KVR Audio

dbmcclain · Post by **dbmcclain** » Mon Oct 31, 2005 1:11 am

I just did several phase inversion tests here to see just how clean this thing really is... In bypass mode - clean as a whistle. No artifacts, except for about 6.5 ms of startup transient... not sure why that is.. I'll have to hunt.

But in peaking Type-1 phase-linear mode with 0 dB gain, and Q=20, quality level 5, feeding it pink noise at 48 kHz SR, (the most difficult test I can imagine)...

I see residual noise floor showing through that measures about -178 dBFS/Root(Hz)) everywhere except very near Nyquist, where it rises dramatically to about -95 dBFS/Root(Hz).

More than likely, this is the dither used in converting back to single-precision floating point (VST standard audio 24-bit mantissa) from the internal 56-bit mantissa used by the DSP core.

It could also be an artifact of the FFT analyzer being used here. Hann windowing with 4096 cells, but the noise figures shown above already add 36 dB for the 4096 cells, and then subtracts 44 dB for the /Root(Hz) part.

--------------------
On another topic, someone asked if I could present a suggestion about what Quality level to use based on the dialed filter settings. That would be really neat! But it is a difficult calculation to make on the fly.

In general, using T60 as a criterion, Quality level 1 is good down to a bandwidth that has a period of 85 samples (frequency in Hz depends on your sample rate, eh?). Quality level 5 is good down to 1365 samples. These were estimated using the blocksize (512 at QL 1 and 8192 at QL 5) divided by 6.

At 44.1 kHz SR, those 85 samples represent about 2 ms, or 500 Hz. So attempting to use QL 1 on anything with a bandwidth narrower than this will give degraded performance. QL 5 will be good down to about 31.25 Hz, using the same reasoning.

So the question then becomes one of "what is my effective bandwidth?". The only time you need to worry about this is when using very narrow BPF, or extremely high-Q LPF and HPF filters. But even then, with a maximum Q of around 20, you would have to be operating below 10 kHz for QL 1 in the extreme, or 600 Hz at QL 5.

In general, most EQ is done with much gentler filtering, like moderate shelves and peak/dip filters. In those cases you almost never have to concern yourself about the Quality level.. Except...

The Quality level also relates to blocksize and hence, how correlated the signal is at the front and back of the data blocks. This has to do with time-aliasing artifacts -- the blending of information from now with a time later on in the stream. (Not the same as frequency aliasing, here we are talking about aliasing in the time domain..)

Using the longest possible blocks gives you a better chance of having smaller correlations between now and later....

Simple, eh?!

bmanic · Post by **bmanic** » Mon Oct 31, 2005 1:16 am

Well I sure didn't understand a single thing of what you just posted (well, actually I did) but what counts is that your eq sounds sweeeeeeeet!

1.45 seems to work well here on this old laptop (yes, the missus threw me into the kitchen, again).

Cheers!
bManic

fritzman · Post by **fritzman** » Mon Oct 31, 2005 1:16 am

Ah yes,

so I made a fool out of myself.

Dunno what I did when I listened to it the first time and when I measured it. It sounded different and I had a difference of -41dB between the two phase cancelled signals. Hmmmm.... Maybe I should just go to bed and sleep the whole nite instead of hanging out in front of the computer.

Good nite, FRitz

Edit: I just measured again and there's no difference between "peaking" in 1.40 and "Pk2" in 1.44 !!! Can this be true? When I phase cancel these 2 signal there's no nothing visible in SPAN except some hi noise from the EQ @ -132.2dB. Is this correct? What's wrong here?

bmanic · Post by **bmanic** » Mon Oct 31, 2005 1:28 am

Fritz, while boosting or cutting? I thought pk2 is only different when cutting? Read that oxford .pdf for info.

I've yet to compare it to 1.40 (as I don't have that one on this laptop, bummer).

Cheers!
bManic

fritzman · Post by **fritzman** » Mon Oct 31, 2005 1:44 am

bmanic wrote:Fritz, while boosting or cutting? I thought pk2 is only different when cutting? Read that oxford .pdf for info.

I've yet to compare it to 1.40 (as I don't have that one on this laptop, bummer).

Cheers!
bManic

Thanx bmanic,

that's right. Boosting is the same and cutting is different. I REALLY need some sleep.

Best wishes, FRitz

dbmcclain · Post by **dbmcclain** » Mon Oct 31, 2005 2:15 pm

I tracked down the source of the startup transient in the phase cancellation cleanliness test...

I implemented a graceful ramp-down / ramp-up in the attenuation so that when changing Quality levels you would not hear a pop in the audio levels.

That ramp-up also happens when the filter first begins its job, and so during that ramp-up the difference between the phase reversed channels is audible.

So I have a question for you pro's... It seems to me that the job of a filter is to do a good job filtering, and not to protect the user from abusive use of it. That has a cost as we see... Shouldn't the precaution be taken that one stops the audioi stream to switch Quality levels, or else accepts the risk of an abrupt pop? That way the filter can do precisely what it is expected to do.

dbmcclain · Post by **dbmcclain** » Mon Oct 31, 2005 2:17 pm

I'm thinking, in particular, about the automated switch in/out of the filter during a mixdown.

You don't want this soft ramp up when you hit the enable, especially if you are trying to hit a particular spot in the edit...

bmanic · Post by **bmanic** » Mon Oct 31, 2005 2:21 pm

Well, I rather have a few ms of ramp up time to avoid clicks when switching filters. One can always edit the automation point to hit a bit early. But then again, there would be total silence in between, right? Hmm.. tough question. Some eq plugins do a marvellous job at switching filters (sonalksis comes to mind) without clicks. Maybe they do a quick crossfade of some sort? I rarely automate changes of filter types and especially in mastering I think this is not really needed (and that's where I would use PLParEQ).

Cheers!
bManic

fritzman · Post by **fritzman** » Mon Oct 31, 2005 4:09 pm

dbmcclain wrote:I tracked down the source of the startup transient in the phase cancellation cleanliness test...

I implemented a graceful ramp-down / ramp-up in the attenuation so that when changing Quality levels you would not hear a pop in the audio levels.

That ramp-up also happens when the filter first begins its job, and so during that ramp-up the difference between the phase reversed channels is audible.

So I have a question for you pro's... It seems to me that the job of a filter is to do a good job filtering, and not to protect the user from abusive use of it. That has a cost as we see... Shouldn't the precaution be taken that one stops the audioi stream to switch Quality levels, or else accepts the risk of an abrupt pop? That way the filter can do precisely what it is expected to do.

Hi David,

I normally don't automate the filter type. If I really needed it I would glue two (or more) rendered versions to get the change. So I basically think it's OK to avoid the clicks. Clicks are more disturbing (and more often experienced in normal use) then an additional edit.

Best wishes, FRitz

dbmcclain · Post by **dbmcclain** » Mon Oct 31, 2005 6:58 pm

Hi Guys, it isn't the filter-type changing that causes the ramping, it is changes to the Quality level. Something that should occur much less frequently in use, and probably never automated. Once the filter is running, it works fine when switching among the filter types. Only about a 200 ms startup ramp to get the transients below -120 dBFS.

bmanic · Post by **bmanic** » Mon Oct 31, 2005 8:57 pm

Oh the quality setting. Well, that is no problem at all.

Cheers!
bManic

DavidGig · Post by **DavidGig** » Mon Oct 31, 2005 9:22 pm

Why am I noticing much more latency with version 1.46 compared to the last one I tried (the DLL dated 10/25/05)?

This is even in Phase Warped mode and lowest Quality. The PDF says this is still supposed to be just 512 samples, but maybe this changed?

bmanic · Post by **bmanic** » Mon Oct 31, 2005 9:28 pm

Umm, what version was the last one? Around 1.30 somewhere it changed so that the plugin has a constant latency of 8192 samples or so, no matter what quality setting. This was due to some hosts crashing.

- bManic

DavidGig · Post by **DavidGig** » Mon Oct 31, 2005 9:34 pm

That must be it. Thanks.

fritzman · Post by **fritzman** » Mon Oct 31, 2005 10:47 pm

bmanic wrote:Oh the quality setting. Well, that is no problem at all.

Cheers!
bManic

Right!

PLParEQ1 and PLParEQ4 1.40 (Are we there yet?)