Quick question about DAC oversampling

[ygg-it]

Hello, in oversampling technique done inside a DAC machine (i.e. via hardware chips) how the additional samples are calculated: by repeating each digit a number of times or by putting zeros in between or by adding samples interpolated in the digital domain in between?

 

[lcsaszar]

All variants are used.

 

[abraxalito]

Usually hardware implemented filters do upsampling in successive multiples of 2. So at each stage, every other sample is set to zero (the sample rate is doubled) and then sent to the filter.

 

[oon_the_kid]

Depends, if you are taliking audio DACs, then they behave very differently from what you were taught in Electrical Engineering in College/University/textbook. They employ what is called a reconstruction filter to fill in the missing bits so you wind up generating the perfect sinewave.

The traditional drawing a straight line which is know as oversampling is when you are sampling at 44.1KHz on a 13KHz signal, the waveform looks like crap and distortion skyrockets cause you only have 3 sample points in one waveform cycle.

I am in the Non oversampling (NOS) camp though. You can read up the various NOS versus non NOS debate on this forum.

Oon

 

[DF96]

The traditional drawing a straight line which is know as oversampling is when you are sampling at 44.1KHz on a 13KHz signal, the waveform looks like crap and distortion skyrockets cause you only have 3 sample points in one waveform cycle.

I am not sure if you are here criticising a naive view of reconstruction or advocating a naive view of reconstruction. Anyway, this is not how it works; you only need (on average) just over two sample points per cycle to get perfect reconstruction so three is enough - no distortion.

Whether DACs work differently from what you were taught in EE school depends on the quality of the school and whether you were listening carefully enough.

 

[lcsaszar]

One gets perfect reconstruction by oversampling/filtering of infinite length periodic signal, that is below fs/2.

 

[plasnu]

ygg-it, you should go to the other forum researching this subject. Too much incorrect information here.

 

[Ken Newton]

Hello, in oversampling technique done inside a DAC machine (i.e. via hardware chips) how the additional samples are calculated: by repeating each digit a number of times or by putting zeros in between or by adding samples interpolated in the digital domain in between?

The additional samples are digitally interpolated in a time-domain view. Which is to say, they are digitally filtered to remove the image bands in a frequency-domain view. Interpolation and image filtering are two views of the same process, which is, partial signal reconstruction.

Zero insertion and such are simply a technique used in conjunction with the digital filter for creating the extra bandwidth for enabling the partial reconstruction process.

 

[MarcelvdG]

Hello, in oversampling technique done inside a DAC machine (i.e. via hardware chips) how the additional samples are calculated: by repeating each digit a number of times or by putting zeros in between or by adding samples interpolated in the digital domain in between?

Theoretically by putting zero samples in between the original samples and then passing the result through a digital low-pass filter, usually a long FIR filter.

Practically by doing the exact same calculation in a more hardware-efficient manner. That is, there is little point in multiplying all the inserted zero samples with the filter coefficients and adding the results, as these will be zero anyway. Instead you can make a shorter filter that only multiplies the original samples with coefficients and adds the results, but the coefficients then need to be cyclically switched between different values.