Yes that's what I wanted to do for my ES9018 project, but I never had time to do it! Getting the opamp stable was always a bit of trouble in the sims. The IV should synthetize an input impedance that gives good damping with the caps, which is difficult without increasing noise at the output...
One comment about Ott, the Black Magic books, Grounds for Grounding, etc., is that a lot of the content is directed at high speed digital applications. Mixed-signal, as we find in dacs, may be another kettle of fish. For example self-damping X7R caps and hysteretic ferrite beads have known benefits in digital circuitry. They can also sometimes cause problems in analog (including analog RF) circuitry. Some of that stuff does not seem to be as well documented in most books and articles I have seen so far. Thus, the designer may more on their own to figure out what works best for a given application. For one example, C0G/NPO bypass caps of sufficiently high voltage rating to assure good linearity may be needed and they may require separate series or parallel damping resistors. Bypass current routing may need to be considered on a case by case basis, etc. That's my take on it anyway.
I've been wondering about that. Okay, X7R distorts a lot if the voltage across it changes, but in a supply decoupling role, it doesn't. Unless it does, and the tiny spikes count.
Since 10x electrolytics in parallel (with 2.5mm pin spacing) will deliver less than 400pH inductance (which means the chip internal inductance will dominate) while having no distortion to speak of, I wonder why this is not used more often...
IME sometimes they do count. Its that I did some experiments clocks, clock buffers, and I/V opamps. X7R didn't fare as well as more linear types. I also have it under advisement that other people tried the same thing before I did, although I didn't know about their work at the time.
Regarding electrolytics, the most sensitive place for caps I have found so far has been AK4499 Vref. AKM did something a bit unusual with AK4499 eval board where they put an RC filter between the Vref regulator and the dac chip Vref terminals. They did it even though they knew it added some HD. Why? IMHO because it had an effect on perceived soundstage, although the HD was presumably an unwanted byproduct. They used a total of 20R and one 470uf electrolytic per dac channel. Since they did that, a bunch of different caps were compared here in that circuit position to see if I could find a type I preferred. Turned out they all sounded different, only some of which could be accounted for by HD. Probably some of the sound was electrolytic DA. Some caps had a sound suspiciously like signal-correlated noise (e.g. new-ish Panasonic FM). Electrolytic cap sound tended to change as they aged. Never did find a solution I like so I abandoned that Vref filter topology and modified the eval board to remove it. Anyway, seems to me 10 electrolytics might be used as you suggest in some cases, but not other cases. How much space they take up might be a factor. FR of electrolytics might matter for analog RF too.
Also IIRC Doug Self found that for electrolytics not to produce HD the AC voltage drop across them had to be no more than somewhere around 60mV. In a lot of situations ripple voltage may be more than that.
It probably varies with more than signal. Default current is different for PCM and DSD modes. Using a filter like the eval board does should have some effect since it add series resistance. Also it would depend on the voltage applied to the noninverting I/V opamp inputs. The main options for that are Vref/2=2.5v, or else use around 1.9v to introduce a DC bias that causes the I/V opamps to perform DC removal from the output music signal. Other than those things I suppose Vref current could be measured depending on signal. But, why? Out of curiosity? Is there some decision I would want to make about a design based on such a measurement (please recall I want to make a listening dac, not a measurement one)?
How would you design a regulator for a load if you don't know anything about the load?
For example ES9018 with a balanced IV draws average constant current from AVCC (which is its VREF). So, besides the ton of HF noise, VREF can be clean. But without an IV, or if the IV is not balanced, then it draws signal-dependent current which is distorted, so VREF can never be clean. So that tells you there's no point in using voltage mode, you have to use a balanced IV.
Are you sure about this? In AK4499EX eval board the RC filter is between Vref regulator and Vcom terminals. Ditto for AK4499EX datasheet.
AK4499EX turns out to have two modes: Measurement Mode and Sound Quality Mode (please see attached). Measurements apparently don't matter so much in SQ Mode.
OTOH for the original AK4499 seems like what they might have considered SQ Mode was implemented with that Vref filter.
OTOH for the original AK4499 seems like what they might have considered SQ Mode was implemented with that Vref filter.
Attachments
IIRC the nominal Vref current in DSD mode is specified as 22mA, so its not a case not knowing anything.
What AKM did in the eval board failed to convince EEs designing dac boards to risk going for a SQ dac verses a measurement dac. The EEs know they will be called out at ASR if the numbers aren't great. In today's marketplace, ASR reviews impact sales.
If AKM has implemented something similar to the Vref filter in DSP for AK4499EX, they probably figure dac board designers will likely give customers the option to use or not. And ASR reviews of the numbers can still look good.
That's my take on it at this time anyway.
If AKM has implemented something similar to the Vref filter in DSP for AK4499EX, they probably figure dac board designers will likely give customers the option to use or not. And ASR reviews of the numbers can still look good.
That's my take on it at this time anyway.
Last edited:
If it is 22mA + 5mA signal dependent, that's very different from 22mA constant...