Follow-up on AK4490 test board measurements
- Comparing cheap LDOs to superb LT3042
- LP2992 shows excellent performance
- AK4493 vs. AK4490
- Even better performance
- Almost pin compatible
- New v1.1 boards in production
- Slightly larger but still ‘wee’
- Supports AK4490 and AK4493, and different LDOs
- Possibly two or three assembly versions with different price and performance
Unfortunately due to busier time in 9-to-5 job and other things the progress has been a bit slower recently. However, new boards are in production and hopefully those will be very close to the final production version.
Earlier measurements recap
Please find the posts about the initial measurements and the first test board measurements for updates what has happened before.
AK4490 performance on the large test board was excellent when using superb LT3042 LDOs. Besides excellent performance, another advantage of LT3042 is that it does not need a large capacitor on VREF. In fact, it may even deteriorate the performance. Downside is the price; especially if using one per channel for optimal PCB layout and performance. For one-off DIY board the price does not matter but as the plan is to make a small batch production run, price is part of the equation. Moreover, optimising the circuit is somewhat interesting – to some extent.
I also tried cheaper but still great ADP7118 LDO but the results were not nearly as good.
Idea of several versions
As LT3042 is pricey, I thought I could make two or three different versions where the best (and most expensive) one would use LT3042s. To support that, I needed to find a suitable cheaper LDO for the cheaper model. In addition, I wanted to test the newer and improved (and more expensive, unsurprisingly) AK4493 DAC for possibly even better model.
LDO tests
As I was looking for a cheap LDO, I decided to pick a few from TI portfolio. Before I had been looking at Analog (ex Linear) parts but they are overall more expensive – if also better performance. Looking at decent output noise and PSRR, and also paying attention to price and package, I picked LP2992, TPS799, and TPS734.
I have hundreds of pages of semi-automated measurement results but here is a brief recap in terms of THD+N at 1 kHz and 0 dBFS:
- LP2992: -109 dB
- TPS799: -99 dB
- TPS734: -99 dB
- ADP7118: -104 dB
- LT3042: -111.5 dB
There is no significant difference in idle noise so SNR remains almost identical 120 dBA between LDOs.
Besides specifications, the difference between the now tested three LDOs is also that the used LP2992 model is a fixed 5 V LDO while the other two are adjustable. Fixed-voltage models have a bypass pin for a noise reduction capacitor which made huge difference in my tests. Adjustable models do not have this which can make all the difference. In fact, noise is indeed the contributor in the THD+N figures above. For adjustable models noise bypass would require extra components around the voltage-setting resistors. Therefore, this comparison may not have been fair but that does not matter – my pick was LP2992 as it can provide superb performance without extra tweaks.
The big surprise here is that LP2992 was very close to LT3042 – and many times cheaper. Of course this is just a single point measurement and there is slightly more to it, especially when looking at lower frequency distortion. THD+N at 100 Hz 0 dBFS shows following, giving LT3042 more advantage with its flat THD+N vs. frequency graph:
- LP2992: -106 dB
- LT3042: -111 dB
One circuit-level difference to note here is that while LT3042 does not like big output capacitor (as is even stated in the datasheet), LP2992 benefits from one.
AK4493 vs. AK4490
AK4493 is the successor for AK4490 with slightly higher performance and price tag. It is almost pin compatible but there are a couple of changes so it is not a drop-in compatible.
Pin compatibility
As far as I know, the physical and feature differences between these are:
- AK4493 requires 1.8 V for digital which can be provided internally or externally
- Pin 16 is LDOE (LDO enable); in AK4490 it is DEM1 pin (consequently AK4493 has only one DEM setting pin).
- If internal LDO is used, bypass capacitor must be provided on pin 1. With external supply this pin is the power input. AK4490 pin 1 is NC.
- There is one more digital filter option but the pins to control them are the same
- There are also some other changes in settings when using I2C control
- For example gain control where user can choose higher signal level and SNR but lose slightly in THD+N
Performance difference
Key specs for AK4490 and AK4493 are following:
- DR and SNR specs
- AK4490: 120 dBA
- AK4493: 123 dBA
- THD+N 1 kHz 0 dBFS spec
- AK4490: -112 dB
- AK4493: -113 dB
While I have seen -112 dB sometimes in my AK4490 measurements, the best results have been mostly -111 dB. Besides performance being always a system parameter consisting of layout and external components, it is good to remember that these are typical specifications. For example for THD+N typical value is the -112 dB but maximum is -105 dB. Therefore, to really characterise this one would need to measure several devices. In my measurements comparing different LDOs and caps I have used the same device so the differences are comparable. However, here comparing AK4490 and AK4493 using only one device each does not really give comprehensive results.
Anyway, I do see difference between these. In fact my THD measurements exceed datasheet specifications so maybe that particular AK4493 was a better than average unit.
- SNR measurement
- AK4490: 120 dBA
- AK4493: 121 dBA
- THD+N 1 kHz 0 dBFS measurement
- AK4490: -111 dB
- AK4493: -114.5 dB
That amazing THD+N figure is achieved using LT3042.
SNR falls slightly short of specification. This may be due to output filter being the bottleneck.
Measurements conclusion
Interesting results again. Besides LT3042, LP2992 shows great potential here – especially considering price. However, again these are test board measurements and final performance will be measured on the final application boards. I am really looking forward to those results and hope they will be close to these results!
Wee DAC v1.1 boards
I designed new v1.1 boards for W-Input and W-DAC. Besides the DAC changes there are significant other changes as well which I will cover in another post later on when listing the features. However, due to extra additions the board size is now 70×50 mm instead of 50×50 mm. Layer count remains four.
Final performance tests need to be performed on these boards and then decide possible versions. If there is enough distinguishable performance differences, there may be a “budget” and/or “high performance” models besides the standard model. All would use the same PCB, the changes being in the DAC chip, LDOs, and possibly output filters.
16 comments
I never thought that the LDO has such an influence on the THD + N!
It seems to be significant at least on these AKM DACs. It is the reference pins that are sensitive to all external stuff. Noise is definitely one important feature of LDO here but there is something else to it as well, and it can be tricky to quantify. But it definitely requires experimentation and measurements if one wants to get everything out of these.
This guy is struggling with the same problems 😉
http://www.dimdim.gr/2017/11/arduino-controlled-dual-mono-ak4490-dac-part-1/
I’ve just run into exactly the same problem so this has really helped thanks. Are you running the LP2992 off the +15V rail? If so, is it getting quite hot – that’s quite a lot of power for such a small package. Are you also running separate LP2992 devices for L and R channels?
I have been actually measuring the W-DAC v1.1 boards now, will try to write updates at the weekend. Unfortunately LP2992 does not show as good performance as on this test board, but still very good. Especially lower frequency THD deteriorates though. LT3042 remains the best option by a clear difference.
I have been running LT3042 off the +15V rail as it has a thermal pad but used +7V for the LP2992. As you said, it is a tiny package so even if the current is small, I don’t feel confident dissipating 10V in it.
On my W-DAC v1.1 boards I have separate devices for L and R channels. The board actually allows for population of LP2992 or LT3042 so I will probably make two different models (using the same PCB) as the performance difference between these is clearer now. Not sure why LP2992 is worse on this board as the layout is very similar to the test board. But of course, it is a lot more complicated board with lots going on.
Overall, it has proven that getting the very best performance requires lots of experimentation and trial and error, and still it can be sometimes tricky to see what makes a difference.
Some quick results of todays experiments in case any of it helps :
– Having a clean VDD isn’t that important – it’s just the VREF pins that matter for THD+N.
– The LP2992 really is very low dropout. I’m running it off +5.25V and it’s generating spot on 5.0V and clearly doing a decent job of regulating. This works partly because I’m only feeding VREF pins so drawing not a lot of current. The reason I’m doing it this way is because at 5.25V everything else on that rail is still well within spec. Having such a low voltage drop and current also dodges the power dissipation problem.
– When you look at the AK4490 datasheet it specs VREF to be above VDD-0.5V which suggests to me that they expect exactly this kind of circuit where VREF is generated from VDD using an LDO low noise regulator.
– Adding more capacitance on the LP2992 output really does help. I’ve got a total of 800uF in ceramic caps on the output (although this is for 4 x AK4490 with 100uF + 100nF next to each VREF pin).
– I’m not seeing any deterioration in THD+N at LF. My sweeps on the AP of THD+N vs frequency are absolutely flat. My guess based on some earlier experiments would be that this is related to the fairly big total capacitance on VREF.
I could probably squeeze a little bit more out of this circuit, but I’m getting about -108dB THD+N which is better than I was aiming at. The LP2992 is a really great find – it’s ideal for this application even though it doesn’t look like it was designed for it.
You’re absolutely right, there’s definitely a lot of experimentation and trial and error involved in this!!
Hope that helps somebody somewhere.
Thanks a lot! Great to hear someone doing similar measurements and have access to AP.
-I have noticed the same with VDD. That’s why I actually have VDD and VREF now coming from the same LDO. Just have one per each channel. On test board I had these separately.
-True. But LDO performance may be better with higher dropout (at least PSRR), there are usually graphs on datasheet. But probably doesn’t change anything here. I may also eventually feed all low voltage LDOs with the same 5.5V supply as that’s the maximum of my 3.3V LDOs.
-I suspect the VDD-VREF difference just comes from internal limitations (parasitic diodes or so).
-Your last point is the most interesting one! I have seen LF deterioration in all of the three PCBs I’ve used, in all LDOs I’ve tried (five?) except LT3042. You mentioned all your C is ceramic? Maybe that’s the thing – I need to try. Because on my last measurement session I was playing with ceramics instead of usual elcos and polyelcos I’ve been trying, and saw some interesting results. But I didn’t try more than 47uF ceramics. What kind of ceramics do you use?
– I agree that most regulators perform better with a larger voltage drop, but reading through the datasheet for the LP2992 it seems really optimised for low current and low voltage drop. That’s part of the reason I decided to try running just VREF (instead of VREF and VDD) so it’s drawing hardly any current, so very low voltage needed for regulation according to the graphs on the datasheet. Although I’m not quite getting the performance you are, bear in mind that my board has well over 1000 components on it (digital and analogue) and is running off a fairly dirty switched mode flyback mains supply so I’m actually quite pleased that I’m even in the ballpark!
– For ADC and DAC decoupling I usually use a 100uF 1206 ceramic in parallel with a 100nf 0603 ceramic. There’s nothing particularly special about the ones I use and given the current problems in sourcing MLCC it’s whatever you can find!! However, I often use a Samsung CL31A107MQHNNNE which is fairly cheap and usually readily available. The interesting thing I found in my experiments was that THD+N carried on improving as I added caps, which I didn’t really expect above 200uF or so. Actually it might improve further above 800uF, but that was just what I already had pads for on the board so I stopped there. I would definitely try stacking up a few 100uF ceramics to see if it solves your LF deterioration.
-I tried stacking several 100u ceramics instead of electrolytics but no improved LF response 😐 I also tried a combination of elco and large ceramics but results were worse. It’s quite an odd one.
-I got some improvement on N part of THD+N at 100Hz by using larger noise coupling cap on LP2992 but it didn’t help on overall figure as THD dominates.
Maybe a very very low ESR electrolytic has the same effect as several stacked 100uF ceramics?
ESR may play role here. I have tried several electrolytics: bulk ones and polymer models. Not much difference between these by the way, in some cases polymer is slightly better if keeping capacitance the same. When adding more capacitance, I can see the overall THD+N level coming down until it reaches some kind of bottom, and then adding more capacitance brings the LF deterioration “cutoff point” lower but not much. If this point is actually more to do with ESR than capacitance, adding parallel caps could help. I haven’t tried parallel elcos as there is no space on my small boards but I could do parallel ceramics. Or even try parallel electrolytics as an experiment. Need to experiment at the end of the week when I hopefully have access to APx555 again.
I have already noticed that replacing small 100u elcos on VDD pins with ceramics actually improved overall performance slightly. As my VREF and VDD are tied together per channel, it may be VREF decoupling that makes the difference instead of VDD decoupling although the cap is next to VDD.
Too many variables here! 🙂 Would be good to understand the root cause and work from there, but it is almost impossible without very detailed input from the chip designers. It is a bit surprising to see deterioration at _low_ frequencies. Before we’ve seen already that LDO noise has some effect, one could of course try to analyse eg. output impedance vs frequency.
At -113dB we are almost at the physical limit. Every little mistake is punished. It will probably be difficult to reproduce these lab values in production.
Most certainly yes. Typical spec for AK4490 is -112dB and max spec -105dB. So even if I will be able to produce something in the order of -113dB in one production batch, it will not be a guaranteed spec as the next batch could potentially be something different. But only time will show how much variation there will be. I’d assume them to be fairly similar within one wafer, for example. And that’s just the DAC chip, something sensitive in the PCB level could have an impact as well, such as the LDOs.
New challenges for Nihtila! AK4499 is out:
https://www.diyaudio.com/forums/digital-source/330506-ak4497-wont-flagship-anymore-6.html#post5786778
Interesting to try one day. It’s not really available yet except samples. But will need to read a bit more and hopefully try one day.
Digikey has a price for it – £68 a pop! That’s almost 20 times more than AK4490!
Comments are closed.