Nonsense. LMK1C110x works very well for digital audio. I have used it successfully for a long time and so have others.
Sure. But how else to get PCM2DSD output to an external DAC? I'd love to go USB out from PCM2DSD to the DAC, but I'm not sure how to do that:
- Use a Linux-based SBC to take I2S on GPIO and go USB out?
- Conjure something up with a MCU like in https://www.nxp.com/docs/en/application-note/AN11819.pdf?
- Some other way?
Actually I hope my DAC does that. It's got ovenized Crysteks. Regardless, the primary purpose is to feed it DSD and I'm impartial to what it does with the clock and signal lines.
I considered that also because I saw IanCanada do it with this HdmiPi Pro. Here was my rationale against, but tell me if I got it wrong: the quad line
DS90LV031A has a channel-to-channel skew of 0.0/0.1/0.5 ns min/typ/max. The single line DS90LV011A has a part-to-part skew of 0.0/0.2/1.0 ns -- twice as large. Would the isolation offered by single channels outweigh that increase in skew?
The NB3L553 datasheet also integrates 12 kHz - 20 MHz to arrive at a typical additive rms jitter of 18 fs for a 100 MHz carrier. For a 156.25 MHz carrier - which should be worse? - the LMK1C110x arrives at 8 fs rms typical additive for the same integration range.
It's a pity that the LMK1C110x phase noise plot only goes down to 1 kHz. They are supposed to supersede the CDCLVC110x which I think has been a popular piece in high quality DACs too. Finally I like that it has 50Ω output impedance out of the gate -- finally something sensical.
Anyway so the decision comes down to taking the trusty NB3L553 or trying the new LMK1C110x. That the phase noise isn't plotted down to 10 Hz does not mean it is a bad part, just that we don't know.
That's interesting. The LMK1C110x even suggests putting one in the datasheet, and the AMB ο1 and γ24 designs also do it - and that's a high quality DAC or certainly used to be a decade ago. Do you know which part you used? Could it have interacted negatively with some neighbouring capacitors causing IR drop where you did not want it?
Thanks, I'll take it.
In my design notes above I wrote why I went into another direction. >99.9% of my listening is 44.1 kHz. So having only one running does not just suit my use case, it also gives the clock a quieter environment by not having noise from that other clock on the supply or ground planes.
Taking some RF design practices and local surface fills is an interesting idea that I will surely entertain as I start designing the PCB.
As for the power supplies, yes I know that having sprinkling LDOs is almost like sprinkling bypass capacitors nowadays. I get you are not a fan of Pi-filters
Honestly I am not to keen on all those LDOs: they take a lot of components, while we don't need the power, and the Pi-filter is better at suppressing noise at the frequencies of interest than a LDO is. And with one clock running, out goes the need for that extra LDO is my thinking.I know it may seem I am pushing back here a bit precisely when I was inviting comments... that's not my intention. I guess that I'm trying to challenge common belief and see about trying something new. But I'm a theorist, not a practical expert!
Which regulators would you recommend? The LT304x are specced better but only so in practice when you get the PCB implementation just right. The ADM715x seems more lenient in that regard and is used in some high quality DACs too, which is why I took it. But let me know what you'd use.
Really I appreciate your input so please take this as me trying to wrap my head around these design choices. Thanks!
I was sort of thinking along the lines of maybe you will design this board so it could be used with your next dac, or maybe it could be shared with forum members who might have different kinds of dacs. That's all.
On the subject of MCLK buffer output skew, it usually doesn't seem all that critical with dacs. We may need to take skew into account to allow for setup and hold times of d-flip flops. That type of thing. If you have a specific a specific case where you think MCLK skew is critical then we could talk in more detail about that.
Regarding Pi filters and or local regulators, IIRC Allo Katana dac used both methods at the same time for various subsystems. For example, each clock ran continuously, each has its own dedicated regulator, and each had pre and post regulator Pi filters. IMHO in the end we need both clean power, and stable, regulated power. In the case of clocks and buffers, my own tests showed that given the rest of the system at the time, it was sufficient to have one optimized regulator for the clocking system (both clocks and both buffers). However that wasn't always the case. Until certain problems were found and fixed, it may have better better to keep clock regulators separate from buffer regulators.
Here is revision A. Changes:
- Changed quad DS90LV031A to single DS90LV011A LVDS drivers (U6-9, C7-C14)
- Separated power between clock and output sections with a second LDO (U13)
- Changed LDO to ADM7154 (U11)
- Added pi-filter for NB3N502 clock doubler (FB1)
- Changed pi-filters to share LDO output capacitors (C25, C30)
- Changed to ferrite beads parts with lower DC resistance (FB2-3)
- Changed bypass capacitors in clock section to 0.22uF (C16-17, C20)
- Added option for 230VAC/5VDC SMPS module (PS1) or supplying external +5V (J6)
When I did this I designed my board so I could do things like switch clocks, and or power buffers just as you have. But I made it so I could do the other experiments too. Thus having tried various options in pretty direct comparison I found out what works and what doesn't. Looks like a mix of both good and bad things in the latest iteration. If you don't believe me maybe you should run your own experiments? If you don't care, that's fine too of course. Your project to do with as you wish.
Last edited:
Regarding Crystek clocks you might want to take a look at this: https://www.diyaudio.com/community/threads/inside-a-cchd-925-output-dead.396310/
That is actually a CCHD 957. Bunch of bog-standard ceramic capacitors and the inner bypass caps seems to be even class II. So what would be the technical explanation for a specific external bypass cap to magically improve the "sound"?
That is actually a CCHD 957. Bunch of bog-standard ceramic capacitors and the inner bypass caps seems to be even class II. So what would be the technical explanation for a specific external bypass cap to magically improve the "sound"?
As explained in LMK1C110x datasheet the reason for using ferrite beads on a clock buffer is to isolate the clock buffer power supply from rest of the digital circuitry (and vice versa). A practice commonly used. If it would have some impact on sound the reason is most probably due to wrong selection of device or poor implementation.
Thanks. I always wondered why Crystek did not specify bypass capacitor requirements in the datasheet. You would expect so if it were a critical performance parameter. I think they do specify one in the datasheets of some of their other clocks, although having no mention of it with the CCHD’s doesn’t have to imply anything.
I was following up here on subjective recommendations. Objectively what I can think of is that increasing bypass capacitance, all other things being equal, can improve noise suppression at lower frequencies.
That is my suspicion as well. Throwing in a bead with just a low-capacitance bypass capacitor can cause peaking: https://www.analog.com/media/en/technical-documentation/application-notes/AN-1368.pdf.
Last edited:
We all know what ferrite beads are marketed and commonly used for. We also know why X7R is typically used for bypass. If I tell you things that are the product of a lot of hard work and because they are not well known, and if you then proceed to ignore/disregard them then why waste my time?
It is normal practice to have external bypass capacitor on oscillators. However the question is why would a certain type of external bypass capacitor have any impact on sound. Adding capacitors of different characteristics typically adds resonances which is the obvious explanation. Even if the added resonance would somehow improve the perceived sound the resonance would be highly implementation-specific (e.g. layout) so definitely not something to be recommended.
I am not sure if and where I triggered any hostilities, and am sorry if I did. Just trying to have a healthy engineering discussion. In my view I have taken and responded to all of your input seriously.
@bohrok2610 so at this stage in the design you would favor having 0.1uF MLCCs there?
Not everybody needs to agree with your recommendations that in many cases are just based on subjective opinions and go against well established engineering practices.
There is some discussion about clocks in another thread. https://www.diyaudio.com/community/threads/return-to-zero-shift-register-firdac.379406/post-7417474
The are reasons for leaving both clocks running at all times. Otherwise they will never sound their best. Clocks used for consumer digital audio are not usually treated that way, except Topping did it with D90 and the Accusilicon clocks. Turning the unit off from the front panel or the remote only turns off the display and the audio output. Everything else is kept warmed up and running. Maybe there is no app notd on that, but its certainly known by clock experts, say, maybe if you want to go read Rubiola. https://rubiola.org/
Regarding bypass for Crystek clocks, why don't you leave 805 pads so you can try both the film cap I recommended and X7R. That is, if you are willing to see for yourself. I don't make this stuff up, and its definitely not snake oil.
That said, there are some people, many of them over at ASR, who believe that if an FFT can't show it clearly, then it doesn't exist or it doesn't matter. IMHO those people don't understand FFTs and PSS measurements all that well. Its more like a matter of faith for them rather than understanding the scientific limits. Not saying that applies to any particular people here, but maybe you will recognize some of it in yourself.
Here is something about typical audio spectral analysis and its limitations: https://purifi-audio.com/blog/tech-notes-1/doppler-distortion-vs-imd-7
Here is something about why you should question ferrites for some sensitive applications: https://purifi-audio.com/blog/tech-notes-1/this-thing-we-have-about-hysteresis-distortion-3
If you don't know why I recommend things, you could just ask instead of speculate about why you think it might be wrong.