AKM’s AK4399. The Best Value 32 bit DAC.

Hi, I’ve just finished building an AKM AK4399 DAC, so here is my page about it. It’s available on Ebay here for US$34. This is the datasheet ak4399_f03e

It’s great value – the best low cost 32 bit DAC there is. It’s also very flexible so you don’t have to follow the output circuit and are free to use anything that will filter and buffer the output because there is no current-to-voltage conversion needed.

The sound signature is very natural and it has great definition everywhere, with deep controlled bass and lots of detail coming through, and yet it manages to stay smooth.  I really like it. Of course, the choice of op amps has a great effect but even humble NE5534 sound good with this DAC. The AK4399 IC is relatively expensive but the total DAC is overall relatively inexpensive to build at around US$100.

However, through headphones, that lovely detail can be a little too much, especially in the very high frequencies, if care isn’t taken building it and choosing components. It needs the 100nF X7R decoupling caps around the DAC IC as close to the pins as possible to prevent over-bright details. You should also avoid op amps like the LME49710 and capacitors like Sanyo Oscon. They are high quality components but will sound bad in combination with this DAC. You’ll see in the pictures that I have tried them but the sound is not pleasant and I have since changed them.  I usually use either Elna SilmicII or Panasonic AM (dark blue caps in the photos; the light blue are Vishay-Philips BC038) for analogue power, and solid polymer caps like Oscon or Nichicon PLE for digital power.

Ground Plane Circuit Layout DAC IC Pin14 is tied high so the DAC is set for parallel; PCM only. It has pads for DEM0 and DEM1 to set the de-emphasis mode (datasheet page 24, table 10, DEM0 high and DEM1 low for off, so leave DEM0 open and short DEM1 for off), and DIF0, DIF1, DIF2 to set the data input; table 7, page 19; leave DIF0-1 open and short DIF2 for i2s 16/24bit; leave DIF2 open for 32bit.

The values on the board are the same as the ones in the AKM EVM module, which is not the same as the datasheet design.The datasheet values will likely sound unnaturally detailed. If you put the EVM values into this calculator, you’ll see the performance can be better, for example by simply lowering the value of Cf to 5600pF. In fact after building it, I decided to add 1000pF across the other 6800pF cap for 7800pF (not shown in the pictures until later). This worked very well and the high frequencies became a little more precise and “sweeter”.ak4399 evm circuit It’s a very high performance DAC as this FFT analysis shows : AK4399 performanceThe parts cost me just US$11. In addition to this, I’ll need two 5V power supplies and one +/-15V supply. I’ll add more about the power at the end.

It isn’t necessary to do this, but the routing and layout can be improved – the black areas are where I have changed it to be sure that the analogue signal is not affected by power lines, and also where the master clock (MCLK) might be affected by power flow.

SMT parts

Op Amps and Caps

As sometimes happens in electronics, something that should work doesn’t. Unfortunately, this is the case here. Debugging is part of the hobby but it’s the least fun part. Oh well. Actually, I build carefully and slowly – I check each part’s value and polarity before I fit it, I check its connection after I fit it, and I check for shorts, solder bridges and solder splatter. I check the power supplies. Then when the power is connected, I check again for shorts and if that is good, I say a little prayer, and switch on. After switching on, I check the output of the DAC for any ACV and DCV.

Here is a photo showing how I have fitted smd 100nF X7R ceramic capacitors as close as I can to the IC pins to make them as effective as possible at reducing HF noise. Since taking this photo, I’ve also added 2 more 100nF X7R, close to the pins for VCML and VCMR, to decouple noise from these pins.

X7R 0.1uF local decoupling

When I started this DAC up for the first time, I expected some mV of both AC and DC, but in this case I have 0.000V. So I went over the board, first checking all power points are correct. They are. I check the input signal connections. All good. I check logic connections – all correct. I try to add hum using my finger on op amps inputs – I get hum. I go over everything again. But still silence. I note that the VCML/VCMR pins have 0V and the analogue outputs have 0V. Dead IC ?

I then start fiddling with the data input. I find I can get noise, crackle, pop, etc, and I can even sometimes get distorted music under a heavy pink noise. So I assumed the DAC IC is getting the signal it needs, and that the op amp circuit is working, and so perhaps the IC simply can’t create music for some reason.

Conclusion ? AK4399 IC is dead, which means I either have to buy another, cut off the old one, desolder its legs, and solder on another, or junk this project. The former is too time consuming and difficult with no guarantee of success, so, its time to reflect on what to do next, if anything.

This would be my first failed project, if I give up. But if I go on, it might still be a failed project and an even greater waste of time and money.

I got a message from the seller that vcml/vcmr should be 0V with no input and 2.25V with music playing. This led me to think my assumption was wrong and the problem was not this DAC but the digital source. Further testing found that my source, a cm6631 usb-i2s,  had a problem and  second source , a Wolfson WM8805 receiver, would only work if the DAC is reset after switching on.

So, the DAC now works and I’ll leave it connected up for a few days so the components can burn-in, and then I can start to improve it. I measured the DC offset on the output and it is very low with almost every op amp I’ve tried.  the worst was -8mV and +3mV.

AVCC uses 56mA playing 44.1Khz/32bit and DVCC uses 17mA. They are both 5V. A good 5v regulator for each is Analog Devices ADP7104. I’m using an AMB Sigma11 for AVCC and a Texas Instruments TPS7A4700 EVM for DVCC. However, this is just to set it up, and I’ll change the power supplies later.

ak4399 evm circuit mkii

The above is an extract of the output circuit given near the top of this page. It shows the second stage of op amps. The first stage has a gain of 1.56, or 3.9dB. I have changed R2 and R3 to 220R – four resistors because the above is duplicated for left and right. This change has increased the gain of the second stage to 2.5 or 8dB, so the voltage output is high enough, about 8V rms, to drive high impedance headphones like Sennheiser HD650, with a suitable op amp like AD8510. However, this increases the load on the preceeding op amp – it now has to drive a 780R load @ ~3V, so some op amps will overheat in this circuit. OPA827 work best of all the op amps I’ve tried in the first stage, and only get slightly warm.

I chose this much gain so I could drop the overall volume and boost the deep bass (50Hz and below) so my HD650 can reproduce these very low frequencies, just like my hifi can, so headphones are not a disappointment by comparison.


I read a technical paper here which suggests adding an AG-grade 600R @ 150Mhz ferrite into the feedback loop near pin2 of the op amp, because this allows the use of current feedback op amps like AD811 and LME49713, without removing the feedback capacitors. They have strong output stages so might drive headphones directly with less THD.

Ferrite in negative feedback

So I did this (removed the screen cut the trace with a wide gap and fitted the ferrite over the gap) and tested it with the above cfb op amps. The AD811 were fine but the LME49713 oscillated, got very hot and messed up heavy bass transients. Unfortunately, the AD811 don’t sound as good as AD8510 so it was an interesting experiment, but I won’t do it again.

I have tested many op amps and found I like OPA827  in the first stage and AD8510 in the second. So, after making an initial choice, it’s time to fine tune it, starting with the power supply. The digital power uses 17mA and I’ve been using an ultra low noise (4.5uV)  reg for this. I wanted to see if  I could get lower noise, so I ran a wire from the +15V op amp input across the top of the pcb ground plane for a clean return current in the ground plane, away from any analogue signal. This went into an RC filter = 100R and 100uF to eliminate ripple and HF noise, and isolate it from the analogue power circuit . This connected to an ADP7104 9V regulator with 15uV noise, which I had mounted to the pcb by removing some of the screen. After this is a 10K VAR potential divider and cap multiplier (10uF Rubycon ZA). This can be adjusted to give exactly 5V output under load. It  reduces the noise from the ADP7104 to the level of the NPN pass transistor; a low noise BC550C, which should be lower than the TPS’ 4.5uV. Here’s a picture to explain, with the circuit diagram in the bottom left corner. It works very well and the sound has even more detail now.

Ultra low noise DVCC

The final change to the power was to use an ADP7104 for the 5V AVCC. I looked for an area that had no signal lines across it, and where I could connect power from the 15V op amp power supply without crossing any more signal lines. The only place is in the centre of the board but even that still requires power to cross the right output signal line, albeit at right angles, but this already happens with this board’s layout anyway. I removed some screen and soldered the gnd pins to the board. I added a 470uF Panasonic AM cap on the input and a 1uF X7R cap on the output.  I didn’t hear any change to the sound quality but it makes the power supplies more convenient because now the board only needs one +/-15V supply. I measured current draw on +15V and it is 115mA.

Regulators fitted

finished underside

A new addition arrived – an isolator / flipflop board. There are two boards and the “slow” (125Mhz) one on the right is destined for this DAC. While I was waiting for it, I removed the sockets and soldered the 4 OPA827 op amps directly into the board. I also improved the power routing for the 5V ADP7104, and I switched the BC550 for a BUP41 so I can use the 5V digital supply to drive the isolated side of the board below.

Acko's isolator

This was designed by Acko and you can read about it in this thread at Diyaudio. I used an ADUM3440 as the isolator, and then upgraded that to a Ti ISO7640M, a 125Mhz Fox Expresso xo as the clock for the flipflops, and SN74AUP1G79DBVR flipflops. It means the output from the CM6631A usb to i2s card is isolated (this adds jitter but blocks noise) and then re-clocked (reduces jitter) so the DAC doesn’t get power supply noise from the PC affecting the sound but the signal is still low jitter so the sound is very good. It worked well for 44.1Khz, but created noise at high sample rates like 192Khz. I did some fault finding and found I needed to improve the clock connections. This removed all the noise for 44.1Khz and its multiples, all the way to 176.4Khz. I use SOX resampler in Foobar to resample all my music to this frequency. It sounds superb.

Temporary hook up :

AK + Acko

And here is the DAC/headphone amplifier, 95% built. The isolator on the DAC side is fed through a 10R resistor, and uses about 45mA. I just need to add a better support for the USB card, and add a couple more filter caps and a dropping resistor for the USB and it is ready to be cased.

AK4399 DAC & Head Amp

And here she is in a case. The transformer is 15-0-15 VAC @500mA (15VA) and the regulators are LM317/337. However, they do not use a resistor at the base to set the output voltage because this increases noise and output impedance. Instead, they use two LM329 7V references in series which gives an output voltage of +/-15.4V, as suggested here. The rectifier diodes are MUR120 fast recovery types. The main capacitors are 2200uF Panasonic AM (audio grade) bypassed with 100nF X7R, and the caps at the base of the 317/337 regs are 22uF tantalum bypassed with 100nF X7R.

Here are some pics of the final build, ready to use as a USB headphone DAC + amp. I really like this DAC !