• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Remaining Considerations on DSD

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
13,345
Likes
20,698
Maybe you should research so called chipless dacs.
 
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
chipless dacs? no thanks.

I mean, I haven't even...

I skim-read Lipshitz but am yet to comprehend it. Then there's Reefman's paper in the same year, and many others we can't access.
 
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
The star wars CDs of ’97, ‘04 and ’15 are controversial. Especially for ‘04 and ’15, the 4-6 trilogy is dsd mastered, possibly at Bernie grundman, but has no difference to the original RCA ’97 (some people did a comparison). What is very likely is that they took the CD or its master, converted or recorded it to DSD, and did SBM Direct, but if so, the noise shaping should have shown up. So it’s a mystery.

Let's assume they did do SBM Direct. What on earth for!? It can only be that Sony Classical or people working there are steeped in SBM Direct and DSD. They keep DSD copies in their archives, so naturally they use SBM Direct before releasing material. One questions how they created those DSD archives in the first place. From PCM in some instances. Terrible if true. Or is it?
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
13,345
Likes
20,698
Chipless dacs use an fpga and send high rate Bitstreams to a resistor and cap at the output. Instead of being dead tech like dsd is a current thing in some circles.
 
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
So Chipless dacs are the future?

I have an aversion for boutique chip-less DSD DACs!

It goes against my principle of audio playback and recording with the highest quality for the lowest cost. What you get with boutique chip-less DSD DACs, unless you’ve made them yourself or someone has released a practical open source chip-less DSD DAC complete with simple instructions, is lowest quality for the highest cost…

Somewhere at a deeper level I’ve a strong hope that the thinking behind DSD, as discussed by Nishio and Massenburg, of simplifying the audio processes, will prevail in the future.

Come to think of it, nobody ever thought ‘hi-fi’ when they saw the words PCM. But when one sees DSD, they think ok, this is probably ‘hi-fi’. As though the format is magic. There’s just no rational basis for such thought. It equates PCM to cassette tapes or worn out records, and DSD to CDs. The difference between the formats could all be down to equipment quality. The more ridiculous PCM = harsh CDs and DSD = near analog softness is a worse analogy. When in reality the difference is close to nil for those without the ultra rare, ultra expensive studio equipment.
 
Last edited:
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
Introduction: In 2021, many USB DACs and portable DAPs are offering "DSD support" and "native DSD playback", which are important keywords in terms of sales. Also in DSD, there are more and more devices that can support formats such as DSD128 (5.6MHz), which is twice as fast as DSD64 (2.8MHz), and DSD256 (11.2MHz), which is four times as fast. It's useful for home USB DACs to be able to support such high sample rates, as it shows off the technical prowess of the manufacturer (although, conversely, there's little way to stand out unless you do). I think that from 2023 onwards it will become the norm for all chips to be able to handle all these high rates.

Classical recording work: It is widely known that DSD is a format that works best for 'one shot' recordings, such as jazz and classical concerts, or digitising old analogue recording tapes. Once recorded, DSD data is cumbersome to edit (or rather, degrades each time it is edited), so high-res PCM is better for studio albums where various takes can be cut and pasted together at a later date. In the case of classical recordings, for example orchestral studio recordings, it was common practice to set up a number of independent microphones for each instrumental section, record each separately (48 tracks or so) digitally, and then cut and paste the recordings later on the computer, adjusting the timbre and volume of each section. This was a common practice.

A recording producer and engineer must be able to understand not only the sound quality, but also the flow and content of the music, for example, "the trumpets should be more prominent here, so let's increase the volume of the trumpet track", or "the violins are covered by the singers here, so let's suppress them a little". It's not just about sound quality, it's about understanding the flow and content of the music and enhancing the musicality. If you look at snapshots of studios of the time, you will often see a producer with a score in his hand and a conductor soundchecking a passage that has just been recorded. In the same way that a film director is obsessed with a single emotional scene, it was commonplace in music to spend two or three hours re-shooting a one-minute passage in order to get it right, and to spend 100 hours editing a one-hour album after the recording session. It was a world in which it was commonplace to spend 100 hours editing a one-hour album.

It is rare to find a violin symphony by a famous performer where the orchestra and the violinist's solo performance were recorded on separate days in different studios. This is a common practice in rock and pop music, the so-called karaoke method. By recording separately, it's easier to edit the song later, and if you don't like the sound of a single note (or if you played it wrong), you can fix it. There is a famous story about a great opera singer who, because of age, could no longer play the high notes of his music, so he had to replace the high notes with another singer's voice. All of this is the result of painstaking multiple recording and careful editing on a generous budget. These days, the music industry, not just classical records, is on a tight budget, and all studios, no matter how good their multiplex recording facilities, want to produce albums in the shortest possible time, as the labour cost of re-recording and editing dozens of times is too high. The members of the orchestra were paid by the hour or overtime, as determined by the trade union, so it costs money to keep them in the studio for re-recording. Also, music fans have been complaining that the edited studio recordings are too perfect and uncomfortable, not at all like the actual live concert experience. If this is the case, it's better to record a live concert performance in high quality and sell it, so that you don't have to edit it later. In the 80's, 90% of new releases on the major labels were studio recordings and 10% were live recordings, but nowadays the situation seems to have reversed!

Most classical concerts are performed three or four times over the course of a week, so for recent albums, I tend to keep the recorder running every day, and cut and paste the best performances of each movement to make a single album (the important ones). (It would be a shame if the whole work was rejected because an audience member coughed. For such simple editing work, DSD is not too difficult. The recording microphones at the concert venue can be pre-balanced using the venue's analogue mixer, and the finished tracks for stereo and surround sound can be recorded on a DSD recorder. All that's left to do is to write the cover photo and liner notes and either make the SACD or sell it as a download on a DSD distribution site.

Why DSD? Whenever the subject of DSD comes up, the question that always heats up the debate is "why bother with DSD and not hi-res PCM? In conclusion, the only answer is that it was the best way to go at the time. Also, unlike audiophiles, many studios use the same recording equipment for 10 or 20 years, because reliability is a priority.

If you go back to around 1999, there was a format war between DSD (SACD) from the Sony-Philips camp and high-res PCM (DVD Audio) pushed by Panasonic, JVC and others because of its higher sound quality than CD. Nowadays, DVD Audio with its high-resolution PCM seems to be the right choice, but at the time it was difficult to get sponsorship from the music industry and so it lost out marketing-wise to SACD. In fact, "better sound quality than a CD" was just a pretext, but the real intention of the media industry was to find a new "copy-protected format" to replace the CD, which had become an all-you-can-copy format with the spread of computers and CD-R drives. At the time, there was a lot of piracy and a lot of focus on copy control CDs and digital rights management (DRM). I think there was a sense of crisis in the industry. As a result, SACD has become the next generation of media, and for classical music lovers who have long been sensitive to 'high quality sound', SACD has become the format of choice, with its ultra-high resolution and surround sound!

The advent of DSD: In order to compare the difference in sound quality between DSD and PCM, let's take a look back at the debut of the DSD format. CDs are recorded at 44.1kHz 16bit PCM, so why was 2.8MHz 1bit DSD born? From a listener's point of view, aside from the theory of sound quality, it seems to me that the biggest turning point for DSD to take root in a commercial sense was the introduction of the SAA7350 series of D/A converter (DAC) chips by Philips in 1991.

At that time, despite the explosion of the CD, pure 16-bit PCM DACs were reaching their limits and the quest for perfection was becoming a costly quagmire. 16 resistive switches were embedded in the DAC chip and each one had to be exactly right. Each resistor has to be exactly right. The DAC chip is made by laser cutting 16 thin copper resistors into small pieces, so that even a few microns of error is not enough to achieve 16-bit accuracy. For example, the Philips TDA1541A and Burr Brown PCM56 were very popular with audiophiles at the time. The less accurate chips were sold at a lower price. For example, the Philips TDA1541A had a crown mark on the chip surface, and the best players had two crowns (double crown), while the rival Burr-Brown had a "K" rank for the highest grade, and "J", "no mark", "L", etc. below that. The price of a single chip was more than 10 times higher than the price of a K-rated chip and the price of an unmarked chip was more than 10 times higher than the price of a K-rated chip. (You can still find fake chips on eBay, for example, where someone has forged the crown mark). For example, the distortion (THD) specs were -92dB for the "K-rated", and -82dB for the "unmarked", so it seemed to me that they were doing their best, almost on the edge, when the logical dynamic range of a 16-bit CD is said to be 96dB.

SAA7350 and bitstream: This is where the Philips SAA7350 series DAC chip comes in, which uses a high-speed 1-bit conversion circuit to convert 16-bit PCM data, eliminating the need for 16 resistors to be laser cut out. With one bit, the analogue conversion circuit is simply a single fast-acting switch. Philips called this the "bitstream method" at the time. In other words, a 1-bit DAC determines the pouring of water or beer (Nishio's example) or milk in a factory by how many times a single tap is opened and closed at high speed, whereas a 16-bit DAC determines the volume of water or beer by how many taps are opened and closed simultaneously. Rather than having to carefully adjust the size and water pressure of all 16 taps at the factory to ensure that all 16 taps are as designed, moving just one tap at high speed is cheaper, easier and more accurate.

The SAA7350 has an astonishing -96dB THD specification as a standard product, even without taking into account rankings and selection, and more importantly, the SAA7350 can be purchased for only about $3, whereas previous high-end 16-bit DACs cost around $20 each. In other words, a cheap CD player with the SAA7350 was "on specs alone" more powerful than a high-end CD player that stuck with an expensive 16-bit DAC. At the time, high-end 16-bit DACs were becoming increasingly costly, with multiple DAC chips installed in parallel instead of just one, to reduce distortion rates.

In the case of 16-bit DACs, it is necessary to use a number of high-grade capacitors to stabilise the switch current, and even if the same DAC chip is used, the cost of the peripheral components will obviously change the sound quality specifications. However, with the SAA7350, as long as the high precision crystal clock was taken care of, the rest of the circuitry was relatively low impact and low cost. It was around this time that the topic of "DAC clock jitter" became popular.

Even then there was a battle between PCM and 1bit CD players in the audiophile world, just as there is now with hi-res PCM v DSD. However, as we all know, the sound quality of audio equipment is not only determined by the specification of the DAC chip, but also by many other factors such as the stability of the power supply and the performance of the analogue circuitry. There was a big difference in sound quality between the PCM CD players, which had evolved into big machines, and the light small 1-bit players.

Apart from the Philips SAA7350 DAC, at the same time Sony was using a 1-bit DAC called the PULSE DAC, and there was also the Panasonic MASH. All of them are excellent chips, but only Philips was able to popularise them in many areas. Burr Brown, one of the biggest 16-bit manufacturers, also introduced a fully 1-bit DAC chip at this time, such as the PCM69, the successor to the PCM54. This means that almost all DAC manufacturers recognised the benefits of 1-bit in some way or another. There are two reasons why the Philips SAA7350 was so important. Firstly, the chip was available in bulk at a low price and was used by many audio manufacturers, not just Philips. It's not surprising, because it was cheap and powerful. The other reason why the SAA7350 is so important in this DSD story is that it has a digital output for 1-bit data.

1-bit dedicated DAC TDA1547: The SAA7350 was certainly a good DAC chip, but from the outset, its top priority was to mass produce it at low cost, and they felt that further improvements in sound quality could be achieved for high-end audio. So instead of releasing a higher version of the SAA7350 chip, Philips opted to take the high-speed 1-bit data from the SAA7350 and convert it to analogue on another high performance chip. This analog conversion chip is the TDA1547. The TDA1547 is a "1 bit data only" DAC chip that cannot handle PCM, and thanks to its accurate and unmatched current switching, it has a distortion (THD) of -101dB and a signal-to-noise ratio of 111dB, dramatically better than the SAA7350 alone. At last, they'd reached an age where DAC chips in ordinary consumer equipment could consistently outperform the 44.1kHz 16-bit of CDs. At the time, it was not common to send or receive 1-bit data, so the SAA7350 and TDA1547 pair was intended to be used as a set. This set was nicknamed the "DAC7" and was widely used in high-end CD players of the time. The DAC7 systems of that era are still the most appealing in terms of sound quality. It was also possible to oversample 44.1kHz PCM data from a CD by a factor of 8 before sending it to the SAA7350 (i.e. 44.1 x 8 = 352.8kHz), so it was possible to use an oversampling chip such as the popular Japan Precision Circuits (now Seiko NPC) SM5803. In other words, if you look at the DACs in Philips CD players at the time, they were a combination of three things: a chip that oversampled the CD PCM 44.1kHz to 352.8kHz, a chip that converted the PCM 352.8kHz to 1bit at high speed, and a chip that converted the 1bit to analogue. The upper limit for the Philips frequency was 44.1kHz x 8 x 24 = 1 bit 8.47MHz , but the actual frequency varies. Around 1995, the TDA1307 chip was developed, combining the first two steps (i.e. the oversample chip and the 1-bit conversion chip), further streamlining the 1-bit conversion of CDs.

SACD players: The configuration of this Philips DAC shows that already in the early 90s, two data formats, "352.8kHz PCM" and "high speed 1bit (i.e. DSD)", were being handled in real time in CD players. This was also the case with the A/D converters used in digital recording, which led to the idea that it would be better to record, store and play back music in an intermediate format, rather than sticking to the 44.1kHz 16-bit format of the CD. I think it is natural that the SACD recording media proposed by Sony Philips was based on 2.8MHz 1bit. From Philips' point of view, a 2.8MHz 1 bit recording can be easily converted to high quality analogue using a single TDA1547 chip. The reason for choosing 2.8MHz (DSD64) instead of 8.47MHz (DSD192), which is the highest speed the TDA1547 can convert, is probably due to the limited capacity of a single SACD disc (approx. 4GB), including surround recordings. In fact, the Marantz (Philips) SA-1 player, which made its SACD debut, has a simple circuit where the PCM data from the CD is converted to DSD via the TDA1307, and the DSD data from the SACD is fed directly into the TDA1547 chip. This is the shortest route to high quality sound that Philips had ever devised. As a side note, Philips in the Netherlands and Marantz in Japan have had a close relationship since the early days of the CD, Philips specialising in the development and manufacture of CD reading lasers and DAC chips, but outsourcing the finished CD player to Marantz, an established audio manufacturer. Many models were sold under the Philips name in Europe and under the Marantz name in Japan and the USA. However, Japan's bubble economy burst and the PCs, MP3s and iPods were coming. When SACD was introduced, the future of the music industry was already bleak in Japan... Marantz did not only stick to Philips DAC chips, but also used Seiko NPC, Burr Brown, and Cirrus Logic's newly developed CS4397 DAC, to which many Philips engineers had moved, at the same time.

Sony: On the other hand, Sony, the other leading player in DSD, used a different 1-bit system from Philips, called PULSE DAC, at about the same time. It started in 1990 with the CXD2552 chip (used in the CDP-X777ES). The method is the same as for Philips, first oversampling 44.1kHz to 352.8kHz with an oversampling chip, then pulsing to 1 bit with a CXD2552 chip, but while Philips uses PDM (pulse density), Sony uses PLM (pulse length) to indicate amplitude. Four years later, in 1994, Sony introduced a bridging "current-pulse DAC" system, in which the voltage pulse output from the CXD2552 is converted into a powerful analogue current pulse by a subsequent chip. When it first debuted, it looked like a gimmick, but now it seems that what Sony was trying to do was a "direct amplification of 1-bit pulses". In other words, for Sony, 1-bit is not just a digital data format, but a neutral entity that treats digital as analog. The S-MASTER amplifier, which has since been used in Sony's AV amplifiers and high-resolution Walkmans, is in principle the same as a current-pulse DAC, with the digital data first converted into PLM voltage pulses, and then driven directly to the speakers or headphones by a powerful current switch. That's how it works. That's where Sony's tradition comes in. Unlike Philips, which supplied DAC chips to many audio manufacturers, large and small, Sony's DAC chips were almost exclusively for its own CD players, so their performance is still largely a mystery. Around the time of the arrival of SACD, Sony, like Philips, ceased to develop dedicated DAC chips. The first Sony SACD player, the SCD-1, used Sony's CXD8594 DAC chip, but from the third generation SCD-XA5400ES, Sony began to use Burr Brown's DSD1796 and other DACs, as did Marantz. The sound quality seems to have changed dramatically during this transition. In 1999, the 2.8MHz 1bit DSD system was a proven high-quality technology in 1bit DACs for CDs, but since then, DAC chip manufacturers such as Burr-Brown and Cirrus Logic have been developing even faster DACs with 128x and 256x oversampling, as well as 4-8bit DACs instead of 1bit. And as a result, 2.8MHz DSD adopted for SACD has been treated as "a format with 'halfway between PCM and multibit' sound quality born in a transitional period". For example, the advanced current segment method used by Burr-Brown since 2003, just after the debut of SACD, combines the best of both PCM and 1bit DAC, with purely resistive conversion for the top 6 bits of 24bit data and high-speed 5 bit delta-sigma modulation for the bottom 18 bits. It's the best of both worlds.

In the end, it is interesting to note that whatever the recording format, the modern D/A conversion process is right in the middle of both DSD and PCM. This means that the debate between PCM and DSD is based on historical evolution and the limitations of the technology 20 years ago. Without getting into the actual human hearing range debate, the idea that "numerical specifications" such as sample rate and dynamic range are "good enough already" is a futile exercise because, like "megapixels" in digital cameras, progress will not stop while there is demand. It's a futile exercise.

The problem with DSD64: One thing that needs to be taken into account when considering DSD is the fact that the DSD64 (2.8MHz 1bit) format used in SACD is "not as high-res as it sounds".

In many high-resolution download shops, DSD albums are priced higher than PCM 192kHz, but in reality they are not that high performance "on spec". (This does not mean that the sound quality is poor). In principle, DSD64 has high frequency noise from around 30kHz due to noise shaping techniques, and this needs to be cut by an analogue filter during playback. Without noise shaping, there would be no high frequency noise, but on the contrary, the noise floor in the audible band would increase (the dynamic range would be worse than on a CD). In other words, the high-frequency range of DSD64 is not the theory that some high-resolution believers claim, such as "reproducing the ultrasonic waves of musical instruments inaudible to the human ear", but is actually a kind of trash bin or storage space for noise shaping, to drive out the audible noise.

At the time, an SACD spokesman for Sony (being a random dude not particularly interested in music) was asked, "How many kilohertz can DSD play? The guy replied, "In principle, it can record up to 100kHz. But we use a technology called noise shaping, which means.... I think that's what... It's just like the hi-rez boom! False. Such misleading brochures were common

Again, the advantage of the DSD method is the simplicity of "direct analogue conversion of digital data", and it was never intended to reproduce high frequencies outside the audible band. In reality, many SACD players and DSD-compatible DACs have circuitry that cuts off frequencies above about 24kHz (or at most 50 kHz) for safety reasons. So, in terms of high frequency reproduction, it's not that different from 48kHz PCM. There are also filters for 30kHz. Of course, the noise generated by noise shaping is not related to music. It may have a healing effect, for example. Some people might think that if the noise is inaudible to the human ear, then it's OK to leave it out without filtering, but that's the danger - even if high frequencies are inaudible to humans, amplifiers and speakers will do their best to reproduce them. There was a recall of Sony speakers that failed due to high frequency noise. Most amplifiers are equipped with a high-frequency filter as a safety measure, to prevent the incoming high-frequency noise from doing anything bad. However, audiophiles are usually averse to unnecessary filters and protection circuits, so the more expensive high-end amplifiers don't have these preventative measures in place. This means that the amplifier amplifies the huge amount of high-frequency noise many times over, which can burn out the speakers. In addition, some manufacturers who develop their own DACs and amplifiers have neglected to conduct sufficient tests, thinking naively, "Our DACs are equipped with high-frequency filters, so the amplifiers we sell in pairs will not need filters. In fact, there was a British manufacturer who had a circuit design that amplified infinitely. In fact, there was an incident where a British manufacturer's amplifier was burnt black by high frequency when connected to a SACD player made by another company.

The most annoying thing about DSD64 is that the starting point of such high frequency noise and the residual noise (so-called noise floor) varies considerably depending on the generation and performance of the DSD recorder. This is especially the case with older DSD converters, such as RCA Living Stereo's DSD remasters, which are quite loose in their shaping, but sound very good. (The noise floor is a remaster of a 1950s tape recording...).

DSD A/D converter: The number of so-called DSD A/D converters, which convert analogue audio to DSD, is limited. It is often misunderstood that since there are many DACs (D/A converters) that convert DSD to analogue, there must also be many ADCs (A/D converters) that convert analogue to DSD, but this is not the case. The reason for this is simple: there is a huge difference in demand between ADCs and DACs. If you release a low-cost, high-performance DAC chip, you can sell 100,000 or 1,000,000 of them because they are used in consumer audio products. In other words, ADCs are a market where you can't spend a lot of money on performance because it's unprofitable. For example, field recorders, MD players and other devices that play and record digitally at the same time need both an ADC and a DAC, so at the time, Sony produced a chip that contained both an ADC and a DAC in one piece. However, if you look at the specifications of these chips, you will find that the ADC is often an order of magnitude inferior to the DAC. The same applies for PC codecs.

In the early days, there were no mass-produced off-the-shelf DSD recorders, so even major labels like Philips used self-made converters. In the early days of SACD, there were only Meitner emmLabs DSD recorders, which Philips loaned to label studios (e.g. TELARC label), and later Prism Sound ADA and dCS 904 converters were popular for both PCM and DSD (RCA Living Stereo's SACD remasters, early Pentatone, etc.), so it's a bit of a fad, and "XX label bought a new converter" is always a topic of conversation in the inner circle. On the other hand, in the US, high resolution PCM recording had been widespread in the film industry before DSD, so many studios were using 192kHz PCM converters such as Pacific Microsonics, which was the mainstream at the time. There were not many of them. Whether it's high resolution PCM or DSD, A/D converters of this period used delta-sigma modulation noise shaping anyway, so there was a lot of difference in sound quality between converters. The problem was that DSD64 was just barely OK in terms of high frequency reproduction and dynamic range, so for studios that had already moved to high resolution PCM, there was little benefit in replacing it.

Nevertheless, the DSD faithful have never stopped, and many studios have continued to use DSD recorders for many years. These days, technology has evolved in its own way, with the simple answer being that if you record at twice the speed (5.6MHz) or four times the speed (11.2MHz), then the high frequency noise can be pushed further out, and the audible noise floor (dynamic range) is better. Of course, this doesn't mean that high-frequency noise disappears completely through noise shaping, but what used to occur from the 30kHz range with DSD64 is now around 80kHz with DSD128, and the aim is to push it to a point where it no longer matters. From a recording point of view, it's like, "I don't have an instrument that can produce such high frequencies, and I don't have a microphone that can record satisfactorily...

Difference in sound quality between DSD and PCM: But what about sound quality? The recent 'DSD vs Hi-Res PCM' debate is exactly parallel to the 1-bit vs multi-bit CD player debate of 25 years ago. Even back in 1991, 1-bit was seen as "delicate, high-resolution, with great air and imaging", and multi-bit was seen as "sturdy, with strong bass and energy". It is rare to record the same performance on both DSD and PCM recorders at the same time, and even if such a comparison could be made, the differences in sound quality between the recorders would have to be taken into account. Also, of course, analog conversion with DACs follows a different path for PCM and DSD, and each manufacturer - ESS, AKM, BB, Cirrus Logic - has its own strengths and weaknesses. It's a difficult topic to reach a clear conclusion on, but the most important thing is to listen. It's easy to make the mistake of concluding, for example, that a certain label uses a certain company's A/D converters, so the sound is harder.

If you ask them what kind of microphones they used for recording, many of them can't answer. In addition, there are factors such as the placement of the microphones and the acoustics of the concert hall, and to begin with, the instruments and performers are different. The A/D converter is the part of the recording process that contributes to the sound quality, but because it is the easiest piece of equipment to talk about in terms of specifications, it tends to be talked about and compared with other pieces of equipment. It's like an engine in a car. One interesting example is an album that was previously sold as SACD or DSD64, but was later sold as a download in DXD, the original master. (DXD is another name for high-resolution PCM 352.8kHz). In this case, one would naturally expect the original PCM 352.8kHz file to have better sound quality than the DSD64 down-converted for SACD, however, comparing the sound quality in the same studio environment, the DXD sounded hard, crisp and "studio monitor-like", while the DSD sounded more realistic, with reduced edges, more depth in the space and a clear emphasis on the "air" around the performer. I'll leave it at that. It was most likely related to the DAC and not the format. There does not seem to be much benefit in digging out obsolete papers on DSD at this point. We simply have to wait and see what new innovations arise in this space.

The sound quality does depend on the DAC you use, so it's a luxury to be able to choose the sound/DAC you like, and the short-sighted debate about superiority that often occurs on internet forums is probably just an argument between people who don't actually listen to much music! I am willing to defend DSD subjectively, and even promote DSD recording, but since none of my dongles make DSD sound awesome, I use both, DSD for 44.1 rates and DXD for 48 kHz rates.

Word count: 4915
 
Last edited:

tmtomh

Major Contributor
Joined
Aug 14, 2018
Messages
1,317
Likes
3,826
Still it’s not clear whether 768kHz is pretty much DSD, without the negative features of DSD.

I think there needs to be some explanation and offer of evidence as to why DSD or 768kHz PCM is in any way necessary or has any benefits above, say 96kHz PCM or 48kHz PCM. Absent that, the ensuing 7,000+ (and counting) words on DSD don't necessarily mean a whole lot.
 
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
It sounds so simple. Forget 96kHz PCM or 48kHz PCM. Actually, Hi-res was a marketing gimmick, and 24/44.1 or 48 or CD is all anyone needs. Actually, we don't need high-end DACs and ADCs, since we have the Apple USB-C adapter, capable of playing 24/44.1 or 48 stereo and recording 24/44.1 or 48 mono. When I refer to 768kHz PCM it is as a format at the final stage such as through HQPlayer or WPUP. On the other hand I am yet to completely dismiss DSD as an ADC format. In this realm there just doesn't seem to be much published research made accessible to the end user. So there is no objective evidence as to why DSD or 768kHz PCM is necessary for studios or consumers.

A sort of 96kHz vs DSD paper does exist.
Breaking the Sound Barrier: Mastering at 96 kHz and Beyond
The SACD paper cited it because Moorer leaned towards DSD a little more than 96kHz. He made it clear that careful conversion algorithms for 96kHz to 44.1kHz have no issues at all. But he added an example first, of where 96kHz fails to be of sufficient mastering quality due to bad filtering. He didn't critique DSD as much as he did for 96kHz PCM. He reckoned DSD has larger bandwidth than 96kHz. Anyway, opinions aside, 96kHz does have a larger bandwidth and does take up less space than DSD so it's actually better for recording and streaming.

In the recording engineer's world, skill is far more important than any format. Or is it. In consumer electronics, recording in DSD is just a marketing story, and so no evidence is provided. The mics are low quality, the chips are fake DSD ADCs, SNR is low...
Probably there is evidence in DIY circles that 768kHz PCM and DSD perform better in certain DACs but I have not seen them. There are plenty of subjective claims of course. We'll deem they don't exist. Absent the evidence we will conclude that PCM and DSD both end up processed into the same sound in consumer level chips.

None of this helps in any way with Chomsky's three challenges facing us, namely future pandemics, nuclear warfare, and climate change.

It's just a diversion, a form of escapism. Some narrow minded self-satisfaction.

Martha Argerich and the Hiroshima Symphony Orchestra held the 2015 An Evening of Peace Concert. The DSD recording was not at all impressive. But the meaning behind the concert included 'the strong belief that the love that music holds within it will weaken people's desire to harm others'.

Glenn Gould became an exclusively recording artist at some point, with a strong interest in recording.

Yo-Yo Ma's Bach Project and bridging culture efforts are well known. Yo-Yo Ma did not re-listen to his own recordings at least.

Haruki Murakami spoke on his radio programme: Today's quote is from the pianist Vladimir Horowitz. 'I trust the musician more than anyone else, because they are too busy practicing every day to do anything stupid.' I see what Horowitz means, but that's just the way Horowitz thought. I think there are a lot of musicians in the world who don't practice at all and are doing something 'stupid'. Musicians, please practice diligently every day. I don't know much about novelists though ...... (laughs).
Murakami also said 'I think that's what music is really about. I think that music which heals people and inspires a sense of kindness is good music.'

Maestro Seiji Ozawa also said (in Absolutely on music : conversations with Seiji Ozawa, Interlude 1: On Manic Record Collectors) somehow, people who were obsessed with things/records rather than the emotion, or even the score, background and interpretation gave him a feeling of distaste. Music to the maestro is more internalized, like 'being able to read foreign literature in the original, rather than in translation.' Schoenberg has said “music is not a sound but an idea,” but ordinary people can’t listen to it that way. Even being able to read a score would be far too little to surmount the high wall between the professional and the novice.

I would conclude from this sample of evidence that, the solution to future pandemics, nuclear warfare, and climate change would definitely involve, at least partly, the listening to of good music, not in a concert hall, but in your own private space. It might even be your nearest doomsday bunker or evacuation point. But if DSD is an actually useful format to convey the concert hall as accurately as possible to the listener, then why not? It'll sound bad for sure, but that's the reality. It's the real sound of the concert hall, however bad the acoustics are. However, it is distasteful by maestro Ozawa's standards for me to continue rattling on with meaningless statements on DSD, as I am indeed obsessed with DSD, much like Andreas Koch's 'DSD=new drug' analogy. Hopefully 1 bit audio is worth more than the amount of words I am writing.
 
Last edited:

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
13,345
Likes
20,698
You have it seems lots of faith in DSD. The problem is faith doesn't need evidence. Yet you keep approaching with faith to provide evidence. It will not work out that way.

96 24 is more efficient with the data than DSD. DSD does not have more bandwidth because the noise shaping requires a filter. There is no advantage to be had there.
 
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
You're right, I am a DSD missionary. I'm here to convert you all to DSD Audio believers.

Sure, I'm trying terribly to provide pretty bad evidence. It is not working out. I get fed up each time, before touching any mathematics or DSP, but I keep approaching to have another crack at solving the DSD mystery. Without using any scientific methods. I take a long time to learn.

Mr Moorer's paper is unhelpful. With respect, the statement on bandwidth was wrong. Sonic Solutions worked with Sony a lot and their Sonic Studio inspired the name SonicStage Mastering Studio for VAIO. The paper overall subtly leaves the reader with the impression that DSD is to be preferred.

The common DSD addict would reply that DSD128 or 256 will 'have more bandwidth', despite the unreasonable data amounts. Or they'd say that high frequency sound (correlated with the audible spectrum) will still be heard by sensitive human ears amongst the 'unhearable unrelated noise'.

I submit that both excuses are wrongly made. 24/96 wins.
 
Last edited:
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
I did find bunpei who presented at Waseda Uni in 2014: 的場文平-san, a DIY audio enthusiast. and Chiaki http://chiaki.cc/

bunpei pointed out several points as "misconceptions about 1-bit audio that audiophiles tend to fall into". Specifically, he mentioned the following points:
  • Many people simplify DSD as "PDM (Pulse Density Modulation)" without Delta-Sigma modulation.
  • They think that they can edit ΔΣ modulated 1-bit audio data with simple bit operations using super techniques.
  • They think that there is only one type of DSD.
  • They think it can be reversibly converted to and from PCM
  • They are always debating the merits of sound quality by simplifying it to "PCM vs. 1-bit audio" alluding to mansr/Miska.
I also found K-1326 '銀箱' 'silver box', brother of K-1327 (SBM Direct).
dal07.jpg

dal08.jpg

Basically you send digital in from Sonoma to play back DSD. Or record from analog in.

They used it to cut records. The master was input directly into the cutting console from Sonoma, a highly trusted DSD workstation, through the analog output of a special SONY K-1326 D/A converter, commonly known as a silver box. No electrical processing such as equalizers, limiter compressors, cut filters, or low frequency phase correction were used. With this simple circuit configuration, they aimed to create a sound that is faithful to the master, emphasizing the freshness and dynamic range of the sound, rather than competing with the cutting level.
 
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
Playing back a DSD/1 bit audio digital signal is very easy if you don't care about the details of the sound quality. It may be an exaggeration to say that a single resistor is all that is needed, but there is no need for an analog amplifier, and the sound can be produced simply by connecting speakers directly to the digital signal. 1-bit audio also contains information about the signal components, so there is no need for D/A conversion, and a mere IC buffer for the digital signal can drive speakers as is.

If the sampling frequency of DSD is increased to 500-768MHz, or more than 100 times the conventional frequency, interesting things will become possible that will overturn the conventional wisdom of DSD. Of course, raising the sampling frequency to such a high level requires ultra-high speed operations, which are processed using an FPGA (a computing device that allows the circuit to be rewritten by a program).

If you really want to do any editing, you need to convert the signal to multi-bit PCM or convert it to analog signal and then process it, which is not pure DSD. However, by raising the sampling frequency, such editing can be possible. The simplest idea is that to mix two signals, you just add them together. The sampling frequency is doubled by filling the data of the 1-bit signal with zeros for every second bit. With the zeros in different positions, you can easily mix by adding up the 1-bit data without any overflow. In this case, the mix will be 1:1 in terms of volume, but of course you want to have a variety of mix balances, not just 1:1. You can change the balance of the mix between two signals, and you can even mix multiple signals. Reverb is achieved by adding signals whose volume decreases with time, so the volume is increased by a certain amount of zeros and then added by shifting the time.

They are still demonstrating the fact that such things can be done, and are not in a position to put them into practical use immediately.

In 1999, Super Audio CD and DVD-Audio were standardised, and in 2001 the 1-bit Audio Consortium was set up by Sharp, Pioneer and Waseda University. Initially, 1-bit audio recorders were expensive, but at the 121st AES Convention in 2006, TASCAM announced the TASCAM DV-RA1000HD as an easy-to-use device, and Korg introduced the MR-1 handheld and MR-1000 portable recorders. This was followed by the KORG MR-2000S (2008), KORG MR-2 (2010), TASCAM DA-3000 (2013), SONY PCM-D100 (2013). Until the early 2000s, the only 1-bit PC audio systems available were expensive DAWs for SACD production such as SONY SONOMA and Pyramix from Merging Technologies. In those days, CPUs were not capable of signal processing, so DSPs for video processing were installed. The launch of the VAIO with integrated SONY "Sound Reality" CXD9872 chip in September 2005 brought 1 bit audio closer to PC audio. VAIO included the Sonic Stage Mastering Studio 2.0 for recording and simple editing and DSD Direct, which upconverts PCM to 1 bit. These were discontinued in 2009 (WIN 7), but the major achievements remain the extension of ASIO to include a DSD mode, the DSF file which has become the de facto standard for DSD distribution, and DSD discs with DSF on DVDs! Korg released AudioGate in 2006 to make 1-bit audio easy to use, and in 2010 released Clarity Recorder/DAW with a dedicated USB I/F. The MR-0808U was presented at the AES Convention, which has 8in/8out and USB2.0. The ability to record at 5.6MHz and the native CPU meant that the number of tracks could be unlimited, depending on the performance of the PC. The Norwegian 2L label was the first in 2009, followed by Blue Coast Records in the USA. OTOTOY and e-onkyo started to distribute 1 bit audio files for download, and this led to the release of DSD compatible USB-DACs by many companies at the same time. Initially, there was no way to play 1-bit audio on MacOS, but in 2012 Andreas Koch of Playback Designs proposed the DoP (DSD over PCM frames) method, which allows DSD to be transmitted over 24-bit PCM frames. Audio streaming was first launched in Europe by Spotify in 2008 with a subscription streaming service, and in April 2015, IIJ (Nishio), Korg, Saidera Paradiso and Sony streamed the Tokyo Spring Festival concert and Berliner Philharmoniker's performance live using DSD 5.6MHz in a demonstration. In December, IIJ launched the service as 'PrimeSeat', offering daily live streaming at 5.6MHz; in 2017, it offered live streaming of the Berlin Philharmonic's subscription concerts at 11.2MHz. All networked audio players use the UPnP AV standard, which was developed in 2002. It is a plug-and-play solution that is ready to use as soon as the device is connected to the home network. It was developed by the UPnP Forum, which included Intel and other companies, so the specification was huge, but the details were left open-ended, which made compatibility difficult. The DLNA has since been disbanded, and companies are now free to use UPnP as they see fit. PrimeSeat is a high quality sound delivery service, but there are many cases where video is more effective (like YouTube). In February 2019, IIJ conducted live streaming of 4K video (H.265) and high-resolution sound (MPEG-4 ALS, 96kHz/24bit) from Berlin Philharmonic concert hall direct to Tokyo via live streaming. The conventional streaming method requires a large-scale distribution system, dedicated hardware/software, and does not support 1 bit audio. IIJ stated that DSD is arguably a good format for live recording and has an affinity with live streaming.
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,061
Likes
9,078
Location
Hampshire
However, by raising the sampling frequency, such editing can be possible. The simplest idea is that to mix two signals, you just add them together. The sampling frequency is doubled by filling the data of the 1-bit signal with zeros for every second bit. With the zeros in different positions, you can easily mix by adding up the 1-bit data without any overflow.
Sorry, but that's just wrong.
 
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
Please have some reading here for basic but under-the-hood understanding of a delta sigma DAC. 1 bit DSD is a burden not the light. Hope you will find this helpful.
https://www.diyaudio.com/forums/digital-line-level/308860-valve-dac-linear-audio-volume-13-a.html

Thank you!

But, not being well-versed in these diy things, the impression one gets from that page is that MarcelvdG's Valve DAC original vs raw dsd are roughly similar, meaning that it doesn't give a strong message that DSD is a burden. His DACs are inherently burdensome, and anyway, it's common knowledge that DSD equipment took years of struggle to come to fruition.

In the meantime, the analog amplifier and D/A converter is replaced with speakers directly fed with the digital signal via an IC buffer. You'd be able to get some sound out of the headphones by simply passing the 1-bit signal straight through the amplifier and then through a low-pass filter. The DSD signal would sound great if loud, but if quiet, it would be buried in noise. Alternatively, a full digital system uses PWM modulation to create a 1-bit digital signal similar to DSD, which is then amplified using a class-D amplifier and an LC low-pass filter.

Normally, frequencies such as 2.8MHz and 5.6MHz are used, but another method uses GHz for recording where basically no analog circuitry is used at all. Instead it uses a simple all-digital circuit configuration. By speeding up the frequency, a more realistic sound can be achieved with a simple circuit. This high-speed sampling makes it possible to record up to the ultrasonic band, and in principle, there are no losses in the amplifier section necessary to drive the device. The S/N ratio around the modulation frequency can be secured by controlling the quantization noise in the desired band. This high-speed 1-bit signal is fed to a parametric speaker without an amplifier and played. It's an acoustic system that uses ultrasonic waves to create sharp directivity of the sound.

Of course, this takes 1 bit audio out of the music sphere and into other fields.
 
Last edited:
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
Sorry, but that's just wrong.
Thank you for reading!

What's 'just wrong' about that simple generalised explanation? Is it the fairly suspicious and unpublicised theory from Waseda Uni that is wrong, or is it the words that don't make sense or are ambiguous? Such editing was demonstrated at Waseda Uni years ago, but given that none of us have heard about editing DSD since then, it's probably a failed idea. They upconvert 1 bit DSD to 500 or 1000 MHz and then somehow slot the two file's data together by shifting one file forwards, and then displacing the empty zeros by adding the files together. I suppose the whole story was not revealed, so this is a questionable method. There are other more recent attempts to edit, given that some computer scientists believe there is nothing we can't do with emerging technologies. How can DSD be uneditable?
 
Last edited:
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
Lipshitz argued that DSD is not light nor editable. It was designed to be cheaper and not better. It was a struggle to get it working, and multibit remains better. He conceded that 1 bit overcame the level matching difficulty but that it was inconsequential. PCM is free of limit cycle oscillations and instability. Since converters are all multibit, converting to 1 bit is an unnecessary degradation. He equates noise shapers with sigma delta converters and shows that PCM can avoid overload while 1 bit overloads when dithered. So full linearity is unachievable with DSD due to inability to dither. Therefore, PCM is the future, not DSD - simply a mistake. PCM requires less noise shaping and 8 bit/176kHz has half the DSD64 data rate but with the same noise density and is overall lower in noise. Overall, from a PCM perspective, DSD is illogical, irrational, and inconceivable.

Reefman argued that recording in the native A/D multibit and converting to DSD is good. DSD can be dithered. 192 kHz is 'bit deficient' and does not cover 350kHz.

With respect, both papers are not convincing for the layperson, appear to contradict each other, and overall Lipshitz has included greater details and a more powerful argument while Reefman does not seem confident enough in their statements. In any case, most statements here are useless if the real way 1 bit is used in practice is a more secreted and perfected way. Further, the papers are old as.
 
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
So in terms of meaningless descriptions:

The sound of DSD has a wide range, with very smooth high frequencies and a wide range of low frequencies. To increase the density of the sound would be to get closer to the ideal sound. Conventional methods of creating an analog sound include using analog comps, analog EQs, and analog tape recorder amplifiers. However, while these methods provide an analog feel, they tend to lose the delicate texture and speed of the sound. With DSD, you can create a thicker sound while maintaining the clarity and looseness that you could not create with analog equipment.

About DSD upconversion of PCM source in KORG AudioGate: Upconverting a PCM source to DSD increases the resolution. The main thing to keep in mind is that not only the subtle nuances but also the electrical distortions are upconverted at the same time. For this reason, distortion may be noticeable when the PCM level is set to its limit. To get the best out of the DSD format in terms of spaciousness, depth, etc., it is necessary to have some headroom. However, if you are recording directly, DSD is more resistant to clipping than PCM, so it is OK to record at a level where the red light turns on momentarily. 5.6MHz has a wider range and more detailed sound, while 2.8MHz is used to increase substance.

An image taken with a cell phone: There are not enough pixels for a designer to use it as is for print data. By upconverting to large format film resolution, lines can be preserved, and contrast and even the color can be adjusted delicately or boldly. Similarly, upconverting a PCM 48kHz 24bit source to DSD 5.6MHz will bring out the breath of vocals and buried grooves that were previously inaudible. It's strange, but it's almost as if the sound that was thinned out during PCM conversion has come back to life. The real beauty of DSD is its realness. The sound rises quickly, and every detail is reproduced well. There is an expressive power that makes it seem as if the music is being played right in front of you. This is because DSD has fewer peaks in the sound and the level can be larger than PCM, and because it rises quickly, the voice comes out naturally. A source processed in this way is unlikely to lose its nuances even when compressed to MP3. Up-converting makes it possible to create a sound that is more detailed and precise. Because the sound conveys more of what the artist wants to convey, the overall level of emotion is also improved.

Engineers and artists describe the characteristics of DSD sound thus: "The sound rises quickly and comes through softly," "It has moisture, depth, and luster," "The depth and breadth are different because you can tell when the sound disappears," and "The nuances of the sound hardly change through recording".

You can't get bolder claims than that right?! What does it mean 'fewer peaks in the sound'? Can't PCM have 'fewer peaks in the sound'?

By conventional standards, this degree of reliance on the format is the reason why DSD ended up with its strange current status.
 
Last edited:
OP
Saidera

Saidera

Senior Member
Joined
Jul 18, 2020
Messages
323
Likes
244
Location
Adelaide, South Australia, Australia
The method is wrong in the same way that you can't mix two regular PCM streams by interleaving the samples.

So how does PCM get mixed in DAWs?
 
Top Bottom