Upsampling allows mathematical filtering which is more accurate than anything in the analog domain. It also tends to randomize differential non-linearity from physical devices.
Anything built in the real world will suffer from differential non-linearity. This occurs in both time (jitter) or in amplitude. (True whatever your DAC topology)
It is simple physics. Build anything with discrete levels (steps) in time or amplitude then you will have inherent physical limitations in accuracy between adjacent levels (steps). The errors are always there whatever you physically build. So building something with many more than necessary steps (levels) means the errors are far more likely to be smaller, benign, inaudible and random.
It is always better to have a higher bit resolution and higher sample rate than actually needed. This means differential errors are more likely to occur outside the audible range (below the noise floor or at higher than audible frequencies).
Think about it.
What is better?
Errors that are close to or within the audible range (16bit 44.1KHz) or errors that are way outside the audible band and can be filtered out (upsampled or high-resolution)
For these reasons most folks find that mathematical upsampling in Roon or other software can help improve the sound from many DACs. This is because it changes the nature of the physical analog errors that absolutely every DAC suffers from.
Latest DAC Chip design tries to leverage this issue by having a hybrid solution - a mathematical random software selects the actual circuits combined to create the amplitude on every cycle - ensuring that no two adjacent samples ever use the same circuit and thus inherent physical differential non-linearity errors are random.
Random errors are several orders of magnitude better than any noise that correlates to a musical signal.
The most accurate DACs like Benchmark DAC 3 are using this hybrid approach. They are equivalent to a 6 bit R-2R DAC running at a very high sample rate. This approach minimizes BOTH the impact of errors in time (jitter) and signal level (amplitude or bits). This approach measures an order of magnitude better than the very best most expensive R-2R NOS design or the very best 1 bit Delta Sigma or DSD.
If you don’t like the sound of accuracy and prefer a bit of distortion then you are part of a large camp. I like tubes too!!!
The above is just a discussion of pure performance accuracy and does not account for tastes.
Anything built in the real world will suffer from differential non-linearity. This occurs in both time (jitter) or in amplitude. (True whatever your DAC topology)
It is simple physics. Build anything with discrete levels (steps) in time or amplitude then you will have inherent physical limitations in accuracy between adjacent levels (steps). The errors are always there whatever you physically build. So building something with many more than necessary steps (levels) means the errors are far more likely to be smaller, benign, inaudible and random.
It is always better to have a higher bit resolution and higher sample rate than actually needed. This means differential errors are more likely to occur outside the audible range (below the noise floor or at higher than audible frequencies).
Think about it.
What is better?
Errors that are close to or within the audible range (16bit 44.1KHz) or errors that are way outside the audible band and can be filtered out (upsampled or high-resolution)
For these reasons most folks find that mathematical upsampling in Roon or other software can help improve the sound from many DACs. This is because it changes the nature of the physical analog errors that absolutely every DAC suffers from.
Latest DAC Chip design tries to leverage this issue by having a hybrid solution - a mathematical random software selects the actual circuits combined to create the amplitude on every cycle - ensuring that no two adjacent samples ever use the same circuit and thus inherent physical differential non-linearity errors are random.
Random errors are several orders of magnitude better than any noise that correlates to a musical signal.
The most accurate DACs like Benchmark DAC 3 are using this hybrid approach. They are equivalent to a 6 bit R-2R DAC running at a very high sample rate. This approach minimizes BOTH the impact of errors in time (jitter) and signal level (amplitude or bits). This approach measures an order of magnitude better than the very best most expensive R-2R NOS design or the very best 1 bit Delta Sigma or DSD.
If you don’t like the sound of accuracy and prefer a bit of distortion then you are part of a large camp. I like tubes too!!!
The above is just a discussion of pure performance accuracy and does not account for tastes.