So a filterless nos DAC won’t add junk in to the audible range?
Here is a very quick description of the D/A process (with no oversampling) from Chapter 3 of
The Scientist and Engineer's Guide to Digital Signal Processing.
The "filterless NOS" conversion represents only the top 2 steps in the left ("a" and "c", time domain) and right ("b" and "d", frequency domain) figures. "Filterless" is a misnomer. The zeroth-order hold conversion (= sample-and-hold stairsteps, as shown in figure "c") is actually a convolution of the digitized samples in the form of an impulse train (figure "a") with a
unit square pulse. In the frequency domain, it is the spectrum of the digitized signal multiplied by the Fourier transform of the unit square pulse, which is the sinc function (spectrum of the stairsteps = the product of the 2 curves in figure "d"). Therefore, the sample-and-hold stairsteps actually are the results of the digitized samples filtered by a filter that has a frequency response that looks like a sinc function.
To correct the frequency response errors of the filtering by the sample-and-hold process (and to remove the residual "junk" above Nyquist due to the incomplete LP filtering), the stairsteps needs to be filtered by a filter with the response that is shown in the right side third from figure "e" (the "ideal reconstruction filter"). After this filter, you'll have the proper analog signal reconstructed (figures "f" and "g"). Without this second filter, the filterless NOS produces a non-flat frequency response in addition to the inadequate removal of the "junk" above Nyquist.