Although this section should be top most, it never existed until today (March 30, 2015) because it was taken for granted that the Arc Prediction Filtering was engaged anyway and was left on for everybody. "No words needed".
The upcoming version of XXHighEnd will contain a new set of filters and this requires some attention.
Since the 1.186-i unofficial version a "Filter Designer" new functionality has been introduced. This is now more official and/but will be extended more in the future.
The filter as set in the Filter Designer will be active when the new "Custom" filter is activated.
Notice : The Custom Filter can not be used together with Phase Alignment.
The [ F ] button you see, brings you to the Filter Designer :
N.b.: Please disregard the Anti Image settings in the right bottom corner.
The fields in the left hand side suggest that "we" can apply all filter settings we like, but this is not so; the number of possibilities are infinite and we all will get crazy. Additionaly, you'd have no means of checking what you imply with the settings, apart from 999,999 out of 1,000,000 sounding wrong.
What was done instead is that "Pre-selections" have been provided, while we over here in the "XXHighEnd labs" can use the fields to create them *and* check with analyzers. So here's what's current in there for Pre-selections :
What you are the settings for the fields you saw in the before last screenshot plus some descriptive information (the part behind the last | character).
What the current new filters all do is actually making use of Arc Prediction which is 100% serving the time domain (no pre- and post ringing) to "shape" that into something which is more serving the frequency domain than what Arc Prediction alone can achieve. Still no visible ringing is in order while the THD (Total Harmonic Distortion) of the higher frequencies vastly improves. Now notice the "trick" here :
If we first assume that Arc Prediction is the very best sounding filter around anyway (and knowing that the number of people who do not and use the Phasure NOS1(a) D/A converter can be counted on one finger at this moment) then it can only get better as long as we can prevent the ringing.
The real thing to let his happen is to disobey the normally taken for granted 20KHz limit of hearing and "shape" more of frequency below that. This assumed that "we" can not hear above e.g. 16KHz anyway and thus we can just as well limit the output level of that. This, while Arc Prediction has a more difficult time just *above* that frequency. So notice that normal Arc Prediction folds back a little into the audio band regarding Aliasing once above 16KHz and let's assume this is audible (it officially *is* distortion, and is now in the audio band, so ...). If we now use the audio band itself for where the roll off of the filter is allowed to start, then we can have very slowly rolling off filters which actually do not ring at all that we can notice (measurement);
If we look again at the last screen shot, we see for example "Mid for 705600/768000 / 2.21% - 3.9dB". The 2.21% indicates the measured distortion at 16KHz. The -3.9dB tells that 16KHz is 3.9dB down in level compared to what it should be (at say 1KHz). Two things happened at the same time here :
1. The distortion is lower than it originally was (was in the 5%);
2. The level of the signal is down meanwhile.
So one and the same means improves from two angles. Notice though that this example assumes that "we" won't be able to hear the 16KHz anyway, so what's happening really ?
For that look at the last part of the description of the example, the 0.22% -0.1dB. This now is the measured distortion at 12KHz and how much down that is. And, we 100% assume now that everybody is able to perceive 12KHz very well (may not be so either but alas). So what we did for the net result is improving on the 12KHz again vastly (think 6-9dB of less distortion in this case) per the "means" of not wanting to hear 16KHz anyway.
Not to forget : this is all without additional ringing which otherwise is impossible.
What we also see in the description of the example is the term "Mid". This means that this is a filter which implies a "mid average" roll off. If we compare with the others then you can see it in the frequency which is set for the -3dB Roll Off point. But watch out, because this is a general term and it is not really true (but depending on the filter). See for example the "Low" of 705600/768000 where this point is set at 12220 while the roll off at 12KHz is still only 0.1dB. Still, distortion figures are lower for this filter *and* 16KHz surely is more attenuated.
For each of the upsampling rates a High, Mid and Low filter has been attempted but for 352800/384000 no Mid is "available". Available ?
Yes, available for no phase changes. So, all the filters provided are 100% Linear Phase which right away is the reason why you can not do this yourself (set randomly nice settings). Notice that it takes "days" to find the precise settings, which in the end are (and were) tuned per 10Hz of the roll off point. And/but if you look at the "High" for the 176400/192000 filter, you see the notation "Phase slightly off". Now, this filter was allowed but with this remark, because no other Linear Phase filter could be found for the "High" setting and otherwise there wouldn't be one (provided).
On a side note, from this we can also learn that the higher the upsampling rate, the more possibilities exist. For example, for the 705600/768000 upsampling, 14 Linear Phase filters were found and we just provided the best regarding the distortion and attenuation combination.