What are the effects of using min / lin phase for oversampling only?

[Colox]

When it comes to oversampling on plugins, some plugins offer to use either linear phase or non-linear phase (minimum) for the oversampling process only. IIR filter plugins can offer to use linear phase for just the oversampling only, but min phase for the "main" processing - so to speak.

Non-linear phase causes phase smearing, while linear phase generates pre-ringing. So if I'm using an IIR EQ plugin, and select to use linear phase for 8x oversampling ... I'm lacking in knowledge to figure out what comes out of that. I'm assuming, since these appears to me to be two somewhat separate processes, that they need to get blended together at a later stage. Meaning, will this mix phase smearing and some pre-ringing?

Is it generally a better idea to use uniform processing settings between main- and oversampled processing, or odd settings? I'd like to understand more about this.

Thanx

[ashcat_lt]

They’re not really “blended together”. They are series processes so more like on top of each other. Oversampling involves (at least) two filters - one on the way in, and one on the way out. So if you’ve got FIR filtering for oversampling wrapped around an IIR filter, you get some preringing, which will be affected by the IIR filter which will also do its phase smearing thing and then more preringing.

I personally avoid FIR processing wherever possible. Everything that was ever recorded before some time in the mid 90s used IIR filters, and it worked fine. I’ve never really heard a track ruined by “phase smearing” unless I was really just doing something stupid to begin with, but I have heard those “robot ghost birds” and they are almost always completely unacceptable. I can’t have it. Even if I can find settings where they’re quiet enough that you can’t really hear them, I know they’re there and it freaks me out.

I also avoid oversampling most of the time, but that’s more because oversampling will ruin the “hard limit” function of the things that actually need it like clippers and saturation. I put a saturator on a track at least partly so that I can know the exact value of the loudest sample that could possibly come out of the thing. Run that now through the reconstruction filter of the oversampling, and all bets are off. It could peak sometimes even a couple of dB higher. The clipping that will cause to me is worse than whatever impact a little aliasing here and there might have.

[Colox]

Originally Posted by ashcat_lt

They’re not really “blended together”. They are series processes so more like on top of each other. Oversampling involves (at least) two filters - one on the way in, and one on the way out. So if you’ve got FIR filtering for oversampling wrapped around an IIR filter, you get some preringing, which will be affected by the IIR filter which will also do its phase smearing thing and then more preringing.

I personally avoid FIR processing wherever possible. Everything that was ever recorded before some time in the mid 90s used IIR filters, and it worked fine. I’ve never really heard a track ruined by “phase smearing” unless I was really just doing something stupid to begin with, but I have heard those “robot ghost birds” and they are almost always completely unacceptable. I can’t have it. Even if I can find settings where they’re quiet enough that you can’t really hear them, I know they’re there and it freaks me out.

I also avoid oversampling most of the time, but that’s more because oversampling will ruin the “hard limit” function of the things that actually need it like clippers and saturation. I put a saturator on a track at least partly so that I can know the exact value of the loudest sample that could possibly come out of the thing. Run that now through the reconstruction filter of the oversampling, and all bets are off. It could peak sometimes even a couple of dB higher. The clipping that will cause to me is worse than whatever impact a little aliasing here and there might have.

Thank you so much for this explanation. This makes total sense to me. I also had this confirmed to me by another knowledgeable person yesterday. So I can believe it.

I have also picked up on pre-ringing, when very few others seem to notice it, but I have never been bothered by phase smearing other than in a couple of clinical comparison experiments. Just like you say, I'd live with that any day, over pre-ringing.
At least as a general principle. If I were to oversample effects on an analogue strings part playing slow pad-like sounds, then the option of using linear phase might cause less artefacts - but also not offer much improvement over IIR.

I was experimenting a lot with oversampling this summer, and I did manage to use advanced filtering to construct way to add plenty of (wanted) distortion effects onto clean tracks, in such a way that the distortion didn't cause any aliasing at all, even when put through clean tone sweep tests. The back side of that kind of distortion, was that every distortion effect invariably got very neat and buttery smooth in character. You could still hear it was distortion, but not much of any raw nastiness remained. Many times, you want that with distortion.
I was as surprised by this, as I was 2 decades ago when I discovered that tracks/instruments that are 100% phase coherent, means they are in mono.

As for level overshoots, which you mentioned, that sounds like intersample peaks to me. I tend to stick to AES/ÄEBU's old recommendation of setting a brickwall limiter to -0.3dbfs on the master out (if I'm exporting a master, which I seldom do). I got Stillwell's Bitter plugin (intersample peak checking) on the system but I never really use it. I used to have it on all the time, for 2-3 months, but it never showed any overshoots beyond the limiter (L2) so I lost interest in using it.

[ashcat_lt]

No, I'm literally never talking about intersample peaks. I have never actually cared about intersample or true peak anything. I'm talking about actual sample values. If you hard clip a signal without any oversampling, you will know exactly what the maximum actual sample level is going to be coming out. No matter what you put in, it can't be bigger than whatever clipping ceiling you set. If you filter afterward, you can't really know anymore, and even though the filter is techinically removing energy, it is also redistributing energy, and individual samples can and will go well past the ceiling you had previously set. It is theoretically possible to create a sort of oversampling plus feedback recursive process that might be able to actually promise its ceiling, but I'm not sure too many plugins are doing that. If you insist on using oversampled "limiters", then you generally have to do that recursive part yourself. Set it, see where it lands, adjust, see if that's better...

[pipelineaudio]

Originally Posted by ashcat_lt

but I have heard those “robot ghost birds” and they are almost always completely unacceptable

and "space monkeys" and "swirling birdies"

We need a whole thesaurus for this