My last post got 'updated' with a long discussion and testing of the Oppo BDP-205 filter choices. I had been using the Linear Fast filter, thinking that was the best of the best. But it seemed to induce about 5 clipping events beyond +4dB headroom I allowed for inter sample overs (ISOs), which shouldn't be happening at all (or at least Benchmark seemed to say it was only necessary to allow 3.1dB headroom for inter sample overs). This is not necessarily the flaw of the Oppo or any of it's digital filters, but a deviance from my previous expectation and a seeming objective variance between the different filter choices. (See update below. I have subsequently determined that my counts of clipping events was wrong. It looks like all the fast filters generated the same number of clipping events. They just looked different.)
The other choice that looked best to me at that point (on technical considerations and examining Archimago's measurements) was the Brickwall. I found that only induced 1 clipping event above +4dB, thereby seeming objectively better, plus having the best numerical specifications on noise, distortion, and loss at 20kHz.
I'm NOT going to choose any of the slow filters even if they eliminated such clipping events (in fact, I'd expect they might) because of their leakage effects.
But there's one other filter that might be the best of all, and it's actually Oppo's choice for the default, so it's apparently what they thought to be the best. It's the Minimum Phase Fast filter.
This filter has the curious effect, like all other minimum phase filters, of moving the ultrasonic ringing past any transient, rather than being on both sides (acausal). Intuitively this seems better to most, including me. I was brushing it off last time as "unnatural" but in fact it is the natural thing you could get with real circuits rather than digital simulations, if you could make those circuits well enough (which in practice isn't possible).
We'll I'm long past thinking about ultrasonic ringing in the first place. I'm more interested in low noise, extended response, and those being equal I'd consider the phase response.
The minimum phase fast HAS higher noise and distortion than the brickwall...but also it appears to have more ultimate bandwidth (a big plus) at least according to Archimago's spectrum graphs (which didn't look right, because the brickwall had the lowest loss at 20k, but on the graph it was cutting out well before 20k steeply). So given the possibility that min phase fast has the widest bandwidth, and nearly as low distortion, that could make it the best.
Anyway it occurred to me it might not have these these above 4dB inter sample overs in the first place, and if so it would be an obvious choice.
But the test shown below shows it has the same single clipping event over +4dB as the brickwall. So it's "equal" in that respect, and better than the linear phase which had 5 such events.
But...looking at that actual clipping event...it makes far more sense than with any of the other filter, to my analytical eye. In fact it makes so much sense, I'm inclined not to "repair" it at all, as there's no repair that would preserve the underlying high frequency transient it is apparently trying to show (which previously I insisted had to be electronic...and it might be...but that hardly matters here) without smearing it.
It looks to me best just left very slightly clipped, for it's clear the clipping is right at the upper bound of where it's going to be anyway. There's likely so little difference made in the single clipping event it's not worth lowering the recording level -0.5dB to avoid it (though a test might be warranted). The clipping looks to be hardly making a difference...the peak would only reach a microscopic amount higher anyway, judging from the trailing ringing which is now concentrated on the "after" side making it easier to understand. The other filters give very messy looking results that defy repair altogether, compared to how this looks.
 |
| Min Fast +4db clip (look for the ringing) |
I'm now leaning towards leaving the Oppo at the factory default Min Fast setting.
Update:
My counts of clipping events may have been wrong. I was just checking visually. This can be misleading and depend on how much time is being displayed and also when exactly it starts. I discovered this by cutting the first section of the Min Fast recording and discovering that there were no clipping events at all, instead of just one. The one had disappeared because of the differing start time. Then when I magnified and scrolled, I saw several events. I probably made that exact same mistake with Brickwall. Lets assume for now they all have the same number of clipping events. (That also explains a recollection of another file I didn't report. It was also Brickwall but had several clipping events, just like linear fast.)
So this number of clipping events was an entirely bogus analysis. What still looks good is the concentrated and natural way the Min Fast makes each inter sample over clipping event look. They just look right, whereas all the others just looked awful.
Right now I have no evidence that the factory default filter isn't the best, and one subjective guess that it looks best. (I doubt I could hear the difference...especially in a double blind test.)
Update May 17
I listened to Min Fast filter last night and thought it sounded great. Pure, harmonic, and no digital artifacted sound.
But now it appears that my bit about the look of the ISO clipping events (once again, the fact that there is clipping is not Oppo's fault, it's mine) and it's damned hard to tell even which one looks best. The Linear Phase Fast filter, which I now see is Archimago's preference, does give very short ringing, shorter even than Brickwall, though it has that annoying pre-ringing also. Archimago simply argued on merits, that linear fast is like previous Oppos and most players, and admitted that he's never heard a difference among filters nor was a difference found in earlier testing of similar filters.
The difference in noise among the Brickwall, Min Fast, and Lin Fast filters is negligible. The Brickwall noise level is slightly better with -119.0 instead of -118.8. The THD numbers are identical at a barely measurable 0.0008%. That indicates none of these three filters have significant leakage. The Brickwall apparently cuts off a microscopic amount faster somewhere above 20kHz resulting in the tiniest bit of extra noise reduction but the benefit is so small, one might as well use the more "natural" looking Min Fast, and maybe that slight added bandwidth is a good thing (especially for someone like me, using supertweeters). The Min Fast may have the widest bandwidth as it does have the lowest loss at 20kHz (-0.19 for the Min Fast vs -0.21 for the Linear Fast vs -0.20 for the Brickwall), and since that's the one and only superior spec for that filter, perhaps it's why Oppo chose it as the default.
Actually none of this matters to me anyway because I only use the Oppo to play high resolution discs, with sampling above 44.1kHz, or SACD's. Standard discs I simply rip to my computer and send the CD quality bits into my system and the DAC in the Oppo doesn't matter at all. With high sampling rates on high resolution discs and SACD's, any of these filters is way more than good enough. I could even go with the slow filters and not suffer significant alias leakage.
I think I'm going to follow Gene Dellasala's advice, which I was trending to anyway.
I also don't have any MQA discs (do they exist?) and when I stream the Oppo DAC isn't involved (and I no longer use a streaming service that supports MQA either). Some apparently like the Apodizing filter with MQA. But for other uses, the Apodizing filter is unappetizing to me in general because it adds a tad of high frequency ripple (almost certainly inaudible, but why have it anyway).
No comments:
Post a Comment