Another example: I created a 50Hz 50% sinewave with Intune. Checking this with a C-Plugs tuner shows that the frequency is exactly 50Hz.
I tried notching this out with PLP. At quality 1, this was not possible. At higher qualities, it was very difficult to remove the hum, and setting it at 49Hz seem to work better than 50Hz.
With Kjaerhus and Ultrafunk, notching out this 50Hz was very easy. I think the way the gain controls work is a factor in this, but I was very surprised. Considering how little CPU Kjaerhus GEQ-7 uses, it's giving PLP a real run for it's money.
10-Band PLParEQ is Here!
-
- KVRist
- Topic Starter
- 236 posts since 5 Oct, 2005 from Tucson, AZ, USA
Aleksi,... I think you misunderstand what I am saying about phase linear mode.
I am using the traditional IIR filters as described by Robert Bristow-Johnson's handy little cookbook. Simple as that.
However, in phase linear mode the data is sent through the filters in both forward and reverse time order. The time-reversal unwinds the phase shifts imparted by the forward-time processing pass.
These are still IIR filters. But when used in this manner, and by means of the required data windowing, you necessarily truncate the "infinite" from IIR, and hence these become precisely FIR in character.
However, unlike traditional FIR filters, our process has found a way of economizing their computation. For example in order to reach 20 Hz with a conventional FIR at say 48 kHz SR, you need to have a filter with at least 7200 taps (at the very least!). And so a traditional FIR will apply 7,200 multiply-adds to every incoming sample.
Our highest quality uses blocks of 8192 samples, but only has to apply approximately 6 multiply-adds per sample for a huge economy of CPU effort.
Despite that, you see that high quality filtering requires very capable CPU resources.
So yes, we are tradiitonal IIR filters, just used in a non-traditional manner for phase-linear mode, and the math shows that these must be equivalent to FIR filters of a sort....
I am using the traditional IIR filters as described by Robert Bristow-Johnson's handy little cookbook. Simple as that.
However, in phase linear mode the data is sent through the filters in both forward and reverse time order. The time-reversal unwinds the phase shifts imparted by the forward-time processing pass.
These are still IIR filters. But when used in this manner, and by means of the required data windowing, you necessarily truncate the "infinite" from IIR, and hence these become precisely FIR in character.
However, unlike traditional FIR filters, our process has found a way of economizing their computation. For example in order to reach 20 Hz with a conventional FIR at say 48 kHz SR, you need to have a filter with at least 7200 taps (at the very least!). And so a traditional FIR will apply 7,200 multiply-adds to every incoming sample.
Our highest quality uses blocks of 8192 samples, but only has to apply approximately 6 multiply-adds per sample for a huge economy of CPU effort.
Despite that, you see that high quality filtering requires very capable CPU resources.
So yes, we are tradiitonal IIR filters, just used in a non-traditional manner for phase-linear mode, and the math shows that these must be equivalent to FIR filters of a sort....
-
- KVRian
- 1442 posts since 30 May, 2005
Hi David,dbmcclain wrote:Hmm... I see a lot of "Bug reports", like Quality always stays at 1... etc.
so what is your favourized way of reporting a "bug"? I thought that after your latency for answering emails and PMs is quite high that making a "bug report" here on the forum might be useful for you. I'm not sure how I should understand your statement.
All the best, FRitz
P.S. Would be cool if you would clarify this. I would have more bugs to report if you would like me to.
In the end will be the word.
Check out some of my music at www.fritzmetal.de
Check out some of my music at www.fritzmetal.de
-
- KVRAF
- 11385 posts since 3 Feb, 2003 from Finland, Espoo
Yes, it would be good to know how to report bugs as there are some left in the code it seems (most notably the problems at 96khz).
Cheers!
bManic
Cheers!
bManic
"Wisdom is wisdom, regardless of the idiot who said it." -an idiot
"They don't ban hate speech; they ban speech they hate." -an oracle
"They don't ban hate speech; they ban speech they hate." -an oracle
-
- KVRist
- Topic Starter
- 236 posts since 5 Oct, 2005 from Tucson, AZ, USA
As for the apparent frequency shift seen between different sample rate processing...
I think I can believe that is happening. We are using traditional Bilinear transforms from the S-plane analog filter designs to the Z-domain for digital processing.
As a result you should probably expect to see some frequency warp occurring, especially toward higher frequencies.
That's, in large part, the reason we upsample the data if it has SR's below 80 kHz coming in. We want the Nyquist frequency to be as far away from the frequency range of interest (0-20 kHz) as possible. You could do even better if you ran at 192 kHz or 384 kHz internally...
But there is a point of diminishing returns... these super high sample rates buy you decreasing quality improvements.
So, I have to examine the math in the Bilinear transformation, but since the delta between 44.1 and 48, or 88.2 and 96 is about 10%, I would expect to see a percent or so shift at the higher frequencies, when you change SR's.
I haven't looked at the situation in detail for several years. Do you find it to be more than I have suggested here?
I think I can believe that is happening. We are using traditional Bilinear transforms from the S-plane analog filter designs to the Z-domain for digital processing.
As a result you should probably expect to see some frequency warp occurring, especially toward higher frequencies.
That's, in large part, the reason we upsample the data if it has SR's below 80 kHz coming in. We want the Nyquist frequency to be as far away from the frequency range of interest (0-20 kHz) as possible. You could do even better if you ran at 192 kHz or 384 kHz internally...
But there is a point of diminishing returns... these super high sample rates buy you decreasing quality improvements.
So, I have to examine the math in the Bilinear transformation, but since the delta between 44.1 and 48, or 88.2 and 96 is about 10%, I would expect to see a percent or so shift at the higher frequencies, when you change SR's.
I haven't looked at the situation in detail for several years. Do you find it to be more than I have suggested here?
-
- KVRist
- 150 posts since 14 Jan, 2004
K v1.57...like the fact that when selecting filter type it automatically enables that band..no more select filter then click enable.
Check out SELF-MADE HELL! myspace.com/selfmadehellmusic
-
- KVRist
- Topic Starter
- 236 posts since 5 Oct, 2005 from Tucson, AZ, USA
Hmmm bug report are probably best handled by sending an email to
support@refined-audiometics.com
But seeing them here is good for everyone else, and lots of times, other users can point out the error's in the ways...
So it's a mixed bag... when I'm really busy, I don't get to check this web site very frequently, and the e-mail will accumulate in a sack of mail waiting for me to review.
Really serious problems ought to be sent to the support address. But other issues are useful grist for conversation and collective improvements!
support@refined-audiometics.com
But seeing them here is good for everyone else, and lots of times, other users can point out the error's in the ways...
So it's a mixed bag... when I'm really busy, I don't get to check this web site very frequently, and the e-mail will accumulate in a sack of mail waiting for me to review.
Really serious problems ought to be sent to the support address. But other issues are useful grist for conversation and collective improvements!
-
- KVRist
- Topic Starter
- 236 posts since 5 Oct, 2005 from Tucson, AZ, USA
-
- KVRist
- Topic Starter
- 236 posts since 5 Oct, 2005 from Tucson, AZ, USA
Version 1.57 is for both the 10-band and the 3-band unit... Maybe you need to download again?
-
- KVRist
- Topic Starter
- 236 posts since 5 Oct, 2005 from Tucson, AZ, USA
Gain controls in PLP come in two flavors...
First of all the Attenuation/Gain control up near the meters affects everything in a global manner and does not show up on any graphs. It is merely a convenience control mounted here, instead of forcing you to lower the gain elsewhere...
Secondly the Gain entries in the individual filters apply in an unusual manner only in that there is a gain applied to LPF, BPF, HPF, and BRF. Those don't normally receive a gain, but it is convenient to have when composite filtering shoves the audio levels down, and you see the composite curve disappearing below the bottom of the graph. You could as well use the Meter gain control, but it wouldn't show up on the graphs.
Finally, the real purpose of Gain is traditionally only for shelving and peaking/dipping filters. There it works as you would expect.
First of all the Attenuation/Gain control up near the meters affects everything in a global manner and does not show up on any graphs. It is merely a convenience control mounted here, instead of forcing you to lower the gain elsewhere...
Secondly the Gain entries in the individual filters apply in an unusual manner only in that there is a gain applied to LPF, BPF, HPF, and BRF. Those don't normally receive a gain, but it is convenient to have when composite filtering shoves the audio levels down, and you see the composite curve disappearing below the bottom of the graph. You could as well use the Meter gain control, but it wouldn't show up on the graphs.
Finally, the real purpose of Gain is traditionally only for shelving and peaking/dipping filters. There it works as you would expect.
-
- KVRist
- Topic Starter
- 236 posts since 5 Oct, 2005 from Tucson, AZ, USA
The mouse-wheel idea for changing Q is interesting.. We used the control-drag combo because one of our ME's in Switzerland suggested this to us.
I'll have to see what it takes to read the mouse-wheel. This GUI stuff is taking me very far afield from where I normally live... .heh!
I'll have to see what it takes to read the mouse-wheel. This GUI stuff is taking me very far afield from where I normally live... .heh!