When talking filters in a DAC, one needs to remember your DAC itself won't produce a smooth output, it will produce a stepped waveform. The filters are there to smooth it out to something a bit more analog like. I've never studied the world of straight PCM dacs (dacs that decode to analog straight from PCM using a resistor ladder rather than 1-bit delta-sigma); they were always complex expensive devices with not much advantage over an oversampling DAC with the exception of not having an additional D-S conversion step.
I guess part of what helps if if you understand how PCM stores it's audio..it's essentially mapping amplitude levels of a waveform. Each sample is a representation of the voltage of an analog signal at that given point; represented as so many bits per sample. The more samples you have, the higher the frequency response...the more bits you have, the lower the amplitude you can represent before "running out of floor"...it also gives you a little more accuracy in the upper end...but with digital audio, it's the lower end of the scale that you really look at. If you have a track that only has a dymanic range that never drops below -6, then you don't need 24-bit to accurately represent that...it's only when you start going in to the real low amplitude levels that your range matters. So, for example, if your audio level was reduced to be an average of -90dBFS, then anything that drops below -94 or 95dB is essentially lost...there aren't enough bits to represent an amplitude that small. IT's why it used to be important to normalize your audio file in the editing stage to prevent quanitzation noise between steps. Basically, there are more bits required for lower amplitudes than higher amplitudes. There's also the old myth of "24-bit doesn't clip"....no, it will. This comes again from the world of audio editing, which is where most people encountered anything more than 16-bit audio. People skipped the fact we are editing/processing using floating point math vs integer data. Floating point allows for a much larger dynamic range than it's integer counterpart; my editor can represent a level of something like +20dbFS over clip before the edited file actually clips. You need that kind of range and overhead in editing though.
Anyway...back to the original point. PCM is basically like a game of connect the dots using a grid. Your vertical scale would represent voltage levels represented by the bits of the sample, the horizontal scale would be how many samples you have. This is how PCM stores it's audio...it's "positional" as in the sample contains all the information needed to map it to a voltage output. Do this 44,100 times a second and you get audio...but it's not exactly clean; that's where the filters come in. They'll remove some of the ultra-sonic junk and "smooth" the waveform out. As I may have (or should have) mentioned; I don't really know much about direct PCM DACs...ones that don't use an oversampling technique. I switched to 24-bit DACs 12-years ago when they first came out...not to mention most study revolved around CD players...and the majority of them used the oversampling technique.
So...let's move on to 1-bit/DSD formats. These do not actually store any samples...after all, you're only dealing with single bits. So how does that all work? 1-bit stores the *changes* involved to make the analog waveform, rather than storing the actual voltage values. You have a sawtooth wave generator that's capable of running at your max sample rate...so for DSD it's 2.882mhz per channel...that's a really fast generator and ultra-sonic signal. But it doesn't run automatically at 2.8mhz, it's just capeable of being flipped at that rate. The bits in a 1-bit format basically control the sawtooth generator. Send it a 1, it changes direction of the sawtooth generator; send a 0, it continues as normal. By controlling how often this thing changes direction; you can ramp up or drop the output voltage accordingly. Of course, it's changing faster than the actual audio...so you have to pulse it to keep it in the approximate area. Surely you know what a sine wave looks like...now picture there were lines above and below that sine wave...now draw a very rapid sawtooth between those lines? See what you get? You get an ultrasonic saw-tooth that "follows" the analog output. You don't hear the very rapid voltage changes becuase they're way above what your hearing can handle; but you are able to pick up on the larger changes in voltage. All you really have to do...is apply a low-pass filter to he audio to get rid of that. There's also distortion from noise-shaping...which the LPF also helps to rid of.
The 1-bit method is actually a lot easier to generate analog from vs having to have a component for each bit and then sum them together. It's not perfect, it has it's limitations...just as analog audio does. And what I gave is a very over-simplified version of it. I don't have a good technical understanding I can fully explain...just the "bare bones differences" between the two. If you really want to be confused, you can read
https://en.wikipedia.org/wiki/Delta-sigma_modulation on the finer details of delta-sigma modulation and
https://en.wikipedia.org/wiki/Digital-to-analog_converter#DAC_types will explain, in moderately more simplistic terms (but not as simple as I attempted to do) the differences and issues with each type of DAC on the market.
I mean..digital is complex...but I think people are making it out to be more complex. One thing that bugged me is the number of things people would claim have an effect on digital...most of which aren't related to the actual medium itself. When I hear people claim that FLAC is "still compression" and that "the extra CPU cycles hurt quality"...I actually want to go in to a rage because it shows a total lack of understanding of the big picture. People that make claims like that usually have a nice DAC but inferior components in the PC...a cheap USB host can require more CPU overhead...which can cause buffering issues and glitches. A cheap motherboard will have all sorts of noise on it's power supply that leaks in. There's a reason I use a specific laptop for audio playback; it contains a USB host with almost zero CPU overhead required...and I don't start getting quality issues till I'm doing something like mixing 32 tracks wit 66 effects...and at that point the CPU just can't render that properly. It's like I've said...if components like your hard drive, ram, or CPU usage cause quality issues...you really need to look at your motherboard than your hard drive, ram, or how you're playing the music back.
But that's all just my $.02 on the subject. I've been messing around with digital audio since long before I was ever interested in analog performance.
) believe they 'hear' differences in things like flac vs. wav, computer implementations and power supply quality - even though the numbers and theory say no-way. Numbers and theory even said digital beat analog from the very beginning, but it certainly didn't sound that way to me. I'm the first to admit however, that the evolution of the digital art has completely changed my view, and I now embrace digital right along with vinyl on it's sonic capability. I personally think format, filter design, processor implementation, DAC chip design, compression (or lack there-of), network implementation, power-supplies, etc. had and have an important role to play in that evolution. I think there is a lot more to this than 'bits-is-bits' and it either works or it's broke - that's what my ears tell me, just like they told me digital wasn't all that back in the 80s when the numbers where used to 'prove' it was all perfect.
