found it.......http://www.tentlabs.com/InfoSupport/...decoupling.pdf

It gives some really good info on how to decouple/bypass chips. But more importantly: why.

I haven't read it in years, but I believe that he also talks about paralleling bypass caps. Long story short, doing that almost undoubtedly reduce the bypass' effectiveness at some frequency range. With luck, it won't be in a range that hurts. Luck isn't something to count on in this realm.

In any case, there is a ton to be learned from that. Read it and digest it. Then think of some questions that pop into your head. There will be lots.......

Pat

[Pete Fowler]

You are not only listening, but comprehending. Both are good, especially together.

Thanks for the vote of confidence on comprehension Pat, but I think I'm a better listener at this point... ;-)

I ended up with 4 browser tabs open - your post, Tent's article, the WM8501 data sheet and a high-rez pic of the Duet's internals. Oof!

So, look at how the bypass caps are arranged on the DAC chip. A small cap in close, and a larger one further out. Both go to the ground plane. Now follow how that energy goes from that point to the ground pin of the chip.

Not a straight line, is it? Or the shortest. (A pet peeve of lots of knowledgeable designers is the positioning of the supply and ground pins on the chips. They are almost never side-by-side.) OK, not much we can do about that, but we can look for ways to make it shorter.

In any case, there is a ton to be learned from that. Read it and digest it. Then think of some questions that pop into your head. There will be lots.......

Pat

You are right - lots of questions. Starting at the 10,000 ft view:

1. Based on Tent, the data sheet for the Wolfson is recommending both good and bad practices?! The good - place bypass caps as close to the DVDD, AVDD and VMID pins as possible and use very low ESR caps.

Bad - they don't mention keeping short paths to the respective ground pins, and they are recommending two caps per pin when Tent indicates only one "correct type" is really necessary or even desirable due to the potential added current paths.

2. Wolfson also does not mention decoupling the AVDD and DVDD lines with resistors or ferrites. So any crap on the rails comes directly into the chip?!

3. Per Wolfson, after I thought about it, they should over-emphasize the need for a clean VMID pin. If the Vref is jumping around due to induced currents it's exactly like injecting jitter directly into the DAC circuits, yes? The transition point will vary randomly with time, depending on the hash on the pin - I believe that's jitter as I understand it?

4. Given the above, the guys who laid out the Duet PCB didn't optimize for minimal current loops (but who can blame them given the Wolfson specs?). As you said, nothing to be done about that, but educational.

Coming down a few thousand feet...

1. As you point out the .01 caps on DVDD, AVDD and VMID all attach to gnd about as far away from their respective GND pins as possible. The secondary bypass caps are even further away. Not good - lots of opportunity for loops.

Possible mod - attach a single(?)bypass cap (type TBD?) directly between power and gnd where possible (esp DVDD and VMID). Remove the existing bypass caps to eliminate current loops. Based on the Wolfson data sheet one could attach an SMD cap directly between pins 10-11 (if the act of soldering wouldn't damage the chip). Also, routing the bypass cap from VMID directly to AGND looks do-able (to my noob eyes)?

All one needs is a good quality solder station and the hands of a neurosurgeon. Oy. ;-)

2. 'Nuther potential mod - if possible desolder or cut the AVDD and DVDD pins and insert a resistance or inductance in there somehow. Again, only neurosurgeons need apply.

3. Then (but only then) think about a very clean dedicated 5V supply for the DAC chip or replace the existing Duet 5V supply with a cleaner one since it doesn't seem to power much more than the Wolfson? Not sure here.

4. Issues like proper ground plane layout and PCB trace routing are off the table since I'm working on an existing design.

Then some mouse-nuts issues...I notice ferrites on the output lines to the analog RCA jacks. Pat, you mentioned removing them from the S/PDIF line to improve the signal, does a ferrite on an analog output matter in your experience?

I also read in the Wolfson specs that it has an internal analog low-pass filter. Since I'm running a tube linestage with transformer-based attenuator I'm not so worried about high freq hash escaping the Duet and trashing my amp. Could most of the Duet's external low-pass filter be removed? My thinking here is that less is more.

I've hit the limits of my understanding for the moment. Thanks for the push!

Pete

[ar-t]

Oy, indeed. But you are a fast learner.

OK, briefly: there are ferrites on both supply pins. Yep, think about a way to make one effective cap go from Vcc to Vss. "Effective" eliminates all audiophile caps.

Here is one to ponder:

They make high-K ceramics that go as high as 22 uF. I know designers who swear by them. And others that swear at them. All have very successful designs on their resume, yet can not agree on bypass caps.

The good point: lots of C, decent self-resonant frequency, so very effective. Only need one. Bad point is that the capacitance changes with voltage. Of course, it does as ceramic is microphonic. (It works both directions.) As some will point out that ceramic is just plain evil.

You have to try and learn on your own. Your experience is more valuable than opinions of experts. In either case, it provides a learning opportunity.

OK, there are 2 points that I have made in other threads. I think Sean has read them, and is biting his tongue. So, in case anyone missed them, here goes again.


I like ferrites, but you have to be very careful where you use them. I personally would never stick one in the audio path, but I am not making millions of units that have to pass EMC standards. You can probably remove them. Make sure that you only do one thing at a time when doing this. (Actually, applies to all mods. Even hot rodding. Real hot rods, which no one has any longer.)

DO NOT REMOVE THE OUTPUT CAPS.

[ar-t]

Oy, indeed. But you are a fast learner.

OK, briefly: there are ferrites on both supply pins. Yep, think about a way to make one effective cap go from Vcc to Vss. "Effective" eliminates all audiophile caps.

Here is one to ponder:

They make high-K ceramics that go as high as 22 uF. I know designers who swear by them. And others that swear at them. All have very successful designs on their resume, yet can not agree on bypass caps.

The good point: lots of C, decent self-resonant frequency, so very effective. Only need one. Bad point is that the capacitance changes with voltage. Of course it does, as ceramic is microphonic. (It works both directions.) And some will point out that ceramic is just plain evil.

You have to try and learn on your own. Your experience is more valuable than opinions of experts. In either case, it provides a learning opportunity.

OK, there are 2 points that I have made in other threads. I think Sean has read them, and is biting his tongue. So, in case anyone missed them, here goes again.


I like ferrites, but you have to be very careful where you use them. I personally would never stick one in the audio path, but I am not making millions of units that have to pass EMC standards. You can probably remove them. Make sure that you only do one thing at a time when doing this. (Actually, applies to all mods. Even hot rodding. Real hot rods, which no one has any longer.)

DO NOT REMOVE THE OUTPUT CAPS. There is lots of HF energy if you do. Now, can you change them? Well, the app note clearly says to use C0G types. Uh, hate to say it but...........them caps that they got on there now sho'nuf look too small for C0G. 'Nuf said.

Bypassing the Vref pin may help. A lot of that will depend on how clean your supply is. I am going to evade answering that directly, as I have also touched on that elsewhere. I know that there are "experts" out there who mine this (and other) forum(s) for free design help. I'll let you try that one on your own, and we can compare notes in 2 months once we see a plethora of mods being sold that have that in them. (It will happen, trust me!)

OK, your turn to digest. My turn to make a $.

Oh, one last point. This will make some folks mad, but tough.

App notes........not always the right place to get accurate design help. As Guido points out in his article, the practice of one 'lytic and one ceramic has carried over from decades ago. A lot of times (and not always, so no hate mail!), companies let junior engineers do these app notes. And most of them, being green, having no real experience and a crappy education to boot, repeat the past. Partly out of ignorance, partly out of not wanting to rock the boat. (Most are not smart enough to rock the boat, so in a sense it is a moot point.) The point being is that app notes can be a gold mine of info and a pile of dung. Sometimes both at the same time. The WM8501 is an example. Typical bypass schemes, possibly inadequate C on the Vref pin, yet smart enough to tell you not to use high-K ceramics on the post filter.

Ya gotta know how to read between the lines on data sheets.

Pat

[Pete Fowler]

OK, your turn to digest. My turn to make a $.

Pat

Pat - many thanks for taking the time to explore all this. I understand your need to make a buck so I'll go dark for a few days and see what I can come up with on my own (with an excellent launch assist).

Being that Mr. Tent seems to know his stuff I was planning on using his shunt regs if and when clean power supplies are needed. I could not get close to his results on my own and you are making a living at this so I think his regs are an acceptable solution.

Regards,

Pete

[ar-t]

I know absolutely nothing about his shunt regs. You may want to try one, assuming his prices are ok, for the DAC. +5 V DC @.....I dunno........10 mA or so. Shunts are only a good choice for low current draw. Unless you like stuff to be really warm.

But as for regs in general......

The simple zener diode on the base of a follower gets pooh-poohed a lot. Usually for no good reason. There are better ways to get a reference voltage, but in any case, if you RC filter it enough, it works 99% of the time. Sure, the output Z is higher than a feedback design, but it is flat throughout the audio band. And to quite a way higher. Eventually, after the alpha cutoff frequency or whatever, it rises. So does a feedback regulator. Which a lot of folks gloss over.

One advantage, although I am not fond of it, is that you don't have to worry so much about how much capacitance you stick on the output. Some guys love the sound they get by sticking 1000 uF on the output of one. Stick something that large on a typical regulator, and you could be looking at severe noise peaking.

So, in a nutshell, a simple follower with a quiet reference on the base is a very handy little circuit for jobs like this.

And cheaper than Guido's! (Sorry, Guido.)

Pat

[Pete Fowler]

They make high-K ceramics that go as high as 22 uF. I know designers who swear by them. And others that swear at them. All have very successful designs on their resume, yet can not agree on bypass caps.

The good point: lots of C, decent self-resonant frequency, so very effective. Only need one. Bad point is that the capacitance changes with voltage. Of course, it does as ceramic is microphonic. (It works both directions.) As some will point out that ceramic is just plain evil.

You have to try and learn on your own. Your experience is more valuable than opinions of experts. In either case, it provides a learning opportunity.

Hey Pat,

Yep - back to ask more extremely basic noob questions...

I've read up on multi-layer low ESR bypass caps (including your suggestion of high-K ceramics) and I think I have a basic understanding of the tradeoffs involved.

EDIT: Never mind - a large capacitance Z5U cap won't come in SMT form (or I haven't found any). So I'm looking at a thru-hole part. Also looking at Z5U since this is intended for bypass, not signal or other use. Doh! ;-)

Looking at the Tent paper (again) I think the trick will be to solder the cap directly to the DAC gnd pin(s) and then to the respective VMID, DVDD and AVDD pins.

Comments?

My thought (now) is to use 22uf bypass cap on the power pins (if I can afford them - not cheap).
I learn slowly...

Rgds,

Pete