TOSLINK DAC

The WM8804 decoder looks like a winner. The audio sounds just as good as before (which is hardly a surprise), but when muted there’s no hiss anymore.

Not only that, but the 8804 seems to be a chip very much in demand today - it’s one of the favorites for adding on to a Raspberry Pi.

So unless there are anymore surprises, I think this may be a final design.

If there’s a possible complaint it’s that the output levels may be just a tad hot. A 1 kHz 0dBFS sine wave winds up being about 6v P-P. Correct line level would probably be closer to what the CS4334 produced - about 4V (keep in mind we’re talking about an absolute peak level here).

For me, I’m satisfied to simply turn down the volume on the amp a little bit. I really don’t want to attempt to reengineer the analog out stage. If I were forced to reduce the levels, I’d probably just add an output pad. But I don’t think I will.  

At this point, I'm pretty much sold on the DAC/analog sections of the design. While it's undoubtedly possible to do better, I'm satisfied with the performance. I'm still a little bit put out by the idle noise, so I am going to try another receiver chip to see if it helps. The new nominee is the WM8804. It's an SSOP-20 package instead of QFN, which is easier to deal with. One big difference is that this chip requires a 12 MHz crystal. Hopefully on PLL unlock it can supply either no clock or a sufficiently fast clock that the DAC won't be noisy.

v1.4.1 is back from OSHPark. It's almost perfect, so far as I can tell. There's just one interesting issue. If you remove the optical input, you get a noticeable white noise output. This is despite the MUTE pin being pulled high by the ERR output of the CS8416. This didn't happen with the DIR9001.

Now, when the PLL unlocks, the master clock output is 750 kHz, which results in something like a 2 kHz LRCK, which is far lower than the minimum input to the PCM1793. But damnit, MUTE is pulled high, so none of that should matter. ERR is driving both the error LED and MUTE, and as a result, it's a little bit low - about 2.8 volts or so. But the PCM1793's minimum threshold for high is 2 volts, so it still should be enough.

I guess a workaround for this would be to supply an actual replacement clock to OMCK. In hardware mode, the CS8416 will supply OMCK on RMCK (and will derive BCK and LRCK from it) if it's available. If not, then you get the unlocked PLL clock of 750 kHz.

I don't think I'll bother, though. If I were to do anything more at this point, I might attempt to replace the CS8416 with an alternative, but I don't think I have a great deal of incentive to do so since the PCM1793 itself isn't rated for higher than 200 kHz sample rates, and I certainly have no use for anything higher than that.

I got v1.4 boards back from OSHPark. This is the version with the CS8146 in a QFN footprint. At first the prototype didn't work, but swapping out one of the PLL filter caps fixed it. I must have picked up the wrong one assembling it. This version also has dual RCA jacks for output. They take up most of the front panel, but they should be much more durable. And, no, I didn't buy gold-plated jacks.

The CS8146 version functions exactly like the DIR9001 version, but has a maximum sample rate of 192 kHz instead of 96 kHz. I suspect that is the maximum sample rate you can expect from TOSLINK. At 192 kHz sample rate, the bit clock is over 12 MHz, and S/PDIF has extra framing, so its data rate is even higher. 

The one thing I am not sure about is whether or not the CS8416 needs a reset circuit. The PCM1793 datasheet pretty strongly implies that it has a power-up reset circuit, but the 8416 sheet says that you have to hold !RST low until all power supplies and voltages are stabilized. I'm going to add a footprint for a MIC803 supervisor chip. It will hold RESET low during power-up and then for a few ms after the 3.3 rail is over 3.0 volts. Along with that, a pull-up on !RST. If nothing else, that will give us a more convenient spot to force !RST low if it's ever necessary.

I took some measurements rather than go to bed on time.

I'm not happy with them, but I am not sure whether it's my measurement technique or the equipment or the device under test.

I sort of have the feeling that the device isn't to blame. If I lift the probe off the output pin, I don't see the noise floor change any, so that tells me that I'm not seeing the device create extra noise on its output.

This screenshot is an FFT with my mac playing a 1 kHz sine wave at full amplitude (I hope) through the AirPlay device I have and into the DAC. AY is at the noise floor and BY is at the peak of the fundamental. The delta between the two is just shy of 90 dB. The specs claim that on a good day this DAC should be able to do 113 dB. But as I said, I don't see the noise floor drop when I take the probe off the output, so maybe the noise floor of the scope is higher than the noise floor of the DAC. Either that, or perhaps the 1 kHz fundamental peak is lower than it is in reality (that seems unlikely).

That said, figure 19 on the PCM1793 datasheet shows a noise floor at -140 dB and a signal peak at -60 dB. That's a difference of... 80 dB, so maybe this is measuring correctly? I'm not that experienced with this sort of measurement, so I can't really be sure. Anyone want to comment?

Version 1.1.1 came back from the fab, and electrically it works perfectly. Mechanically, there are still some issues.

The biggest one is that the TRS jack just doesn't stick out far enough. Attempting to plug into it doesn't go in far enough before it hits the front panel, so it makes poor contact.

I've decided to bite the bullet and go with dual RCA jacks instead. Surely those will stick out far enough. While I'm at it, I decided to move the TOSLINK jack just a bit further back. Mating hasn't been an issue thus far, but I want to make it even less of a potential one. I might wind up deciding to go with OSHpark's half-thickness boards for the front and back panels as well. We'll see.

The two RCA jacks and the "error" LED barely fit on the front panel of the chassis I sourced, but I am fairly confident that fit they will.

The sounds is just as beautiful as it was before. I'm going to have to find a way to generate some optical test signals and try to measure the S:N of the prototype to see if all of my noise reduction precautions have been sufficient. The plan is to attempt to generate sine tones and look at the output with the FFT mode of my scope to look for spurs and measure the S:N.

The boards for the DIR9001 + PCM1793 aren't back yet, but the DIR9001 only going to 96 kHz only allows for half of the DAC's sample rate range to be used. Swapping the DIR9001 out for the CS8416 would fix that. They're backordered until the next ice age, but I have thrown together a design that puts the 8416 in hardware mode and configures it how I think it ought to work for the 1793.

Meanwhile, I have every confidence that the boards that arrive this week will work. I also have some confidence that I can put this on Tindie at $50 (board only) and in comparison to other products on Amazon that are using these same chips, I think the pricing may not be completely out of line.

EDIT: It bugged me a little that the oscillator frequency of the MAX1721 is about 125 kHz. For the 96 kHz variant, that doesn't bother me too much, but I really want the charge pump to run faster than the maximum sample rate for the board. I've discovered the SP6661, which runs at just under a megahertz, so that's going to be the device for the CS4816 variant.

I hacked together a charge pump on a little SOT23 breakout board and wired that into the prototype. It sounds really good. I certainly can’t hear any artifacts from the charge pump in the audio. And the chip I used had quite a low oscillation frequency. I want to use the MAX1721 in the final design so that hopefully the switching frequency will be too high to impact the audio.

I tested with my AirPods Max so as to try to maximize my ability to pick up any noise or distortion, but didn’t hear anything untoward. That’s not a great test, of course, but it’s a start.

I also tested against the other two TOSLINK DACs I have handy - a $10 one from Amazon and the one built with the CS4334. The other two sound distinctly flat to me by comparison. I am not sure if it’s a placebo effect or not (and I’m not sure I would be able to by definition), but the more expensive DAC does seem to make a difference.  

I built the next prototype. While the boards were being fabbed, I found an error in the schematic - for the left channel, the + and - differential outputs from the DAC were reversed. I cut the traces on the board and bodged them back.

But the audio of the prototype sounded bad. After spending the afternoon trying to figure it out, it turns out that the problem was that I gave the final amplifier a unipolar power supply (that is, the V- pin of the amp was grounded.

Lifting that pin and temporarily hacking in a 5 volt wall wart with - to V- and + to ground gave correct audio (albeit with a nasty 60 Hz hum).

The problem now is that everybody I've found on the Internet who has made the schematic of their DAC public has a bipolar power supply.

Well, you can make -5 volts from +5 with a charge pump. The MAX1721 has a switching frequency of 125 kHz, so ostensibly above anything audible. Hacking in a MAX1720 on the prototype did make it work. But if it's possible, rebiasing the output amp to operate on a 2.5 volt offset, reducing the gain and adding a DC block on the output would probably be better for the THN spec of the result.

The next prototype board (boards?) will test one or both of these options.

I'm still not entirely sure this project is worth productizing or not. But if it is, there is probably room at the high end for something particularly good.

In looking for interface chips, it appears that the DIR9001 is the best overall choice.

For DACs... well, it's all over the place. The PCM1730E is an excellent choice... at $20 a piece.

The PCM1793 is much more reasonably priced, at about $6.25 each. Couple that with an OPA2134UA at about the same price to be the output LPF and buffer... and we might have something worth actually trying to sell.