I will summarize some concepts into a couple of points, so that the reader knows what is necessary or required for a consumer to choose his dac:
- know that 99% of audio signals used are either pcm or pdm:
that is because pcm covers nearly 99% of audio formats, every format you ever encounter is modulated in pulse code and has pcm containers, the major 1% exception is DSD which uses pdm, this point is essential for one to understand how different DACs function and base his preference upon.
-know that 99.9% of dacs are either delta-sigma or resistor based:
-90% of dacs use commercial chips manufactured by vendors like ESS, AKM, Burr Brown and Cirrus logic/Wolfson.
-99% of these commercial chip dacs use delta sigma modulators
-so in essence ~90% of audio marketshare is dominated by commercial chip dacs with delta sigma modulators.
1-Commercial chip delta sigma DACs:
what these dacs do is:
stage 1: fetches data using the interface then manages the data and sync clocks for the waveform of the file.
stage 2, scenario a: any incoming signals will get converted (up-quantized and upsampled) to high sample rate pcm signal (pcm signal used by the dac internally using digital interpolation filter (to 24-192khz or higher sample rates, not necessarily the maximum pcm signal the dac can accept, this means the DAC marketed sample rate is simply just it's data input capability).
Yes, all your pcm music gets converted in your regular dac, all your pcm music gets upsampled, hence the "upsampling is bad for your audio and you should never upsample with software!" argument is stupid and redundant since your audio will already be upsampled inside your hardware.
stage 2, scenario b: incase your chip dac is marketed to have "dsdxxx native" or "dsd path" or anything dsd related (99% of commercial hifi dacs) it means it has a separate path for dsd files, the signalleft untouched until it reaches the core of the DAC and uses only a single bit DAC switch, only if the input signal is in dsd (pdm containers like .dsf or .dff)
the true power of delta-sigma over other DACs is really just DSD.
bypasses everything to stage three and gets converted to multibit pdm entering a delta-sigma modulator, so it's only oversampled once.
stage 3: (DAC core) the upsampled pcm signal in a delta sigma modulator, a ddc and the modulator are the "dac chip" itself, which they essentially yet again convert (downquantize and oversample at the same time) the interpolated pcm signal from the ddc to a multibit pdm signal like 5-bit (yes, all your pcm files are processed using the same modulation of dsd, although multibit and to a sample rate limited and specified by the dac architecture itself like 2.8 mhz)
stage 4: the delta-sigma boy (DAC) then uses this 3, 4 or 5 bit pdm signal (corpos call it and market it as bitstream) to operate multibit like switches (called current switch circuits or current steering) and delta-sigma negative feedback loop ensures all the details are correct in concurrency for this operation.
only after this step, you get an analog waveform.
stage 5: a variety of analog filters are applied, like LPF/reconstruction filters to smooth the hell out of that innacurate jagged mess, means everything is automatically smoothed out even if it means sacrificing some genuineness.
Q/ but what does all this mean?
A/ All commercial delta-sigma chip dacs, even the hifi ones or the ones branded for dsd capabilities are really 3,4 or at best, 5 multibit pdm dacs in function.
these real bits of operation are called "quantizer bits".
what they hide from you is the important stuff, how many quantizer bits are actually used in this operation, that they they will never specify because it's a major turnoff and would spin the consumer into a rabbithole of research to understand the what and why.
2-Mutlibit chip dacs
there are two variants of these
a- commercial delta-sigma chip dacs branded as "multibit": multibit as a standalone term for any DAC device that uses delta-sigma ess or akm or burr brown or cs/wolfson chips, is mostly just a marketing term, as we concluded before all these dacs are multibit delta-sigma.
You will occasionally stumble upon those commercial "multibit" dacs like topping e50,
it is just all marketing since it uses ess9068as, a delta-sigma dac.
however there are exceptions:
b- segmented r2r integrated chip multibit dacs (no delta-sigma):
essentially chip DACs, often with multiple paths just like chip DACs however they use chip integrated multibit design, like laser trimmed segmented r2r, inferior to pure r2r in cost and complexity, this is just things that you don't need to consider thinking about, it's the manufacturer's way of saying "our dac doesn't use a delta-sigma modulator and it's more exclusive and harder to make"
however all these "true multibit" chips are cheap and not completely bespoke regardless, since they use integrated and segmented r2r like AD5547, to be fair to schiit, ad5547 costs like $35 while ess9068as costs like $18, you're paying these dac vendors in excess of $200-$1000 for something so not exclusive and measures worse than delta-sigma, a regular delta-sigma dac like smsl su-1 for $80 would be sufficient, anything more expensive would be low noise, high dynamic range, and measurement chasing only.
which is kind of meaningless considering that it is essentially more or less the same principles of operation regarding modulation.
it's just a manufacturer's way of saying "we're not using commercial chip DACs, but our design is not completely bespoke because you're broke and on a budget, you get for what you pay"
3- Multibit resistor DACs (bespoke architecture):
DACs which don't use Delta-sigma modulation or ICs for the DAC core, instead they use direct resistor approach.
when NOS is activated, these DACs don't oversample, however conversion still happens as they truncate/downconvert PCM input to lower bit depth if the bit depth exceeds the ladder network bits, they usually have 16 bits, this is the limitation of these DACs since the highest I could come across for R2R ladders in high end R2R is 20 bit (I hope I'm wrong) I also read somewhere that more bits above this limit increase distortion.
and since they use multibit resistor architecture, the pcm file will get converted to analog in it's native sample rate as long as it is lower than the ladder network bit depth, it is the delta-sigma dac equivalent of dsd path, but for pcm instead of dsd
as for dsd playback, the cool thing about these DACs is that they can have a separate 1-bit DAC path for DSD handling without any conversion,
there are three main architectures I remember studying when I was in computer engineer college and as I remember the main difficulty with these DACs stems from the resistor values.
1-Binary weighted:
the most simple in design and circuity, the hardest to implement since it requires precise resistor values.
This is for 6 bit dac, the problem is clear:
every single resistor used here is a different value, and it needs to be double of the previous one.
the problem is the values, for example if 2R isn't near exactly twice the value as R, it's a problem, if 32R isn't near exactly twice as 16R, problem, if 32R isn't exactly near 32 times the value of R, problem.
this is called resistor tolerance and matching and for such a design it needs to be unrealistically and impossibly low and precise otherwise you will have all sorts of problems, for high bit applications like consumer audio which uses atleast 16 bit, this architecture is impossible.
2- Resistor string DAC:
while this looks much practical than the previous architecture since you only see R so all resistor values match, what you see is only a 3 bit resistor DAC, that is because resistor string DACs use 2^n-1 count of resistors, that means for 16 bit DAC, you need 2^16-1 resistors, 65535 resistors!!!
clearly impractical for audio, which brings us to:
3- R2R (resistor ladder) architecture:
this is clearly the best implementation for resistor based audio DACs.
why? because it needs 2N resistor count
so for 16 bit dac, you need 32 resistors
but with a minor caveat, half of these resistors should be R, and the other half should be twice that value, hence the name R2R.
this causes some difficulty regarding tolerance, however not as much as binary weighted where you have all resistors in multiples of previous ones. also it keeps the architecture compact and simple, needing only 32 resistors instead of 65535 resistors for the resistor string.
it is still difficult to get two groups of resistors in which one is double the value of the other, hence why they are more expensive, their price boils from the resistor manufacturing and precision afterall.
that is why R2R DACs like denafrips are so expensive, even if they perform worse than delta-sigma DAC for the same price.
But do they really perform worse???
4- Custom FPGA DACs (bespoke architecture):
these are DACs that use bespoke resamplers and fpga modulators instead of delta-sigma chips where everything, even DSD, gets converted to a fixed format. since they are like 100% bespoke, they will empty your wallet, and your savings.
they essentially do the same thing a delta-sigma modulator does for pcm: convert to a unified format, but the conversion happens for everything, for most of them, no seperate paths are used but the conversion technique is better in theory, chord dave uses "pulse-position modulation", mola mola uses a gigantic 32 bit 3mhz pwm format, which is a way of saying "you'll pay $10000 for an upsampling DAC".
it's also not only bespoke for sampling design, those FPGA DACs can have weird design choices and philosophy, like having and operating only on a 1-bit only DAC, things get weird because that sounds too old and lazy, but recent DACs like PS audio directstream mk2 are implementing these old designs with custom FPGAs for a purpose and philosophy.
FPGA single bit/directstream DACs are a single bit only, there is no delta-sigma modulation at all.
so what happens? like all fpga DACs, all none rate matched files, whether PCM or DSD will get upsampled and up-quantized to a fixed higher DSD format using fancy, wallet emptying fpga design, and the single bit DAC will operate on this signal.
WITH MATCHED DSD INPUT: same thing as dsd path on delta-sigma modulators.
Regardless I'm not a fan of these DACs since everything gets converted, with hifi resistor based or even 90% of delta sigma DACs, you get atleast one path that is native with no oversampling or conversion, however unified formats are said to have "uniformity" meaning everything is passed or converted as a single type, making all filters and whatever algorithms inside the chip highly optimized for said internal format, I don't believe this because they measure just as delta-sigma DACs most of the time (ps audio even measures worse) I think the conversions add more complexity at the design and at the same time it can be considered a false, lazy logic since it saves the hustle of applying different filters and algorithms to different types of input compared to other high end R2Rs and delta-sigma DACs.
So, What would be the perfect "one for all" DAC while being affordable as possible?
Hypothetically speaking, a discrete NOS R2R bespoke DAC with separate path for native DSD (single bit) sets (64 up to 1024)
this philosophy makes the most sense, 99% of all music files are either pcm or pdm, it would make sense to have a DAC that operates these files directly without any oversampling or conversion, well atleast up to 20 bit for R2R (as 24 does not seem feasible for current materials and engineering).
such DACs exist but since R2R is expensive in nature, we're talking about a starting point of $800 with the denafrips ares, gustard r26 as well, but these are low performers for both DSD and PCM.
high end unicorns like holo may which excel in every audio process path would cost like 5 grands.
Ares doesn't have native DSD handling, it uses 6 bit FIR, other R2R ladder DACs use unary, none of which are true single bit DACs.
this is where the line is drawn and makes delta-sigma look arguably better, since it the most affordable and cheep between all these imperfect DACs, also achieving very low noise and high SNR is easy with these DACs.
but then, even delta-sigma DACs will make you think twice, as seperate DSD paths will still convert DSD to multibit PDM.
The conclusion is: No DAC is perfect, and all for one DACs like holo may are insanely expensive, ranging from 5 grands all the way to $20000.
I would make corrections by quoting if I found that any information given here is innacurate, I hope this thread serves as a guide for anyone who's interested about audio science and processing paths as well as buying/choosing a DAC from one of these categories.
- know that 99% of audio signals used are either pcm or pdm:
that is because pcm covers nearly 99% of audio formats, every format you ever encounter is modulated in pulse code and has pcm containers, the major 1% exception is DSD which uses pdm, this point is essential for one to understand how different DACs function and base his preference upon.
-know that 99.9% of dacs are either delta-sigma or resistor based:
-90% of dacs use commercial chips manufactured by vendors like ESS, AKM, Burr Brown and Cirrus logic/Wolfson.
-99% of these commercial chip dacs use delta sigma modulators
-so in essence ~90% of audio marketshare is dominated by commercial chip dacs with delta sigma modulators.
1-Commercial chip delta sigma DACs:
what these dacs do is:
stage 1: fetches data using the interface then manages the data and sync clocks for the waveform of the file.
stage 2, scenario a: any incoming signals will get converted (up-quantized and upsampled) to high sample rate pcm signal (pcm signal used by the dac internally using digital interpolation filter (to 24-192khz or higher sample rates, not necessarily the maximum pcm signal the dac can accept, this means the DAC marketed sample rate is simply just it's data input capability).
Yes, all your pcm music gets converted in your regular dac, all your pcm music gets upsampled, hence the "upsampling is bad for your audio and you should never upsample with software!" argument is stupid and redundant since your audio will already be upsampled inside your hardware.
stage 2, scenario b: incase your chip dac is marketed to have "dsdxxx native" or "dsd path" or anything dsd related (99% of commercial hifi dacs) it means it has a separate path for dsd files, the signal
bypasses everything to stage three and gets converted to multibit pdm entering a delta-sigma modulator, so it's only oversampled once.
stage 3: (DAC core) the upsampled pcm signal in a delta sigma modulator, a ddc and the modulator are the "dac chip" itself, which they essentially yet again convert (downquantize and oversample at the same time) the interpolated pcm signal from the ddc to a multibit pdm signal like 5-bit (yes, all your pcm files are processed using the same modulation of dsd, although multibit and to a sample rate limited and specified by the dac architecture itself like 2.8 mhz)
stage 4: the delta-sigma boy (DAC) then uses this 3, 4 or 5 bit pdm signal (corpos call it and market it as bitstream) to operate multibit like switches (called current switch circuits or current steering) and delta-sigma negative feedback loop ensures all the details are correct in concurrency for this operation.
only after this step, you get an analog waveform.
stage 5: a variety of analog filters are applied, like LPF/reconstruction filters to smooth the hell out of that innacurate jagged mess, means everything is automatically smoothed out even if it means sacrificing some genuineness.
Q/ but what does all this mean?
A/ All commercial delta-sigma chip dacs, even the hifi ones or the ones branded for dsd capabilities are really 3,4 or at best, 5 multibit pdm dacs in function.
these real bits of operation are called "quantizer bits".
what they hide from you is the important stuff, how many quantizer bits are actually used in this operation, that they they will never specify because it's a major turnoff and would spin the consumer into a rabbithole of research to understand the what and why.
2-Mutlibit chip dacs
there are two variants of these
a- commercial delta-sigma chip dacs branded as "multibit": multibit as a standalone term for any DAC device that uses delta-sigma ess or akm or burr brown or cs/wolfson chips, is mostly just a marketing term, as we concluded before all these dacs are multibit delta-sigma.
You will occasionally stumble upon those commercial "multibit" dacs like topping e50,
it is just all marketing since it uses ess9068as, a delta-sigma dac.
however there are exceptions:
b- segmented r2r integrated chip multibit dacs (no delta-sigma):
essentially chip DACs, often with multiple paths just like chip DACs however they use chip integrated multibit design, like laser trimmed segmented r2r, inferior to pure r2r in cost and complexity, this is just things that you don't need to consider thinking about, it's the manufacturer's way of saying "our dac doesn't use a delta-sigma modulator and it's more exclusive and harder to make"
however all these "true multibit" chips are cheap and not completely bespoke regardless, since they use integrated and segmented r2r like AD5547, to be fair to schiit, ad5547 costs like $35 while ess9068as costs like $18, you're paying these dac vendors in excess of $200-$1000 for something so not exclusive and measures worse than delta-sigma, a regular delta-sigma dac like smsl su-1 for $80 would be sufficient, anything more expensive would be low noise, high dynamic range, and measurement chasing only.
which is kind of meaningless considering that it is essentially more or less the same principles of operation regarding modulation.
it's just a manufacturer's way of saying "we're not using commercial chip DACs, but our design is not completely bespoke because you're broke and on a budget, you get for what you pay"
3- Multibit resistor DACs (bespoke architecture):
DACs which don't use Delta-sigma modulation or ICs for the DAC core, instead they use direct resistor approach.
when NOS is activated, these DACs don't oversample, however conversion still happens as they truncate/downconvert PCM input to lower bit depth if the bit depth exceeds the ladder network bits, they usually have 16 bits, this is the limitation of these DACs since the highest I could come across for R2R ladders in high end R2R is 20 bit (I hope I'm wrong) I also read somewhere that more bits above this limit increase distortion.
and since they use multibit resistor architecture, the pcm file will get converted to analog in it's native sample rate as long as it is lower than the ladder network bit depth, it is the delta-sigma dac equivalent of dsd path, but for pcm instead of dsd
as for dsd playback, the cool thing about these DACs is that they can have a separate 1-bit DAC path for DSD handling without any conversion,
there are three main architectures I remember studying when I was in computer engineer college and as I remember the main difficulty with these DACs stems from the resistor values.
1-Binary weighted:
the most simple in design and circuity, the hardest to implement since it requires precise resistor values.
This is for 6 bit dac, the problem is clear:
every single resistor used here is a different value, and it needs to be double of the previous one.
the problem is the values, for example if 2R isn't near exactly twice the value as R, it's a problem, if 32R isn't near exactly twice as 16R, problem, if 32R isn't exactly near 32 times the value of R, problem.
this is called resistor tolerance and matching and for such a design it needs to be unrealistically and impossibly low and precise otherwise you will have all sorts of problems, for high bit applications like consumer audio which uses atleast 16 bit, this architecture is impossible.
2- Resistor string DAC:
while this looks much practical than the previous architecture since you only see R so all resistor values match, what you see is only a 3 bit resistor DAC, that is because resistor string DACs use 2^n-1 count of resistors, that means for 16 bit DAC, you need 2^16-1 resistors, 65535 resistors!!!
clearly impractical for audio, which brings us to:
3- R2R (resistor ladder) architecture:
this is clearly the best implementation for resistor based audio DACs.
why? because it needs 2N resistor count
so for 16 bit dac, you need 32 resistors
but with a minor caveat, half of these resistors should be R, and the other half should be twice that value, hence the name R2R.
this causes some difficulty regarding tolerance, however not as much as binary weighted where you have all resistors in multiples of previous ones. also it keeps the architecture compact and simple, needing only 32 resistors instead of 65535 resistors for the resistor string.
it is still difficult to get two groups of resistors in which one is double the value of the other, hence why they are more expensive, their price boils from the resistor manufacturing and precision afterall.
that is why R2R DACs like denafrips are so expensive, even if they perform worse than delta-sigma DAC for the same price.
But do they really perform worse???
4- Custom FPGA DACs (bespoke architecture):
these are DACs that use bespoke resamplers and fpga modulators instead of delta-sigma chips where everything, even DSD, gets converted to a fixed format. since they are like 100% bespoke, they will empty your wallet, and your savings.
they essentially do the same thing a delta-sigma modulator does for pcm: convert to a unified format, but the conversion happens for everything, for most of them, no seperate paths are used but the conversion technique is better in theory, chord dave uses "pulse-position modulation", mola mola uses a gigantic 32 bit 3mhz pwm format, which is a way of saying "you'll pay $10000 for an upsampling DAC".
it's also not only bespoke for sampling design, those FPGA DACs can have weird design choices and philosophy, like having and operating only on a 1-bit only DAC, things get weird because that sounds too old and lazy, but recent DACs like PS audio directstream mk2 are implementing these old designs with custom FPGAs for a purpose and philosophy.
FPGA single bit/directstream DACs are a single bit only, there is no delta-sigma modulation at all.
so what happens? like all fpga DACs, all none rate matched files, whether PCM or DSD will get upsampled and up-quantized to a fixed higher DSD format using fancy, wallet emptying fpga design, and the single bit DAC will operate on this signal.
WITH MATCHED DSD INPUT: same thing as dsd path on delta-sigma modulators.
Regardless I'm not a fan of these DACs since everything gets converted, with hifi resistor based or even 90% of delta sigma DACs, you get atleast one path that is native with no oversampling or conversion, however unified formats are said to have "uniformity" meaning everything is passed or converted as a single type, making all filters and whatever algorithms inside the chip highly optimized for said internal format, I don't believe this because they measure just as delta-sigma DACs most of the time (ps audio even measures worse) I think the conversions add more complexity at the design and at the same time it can be considered a false, lazy logic since it saves the hustle of applying different filters and algorithms to different types of input compared to other high end R2Rs and delta-sigma DACs.
So, What would be the perfect "one for all" DAC while being affordable as possible?
Hypothetically speaking, a discrete NOS R2R bespoke DAC with separate path for native DSD (single bit) sets (64 up to 1024)
this philosophy makes the most sense, 99% of all music files are either pcm or pdm, it would make sense to have a DAC that operates these files directly without any oversampling or conversion, well atleast up to 20 bit for R2R (as 24 does not seem feasible for current materials and engineering).
such DACs exist but since R2R is expensive in nature, we're talking about a starting point of $800 with the denafrips ares, gustard r26 as well, but these are low performers for both DSD and PCM.
high end unicorns like holo may which excel in every audio process path would cost like 5 grands.
Ares doesn't have native DSD handling, it uses 6 bit FIR, other R2R ladder DACs use unary, none of which are true single bit DACs.
this is where the line is drawn and makes delta-sigma look arguably better, since it the most affordable and cheep between all these imperfect DACs, also achieving very low noise and high SNR is easy with these DACs.
but then, even delta-sigma DACs will make you think twice, as seperate DSD paths will still convert DSD to multibit PDM.
The conclusion is: No DAC is perfect, and all for one DACs like holo may are insanely expensive, ranging from 5 grands all the way to $20000.
I would make corrections by quoting if I found that any information given here is innacurate, I hope this thread serves as a guide for anyone who's interested about audio science and processing paths as well as buying/choosing a DAC from one of these categories.
Last edited:
