The second point that I'd like to talk is understanding of DXD sources.
As all of us may know, an audio data format called "DXD" is just a 352.8 kHz/24 bit version of LPCM. It has no direct technical relationship with DSD. The DXD audio data is released in an usual WAV format file.
I guess DXD audio sources released by 2L are originally recorded in DXD format using Digital Audio Denmark(DAD) AX24 ADC/DAC modules connected to ProTools. In this case, an anti-aliasing filter would be set to remove high range components over 180 kHz. In this configuration, audio signals in 0 - 176.4 kHz range are to be treated as meaningful ones.
If you filter out high range signals from DXD audio sources, that filtering means you throw away the most valuable part in the sources. I do not think it's a respected way in principle.
I think you can understand it in this way. You obtained an expensive food product containing a nutritive component that can be only cooked effectively in higher temperature above 200 degree. You have just a cooker of 100 degree. In this low cooking temperature, the component tastes bitter. You have filtered away the expensive component as you hate the bitter taste.
However, there can be various approaches in this audiophile world. If you regard higher range signal components over 100 kHz in DXD sources are harmful noises, it's your way. Frankly speaking, I have never found meaningful sonic signals above 50 kHz in actual musical recordings.
Anyway, I appreciate your finding very much as "genuine/fake" DXD checker for 2L sources.
As all of us may know, an audio data format called "DXD" is just a 352.8 kHz/24 bit version of LPCM. It has no direct technical relationship with DSD. The DXD audio data is released in an usual WAV format file.
I guess DXD audio sources released by 2L are originally recorded in DXD format using Digital Audio Denmark(DAD) AX24 ADC/DAC modules connected to ProTools. In this case, an anti-aliasing filter would be set to remove high range components over 180 kHz. In this configuration, audio signals in 0 - 176.4 kHz range are to be treated as meaningful ones.
If you filter out high range signals from DXD audio sources, that filtering means you throw away the most valuable part in the sources. I do not think it's a respected way in principle.
I think you can understand it in this way. You obtained an expensive food product containing a nutritive component that can be only cooked effectively in higher temperature above 200 degree. You have just a cooker of 100 degree. In this low cooking temperature, the component tastes bitter. You have filtered away the expensive component as you hate the bitter taste.
However, there can be various approaches in this audiophile world. If you regard higher range signal components over 100 kHz in DXD sources are harmful noises, it's your way. Frankly speaking, I have never found meaningful sonic signals above 50 kHz in actual musical recordings.
Anyway, I appreciate your finding very much as "genuine/fake" DXD checker for 2L sources.
There are a few errors or misunderstandings in your comments.
352.8 kHz sampling must completely remove high range components over 176.4 kHz. If you set the anti-aliasing filter higher, at 180 kHz as you suggest, then frequencies between 176.4 kHz and 180 kHz will alias down into the range between 172.8 kHz and 176.4 kHz, mixing with the original content at those frequencies. Assuming that we can hear frequencies that high, the audio would be horribly aliased. In practice, the anti-alias filter is set significantly below 176.4 kHz so that the dB/octave slope of the filter will have sufficiently complete attenuation at 176.4 kHz, and thus the -3 dB corner of the filter will be lower. In other words, your 180 kHz figure is completely bogus. If I am missing some documented fact, please cite a reference.
You speak of DXD audio sources as if they have higher frequencies than 180 kHz that might be lost, but they do not. The Nyquist-Shannon sampling theorem states that there can be no frequencies contained above the half-way point, and in practice there is amplitude loss even below that. Nothing is lost in the DXD files, because there are no 180 kHz tones in a 352.8 kHz file, unless you're talking about something other than the DXD format.
Perhaps you are referring to the DSD format, for which DXD was designed as an editing format. You might think that DSD allows frequencies up to 1.4112 MHz, but it does not. At best, delta-sigma would only allow a very faint amplitude at that frequency, say -144 dB, because such high frequencies would only have a maximum amplitude of 1 bit. Even when used at it's ideal maximum bandwidth, DSD would have a gradual slope low-pass response, probably 1st order 6 dB/octave. In practice, most DSD equipment already has a 50 kHz low-pass, if not on the input ahead of the A/D, then certainly on the output. This has been measured. SACD players have no response above 50 kHz, it is only the marketing brochures that speak of the higher sampling rates as if they actually carry audio content - because such braggadocio sells expensive equipment.
Finally, you should realize the DSD works by generating a great deal of quantization noise by operating in 1-bit sampling mode, but then shifts most of this noise up into the higher frequencies above 20 kHz. It is not a matter of merely regarding these higher frequencies as harmful noises, it is an actual fact that there is significantly increasing quantization noise in each higher octave. This is the primary reason that SACD players have a 50 kHz low-pass on the output - so that the noise is not heard.
If you would care to explain why the frequencies above 180 kHz are "the most valuable" then perhaps we can get to the bottom of this. I think part of your problem is that 24/352.8 kHz DXD holds four times as much information as 1/2.8224 MHz DSD. There is a theoretical capacity for very faint amplitudes above 176.4 kHz in the DSD stream, but these are all removed in the analog realm before the DSD stream is created. There is a great deal of literature from the Audio Engineering Society on this topic, so I will not repeat it here.
352.8 kHz sampling must completely remove high range components over 176.4 kHz. If you set the anti-aliasing filter higher, at 180 kHz as you suggest, then frequencies between 176.4 kHz and 180 kHz will alias down into the range between 172.8 kHz and 176.4 kHz, mixing with the original content at those frequencies. Assuming that we can hear frequencies that high, the audio would be horribly aliased. In practice, the anti-alias filter is set significantly below 176.4 kHz so that the dB/octave slope of the filter will have sufficiently complete attenuation at 176.4 kHz, and thus the -3 dB corner of the filter will be lower. In other words, your 180 kHz figure is completely bogus. If I am missing some documented fact, please cite a reference.
You speak of DXD audio sources as if they have higher frequencies than 180 kHz that might be lost, but they do not. The Nyquist-Shannon sampling theorem states that there can be no frequencies contained above the half-way point, and in practice there is amplitude loss even below that. Nothing is lost in the DXD files, because there are no 180 kHz tones in a 352.8 kHz file, unless you're talking about something other than the DXD format.
Perhaps you are referring to the DSD format, for which DXD was designed as an editing format. You might think that DSD allows frequencies up to 1.4112 MHz, but it does not. At best, delta-sigma would only allow a very faint amplitude at that frequency, say -144 dB, because such high frequencies would only have a maximum amplitude of 1 bit. Even when used at it's ideal maximum bandwidth, DSD would have a gradual slope low-pass response, probably 1st order 6 dB/octave. In practice, most DSD equipment already has a 50 kHz low-pass, if not on the input ahead of the A/D, then certainly on the output. This has been measured. SACD players have no response above 50 kHz, it is only the marketing brochures that speak of the higher sampling rates as if they actually carry audio content - because such braggadocio sells expensive equipment.
Finally, you should realize the DSD works by generating a great deal of quantization noise by operating in 1-bit sampling mode, but then shifts most of this noise up into the higher frequencies above 20 kHz. It is not a matter of merely regarding these higher frequencies as harmful noises, it is an actual fact that there is significantly increasing quantization noise in each higher octave. This is the primary reason that SACD players have a 50 kHz low-pass on the output - so that the noise is not heard.
If you would care to explain why the frequencies above 180 kHz are "the most valuable" then perhaps we can get to the bottom of this. I think part of your problem is that 24/352.8 kHz DXD holds four times as much information as 1/2.8224 MHz DSD. There is a theoretical capacity for very faint amplitudes above 176.4 kHz in the DSD stream, but these are all removed in the analog realm before the DSD stream is created. There is a great deal of literature from the Audio Engineering Society on this topic, so I will not repeat it here.
Dear rsdio,
I appreciated your comments and your kind corrections very much.
First, my reference on frequency 180 kHz was not quite adequate. When we talk about anti-aliasing filter functions, 176.4 kHz is correct. I should have been rigorous on this point.
The second point is;
What I wanted to regard as "valuable" was the frequency range from 96 - 176.4 kHz where DXD sources are supposed to have its unique contents. The filtering in the context does not mean the "anti-aliasing filter" over 176.4 kHz, but a high cut filter, cut-off around 20000 Hz, referred by SunRa.
I hope you understand well what I mean.
Thank you very much for your comments again.
Best regards,
Bunpei
I appreciated your comments and your kind corrections very much.
First, my reference on frequency 180 kHz was not quite adequate. When we talk about anti-aliasing filter functions, 176.4 kHz is correct. I should have been rigorous on this point.
The second point is;
What I wanted to regard as "valuable" was the frequency range from 96 - 176.4 kHz where DXD sources are supposed to have its unique contents. The filtering in the context does not mean the "anti-aliasing filter" over 176.4 kHz, but a high cut filter, cut-off around 20000 Hz, referred by SunRa.
I hope you understand well what I mean.
Thank you very much for your comments again.
Best regards,
Bunpei
Thank you for the clarification.
Yes, 96 kHz to 176.4 kHz (or even 192 kHz with 384 kHz audio) is what distinguishes super high resolution audio from standard CD quality audio. If you filter out the content above 20 kHz, then you might as well just save the disk space, use 44.1 kHz, and let the DAC handle all of the oversampling.
However, it all depends upon the front end. As I understand it, DXD files are intended as an editing format for DSD recordings. However, DSD recordings may not contain much audio content above 20 kHz anyway. All of the DSD content that I have analyzed contains nothing but quantization noise in the upper octaves beyond 20 kHz. This is in contrast to PCM front end sampling at 96 kHz or 192 kHz, where there is much less noise and you can clearly see audio content at 30 kHz, 50 kHz, or even 60 kHz to 70 kHz in brief transients.
Can anyone point to DXD equipment that has a PCM front end instead of DSD?
Please look at this page
DAD - Products AX24
and this page.
High Resolution Music DOWNLOAD services .:. FLAC in free TEST BENCH
Thanks.
DSD is apparently only equivalent to 20-bit PCM between 20 Hz and 20 kHz, with even less than 20-bit resolution (more noise) at higher frequencies.
The DXD equipment looks very impressive. It uses a 5-bit delta sigma converter, which avoids the biggest issue of DSD: 1-bit delta sigma cannot be properly dithered because 1 bit leaves no room for addition or overflow, all multi-bit A/D can be dithered sufficiently to completely removed such that there is absolutely no distortion due to quantization. DXD also uses oversampling to go beyond the DSD 20-bit limit.
I think it is important to consider the source, and also to analyze the data, because some of the noise issues might be explained by the unique noise signature of certain delta sigma A/D technologies. DXD seems to not suffer from the worst of these.
DSD is apparently only equivalent to 20-bit PCM between 20 Hz and 20 kHz, with even less than 20-bit resolution (more noise) at higher frequencies.
The DXD equipment looks very impressive. It uses a 5-bit delta sigma converter, which avoids the biggest issue of DSD: 1-bit delta sigma cannot be properly dithered because 1 bit leaves no room for addition or overflow, all multi-bit A/D can be dithered sufficiently to completely removed such that there is absolutely no distortion due to quantization. DXD also uses oversampling to go beyond the DSD 20-bit limit.
I think it is important to consider the source, and also to analyze the data, because some of the noise issues might be explained by the unique noise signature of certain delta sigma A/D technologies. DXD seems to not suffer from the worst of these.
Sorry for the late posting, I've been in a field trip for the last 5 days.
Thanks for the comment on the DXD issue, the info you just gave is by any means better and more concise than everything I could find on the net.
Now, to clarify some aspects on the issues I have with the DXD playback on AK4396:
1. The low pass filter I applied was for experimental purposes. I simply wanted to show that maybe I don't have a hardware problem but rather a file format problem. I am not willing to play DXD files low passed at 20Khz as I don't find this desirable.
2. Nobody really explained until now why can I play upsampled 384Khz files without any noise coming out (using the 24Mhz clock of the U2I and not a 40Mhz clock as it would be appropriate) and why the DXD files should present the noise because of the slower clock I use and not because of the additional high frequency content it has.
I am aware the DXD files are standard PCM files however they use a different anti-aliasing filter than standard digital recordings, specifically needed for DSD conversion as DXD is an editing format for DSD. Furthermore, because of this you could say that the two formats should have a technical connection.
Thanks for the comment on the DXD issue, the info you just gave is by any means better and more concise than everything I could find on the net.
Now, to clarify some aspects on the issues I have with the DXD playback on AK4396:
1. The low pass filter I applied was for experimental purposes. I simply wanted to show that maybe I don't have a hardware problem but rather a file format problem. I am not willing to play DXD files low passed at 20Khz as I don't find this desirable.
2. Nobody really explained until now why can I play upsampled 384Khz files without any noise coming out (using the 24Mhz clock of the U2I and not a 40Mhz clock as it would be appropriate) and why the DXD files should present the noise because of the slower clock I use and not because of the additional high frequency content it has.
I am aware the DXD files are standard PCM files however they use a different anti-aliasing filter than standard digital recordings, specifically needed for DSD conversion as DXD is an editing format for DSD. Furthermore, because of this you could say that the two formats should have a technical connection.
Your assumption is correct, I am using the 22.5792 clock provided by the exaU2I device.
I agree entirely with you! I also can't understand why on earth I have a noise free playback of 384 and 352.8Khz files. I am referring of 192Khz upsampled files and also two 352.8Khz/32bits files marketed as master files - I believe now these 352.8Khz files either are recorded at 352.8Khz using a different technique than DXD recording, either these are simply upsampled, the same way as I did.
The only abnormal play (as is, with noise) to my years is with 352.8Khz DXD files. This and the little experiment I did with the low pass filter leads to the conclusion that files recorded on DXD specs are somehow different than standard, PCM files at 352.8Khz.
I completely agree with this opinion.
My recent powerful tool for this purpose is free software, WaveSpectra, a FFT analyzer program for WAV files or Windows audio input. This program was developed and released by efu in Japan.
(There is no manual written in English. However, you will be able to use it without manual documents.)
http://www.ne.jp/asahi/fa/efu/soft/ws/WS140.ZIP
They say, on Windows Vista/Windows 7, 352 kHz/24 bit WAV can be analyzed with this analyzer.
Unfortunately, as my PC is of Windows XP and I made a trick. I hacked sampling frequency value in a DXD file from 352800 Hz to 176400 Hz and removed JUNK chunk in its WAVE header. The analyzer software could handle the tweaked DXD file as 176.4 kHz WAV and show right spectra with 1/2 frequency scale.
I'd like to recommend that those who are interested in hires audio sources examine the unique noise signature in both digital source data and output analog signals of their DAC. They will find enough realities there.
An interesting comparison with graphs on different audio formats (DSD, DXD, redbook, plain PCM at higher sampling rates) from DAD: The advantages of DXD for SACD
edit: Thanks for the last Link Bunpei, I've also downloaded Sonic Visualizer and I will compare the spectrum of the various 352.8Khz files I own and play differently on the ak4396.
edit: Thanks for the last Link Bunpei, I've also downloaded Sonic Visualizer and I will compare the spectrum of the various 352.8Khz files I own and play differently on the ak4396.
Last edited:
It is even much stranger than that
If you take a look on the clock chips on the exaU2I you will discover that the MCLK for the 44.1k modes (including DXD / 352.8k) are NOT 22.5792MHz, but 11.2896MHz
Hi RayCtech, I am very well aware the board is using two clocks for the two frequency domains, I just looked at Bunpei's reply and I was just wanting to confirm that indeed the DAC is working with a much lower clock. It was just a typo
edit: it is still weird though that using a 22 MHz clock I can play 384Khz files
edit: it is still weird though that using a 22 MHz clock I can play 384Khz files
Hi RayCtech, I am very well aware the board is using two clocks for the two frequency domains, I just looked at Bunpei's reply and I was just wanting to confirm that indeed the DAC is working with a much lower clock. It was just a typo
edit: it is still weird though that using a 22 MHz clock I can play 384Khz files
It is weirder that a 11.2896MHz masterclock "can" play DXD / 352.8k files.
With 384k files the masterclock are 24.576MHz and only half-weird in comparison...
As I commented earlier in this thread I expect your DAC are downsampling in some kind of strange fashion..
Hi all,
I am coming back with some spectrograms.
1. 352.8k DXD file, plays with audible noise
2. 352.8k DXD file above, low passed with a second order filter at 20k, plays with no noise
3. different 352.8k file, plays with no noise.
4. 352.8k DXD file, the same as the first two, this time low passed with a 1st order filter at 20k - I get audible noise on this.
I am coming back with some spectrograms.
1. 352.8k DXD file, plays with audible noise
2. 352.8k DXD file above, low passed with a second order filter at 20k, plays with no noise
3. different 352.8k file, plays with no noise.
4. 352.8k DXD file, the same as the first two, this time low passed with a 1st order filter at 20k - I get audible noise on this.
Attachments
Last edited:
Thanks for that link. The information is interesting, but I think it may not present the full story.
Some of the comments in the article are not exactly true, such as the claim that DSD has a higher bandwidth. That may be theoretically true, but I have yet to see evidence that any actual DSD recording contains more than 50 kHz of audio bandwidth. In other words, if you think of the output of an SACD player as 20/96 plus noise, then you can fairly compare with 24/96 to 24/352.8 or 24/384.
There is also the comment that editing is compromised in quality by being limited to 1-bit. Again, I have seen no evidence that it is even possible to edit a 1-bit stream, so that comment seems to come out of thin air. If you've ever looked at the gate level implementation of a processor, you would be aware that all math operations in a CPU are performed on multi-bit values. These adders are available as discrete parts and can also be implemented in FPGA. However, I have never seen evidence of 1-bit arithmetic units, and thus I do not understand how an editor could even be designed. At the pure mathematical level, 1-bit only contains two numbers: +1 and -1 (or +0.5 and -0.5 in much of the hard core literature), and thus there is no way to add (mix) or subtract (invert) or multiply (gain) the samples. If it were to exist, a 1-bit ALU would need to store a long state sequence of stream bits just to do simple operations. My understanding is that all editing of 1-bit audio streams is accomplished by converting to PCM, regardless of whether the file being editing is DSD or DXD. If anyone has evidence of direct 1-bit processing, please cite references.
The article eventually mentions the difficulty of 1-bit editing, but leaves the implication that this was only difficult in the beginning before editing formats were available. The problem is that 1-bit editing is still impossible, so I find it misleading to say that it was only a problem in the early development of the format.
Figure 1 bothers me. If you ignore the DSD sampling, and only look at the analog anti-aliasing filter, then Figure 1 might be accurate. However, the DSD 1-bit process itself filters the data by nature of its slew limiting. The article as much as admits that these frequency response graphs are made from computer simulations based on documentation, and that they do not come from actual measurements of the DSD equipment. AES members have measured actual, official DSD converters, and the response is much worse. Sony's public documentation of DSD leaves out many 'proprietary' details which make the computer simulations of Figure 1 entirely inaccurate. At the very least, a perfect, sharp impulse as shown in Figure 1 cannot be possible because DSD can only increase the output one step at a time. Assuming 20-bit performance of DSD, it would require half a million samples to reach half scale, and a million samples peak to peak. That is not a sharp spike. If you were to attempt to achieve 24-bit performance from DSD, the slew rate would be such that eight to sixteen million samples would be required. In contrast, DXD and all PCM formats can change from 0 to full scale in a single sample. There is no slew rate in the PCM format itself, unlike DSD. Figure 1 does not represent the actual performance of DSD, because at least two major factors are ignored.
Also disappointing is that the article mentions the smooth slope of the DSD anti-alias filter, but never specifies the frequency of this filter!
Figure 2 is wrong, since measured performance of many DSD players shows frequency response to be less than that of a 96 kHz PCM recording. Perhaps there are DSD players with better frequency response, but I have not seen actual measurements.
Fortunately, Figure 3 seems to be correct. DSD does have a great deal of noise - more than any other format. This noise is still present even when the DSD original is converted to DXD or 24/96 HD downloads. The Scarlet Book noise limit is equivalent to 3-bit resolution in the high frequency range, down from an already disappointing 20-bit resolution in the 20 Hz to 20 kHz range.
The articles cites a single AES paper "Why DSD is the best choice as a digital audio format." For contrast, I would like to cite the AES responses to that paper:
Why Professional 1-Bit Sigma-Delta Conversion is a Bad Idea, AES paper 5188, Sep 2000
Towards a Better Understanding of 1-bit Sigma-Delta Modulators, AES paper 5398, May 2001
- Part 2, AES paper 5477, Sep 2001
- Part 3, AES paper 5620, May 2002
Sorry, no links. These papers are available for purchase at the AES.org for $20 each, $5 each for AES members, and free for literature library subscribers.
P.S. None of my comments here are intended to say that DXD is not better than DSD. I will agree that DXD is certainly better than DSD. However, the cited article includes marketing propaganda in many places instead of engineering and mathematical facts, and therefore I suggest that it represents a disservice to audio engineers.
Would you tell us actually measured and confirmed frequencies of Bit Clock and Master Clock when you guys play 352.8 kHz or 384 kHz audio sources?
If you get 22.5792 MHz Bit Clock in spite of 11.2896 oscillator, the FPGA must have an internal (PLL) multiplier.
A short comment on the spectograms above. At the risk of sounding like a broken wheel, I believe they support my findings.
1. the genuine DXD file has a large noise content up to 170Khz, as expected. I will compare this to a master 192Khz file I've been informed it had been recorded at this sample rate to check if it has the same noise distribution. I believe it won't.
2. the low passed DXD file has little information above 20Khz, therefore no audible noise effects. Ofcourse this is not desirable as there might be relevant audio information up to 40Khz. I don't know what to believe about the idea that content above 22Khz has any importance in the audio band however the bandwidth has an impact on the impulse response and this is a reason good enough to perfect 352.8Khz playback in my system.
3. the first 352.8k I played in my system (with no noise), was sourced from a site or recording studio I can't rememebr of. I can't explain the sudden low pass at 88Khz. I believe it's either an upsampled 192Khz file, either a genuine 352.8Khz recording which employs a very sharp low pass at that point.
4. A shalow low pass (6db) on the DXD file lets some amount of noise density in the high frequincy area. I believe this noise interacts with the DAC and produces the noise/hiss I hear in the audioband. This noise is fainter in intensity than on the original DXD file.
These aspects combined with the fact that Anagram technologies had a module working with AD1955 and wolfoson DACs capable of 384 (and above) playback, implemented in Cambridge Audio cd-players, makes me believe there's a chance my DAC actually passes high resolution content. Interesting enough the upsampling module at the core of this system was working with a 24Mhz clock.
Also there's a recent APL usb streamer (check 6moons) capable of 384Khz using AKM4399, a DAC similar to AK4396 in it's clock requirements.
I will try to figure out how the different internal modules of the AK4396 are working and how do they contribue to the normal operation modes of 64fs and 128fs and check if the presence of a 352.8 or 384Khz signal simply makes the DAC bypass some of these modules.
1. the genuine DXD file has a large noise content up to 170Khz, as expected. I will compare this to a master 192Khz file I've been informed it had been recorded at this sample rate to check if it has the same noise distribution. I believe it won't.
2. the low passed DXD file has little information above 20Khz, therefore no audible noise effects. Ofcourse this is not desirable as there might be relevant audio information up to 40Khz. I don't know what to believe about the idea that content above 22Khz has any importance in the audio band however the bandwidth has an impact on the impulse response and this is a reason good enough to perfect 352.8Khz playback in my system.
3. the first 352.8k I played in my system (with no noise), was sourced from a site or recording studio I can't rememebr of. I can't explain the sudden low pass at 88Khz. I believe it's either an upsampled 192Khz file, either a genuine 352.8Khz recording which employs a very sharp low pass at that point.
4. A shalow low pass (6db) on the DXD file lets some amount of noise density in the high frequincy area. I believe this noise interacts with the DAC and produces the noise/hiss I hear in the audioband. This noise is fainter in intensity than on the original DXD file.
These aspects combined with the fact that Anagram technologies had a module working with AD1955 and wolfoson DACs capable of 384 (and above) playback, implemented in Cambridge Audio cd-players, makes me believe there's a chance my DAC actually passes high resolution content. Interesting enough the upsampling module at the core of this system was working with a 24Mhz clock.
Also there's a recent APL usb streamer (check 6moons) capable of 384Khz using AKM4399, a DAC similar to AK4396 in it's clock requirements.
I will try to figure out how the different internal modules of the AK4396 are working and how do they contribue to the normal operation modes of 64fs and 128fs and check if the presence of a 352.8 or 384Khz signal simply makes the DAC bypass some of these modules.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.