[Whitigir]

Very interesting for sure, low bit DAC Pro is the ability to achieve better linearity and higher original dynamic range theoretically, but it correlations between quantization noises would be astronomical as Con. This is why multi bit was born. It is interesting to see the original form of DAC are being manufactured again. However, with a better understanding at mathematical, algorithms and better fabrication techniques and components this time around. How could I stop myself from exploring this !!!!

Curiosity kills my pockets

 

[Z_Showmaster]

It's very interesting how Ed Meitner in 1992 said he doubted a 1-bit DAC would have sufficient resolution, yet years later his EMM Labs products all appear as 1-bit at the point of conversion. Bruno Putzeys had been an outspoken critic of DSD as a marketing gimmick and provenance of the recording chain – not the technology – but I was still really surprised to find out Mola-Mola builds 1-bit DACs.

I think the foundation of their change of stance is their own DSP work making DSD conversion a high-performance reality. I too am bitten by the technological bug because while this wave has been going throughout Hi-Fi since the 2010s, this is the first time Head-Fi has counted on all this original work in proprietary code and discrete DACs.

Kudos to the likes of iBasso, Cayin and HiBy who never settle and always push the boundaries of what's possible.

 

[Whitigir]

So, from reading over these techs posts regarding the R2R, together with the mentioning of Cayin N7 by @hongky . I think I found some more assumed conclusions, just based on what I know and think I know, this is not definite technical information. Cayin R-01 module, Ru-7, N-7, are all seemingly sharing the same technical designs with different implementations, whether it is into a dongle form, or a compact whole motherboard, or a complete DAP solution. The D16 DAC conversion from Ibasso is similar but taken to another step further with it being a dedicated and stand alone DAC solution.

Basically, what I got out from all of these is that The R2R ladder is one of the Low Pass filter Stage. I also think that because DAC is a little complicated, with different marketing, people get even more confused. Therefore, from what I know, I can say this. All of the DAC is whether Over Sampling or Non Over Sampling. It could be whatever Topologies, whether it is R2R or Sigma-Delta, it wouldnt matter.

The confusing part is that the earlier DAC technologies was more of a "discrete design" or we would call it a Resistor ladder converters. When we mention about Resistor Ladder or another term of Resistors to Resistors, we simply sum up all of them to carry R2R designs. Now, let dissect into these confusions

1/ R2R converter, this is the conventional Resistor ladder conversions. Basically, it took all of the binary coding, and pass through a series of resistors in sequential orders, and during this process, especially for D/A and not A/D, the conversions can follow Nyquist standards to stick with the most minimal bit that we have, a 16 bits. This means that we will have to deal with Most significant and Least significant bit. The differences between them, the voltage crossing at a certain gated threshold can cause significant distortions and errors. Therefore, technically speaking, the conventional design is Non Over Sampling, with a whole lot of distortions, floor noises, errors....etc... you name it. This is where people also love conventional R2R converter. The same as people who love tubes and it distorions .

My argument is that, music in itself is just noises and harmonic distortions, right ? LOL. I love conventional R2R conversion, and also love Tubes

Now, because of this, the engineers found out that if they were to have feedback loops implemented during the conversions, it would improve the resolutions, but in trade off for more components and spaces needed. This is where OverSampling topology came in.

2/ The modern Solution of R2R topology: This is no longer a conventional R2R, or following the NOS topology, it is OS topology. Because having the whole array of resistors of 16 bits together with OverSampling in discrete lay out will need a lot of spaces, and components, together with the tolerances in between them. This never came to light until recently where FPGA and programmable chip was a thing. FPGA solves many problems for the engineers, from having to fabricate their own chips, to just written algorithms with Oversampling factors and specially order it to be programmed with FPGA to solve the problems. Since it is programmable, engineers have the options to whether or not to compiling the whole conversion on a chip, or splitting it out. One of the Example of a whole conversion would be Chord as a brand, they are touted to known with their FPGA technology.

Beside using FPGA, of course you can use other meaning to convert digital signals as well, by using computational power and a written algorithm, then stored it into a file format of DSD. We call this Offline Conversion. This leaves all of the DAC systems to be as an Live conversion, because it takes in multibit PCM and convert it......you guessed it, toward to 1 bit DSD. Are you confused yet ? simply put, all of the Digital coding, before becoming analog signals, will have to be a 1 bit signals, and then passing by Low pass filter networks, before becoming Analog.

Then there is DXD to add into the confusion, It is just glorified PCM multibit and isnt directly related to 1 bit DSD

3/ Why Cayin, Ibasso is claiming 1 bit DAC as a title ? Isnt all DAC converting multibit into 1 bit before passing to LPF (Low pass filter) ? That is right, and they claimed it so, because They are implementing a network of Resistor ladder for the filtering processing, instead of only be using an opamps with feedbacks for example.

Is it R2R still ? Well, it is R2R alright ? but it isnt the conversion itself. Rather it is the filtering or the aftermath of the conversion

So what is really a conversion here ? well, it is the SRC or Sample Rate converter, or an FPGA if they used it instead.

So then, it means people who seeks for FPGA can rejoy with this ? Basically, Yes, it is similar to Sony/Chord topology, but with a tweak, and of course, the algorithm and firmware is really all that matters

4/ Finally, is it R2R ? is it Sigma Delta ? is it Over Sampling ? is it Non Oversampling ?

Again, there are only 2 topologies, NOS and OS. Because NOS in a typical/original Resistors ladder design cant be adequate for high resolution, the new innovation solutions is to adopt Over Sampling but hybridizing it with R2R analog filtering . It is Safe to assume that all of the modern R2R solutions are playing it safe by incorporating OverSampling. Because it is OS, it is the same as Sigma Delta

Can it be NOS as well as an option ? Good question, but it depends on the point of engineering. FPGA/SRC chips have it own limitations. If it was originally designed for a NOS as an option, then it could be incorporated, otherwise it is not going to be an option. Some FPGA solution can go for using an Oversampling with 1X factor instead, which make it an Oversampling yet without Oversampling....we may think of it as a NOS, or more precisely a +NOS feature.

I am not sure if the D16 will incorporate NOS as a feature, or what exactly is the rate of SRC for OS ? it would be very interesting to find out. Also, the measurements and it specifications can tell for a story by itself, and technically as a Converter, it is a primary goals. Then how it sound ? the main goal and objective ? to be subjective ? only one way to find out

For all that said, I am extremely interested in D16. Something to compete against the like of Sony/Chord ? why not ? or we can trim down the herd and .... selectively be saying, Sony only has Walkman with Class D output and no true line out, while Chord is an all in one solution of FPGA together with a lot of IC-Opamps design rather than resistive ladder for LPF....this really puts Cayin/Ibasso into the spot lights. Before I forgot it, why is R2R LPF a point of interest ? Because with this, the analog tuning can be stretched further . Because, technically, and honestly, There can only be so much our ears can hear, whether it is floor noises or distortions, even when we say the better the measurements is the more desirable, but the tuning of the signature across the board is actually an impactful factor. Therefore, having a better way to tune your own analog signature, is absolutely desirable.

Why am I more gravitated toward the D16 ? well, because it is a dedicated solution toward high-end system with dedication, and not just aiming toward compact solutions like the Ru7.

Last and not Least, will Cayin be using the same Technicalities as Ibasso ? I personally Dont think that they are similar. There is zero way to tell exactly, and I don't think either of them want to disclose it either, as said above, having their own Discrete Low Pass Filter is having their own impact into their own Canvas and Arts which will yield unique results. The possibilities are endless, they could be using different factor of feedbacks, different set of values...etc....So, they are similar but also very Uniquely Different. Asking this question is like asking if both Picasso and Davinci were both using paints/canvas/brushes, would they be having the same arts

 

[Z_Showmaster]

Read your musings and it is a nice summary of what's past and also to come, I think there's so much in miniature digital that promises so much.

Why Cayin, Ibasso is claiming 1 bit DAC as a title ? Isnt all DAC converting multibit into 1 bit before passing to LPF (Low pass filter)

Just clarifying this though that technically clear distinctions must be made between 1-bit, and anything >1-bit. Just because 1-bit is PDM, and 2-bits is PCM. Anything mutli-level can no longer be low-pass filtered because of several elements switching between 1 and 0 simultaneously – no matter if there is noise-shaping involved (as your discussion into oversampling went).

Because 1-bit represents a single element turning on and off, very fast, at any one time, DSD/PDM is digital's closest point to analog, and can be low-pass filtered easily to derive analog. Not so with 2-bits anymore, when the numerous elements need to be recombined dynamically with element matching. In there is the pitch of 1-bit/DSD/PDM – there are no errors, and it is magnificently simple to decode as a digital signal to analog.

In trying to distill the difference between a 1-bit PDM and >1-bit PCM decode, 1-bit's low-pass filter will be implemented in the analog domain (since no DSP can be used to treat 1-bit). >1-bit's low-pass filter can be implemented in the digital domain (the noise-shaping process).

Not every DAC converts delta-sigma/PWM to 1-bit, especially not if we're talking normal chip DACs. And then there are the R-2R designer junkies and noise-shaped oversampling camp that would not tolerate going anywhere near DSD because they do not believe it a viable format for highest-performance digital still.

 

[Whitigir]

Read your musings and it is a nice summary of what's past and also to come, I think there's so much in miniature digital that promises so much.



Just clarifying this though that technically clear distinctions must be made between 1-bit, and anything >1-bit. Just because 1-bit is PDM, and 2-bits is PCM. Anything mutli-level can no longer be low-pass filtered because of several elements switching between 1 and 0 simultaneously – no matter if there is noise-shaping involved (as your discussion into oversampling went).

Because 1-bit represents a single element turning on and off, very fast, at any one time, DSD/PDM is digital's closest point to analog, and can be low-pass filtered easily to derive analog. Not so with 2-bits anymore, when the numerous elements need to be recombined dynamically with element matching. In there is the pitch of 1-bit/DSD/PDM – there are no errors, and it is magnificently simple to decode as a digital signal to analog.

In trying to distill the difference between a 1-bit PDM and >1-bit PCM decode, 1-bit's low-pass filter will be implemented in the analog domain (since no DSP can be used to treat 1-bit). >1-bit's low-pass filter can be implemented in the digital domain (the noise-shaping process).

Not every DAC converts delta-sigma/PWM to 1-bit, especially not if we're talking normal chip DACs. And then there are the R-2R designer junkies and noise-shaped oversampling camp that would not tolerate going anywhere near DSD because they do not believe it a viable format for highest-performance digital still.

Yes, good point to further clarifying 1 bit PDM vs multi bit PCM and in which domain they are lying at

I think R2R with noise shaping people claimed that toleration due to the limitations of the physicality on the resistors and capacitors limits as the components

Don’t normal chip DAC convert to 1 bit in digital domain (sigma delta Oversampled) while Cayin and Ibasso convert it to 1 bit as well but forward it to the Analog domain as you said ? I think I usually call it a different way to skin the cat.



I think I have asked Cayin previously about NOS topology, but I don’t think Cayin had the ability to incorporate their NOS topology into their designs. I asked the same question toward Ibasso, and awaiting for their answer. It would be very interesting to know. However, I do believe that D16 has the potential, since it is using FPGA rather than SRC chips similar to Cayin. There is highly likely possibly that the FPGA could accommodate this. If so, then the D16 will be even further more interesting

 

[Whitigir]

Additional side of a debate between the integrated chips is that fabrication of such IC and wafer, the industry is able to fabricate silicon dies that is 100x or higher in the more accurate tolerances precisions together with much tighter PPM as part purity is at the absolute highest purity as a manufacturer limits can be. Hence bringing down the thermal coefficient to the lowest rather than traditional capacitors, film foils/ wire wounded resistors. I think this is where Luxury&Precision is still one step ahead, by using R2R topology but from an IC wafer chip itself which took this advantages and implemented it in their players. Too bad, they don’t have a standalone DAC which I am more interested in like the D16

Also, looking at this side of the debate, we will be back at the point of fabricating the FPGA chip and it quality. Different chip makers/brands have their own fabrication techniques, which results in different chips much like AMD vs Intel whatever. My points is that the FPGA will be the core of the sound performances. I think Chord uses Xilints which is now Intel ? I haven’t been keeping up with these merging stuff

Oh ghost, all this stuff reminds me of Walkman M2 and their FPGA for DSD remastering (another sigma delta OS topology) …. well, too bad, Sony also don’t have portable DAC! Back to the D16 we are !!!!

 

[ahossam]

On another note, PB5 & D16 is not limited run right?

 

[Whitigir]

On another note, PB5 & D16 is not limited run right?

I hope it isn’t but in fear of missing out….grab them while they are hot instead

 

[Z_Showmaster]

First of all I've got to say @Whitigir that it's great to be having these discussions with you, that's how the knowledge is furthered. It's nice there's so much curiousity about new technology. Technology has to have an audience in a consumer product; otherwise its existence is meaningless even if it's technically better.

There are different levels to DAC architectures, which I'll try to summarise with examples here:



Chip DACs: 90% of run-of-mill DACs, using DAC chips and the digital filters onboard

Chip DACs with custom software: Using DAC chips as a pure decoder and bypassing the filters aboard with their own written in-house, like the way Auralic, Ayre, Weiss, Benchmark and Ferrum Wandla use ESS. ROHM also gives manufacturers that function, AKM added it recently

Discrete multi-level, oversampling noise-shaped DACs: A DAC built from scratch with individual parts and proprietary digital filters and code working in 2-7 bits, like Chord and dCS (RING DAC is a 5-bit delta sigma design), new Schiit Gungnir 2. R8II falls into this category

Discrete 1-bit DAC: Converts PCM to DSD to be decoded as a 1-bit signal. Cayin's N7 and RU7 work like that. Gear from PS Audio, Playback Designs, EMM Labs, Nagra and Mola Mola works like this etc

NOS R2R: Non-oversampling R-2R networks work above 16 bits without a digital filter like Holo Audio, totalDAC, and selectable by Schiit True Multibit, Denafrips, HiBy RS6 and RS8 etc

OS R2R: Oversampling R-2R networks work above 16 bits with a custom digital filter like Rockna and selectable by Schiit True Multibit, Denafrips, HiBy RS6 and RS8 etc

Just a quick breakdown I made in the HiBy R8ii thread.

Don’t normal chip DAC convert to 1 bit (sigma delta Oversampled) with a built in integrated LPF of themselves and output analog signal (industrial secret trade guarded) ?

In there I definitely see the grounds for confusion with the interpretation of the diagram. Maybe with DSD, we start by the fundamental that it is a 1-bit format, and that under no circumstances can pure 1-bit be edited digitally. That means things like no digital volume control or EQ etc.

The next thing is that, in delta-sigma chip DACs, they're composed of five or six 1-bit elements at the point of conversion. The PCM and DSD pass through the same 1-bit stage, albeit fundamentally differently because 1-bit DSD and 2-bit PCM are night and days apart in terms of code.

The ES9038Pro is a 6-bit PWM chip. An AKM 4499EXEQ is a 7-bit delta-sigma chip. It consists of 6 or 7 1-bit elements, all 6-7 switching together at once with their outputs combined by element matching. PCM doesn't go through shrunk as 1-bit; it goes through as 6-7 1-bit paths, all together. That's because to change the 6-7 bits to 1-bit would require more software (when you reduce bit depth by 1 and double the sampling rate, you preserve digital resolution); the hardware 1-bit elements cannot decode 6-bits. Each bit is weighted and treated as PCM, not DSD.

The PCM LPF lies in the digital domain on the other hand in oversampling DACs. Reconstruction filters basically build the waveform and reject frequencies above Nyquist, so ideally the frequencies that pass through are half or less.

Not so with DSD. It goes through the same converter in a DAC, but that gets let through as 1-bit (you can't have more bits than the signal has unless digitally altered of course, meaning non-DSD DACs like Schiit need to turn the signal into PCM). Then the point of conversion becomes an analog low-pass filter, simply removing the quantisation white noise from the signal (DSD64's huge noise at 30Khz comes from this).

So by seeing noise-shaped PCM as its 5-bit weighted whole, and DSD as a 1-bit simplification, we see how engineers of both chip and discrete DACs attempt to decode each.

If you see a DAC claim to process DSD natively without alteration but offer you digital volume control or EQ, they're lying. Pure DSD can't be altered digitally. That's why you've seen all ROHM implementations I'm aware of come with analog volume controls. ROHM doesn't process DSD via any PCM process. AKM offer such a DSD passthrough as an option. ESS are a mystery to me still despite them releasing their datasheets officially a few years back.

I hope this sheds some light on how DAC chips operate with their single 1-bit DC elements working together but in wholly different ways to decode PCM and DSD.

 

[Whitigir]

First of all I've got to say @Whitigir that it's great to be having these discussions with you, that's how the knowledge is furthered. It's nice there's so much curiousity about new technology. Technology has to have an audience in a consumer product; otherwise its existence is meaningless even if it's technically better.



Just a quick breakdown I made in the HiBy R8ii thread.



In there I definitely see the grounds for confusion with the interpretation of the diagram. Maybe with DSD, we start by the fundamental that it is a 1-bit format, and that under no circumstances can pure 1-bit be edited digitally. That means things like no digital volume control or EQ etc.

The next thing is that, in delta-sigma chip DACs, they're composed of five or six 1-bit elements at the point of conversion. The PCM and DSD pass through the same 1-bit stage, albeit fundamentally differently because 1-bit DSD and 2-bit PCM are night and days apart in terms of code.

The ES8038Pro is a 6-bit PWM chip. An AKM 4499EXEQ is a 7-bit delta-digma chip. It consists of 6 or 7 1-bit elements, all 6-7 switching together at once with their outputs combined by element matching. PCM doesn't go through shrunk as 1-bit; it goes through as 6-7 1-bit paths, all together. That's because to change the 6-7 bits to 1-bit would require more software (when you reduce bit depth by 1 and double the sampling rate, you preserve digital resolution); the hardware 1-bit elements cannot decode 6-bits. Each bit is weighted and treated as PCM, not DSD.

The PCM LPF lies in the digital domain on the other hand in oversampling DACs. Reconstruction filters basically build the waveform and reject frequencies above Nyquist, so ideally the frequencies that pass through are half or less.

Not so with DSD. It goes through the same converter in a DAC, but that gets let through as 1-bit (you can't have more bits than the signal has unless digitally altered of course, meaning non-DSD DACs like Schiit need to turn the signal into PCM). Then the point of conversion becomes an analog low-pass filter, simply removing the quantisation white noise from the signal (DSD64's huge noise at 30Khz comes from this).

So by seeing noise-shaped PCM as it's 5-bit weighted whole, and DSD as a 1-bit simplification, we see how engineers of both chip and discrete DACs attempt to decide each.

If you see a DAC claim to process DSD natively without alteration but offer you digital volume control or EQ, they're lying. Pure DSD can't be altered digitally. That's why you've seen all ROHM implantations I'm aware of come with analog volume controls. ROHM doesn't process DSD via any PCM process. AKM offer such a DSD passthrough as an option. ESS are a mystery to me still despite them releasing their datasheets officially a few years back.

I hope this sheds some light on how DAC chips operate with their single 1-bit DC elements working together but in wholly different ways to decode PCM and DSD.

Thank you for this post. I also love to have discussions like this and further my knowledge. I didnt have a clue about what multi bit the AKM and ESS were sitting at, and now I know !! Thank you very much. Yes, IC Dac chips is doing this in digital domains, and Ibasso/Cayin as I understand is doing it in Analog Domain, which completely satisfy the scenarios of such statement as a pure 1 bit DAC rather than Multibit here.

My guess is Portable Balanced, and the D16 I know to reference its 16 FIR element stage that acts as the low-pass filter decoding the 1-bit stream.

So, instead of having only 6-7 elements of 1 bit, we are now arriving at 16 elements, correct ?

 

[Z_Showmaster]

Yes, IC Dac chips is doing this in digital domains, and Ibasso/Cayin as I understand is doing it in Analog Domain, which completely satisfy the scenarios of such statement as a pure 1 bit DAC rather than Multibit here.

That's right! Something to distinguish still is that while delta-sigma chips do it digitally, some discrete DACs like R8ii, Chord before them, dCS etc are all also running digital low-pass filters for anti-aliasing etc.

Then we have the DACs that are hybrid R2R and delta-sigma like TI. And Cees' work at Metrum and then Sonnet. Both of them work to split the LSBs from the MSBs when decoding PCM, meeting the requirement of decoding each better.

As an aside, as I've said in the R8II thread, I consider ESS's Hyperstream technology, now in its fourth generation, the best of its kind for chips. That's why so so many Hi-Fi manufacturers use ESS, but bypassing their digital filters for a degree of customisation.

So, instead of having only 6-7 elements of 1 bit, we are now arriving at 16 elements, correct ?

Yup, except the D16's 16 1-bit elements are not built to decode PCM; they're therefore not binary weighted, but unitary weighted. Simple, mere low-pass filters without dynamic element matching. And each is run in time according to the FPGA 2's clock cycles programmed aboard at megahertz level to reconstruct the analog audio waveform.

 

[Whitigir]

That's right! Something to distinguish still is that while delta-sigma chips do it digitally, some discrete DACs like R8ii, Chord before them, dCS etc are all also running digital low-pass filters for anti-aliasing etc.

Then we have the DACs that are hybrid R2R and delta-sigma like TI. And Cees' work at Metrum and then Sonnet. Both of them work to split the LSBs from the MSBs when decoding PCM, meeting the requirement of decoding each better.

As an aside, as I've said in the R8II thread, I consider ESS's Hyperstream technology, now in its fourth generation, the best of its kind for chips. That's why so so many Hi-Fi manufacturers use ESS, but bypassing their digital filters for a degree of customisation.



Yup, except the D16's 16 1-bit elements are not built to decode PCM; they're therefore not binary weighted, but unitary weighted. Simple, mere low-pass filters without dynamic element matching. And each is run in time according to the FPGA 2's clock cycles programmed aboard at megahertz level to reconstruct the analog audio waveform.

Wow!! Now I am even more excited to see how special the D16 can be. Could it bring back my Nostalgia memories with Sony/JVC Old 1 bit DAC ? Staring at my crying wallets

 

[Whitigir]

Both of you are on to something in discussing D16's DAC. To begin differentiating R-2R from a 1-bit resistor low-pass filter I think we first have to discuss PCM as a multi-bit format (>1-bit) whereas DSD is always a single 1-bit format.

Therefore, with a multi-bit PCM format we begin dealing with MSBs and LSBs, with Cees Ruijtenberg and Jeff Zhu holding particular patents treating each differently in their R-2R DACs. Whereas with 1-bit DACs we're only ever dealing with two states of a single DC element at one time: 1 or 0, on or off. DSD's key is its simplicity and lack of error.

The most concisely comprehensive explanation between conventional R-2R for PCM and a 1-bit DAC I've found has been cited here.

The concept of the D16 is fundamentally the same as N7 in that it decodes all digital to analog as DSD. How D16 treats the incoming bits via DSP is fundamentally different. All formats, including DSD, is processed in very high-rate DXD like PS Audio or Mola-Mola or EMM Labs (explaining the presence of their digital volume controls) before conversion to analog. N7 does not pre-process DSD. Only PCM is converted to DSD – DSD passes straight through unaltered without SRC or upsampling on N7 and RU7.


This is my interpretation, not information. D16 is different to the way N7 works, yes, and I believe D16 use the 16 hardware FIR filters low-passing 1-bit at intervals of one clock cycle between each, run by their new FPGA 2.0 algorithm first described in DX260's presentation. We are probably seeing much more accurate waveforms produced as a resort because of how each stairstep in time is plotted a lot more precisely.

Does this matter if you use USB C or Coax at all ? or all DSD will be processed into DXD regardless

 

[Z_Showmaster]

Yup, all digital inputs feed to D16's DSP that treats all incoming formats with a DXD process.

 

[yannis]

From Weibo:

PB5 is 3 days away from the countdown to the start of sales, and it will officially debut on January 23rd! Dual Nutube tube amps take you to enjoy the rich bile.