Improving AK4490 THD(+N) figures

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Your issues would seem to reflect LP3878 noise levels, which are quite high below 1-2 kHz in particular according to the datasheet. In hindsight (which proverbially is 20/20, I know), maybe not the best candidate for +5VA. I'd suggest trying a larger capacitor on the BYPASS pin, which would seem to be filtering the voltage reference. The datasheet insists on something low leakage - if you have some 220-470 nF films around, try one of those.

It is unfortunate that the capacitor in parallel to R1 is already used for some pole/zero trickery, but what you could try is another RC combo in parallel to both - an electrolytic in series with maybe the same value as R2 (or half that). This would make the regulator act like a 2 V (or 1.5 V) one for AC, with correspondingly reduced noise gain. I'd try 1 kOhm + 10-22 µF (low ESR not needed for obvious reasons... quite the contrary, you want lowish leakage).

With as much capacitance as you have on the VREF pins now, a bit of series R (as mentioned before) is in fact also worth a shot. RC filtering always works much better if R isn't just regulator output impedance, for obvious reasons.

Indeed, it is not the best regulator. In fact, seems LT/Analog have so much better choices for demanding applications although they are more expensive as well.

Thanks for suggestions, will see if I play with it but if I move to on-board regulators, the model will be something else.
 
Hi ,


this is from a well known product.

I find it intresting....
 

Attachments

  • AKM4990 decoupling.PNG
    AKM4990 decoupling.PNG
    83.8 KB · Views: 428
Last edited:
Hi ,


this is from a well known product.

I find it intresting....

To me it looks like they don't really know what they're doing, just fire all types of caps there thinking each one cures different type of issues. I suspect could remove most of them without much impact.

In general it is not a good idea to put so many different value caps in parallel, especially not so close values. You get nasty resonance spikes in the impedance curve. For digital decoupling you basically want many equal value caps in parallel to bring the parasitic inductance down but without creating these nasty spikes. In analog/low-f you want big caps and maybe one small in parallel.

I may be wrong of course :) Maybe they have evaluated that. But for example Henry Ott has good material on measurements of parallel caps, and showing that different values often create more issues than benefits.
 
Member
Joined 2007
Paid Member
Try treating Vref as a low impedance node where packets of charge are taken from rather than something held up by capacitance. Firstly low pass filter your voltage reference to remove high frequency noise, then feed into an appropriate bandwidth (select based on converters true conversion clock) low noise opamp which feeds the Vref pin on your converter. http://www.ti.com/lit/ml/slyc147/slyc147.pdf (see figure 6).
 
Getting wideband low impedance is surprisingly nontrivial. Looks like that's what they were shooting for here. Pretty much a cap every decade sounds about right if you don't want any nasty surprises. (And look at all the unpopulated spots.) I'd imagine this should get the job done well into the dozens of MHz. Not sure why Vref needs this (I mean, the actual DAC is operating in the double-digit MHz, but still), but the datasheet of another AKM DAC puts great emphasis on VREFH-VREFL having to be very clean (up to the point of using 10 ohm resistors in series with both to exploit the bypass caps for RC filtering), including keeping all kinds of clocks away from them. So clearly "Vref hygiene" is Important Stuff.
 
The figure 6 below looks a lot like what we use for AVCC (the dac reference voltage) for ESS mobile dacs (ES9038Q2M). Except we use LTC6655 as the reference, and the opamp is OPA1612 or AD797. It works very well for that application.

EDIT: To be a little more clear, we use one OPA1612 or two AD797. One opamp for each dac channel maintains best stereo separation.
 

Attachments

  • !fig6.jpg
    !fig6.jpg
    140.2 KB · Views: 380
Last edited:
Ok, interesting to see that in a high volume product. I would imagine they have done their tests then. I don't doubt it not working but think there could be an easier and cheaper way for the same result - if all those caps are even populated.
We can only hope they probed the impedance right at the chip's pins with a network analyser...
In general, I agree with you. Paralleling ceramics (or any low-ESR caps) of different values staggered in more than 3:1 rations most often leads to disaster.
 
The figure 6 below looks a lot like what we use for AVCC (the dac reference voltage) for ESS mobile dacs (ES9038Q2M). Except we use LTC6655 as the reference, and the opamp is OPA1612 or AD797. It works very well for that application.

EDIT: To be a little more clear, we use one OPA1612 or two AD797. One opamp for each dac channel maintains best stereo separation.

Interesting they're using that over something like the LTC3042 or its brethren. Then again, mobile application is going to have different needs, and be originally running from a battery, so the bandgap vref is probably the noisiest thing around!
 
I guess one could use LT3042 as a reference, but not as the buffer. People try that frequently and always say it sounds much better with an opamp buffer. In fact, a few people have stopped using LDO regulators for anything having to do with dacs. That sure looks like an over reaction, as other people report very good results using ADM7150 or ADM7154 for AVCC. One regulator for each channel, of course. Have to try those ADM715x devices sometime myself.
 
Last edited:
Try treating Vref as a low impedance node where packets of charge are taken from rather than something held up by capacitance. Firstly low pass filter your voltage reference to remove high frequency noise, then feed into an appropriate bandwidth (select based on converters true conversion clock) low noise opamp which feeds the Vref pin on your converter. http://www.ti.com/lit/ml/slyc147/slyc147.pdf (see figure 6).

Can you (safely) actively drive the reference pins of the AKM DACs? I don't think the datasheet says much about it.
 
I guess one could use LT3042 as a reference, but not as the buffer. People try that frequently and always say it sounds much better with an opamp buffer. In fact, a few people have stopped using LDO regulators for anything having to do with dacs. That sure looks like an over reaction, as other people report very good results using ADM7150 or ADM7154 for AVCC. One regulator for each channel, of course. Have to try those ADM715x devices sometime myself.

You know me and my incredulity around the power of suggestion, proud papa syndrome, and anecdotes, so we'll leave that to be, if you don't mind. :) That said, I'm curious more about electrical reasons why that may be the case (layout would be my first hypothesis, especially with people using patched-together Chinese-sourced boards vs a all-on-one board of their own making like the OP), as the LDO's being quoted should outperform the circuit you linked (see superregs vs newer LDO's comparisons).

Thanks for pointing out those ADM parts!
 
That said, I'm curious more about electrical reasons why that may be the case...

So am I. The reports come from various unrelated people working on different projects who don't follow or talk to each each other. They try LT304x parts because they are quick and easy, and look good on paper. Eventually, maybe they happen to try something else, or maybe their dac doesn't sound as good as some other dac they heard and they start experimenting to see what can be done. In other words, different stories from different people.

Most likely if someone who knew what they were doing decided to look into it they could figure it out. I don't have time. You want to do it?
 
Mark.. You say lt30xx is quick & easy.
I would say nothing is farther from being quick & easy than these.. supposing one wish to build to maximum yield, from scratch.
It is so as you say only when you talk about using prepackaged modules. So we are back to the situation what DPH described. Which is cancelling most of the chip's capabilities...

So I am a bit perplexed as well..
Especially considering the strong technical shortcomings of the suggested topologies, with respect to these chips..

Ciao, G
 
Some of the prepackaged modules are laid out exactly per the data sheet recommendations. Doesn't seem to make any difference in the reports. Could be it is because they have unipolar outputs, they overshoot or ring with the loads because the load behavior with the supply is not well understood, etc. What we have at this point are subjective reports rather than measurements. If we could correlate subjective listening impressions with measurements we might actually learn something useful out of the exercise.

In a somewhat analogous situation, I recently rediscovered something that has been known on and off for a long time, which is that rather modest power supplies supplemented with about 100uf of good quality film caps on the output stage power rails reduces a bit of remaining audible low level distortion. Experiments with electrolytics were unable to duplicate the same audible effects. Why? Unfortunately, I don't have a well-equipped lab here in retirement. There is a decent scope, a DVM, an LCR meter, and a few other pieces, but hardly what I would call well-equipped by some standards. I do however happen to have a very low distortion dac, an ultra-low distortion power amp and some usable speakers. In addition, I am well-trained in how to use them, and I know how to use double blinding when comparing small effects. So, I can perform some experiments and report findings, but not correlate them with good measurements. All that being the case at this point in time, I can make some progress with improving sound quality of circuits, but not provide my fellow engineers with the information they would like to have to understand what is going on in technical terms.

Another thing I may as well mention since we are talking about some stuff here, which is that with the advent of ulta-low noise LDO regulators, it seems like there has been an over emphasis on power supply noise above all other specifications. Low noise is great for low level preamps and such, but is not the only thing needed to make for a good power supply for audio gear. Some of the well known old circuits that work very well for many hi fi applications, such as the Jung regulator, seem have fallen by the wayside as everybody is low noise LDO bandwagon.
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.