cliveb wrote: > John, thanks for trying to explain, but I'm still confused. Your very > first paragraph is completely at odds with what I have always understood > to be the purpose of oversampling in a DAC: > > > I thought that the effect of oversampling (ie. zero-stuffing) was that > it moves the aliasing artefacts up the frequency spectrum. If you 4x > oversample a 44.1kHz signal, then the aliasing artefacts will begin at > 88.2kHz instead of 22.05kHz. Hence you can use a much gentler > reconstruction filter - indeed, you can in this case use EXACTLY the > same filter that you would use on a non-oversampled 176.4kHz signal. > > Are you saying that I've been misunderstanding the purpose of playback > oversampling all this time, and that when you oversample the aliasing > artefacts are NOT moved up the frequency spectrum? If that is the case, > then what IS the purpose of oversampling?
I hope I can try and do this without pictures, I don't have time right now to draw some nice pictures, so I'll try and be clear with the words. So what causes the aliases in the first place? Think of a spectrum of the output of a good old fashioned ladder DAC chip running at 44.1, each time a new sample comes along the output (almost) instantaneously changes to a new value, the infamous "stair step" output. What does the spectrum of this look like? Each of those sharp edges going from one value to the next requires a series of high frequency harmonics to implement the sharp edge. These high high frequency harmonics beating with the audio signal frequencies are what create the aliases. It is purely a byproduct of the "sharp edges" in the stair step. Now lets try a 4X oversampling, we create 4 times as many samples, every fourth one being an original value and the rest being zero. Now take the spectrum of that, the sample rate is now 4 times higher, but the data only changes every 4 samples, the harmonics are at exactly the same frequencies, the amplitudes are MUCH greater because the sharp edges are now going from zero to the full sample value, not just the difference between sample values. So this step by itself has made things WAY WAY worse. The magic is what happens when you run this through the FIR filter. The filter in a nutshell puts values in those zero slots to produce a smoothly varying curve between the original samples. So what does this look like in the frequency domain? Well look at the spectrum of the zero stuffed signal, audio data up to 20KHZ and aliases and harmonics above 22.05KHz, what do you have to do? Get rid of all that stuff above 20KHz. of course! You need a filter that passes everything up to20KHz but blocks everything above 22.05KHz, this is the infamous brick wall filter. This process fills in the spaces between the original samples, it still has stair steps, but those staps are now coming at 176.4 and the height of the steps is much smaller. The spectrum of this shows audio up to 20KHz, then nothing up to 88KHz then harmonics of the sampling frequency going up from there, but these are now much lower in amplitude than the harmonics from the original 44.1 signal. The oversampling and zero stuffing BY ITSELF does not fix the situation, it just allows you to implement a filter which can filter out most of the above audio band stuff caused by the "stair step" in the original data. Again I hope this makes sense, it's a lot easier to grasp with the right pictures. John S. ------------------------------------------------------------------------ JohnSwenson's Profile: http://forums.slimdevices.com/member.php?userid=5974 View this thread: http://forums.slimdevices.com/showthread.php?t=94352 _______________________________________________ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/audiophiles
