What About DXD? Surprise!
One of the new digital formats on the block is called DXD. It stands for Digital Extreme Definition. It was initially develop by Merging Technologies for their Pyramix DSD workstation because of the difficulties associated with amplitude changes, EQ and crossfades. It was endorsed by both Sony and Phillips as a viable tool for the preparation of SACD masters. There has long been a close association between DSD and DXD…but there shouldn’t be. DXD is nothing but 24-bit PCM with a very high sampling rate of 352.8 kHz. It would be nice if it was named accurately.
There are a few modern DACs that support the format (and it’s ultra HD PCM flavor of 384 kHz sampling rate at 24-32-bits) and even a few digital music stores that make DXD files available for download. My friend Morten Lynberg of 2L Records uses the format to capture his amazing source recordings and then downconverts the original DXD files to a variety of flavors including 192 kHz/24-bit FLAC and DSD 64 files. He also sells the original DXD files for a very premium price.
There are several important questions that need to be answered when we start to consider Ultra HD-Audio standards (I’ve borrowed the Ultra HD-Audio name from the next generation video standard, which was recently renamed Ultra HD from 4K video). The first question is do we actually need the additional octave or two that moving the sampling rate so high provides? Is there really any meaningful sound up there? Maybe the reason is because the filtering needed at lower sample rates becomes unnecessary when you get to 352.8 or 384 kHz.
My general feeling is that we haven’t yet taken advantage of the bandwidth available using a sampling rate of 96 kHz so why bother quadrupling the file size of a soundfile that sounds the same…or worse. I decided to take a look at some DXD files and see what I could see.
I got a hold of a couple of DXD files and opened them up in Adobe Audition. The program complained that it wouldn’t be able to play back the files AND I do not currently own a DAC capable of playing a native DXD file. But since these are in reality PCM files, I clicked OK on the warning screen and presto…the files opened just like any other PCM soundfile. And the program displayed a spectragraph of the DXD files.
I was astounded at what I saw. Knowing that these were PCM files, I would have expected to see a rich area of amplitude in the 20-20 kHz range with some additional sound (the upper most partials) drifting past 25kHz to around 40-50 kHz. Anything higher than that would be visually and aurally invisible. But as you can see from the plot below, the ultrasonic frequencies exhibited the same “purple haze” that is the signature of DSD files (of any sampling rate). Could it be that Pyramix is really using a multibit DSD technology and merely calling it DXD?
Figure 1 – The spectragraph of an original DXD file and the derivative PCM files at 192, 96 and 44.1 kHz. (Click to enlarge)
To be fair, the haze in this file begins in ernest around 50 kHz and rises to a very troubling -70 dB at its peak. This ultrasonic noise is never going to be heard but why would you want to pay big dollars for an oversized sound file only to discover that the meat of the music stops at around 35 kHz…well within the range of a PCM file at 96 kHz!
The PCM files that are derived from the DXD source don’t display the same problems as the DXD within their own frequency ranges. The green line of the 192 kHz stuff is that same noise. Even the 96 kHz file shows hints of the same HF noise. The only file that doesn’t have the noise is the lowly CD, which might after all be the best sonically.
I’m going to do some additional research on the DXD format. I can think of no reason why the ultrasonics would contain this DSD-like noise unless it is somehow a DSD 256 format. Then what’s the point of having a DXD format at all? And why are sites that sell DXD files often 30-50% more expensive than HD-Audio PCM files at 96 kHz? The files are huge and you seemingly get not benefit.
I’ve decided I will have to reach out to a hardware vendor and record a new file at Ultra HD-Audio specifications…just to see if the HF looks like this DXD plot. Interesting…
Hi Mark,
I agree with your observation that the samples DXD files exhibit the same high frequency noise signature as DSD files converted to PCM (using Audiogate).
By the way, did you have a chance to look at the various spectrograms of “Jimmy And The Crows” DSD file converted to PCM at http://leblogdejn.wordpress.com/spectrum/ ? It tends to show that the shape of the spectrum in the high frequency range depends on the LPF filter used to convert DSD to PCM.
Cheers,
JN
I did look at the spectragraphs that you posted and I think you’re probably right that the high end roll is the result of deliberate use of LPFs. This seems to be the pattern for any files that are derived from DSD sources.
Thank you for looking.
There is indeed no way to get the spectrum of a DSD encoded audio signal without going first through an LPF, whether digital (the result is PCM) or analog.
There might be an exception to this – although I am not sure – in the LampizatOr DSD DAC which claims to use the good old “FM radio principle for tuning into the data stream and extracting music modulation from the carrier wavefrom”. Read more about the LampizatOr DSD DAC at http://www.lampizator.eu/newdac/lampizator/DSD_DAC.html
I am wondering whether “extracting music modulation from the carrier waveform” boils down to applying a LPF, or not… Anyway this is the only DSD only DAC that I know of.
Hi, Mark!
I just visited your pages for the first time to read your posts on RMAF 2013.
I am not a professional of this industry but a humble audiophile who is merely interested in such high resolution sources as DXD and DSD256.
I checked a dozen of FFT spectra of DXD sources both downloaded from Norwegian sites and purchased from their local distributor. The increasing slope of noises at around 50 kHz is observed uniformly in all the DXD sources. I guess the noises are typical quantization noises introduced by a delta-sigma modulator adopted in an ADC device employed in their recordings. I completely agree with your concerns.
However, at the same time, through my actual experiences of listening DXD and DSD256 sources with my ESS ES9018 based DACs, I have very positive impressions on resulting sounds from those high resolution sources. (Of course, LPF filters are applied in the DAC systems explicitly and implicitly.)
My current understanding is “A human listening function is not a rigorous FFT analyzer. ” I think we need more objective experiments and case studies in this field.
Bunpei
Hi Mark,
Correct me if I’m wrong, I believe the sampling rate (44.1kHz, 48kHz, 88.2kHz, 96kHz, 176.4kHz, 192kHz, 384 kHz) is not only pointing to sound frequency that you can record. It’s also how many Pulse (PCM – Pulse Code Modulation) per second a digital audio converter can use to detect or reproduce a sound, the more pulse you have in a second is more accurate compared to the one has lesser pulse. Just like how you plot a graph, the more dots you have, the more accurate your graph is.
Trimi
Only four sample for the 10khz sinus cycle ( 16/44100 ). So any smooth cymbal sound turn into a harsh buzz
Digital sampling theory (and practice) states emphatically that two samples per waveform is enough to recover all of the information. There is no harsh buzz on the cymbals or anything else.
What you are looking at here is the delta sigma noise in the converter. Almost all professional ADCs use delta sigma conversion. The most popular source for DXD files is probably 2L. The use a Denmark AX24 converter, if I am not mistaken. If you go to Denmark’s website, the brochure for the AX24 has a graph showing the noise shaping performance of their converter, which is consistent with most modern multi-bit delta sigma converters.
You’re right about the Delta Sigma noise…although the converters running at 192 kHz don’t exhibit the same noise signature. When I was contacted by the people from Merging Technologies, they said the problem was the converters and that the new Horus converters would solve this problem.
Mark,
Is that why the amplitude of the signal looks so low?
It only appears low compared to the entire range…and it’s a classical recording.
Hi,
Just come across this post (during a search for DXD capable USB interfaces).
You should take a look at Digital Audio Denmark’s explaination of DXD from 2004, Digital Audio Denmark the people providing the most popular and expensive DXD capable A/D converters in studios around the world (including the tech in the Horus mentioned above).
Going this high is not about extended octaves or even more detail, it’s about moving the necessary low pass filters so far away from the audio band that you can use soft filters which move away from Nyquist theory without effecting the audio band. It’s also about the fast impulse response at these high rates and hence more accurate transients etc.
They choose 352KHz (not 384) because DSD datarates are based on the 44.1KHz PCM datarate mulitples and so you can convert between DXD and DSD simply. One thing about DSD is you cannot edit levels or edit anything due to the nature of the format – you can only edit by converting to PCM first, then converting back to DSD after the editing. This ruins the DSD sound signature and you get DSD which sounds like PCM. Therefore you might as well just record straight in DXD and from there derive what you need for each physical format (SACD, CD etc). Only purist, archival recordings with nothing more than cross-fades, fade-ins and fade-outs can remain as DSD (with those fades patched in with DSD->PCM->DSD conversion at only those points).
Here is the link to that document I mention at the beginning – http://www.lindberg.no/english/collection/004.pdf
I hope that’s helpful for understanding DXD!
I’ve seen the Digital Audio Denmark article and read the information on their website. There explanation doesn’t hold water with regards to high rate sampling. The fact is that so-called “brickwall” filters operating in a 96 kHz system are capable of doing a great job with regards to aliasing, imaging, and even transient response. The audio band (take your pick 20-20 kHz or 20-40kHz) is in tact using a mere 96 kHz. The filters remove anything higher than the Nyquist (especially when you consider the amplitude of any potential higher frequencies…which are greater than 120 dB down!) and the supposed advantages for “pre-ringing” is a red herring too. The pre-ringing (and post ringing) exist at frequencies just below the Nyquist (48 kHz) and are very far down in amplitude…no one “hears” pre-ringing. And modern apodizing filters do a great job of making them disappear.
There is absolutely no reason to increase the sampling rate of PCM beyond 96 kHz! There are no sonic benefits, no mechanisms for commercial producers to use anything higher, and the reality is that current production methods (plug ins etc) are limited to 96 kHz.
We’re in an irrational race to higher sample rates and longer word lengths. The DAD group is perpetuating the “higher is better” myth because they sell high rate equipment. When I hear a recording that delivers more than what I have done for 15 years at 96 kHz/24-bits, I may change my mind. So far, I haven’t.
The fact that a company promotes a thing because they make equpiment to do that thing, coupled with an argument therefore their claims shouldn’t be listened to is logically a zero-sum argument – if they are correct then they are making the gear precisely because of all the reasons they claim it is great and one would expect them to be so enthusiastic as to strongly promote their own equipment.
There is no denying however that the DXD format was coined for DSD editing and NOT actually primarily as a sonic advantage and therefore you mis-understand and mis-characterise Merging’s role in that. Merging’s concern was about introducing a DAW on which to edit DSD and there are still limited options for doing this 10 years later!
Remember too that people like Dan Lavry enthusiastically promote their (very good) 96KHz limited A/D and D/A conversion and similarly promote their own white papers on how 96KHz is the highest one should record at and constantly tell people that they should only invest in the best quality 96k convertors… which they make. Of course they would say that!
We’re not on an irrational race to higher sample rates because even the DAD founder can’t see any reason to go to 768KHz because 384 or 352 already are high enough to achieve their technical aims and so will concentrate their efforts elsewhere. Those recording engineers enthusing about high rates will be the big money buying or rejecting the converters which provide the sources and when they find that they will have to replace the whole digital side of their studio to accomodate higher than 384KHz sample rates but for no sonic benefit they can hear, then there will be next to no adoption and therefore no source material.
There are MANY great and top plugins laready working at the higher rates and technically plug-ins can work much better at the higher rates.
I really don’t see a problem with going to DXD et al, especially if it’s simply the original recording format for ease of conversion to DSD – why bother SRCing it to lower rates for distribution? Only for convenience for consumers as consumer DACs catch up. And yet many say that subjectively DXD downsampled to 44.1k or 96k sounds better than 96k or 44.1k recorded sources anyway..
I do see a problem though in obstinately refusing to increase sample rates due to a belief or a technical opinion. Only if higher rates are perceived to degrade the sound should one refuse and yet that can well be converter dependent and it should be a personal choice for your own recording or engineering work as with any recording process or equipment choice. Also the investment needed to upgrade a studio geared around 96k should also be considered if you’re already producing great results – yet still shouldn’t preclude the idea of new studios or those with investment capital to move to the higher rates.
I myself am just a home-studio enthusiast couple with audiophile leanings. I am going to be buying an AX24 in the new year with no financial loss – it being secondhand – and being simply a cost-effective tool at secondhand prices (cheaper and much more cost-effective than say the 4 pairs of Lavry Gold A/D D/A conversion I’d need to compete, coupled with a great 8 channel mic-pre etc etc which would come to ~ £10,000 secondhand! ) so if I can afford that on a low wage… then where’s the problem?
Anyway, don’t take my rambling comment too seriously and hope you had a lovely Christmas Day!
Thanks John for the thoughtful comments. Happy Holidays.
Looking around, older Lavry AD122s and DA924s for an 8 channel set-up would cost me nearer £16,000 secondhand.. then there would be the mic pres. The more I look at Lavry prices, the more I think Lavry really needs to hook people in to his brand on an ideological and psychological level !
Too much money…!
Pingback: I Have A Theory About Why I Like Linear Phase Filters
Se não estou enganado eu já estudei este assunto e parece que se trata de transferir o ruído de dither/quantização para altas frequências ultrassônica, onde ele acaba não se misturando com as frequências audíveis, assim resultando em um som mais limpo. E quanto ao DSD, parece que esse formato tem um pulso mais preciso, e se assemelha a resposta de pulso analógico, sendo que o DSD usa uma AD/DA própria para o formato, porque ela não é só o arquivo de formato DSD, ela é o hardware conversor para que funcione corretamente. Eu disse pulso, mas não sei ao certo se ele é pulso ou impulso de resposta, mas você intendeu, me refiro ao ciclo, frequência e seu tempo de resposta, e se ele tem distorção harmônica. A resposta do pulso no DSD tem amplitude, curto tempo de attak e decay, e sem distorção harmônica, comparado com os formatos e AD/DA tradicionais, este está mais limpo, preciso e próximo a resposta elétrica de um sinal analógico. Como o DSD é um formato que não possibilita edição, veio o DXD para solucionar, mas ambos é necessário algo para filtrar essas frequências altas, e por isso é que usam conversores específicos, pois fogem do sistema tradicional PCM, onde o ruído de quantização está por todo o espectro.
Não tem como tirar conclusões com analizadores de spectro usados para PCM, pois os formatos DSD e DSX vão apresentar o ruído de quantização e que vão lhe confundir se você não souber do que se trata essa curva esquisita no final do espectro.
Veja esses links e entenda melhor e não se confundir com minha explicação enrolada:
http://3.bp.blogspot.com/-Z25yXSVstkg/VRyNncNSJLI/AAAAAAAAEYA/JwlubBQPU28/s1600/dsdresponse_big.png
https://samplerateconverter.com/sites/default/files/u1/dsd-pcm-principles-comparing.png
https://samplerateconverter.com/educational/dsd-pcm