Go Back   Home > Forums > >
Home Forums Rules Articles diyAudio Store Blogs Gallery Wiki Register Donations FAQ Calendar Search Today's Posts Mark Forums Read

Digital Line Level DACs, Digital Crossovers, Equalizers, etc.

Improving AK4490 THD(+N) figures
Improving AK4490 THD(+N) figures
Please consider donating to help us continue to serve you.

Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
Reply
 
Thread Tools Search this Thread
Old 9th January 2019, 09:25 PM   #1
nihtila is offline nihtila  Finland
diyAudio Member
 
Join Date: Jan 2006
Location: Edinburgh, Scotland
Default Improving AK4490 THD(+N) figures

Hi

First a small intro..

I have designed a tiny DAC based on AK4490 which is part of my small modular DAC design. I can tell (or open another thread) about the project later but here are photos to show what it is about. So the DAC itself is only 50x50mm. The boards are stackable with Input board, DAC, and balanced output board, and breakout and PSU board as base board.

Click the image to open in full size.

Click the image to open in full size.

I built the prototypes during holidays and everything is functional. At test bench the first THD figures I got out of it were not impressive so I have been playing with it this week to improve the numbers. However, I am still not close to the best results I have seen so I would be keen to hear experiences on how to improve THD+N. I know many have played with this IC and some have been meausring it as well. I know some designs have ended up in commercial products so I do not expect you to share all secrets but any tips are welcome. Likewise, I will share my experiments. The idea is that this board also ends up for sale; however, I will keep the schematic public and write about the development and results.

I have access to AP at work so I have the big boys' toys to play with.


Something I have found out so far:
  • 5V line (goes to VDDR/L and VREFR/L) is a sensitive one; my initial THD at 0dBFS was -93dB and I saw weird sidelobes on FFT. Apparently the bench PSU providing this was not good enough, after changing to the PSU baseboard I got rid of those. I suspect VREFs are the more sensitive ones. Need to do more tests on this.
  • Adding large capacitance on VREF improves THD+N significantly (as is known - although datasheet does not say it anymore!). However, my THD+N vs. frequency still shows elevated distortion at lower frequencies.
  • In fact, VREF capacitance seem to matter the most, even removing other caps do not change performance much.
  • I see THD+N sweet spot at around -6dBFS.
  • Increasing VREF (I've only tried 6.2V as my caps were 6.3V) improves THD+N ratio slightly at lower signal levels, close to full-scale things get worse. THD+N level actually remains constant but ratio improves due to increased signal level. Obviously this improves SNR directly.
  • Playing one channel at a time shows an improvement of around 2dB to stereo.

So I have got from -93 dB to -108dB (0dBFS), the latter with 2200uF on VREF, voltage kept at 5V. Obviously these have been ugly mods and it can be tricky to fit lots of capacitance on such a tiny board. I also have some layout changes I will do for next revision. I probably will also make a bigger board for experimenting but will see now what I can get out on that form factor.

The caps I have tried so far are basic bulk caps. I have ordered some polymer caps and will try when I get them.

Things I'd been keen on hearing experience on:
  • Capacitor type/quality, capacitance seem to be key factors here but what about the type of cap, its ESR and so on?
  • Separating VREFs from VDDs, separating channels?
  • Separate regulators for VREFs and VDDs, having them locally close to the IC?
  • Any specific layout tips?

I have checked AKM evaluation board (manual) in detail, and it seems a bit off if they have really got the numbers and figures on that.

And while I am open to all feedback, I am really after measured differences here.

----------

And finally some figures where I am now at:

Key figures:
  • Full scale output level: 2Vrms
  • THD+N at 0dBFS: -108dB
  • SNR/DR: 119 dBA
  • Crosstalk at 10kHz: around -130dB

THD+N ratio vs amplitude (I like dBs but forgot this to %), THD+N level vs amplitude, and THD+N vs frequency, and FFT 1k 0dBFS plots below.

I also put schematic. Basically on board I have 100n+100u on all supplies, except now in these mods I have soldered extra caps. And as a sidenote, output filter is not a limiting factor here, I have also measured directly from DAC output.

Lots of info, but thanks if you read and can help!
Attached Images
File Type: png THDN_ratio_d3.png (72.2 KB, 384 views)
File Type: png THDN_level_d3.png (64.3 KB, 382 views)
File Type: png THDN_frequency_d3.png (63.6 KB, 375 views)
File Type: png FFT_1k_0dBFS_d3.png (76.1 KB, 96 views)
File Type: png schematic.png (294.1 KB, 161 views)
__________________
http://nihtila.com - Vegan DIY audio electronics, does not contain snake oil
  Reply With Quote
Old 9th January 2019, 10:10 PM   #2
nihtila is offline nihtila  Finland
diyAudio Member
 
Join Date: Jan 2006
Location: Edinburgh, Scotland
Oh, and here is schematic of the PSU baseboard. So all power lines go through pinheaders which is of course not ideal.

As I mentioned, first round of measurements were done with bench PSUs powering all supplies. See the resulting 1kHz FFT! I haven't seen such FFT before.
Attached Images
File Type: png W-PSU_sch.png (237.0 KB, 165 views)
File Type: png FFT_1k_-10dBFS.png (69.2 KB, 141 views)
__________________
http://nihtila.com - Vegan DIY audio electronics, does not contain snake oil
  Reply With Quote
Old 9th January 2019, 11:09 PM   #3
JensH is offline JensH  Denmark
diyAudio Member
 
Join Date: Jul 2009
Improving AK4490 THD(+N) figures
It looks pretty good.
I would use a lower noise regulator and place it adjacent to the DAC on the same PCB.
And as you already tried, add some large capacitors. I used polymer types.

The FFT from the first round of measurements looks familiar! I saw something similar at one point, where the LDO was not set up correctly, so the ripple rejection was bad.
You have some noise at 1 kHz +/- 50 Hz, +/- 150 Hz and +/- 250 Hz. These are "mixer products" that you get from the 1 kHz tone and hum on the VREF supply.
  Reply With Quote
Old 9th January 2019, 11:31 PM   #4
bnae38 is offline bnae38  United States
diyAudio Member
 
Join Date: Aug 2015
Location: WI
Can try some series R to vrefs, should not sag to speak of..

24.9ohm is common in equipment I've worked on.
  Reply With Quote
Old 10th January 2019, 03:37 AM   #5
BesPav is online now BesPav  Russian Federation
diyAudio Member
 
Join Date: May 2011
Location: Russia, Moscow
Default Improving AK4490 THD(+N) figures

1. R4 to R7 are too high, loads DAC outputs and rise THD. Try to increase them at least to 2k7-3k9. Yours value mean that high-resistance Sennheiser HD800 could be used straightly with DAC outputs while coil inductance would filter out aliases...
2. Try one stage 3-order Bessel envelope filter based on the fully-differential THS4131.
Page22:
http://www.ti.com/lit/an/sloa054e/sloa054e.pdf
3. Series supply-regulators are ugly, go to shunt ones.
  Reply With Quote
Old 10th January 2019, 07:28 AM   #6
JensH is offline JensH  Denmark
diyAudio Member
 
Join Date: Jul 2009
Improving AK4490 THD(+N) figures
1. No. The DAC will not see a load equivalent to the resistor values. It will be much higher. I would rather reduce the values to reduce the noise. I used 220 ohm in the RTX6001.
2. I don't think this will improve the distortion. It will give a higher load on the DAC.
3. As long as the output impedance is low enough and the noise is very low, the regulator type should not matter. Can you measure the difference as nihtila asked for? I used an LT3042 in the RTX6001.

@ nihtila
I also found that increasing the VREF increases the distortion when close to full scale. Below -7dBFS the distortion was actually lower with a 7 V suply compared to a 5 V supply!
  Reply With Quote
Old 10th January 2019, 08:44 AM   #7
nihtila is offline nihtila  Finland
diyAudio Member
 
Join Date: Jan 2006
Location: Edinburgh, Scotland
Thanks for the comments.

@bnae38
Have you had the cap after the R, to make RC filter? Or probably cap on both sides.

@BesPav
1. The impedance of the filter caps are quite high at lower (=audio) frequencies so the loading of the DAC should be within specs; and at DC there is basically no loading at all. As Jens said, going higher resistances will introduce extra noise. Even if this is nothing extremely quiet, -112 dBV or so noise at audio band is quite low and easily increases when resistors go to k-range.
2. Fully differential circuits load even more as you need to have low resistances to get noise low. Unless you buffer the outputs, and then you don't win anything anymore. Fully differential opamps seem to be more expensive as well.

@JensH
How was your THD+N vs frequency? Did you get it nearly flat? If I would get it nearly flat with similar level to around 5kHz, that would be excellent. In various AKM datasheets they have this shape of graph for small VREF cap values. Mine was not small but it was a massive 2200u cheap bulk with ugly mod. I hope better quality caps in smaller form factor would improve this.

Local regulators are something I do want to try, but probably not possibly on this board.
__________________
http://nihtila.com - Vegan DIY audio electronics, does not contain snake oil

Last edited by nihtila; 10th January 2019 at 08:47 AM.
  Reply With Quote
Old 11th January 2019, 02:06 AM   #8
Shinja is offline Shinja  Japan
diyAudio Member
 
Join Date: Jul 2008
Sallen-key type (VCVS) filter circuit have unequal impedance, seen from opamp's +input and -input and it may introduce distortion at opamp's input which is not cured by NFB.
Solutions are choose appropriate opamps or insert same impedance to the +input side , or bootstrap whole opamp i.e. bootstrap their rails.
  Reply With Quote
Old 11th January 2019, 09:34 AM   #9
ojg is offline ojg
diyAudio Member
 
Join Date: May 2003
Location: Norway
Improving AK4490 THD(+N) figures
I agree with Jens suggestion to use a better regulator and place it adjacent to the DAC on the same PCB using +15V as the source. I think LP3878 probably has noise levels low enough, but the ripple rejection >1kHz is nothing to write home about.

In the meantime you could try out remote sensing for the regulator you have. Connect R5 and C10 on the regulator board with a (short) wire to +5VA on the DAC board.

Also, in general, MFB type filters have lower distortion than Sallen-Key types, due to avoidance of common-mode distortion in the opamp. However with LM4562 this should not be a major issue.

And having the "sweet-spot" for THD at -10 to -6dBFS in very normal for delta-sigma converters.
You have reached -110dB at -10dBFS, I think that is quite respectable compared with the datasheet, how low distortion do you aim to get out of this chip? Which AP analyzer are you using? You may be getting close to its limits as well.
  Reply With Quote
Old 11th January 2019, 10:30 AM   #10
nihtila is offline nihtila  Finland
diyAudio Member
 
Join Date: Jan 2006
Location: Edinburgh, Scotland
I know my regulators and powering scheme is not the best but it's also difficult to guess what makes a difference. I looked at some other regulators, like LT3042 Jens mentioned, and the figures are from a different planet indeed. Especially noise and PSRR. Noise may not be an issue as ojg said, but there's massive difference in PSRR between those regs. I need to study this a bit more, I didn't even know there are such low figures available.

I'm not a filter expert, Shinja's point is noted and something I have not been considering. However, I still think it's not an issue here. CM distortion in SK filters is something I am aware of but it should not be an issue with LM4562 which handles CM very well (according to Douglas Self's measurements). I have characterised the filter part with AC signal, I still need to do that with the DC level as well.

The last measurements I did I think was with massive bulk caps at VREF and there I didn't really see THD+N deteriorating close to maximum level anymore. That is the graph I have attached in the first post. So minimum is around -108dB.

I am very happy with these results and if I can implement mods on the PCB that get me these figures, it is excellent for this project. But there is of course some academic interest in what can I do to push the performance even further. And then possibly use that knowledge on another board.

I should get polymer caps today or on Monday, then do some cap experiments. I will do a bit of research on regulators during the weekend and order in some and experiment with those later next week. Then see what I can implement on this board.

I am using APx555 so there is still some headroom left.
__________________
http://nihtila.com - Vegan DIY audio electronics, does not contain snake oil
  Reply With Quote

Reply


Improving AK4490 THD(+N) figuresHide this!Advertise here!
Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump

Similar Threads
Thread Thread Starter Forum Replies Last Post
Lissajous figures rtarbell Everything Else 1 28th October 2005 02:23 AM
Noise Figures northsiderap Solid State 12 7th May 2005 09:48 AM
Xmax figures to actual exurcsions? TRANSDUCEReng Subwoofers 20 21st September 2004 09:19 PM
DC offset figures johnm Chip Amps 7 6th February 2004 02:01 PM
help, need some figures Alex_B Multi-Way 54 9th December 2002 01:36 AM


New To Site? Need Help?

All times are GMT. The time now is 05:40 AM.


Search Engine Optimisation provided by DragonByte SEO (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.
Resources saved on this page: MySQL 14.29%
vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.
Copyright ©1999-2019 diyAudio
Wiki