It seems like you're on the right track. My only confusion is that the 4 images you attached do not seem to show any of your low-pass filtering, and thus I cannot correlate your descriptions with the images. Perhaps it is merely that the software you are using is different than what I use.
While I'm on the topic, I suggest that you use linear frequency scaling for this sort of analysis. Logarithmic frequency scales are great for musical interpretations of audio, or for creative equalization, since they weight the frequencies similarly to how we hear them. However, when you're looking for aliasing or evidence of upsampling, such mathematical processes are easier to recognize on a linear frequency scale. I've certainly discovered serious aliasing in 'free' high definition audio downloads, and I never would have noticed with the typical mastering spectrograph.
I have reproduced your observations and obtained similar results on my system, a combination of SDTrans192 and AK4399. I also found similar spectral profile for some 352.8kHz sources. Your spectrographs are very exact.
As long as your purpose is to play 352.8kHz sources in a tricky exploiting way, your approach is very interesting. I agree with your speculation that high frequency components in those source files causes the "hiss" noises. If you find your highest priority in just playing 352.8kHz with your current configuration, stripping off high frequency components by low-pass filtering is a clever way.
However, my basic understanding is that you are using the DAC chip in unintended way. When you want to get a fair result on 352.8kHz plays, I think you must apply 45.1584MHz Master Clock to the chip though it is not within the upper limit in its datasheet.
I believe the under-frequency Master Clock you apply is causing malfunction of 8x OSF in the chip and that is producing hiss noises by interacting specifically high frequency components in the original audio data. You can't blame anyone else for this noise because you run the chip in unintended way and the 2L DXD sources are valid enough.
Frankly speaking, I appreciate your finding very much because it is just very interesting. Your finding made me check spectral profiles on DXD sources and that brought me useful considerations.
Thanks rsdio. Do you have a link to the software you are using? I'm not sure mine lets me change the scale.. By the way, how should I look after aliasing in a spectogram? I see only noise 
From the four images, only the second and the last ones are low passed by me.
For example, comparing the first image (an original DXD file) with the second image (the same DXD file low passed with a 4th order filter) shows the masive wipeout it does in the higher region.
The last image is the same dxd file with only a first order filter, the attenuation seems much lower.
edit: Thanks for your feedback Bunpei, much appreciated. I don't really intend to play 384k files in an inappropiate manner, it's not my intention as I've emphasised enough in other posts. I am just trying to figure out what is happening.
(I agree though that probably I won't touch a buffalo dac too soon and that if I'll manage to make the AKM part work in a proper way with high resolution material it would be great).
I was just wondering, is there any chance that the ESS DAC has some kind of low pass filter at a very high frequency and this is the reason you are not hearing the noise I get? If you have an ocilloscope I guess you could check that by looking at the outputs of your bufalo DAC.
I admit though that your explanation for the reason I am experimenting the noise is some malfunction that becomes apparent with sources having large amounts of noise above 90Khz (I have no such problems with 192Khz files and low passed 352.8Khz files low passed at roughly the nquist limit, as the audio file in the third image from my previous post)
From the four images, only the second and the last ones are low passed by me.
For example, comparing the first image (an original DXD file) with the second image (the same DXD file low passed with a 4th order filter) shows the masive wipeout it does in the higher region.
The last image is the same dxd file with only a first order filter, the attenuation seems much lower.
edit: Thanks for your feedback Bunpei, much appreciated. I don't really intend to play 384k files in an inappropiate manner, it's not my intention as I've emphasised enough in other posts. I am just trying to figure out what is happening.
(I agree though that probably I won't touch a buffalo dac too soon and that if I'll manage to make the AKM part work in a proper way with high resolution material it would be great).
I was just wondering, is there any chance that the ESS DAC has some kind of low pass filter at a very high frequency and this is the reason you are not hearing the noise I get? If you have an ocilloscope I guess you could check that by looking at the outputs of your bufalo DAC.
I admit though that your explanation for the reason I am experimenting the noise is some malfunction that becomes apparent with sources having large amounts of noise above 90Khz (I have no such problems with 192Khz files and low passed 352.8Khz files low passed at roughly the nquist limit, as the audio file in the third image from my previous post)
Last edited:
I also played the same sources with my Buffalo II with 96MHz Master Clock.
In this configuration, I listened no noises on both OSF/ON mode and OSF/OFF mode.
Two points we can consider on ES9018.
1. Low pass filtering
ES9018 has IIR digital low pass filter after FIR OSF. Its highest cut-off frequency is 60 or 70 kHz or so. It must be effective only on OSF/ON mode. On the other hand it is not effective on OSF/OFF mode.
2. Master Clock Frequency
When I applied 80MHz Master Clock to ES9018 before, I listened similar noises on 352.8 kHz plays. Currently with 96MHz Master Clock, no noises.
That's why I emphasize under-clocking in your configuration.
As I can generate a sweep test signal ranging to 176.4 kHz, I will observe and compare analog outputs of both ES9018 and AK4399. I think I can check the effect of internal low-pass filters.
Monitoring raw audio output on AK4399 chip
I just wanted to listen to "raw" audio output of AK4399 chip without any additional analog stage.
I connected my tweaked, "wired in balanced mode", ER-4S earphone directly to + - output pins of the DAC LSI and was much surprised!
(Input impedance of the bare driver is approximately 5 ohm and an additional variable resister in series.)
The sound I listened was amazing! Enough power or energy with very detailed and exact reproductions. Bass is enough but not boomy. I was immediately attracted to this DAC!
(DAC chips that I have been familiar with are, WM8741, ES9018 and PCM1795.)
I just wanted to listen to "raw" audio output of AK4399 chip without any additional analog stage.
I connected my tweaked, "wired in balanced mode", ER-4S earphone directly to + - output pins of the DAC LSI and was much surprised!
(Input impedance of the bare driver is approximately 5 ohm and an additional variable resister in series.)
The sound I listened was amazing! Enough power or energy with very detailed and exact reproductions. Bass is enough but not boomy. I was immediately attracted to this DAC!
(DAC chips that I have been familiar with are, WM8741, ES9018 and PCM1795.)
Thanks Bunpei for checking this out. I am glad to know that AK4399 is also working because I want to use it also. It's 32bits capable and this makes it interesting. It's also newer than ak4396. Unfourtunately I can't source it here in Romania.
I am running the ak4396 through a capacitor, unbalanced and directily into an amp. I enjoy it very much but I want to compare it with some output stages also.
I had this funny thought the other days. Is there any chance that running the DAC balanced, as you did, some of the noise cancelles out? The noise should be the same on both +- branches but opposite in phase. That would be funny
I am running the ak4396 through a capacitor, unbalanced and directily into an amp. I enjoy it very much but I want to compare it with some output stages also.
I had this funny thought the other days. Is there any chance that running the DAC balanced, as you did, some of the noise cancelles out? The noise should be the same on both +- branches but opposite in phase. That would be funny
Last edited:
Line transformer for AK4399
Before I tried the direct connection of Etymotic Research ER-4S, we listened to an output of prototype line transformer developed by a Japanese maker.
The raw balanced output signals of AK4399 were connected to the line transformer and a single-ended output of the transformer was connected to a high impedance line input terminal of a conventional tube amplifier. The line transformer includes low-pass filtering function. The transformer is newly designed for the DAC chip on a different concept from that was found in European makers. You may find a new model in the near future.
Before I tried the direct connection of Etymotic Research ER-4S, we listened to an output of prototype line transformer developed by a Japanese maker.
The raw balanced output signals of AK4399 were connected to the line transformer and a single-ended output of the transformer was connected to a high impedance line input terminal of a conventional tube amplifier. The line transformer includes low-pass filtering function. The transformer is newly designed for the DAC chip on a different concept from that was found in European makers. You may find a new model in the near future.
Ah, now I finally see it. Because the dB color scale is along the left edge, I though that amplitude was height. I now see that frequency is height, and amplitude is color. I'm less confused.Thanks rsdio. Do you have a link to the software you are using? I'm not sure mine lets me change the scale.. By the way, how should I look after aliasing in a spectogram? I see only noise
From the four images, only the second and the last ones are low passed by me.
For example, comparing the first image (an original DXD file) with the second image (the same DXD file low passed with a 4th order filter) shows the masive wipeout it does in the higher region.
I should also explain that I do not actually use the spectrogram feature of the plugin that I have, but rather I look at the FFT graph. I use AULab as a host application, along with several plugins for metering. The plugins are from Elemental Audio, a company that is no longer in business. They did sell their plugins to another company, so perhaps they are still available. In addition to FFT, I have a level meter that implements Bob Katz' K-20, K-14, and K-12 scale, a BitViewer that shows whether the material is 16-bit, 24-bit, or 32-bit, and a new plugin from SSL (Solid State Logic) that depicts the bit depth in a different manner along with a simulation of whether the analog output of the DAC will be clipped due to inter-sample overages or anti-alias filtering.
I'm using an 8192-point FFT, and the plugin graphs left and right on the same grid with different colors. The display is a line graph representing the frequency response, with a peak, average, and instantaneous in different shades. It is the FFT where I select a linear frequency grid, and I also select the 140 dB amplitude window so I can see the quantization noise floor of sources such as DSD.
The Elemental Audio plugin that I'm describing is called Inspector XL "IXL Spectrum Analyzer."
Have you tried this direct connection with a current output DAC? I would think that the sound would be even more precise from a current source (see First Watt F1 & F2, F1J & F2J)
Yes. For ES9018, it is feasible for my own use but I hesitate to recommend it to other people. Power is not enough. For WM8741, it is not feasible. The signal is too weak.
A couple of bipolar transistors could possibly boost the native current output to a strength sufficient to drive your monitors.
I created a sweep sine wave of frequency range from 22.05 kHz to 176.4 kHz in DXD 352.8 kHz/24 bit WAV format. I played it with my SDTrans192 Rev. 2.1.
In the case of Twisted Pear Audio Buffalo II DAC (ES9018) with 96MHz master clock, perfect sine waves could be observed as a DAC output analog signal over the entire sweep range. The ES9018 was set to OSF/ON mode.
On the other hand, for AK4399 SB Evaluation Board, in the frequency range less than approximately 60 kHz, a normal sine wave was observed. However, above 60 kHz, the output wave shape became collapsed and at near 176.4 kHz range, the output wave were observed as a very low audible frequency sine wave. That reminded me of a kind of "aliasing". In my case, master clock frequency is 22.5792 MHz.
Thus, my conclusion is;
Using master clock frequency 22.5792 MHz and I2S signal of 352.8 kHz/24 bit, high frequency components above 50 or 60 kHz can't be played correctly on AK4399. Particularly, near 176.4 kHz component is played as incorrect low frequency audible sound. This might be recognized as "hiss noise" in actual plays of DXD recorded sources.
To SunRa,
If you want the sweep sine test signal that I created, please give me a private message.
Do you mean "low audible" as in low amplitude? That would not be aliasing, but rather it would just be simple low pass filtering with some accumulative dB loss per octave.
Do you mean "low frequency"? What is the ratio of input frequency to output frequency? If you know the ratio, then you will know more about the cause of the aliasing.
I observed descending low frequency non-collapsed sine wave. I did not measure its frequency. The frequency might be (176.4 kHz - ft) or (176.4 kHz - ft)*n, where ft is the frequency of the original test sine wave.
352.8 kHz/24 bit play on PCM1795
I had a chance to hack Onkyo DAC-1000(s), TI BB PCM1795 Dual Mono DAC.
I injected I2S signals to those PCM1795 chips and controlled them from Arduino.
As SunRa tried on AK4396 DAC, I played a few 352.8 kHz/24 bit sources in the same manner for 192 kHz/24 bit sources. Master clock frequency was 22.5792 MHz.
As a result, I confirmed output sounds were very similar to those obtained on AK4399. For only 2L sources I had the same kind of noises.
I have not tried a "Digital-filter bypass mode".
I had a chance to hack Onkyo DAC-1000(s), TI BB PCM1795 Dual Mono DAC.
I injected I2S signals to those PCM1795 chips and controlled them from Arduino.
As SunRa tried on AK4396 DAC, I played a few 352.8 kHz/24 bit sources in the same manner for 192 kHz/24 bit sources. Master clock frequency was 22.5792 MHz.
As a result, I confirmed output sounds were very similar to those obtained on AK4399. For only 2L sources I had the same kind of noises.
I have not tried a "Digital-filter bypass mode".
Hy guys,
Just wanted to tell that I am still around, unfourtunately I had to adress some personal matters in this period and couldn't make any progress on the DAC board itself, nor following this discussion.
I wanted to thank Bunpei for taking the time and testing my claims, I didn't had the means to further probe them and the fact that you jumped in and tested this with AK4399 and BB PCM1795 is great.
Also I believe that your test at post 112, preety much settles this down. The AK4399 (and possible similar DAC chips such as Burr-Brown) are capable of playing only up to around 60Khz of audio content. This might be a chip design problem or it could be addressed by an increase in the clock frequency. This is yet to be found out.
Regarding the development of this board, to be honest I am very interested in the AK4399 and requested for two samples, unfortunately the initial response is not too great. I hope I'll source these parts eventually.
The next steps are as follows:
1. Finish the updated schematic with new regulators (surprise, surprise) and op-amp buffer for AK4396. I am still in the process of making the parts in Eagle for the new components.
2. Finish the PCB design with the new schematic.
3. Migrate the new design to AK4399, if I get a sample (it's fairly simmilar to AK4396 in terms of design, the package differs significantly)
4. Re-work the entire PCB as AK4399 is a different beast.
That would be all, I'll let you know how this turns out. Thank you again for your support
Just wanted to tell that I am still around, unfourtunately I had to adress some personal matters in this period and couldn't make any progress on the DAC board itself, nor following this discussion.
I wanted to thank Bunpei for taking the time and testing my claims
, I didn't had the means to further probe them and the fact that you jumped in and tested this with AK4399 and BB PCM1795 is great. Also I believe that your test at post 112, preety much settles this down. The AK4399 (and possible similar DAC chips such as Burr-Brown) are capable of playing only up to around 60Khz of audio content. This might be a chip design problem or it could be addressed by an increase in the clock frequency. This is yet to be found out.
Regarding the development of this board, to be honest I am very interested in the AK4399 and requested for two samples, unfortunately the initial response is not too great. I hope I'll source these parts eventually.
The next steps are as follows:
1. Finish the updated schematic with new regulators (surprise, surprise) and op-amp buffer for AK4396. I am still in the process of making the parts in Eagle for the new components.
2. Finish the PCB design with the new schematic.
3. Migrate the new design to AK4399, if I get a sample (it's fairly simmilar to AK4396 in terms of design, the package differs significantly)
4. Re-work the entire PCB as AK4399 is a different beast.
That would be all, I'll let you know how this turns out. Thank you again for your support
I'm looking for an AK4396 kit that is exactly that: a kit. Only the SMD parts would be soldered in on the pcb, as they are the harder to do right with a DIY solder pen.
The only AK4396 kit I could find on eBay is all assembled, but upgrading the parts seems to be taking this DAC to higher quality levels.
Does anyone know of such a kit as I'm looking for?
The only AK4396 kit I could find on eBay is all assembled, but upgrading the parts seems to be taking this DAC to higher quality levels.
Does anyone know of such a kit as I'm looking for?
I never heard of any. I searched the net for days.
For soldering SMT parts I have to tell You it is not so hard as it seems.
There are tools which is needed for them but not many knows it. It is the flux. The first I tried it I was astonished how easy the soldering was with these. I even soldered a TSSOP-28 chip with it successfully.
So this is the trick. Try this one:
The name is: ANC559TF
I put some really thin layer with a toothpick to the solder pads, and a really tiny bit of solder to the iron (I hardly can see it) and touch the resistor legs. Only touch it. You will see how the flux reacts, and You are ready. Only 1sec. Of course You have to hold the resistor until You are ready. Usually it takes more time to position it than soldering.
For soldering SMT parts I have to tell You it is not so hard as it seems.
There are tools which is needed for them but not many knows it. It is the flux. The first I tried it I was astonished how easy the soldering was with these. I even soldered a TSSOP-28 chip with it successfully.
So this is the trick. Try this one:
The name is: ANC559TF
An externally hosted image should be here but it was not working when we last tested it.
I put some really thin layer with a toothpick to the solder pads, and a really tiny bit of solder to the iron (I hardly can see it) and touch the resistor legs. Only touch it. You will see how the flux reacts, and You are ready. Only 1sec. Of course You have to hold the resistor until You are ready. Usually it takes more time to position it than soldering.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.