Hi Andrew,
The ripples you are showing occur at -144dB. I don't think it should be considered as a problem as is.
As the HP and LP will likely have different length and window, iterative optimization will give different result, so a subtractive solution would probably not give the best result for a given impulse length...
As for the ripples you showed me starting post #9 (low level ripples at the Nyquist frequency), I have found a way to reduce them using oversampling (and double the number of taps) and subsequent downsampling (with amplitude derivation being corrected during the fist iterative optimization step).
Not sure I will implement it in the next release though, as I would like to focus on measurement import first.
Why not upsample to the nominal sampling freq before outputing to the DAC?
(interpolation might require an FFT though
)I haven't been following this thread. But what exactly do you mean by clever marketing? the wavio device has two sets of i2s outputs, 1 set is isolated by NVE IL715 isolator chip and are on pin headers (you provide 'dac side' power for the i2s output), the other set of i2s outputs is non-isolated and on u.fl connectors.
What exactly is 'marketing' about that? It's fact as far as I can tell?
SHARC 21469 EZ-BOARD FOR THE ADSP-2146X SHARC FAMILY. It has the most powerful Sharc processor, ADSP-21469 (450 MHz, 5 mbit on chip sram) as well as a multichannel AD1939 DAC/ADC with balanced connectors (no i2S fiddling needed). There is also an SPDIF input. I believe this board is good enough to verify if it is possible to build a stereo 4 way filter using FIR convolution on the Sharc.
You may think it isn't problem, a simple one to fix. Calculation of high pass should be done buy calculating low pass and subtracting normalized value of one from peak sample. Additionally, linear phase version filters for IIR family is faster and more accurate to calculate IIR with 1/2 gain/slope, convolve with time reversed copy, trimming to desired tap count, and applying symmetrical fade in and fade out at ends.
Fare and away at moment is indeed import of user IR for response modeling, followed by solver for optimizing. Even REW approach is tedious compared to good technique with direct inversion.
Making the downsampled stop band wide enough to allow a slow rolloff antialiasing filter limits the downsampling ratio. Between that, the cost of the antialiasing, and the minor impulse blurring from the pre-DAC antialiasing round one likely ends up with a more complex implementation and little or no computational savings for three ways and a good fraction of four ways.
Downsample+upsample might let you fit a tweeter+mid+high sub+low sub four way into a DSP that's more of a three way size, though. That's the kind of thing which wants a case study. It's not something I've looked into as three ways are more than capable of delivering SPLs I find uncomfortably loud.
Subtracting is not the way it is done in rePhase, so 100% complementarity cannot be guaranteed. But you know what you get when looking at the ripples, and frankly a -144dB complementarity error is not something that would worry me. How is the complementarity of the acoustic filters?...
You can also use more taps for the given frequency and these complementarity problems will go even deeper in level as the result curve will fit the theoretical one better (and theoretical filters generated in rephase are 100% LP/HP complementary).
The IIR=>FIR generation method using cooledit is good, but only work for textbook filters whereas here any function (for both magnitude and phase) can be generated (Linkwitz-Riley, Horbach-Keele, reject high/low, overlapping, etc...).
There are room for many tools and methods for FIR filtering (automated inversion, integrated tools, etc.), and rePhase is just one of them, with a specific approach (manual correction and wysiwyg result).
Hello there and congrats for the ideea,
But......many of are not able to figure out how exactrly to work with the app...sooo..
it there any tutorial for using this program with...foobar for example?
Thank you, and good luck with your project!
Hi rvrazvan,
Thanks for the kind words
It depends on what you want to do with rePhase: correct the phase of an existing loudspeaker? Correct both phase and magnitude? Build the whole crossover with rePhase?...
There are some tutorial online, for example Thierry's here, with a melting pot of all the possible approaches, with a focus on PC VST convolution:
http://www.diyaudio.com/forums/pc-based/223805-easy-fir-crossover-pc-based-drc.html
Here is Another one in French by Jimmy Thomas, explaining how to build an entier crossover with rePhase and Jriver:
http://jimmy.thomas.free.fr/Jriver/Tuto-Jriver-RePhase-HolmImpulse.pdf
If you read French you will find additional information along the way on this topic:
rePhase: linéarisation de phase, EQ et filtrage FIR - Enceintes
Explain you situation and I will see how I can help (and this might be the first step to another tutorial).
HI Rvrazvan, i made a short description of my simple way to use RePhase as phi équalizer in Foobar (with foo-convolve.)
in French, but may help i hope.
crd
Egalisation fir avec RePhase - Le blog de jimbee
in French, but may help i hope.
crd
Egalisation fir avec RePhase - Le blog de jimbee
Hi jmbee
Yes, your tutorial is more in line with what Rvrazvan seems to be looking for!
Especially the second part, "Test avec filtre iir standard L-R. trois voies", as the first part is dealing with a system that is already quite linear in phase (using a modified DCX2496 for substractive filtering)
By the way, when using the foobar convolver plugin, it is best to uncheck the "Auto level adjust" checkbox.
Yes, your tutorial is more in line with what Rvrazvan seems to be looking for!
Especially the second part, "Test avec filtre iir standard L-R. trois voies", as the first part is dealing with a system that is already quite linear in phase
(using a modified DCX2496 for substractive filtering)By the way, when using the foobar convolver plugin, it is best to uncheck the "Auto level adjust" checkbox.
I don't think bypassing the convolver is a good idea. I generated an impulse that makes no correction and use it as the alternate to the 'fixed' impulse. So, the convolver runs for both the cases.
The big thing is not knowing what is being played when you do the listening.
How about this:
1. Play a song in Foobar with the corrected impulse and record the output of the soundcard.
2. Play the same song with uncorrected impulse with convolver running perfect impulse and record output of soundcard.
3. Play both recorded files through speakers using the ABX plugin in Foobar.
Will this capture the change correctly?
The big thing is not knowing what is being played when you do the listening.
How about this:
1. Play a song in Foobar with the corrected impulse and record the output of the soundcard.
2. Play the same song with uncorrected impulse with convolver running perfect impulse and record output of soundcard.
3. Play both recorded files through speakers using the ABX plugin in Foobar.
Will this capture the change correctly?
this is the simplest way to adjust level.
time to switch "flat impulse" to "corrected impulse" do not allow an accurate setting.
auditive memory is not required,when bypassing on the fly.
yes,you can convolve "offline" and record.
why don't use multiple instance of foobar instead ? play/pause for each instance.
Hi jmbee
Yes, your tutorial is more in line with what Rvrazvan seems to be looking for!
Especially the second part, "Test avec filtre iir standard L-R. trois voies", as the first part is dealing with a system that is already quite linear in phase
By the way, when using the foobar convolver plugin, it is best to uncheck the "Auto level adjust" checkbox.
Hi Pos,
Thank you for the advice, i have corrected.
Even with the DCX in subtractive-delay, Rephase alows to improve low frequency response, less boominess, more focused.
crd
Offline convolution looks like a good solution indeed.
SoX should be able to do that using the txt impulse format (never tried though).
What are you loudspeakers and correction settings?
What improvements are you hearing (halucinating?
It's a simple two-way with vented Deltalite2515 and SEOS12. I have large tractrix horns, but they are down for repair at the moment.
I corrected the vented response and the 6th order crossover at about 1200 Hz.
On one particular song, there is percussion with flute. With corrected phase, I seem to latch on to the flute, and everything makes sense. Without corrected phase, I latch on to the bass and the music doesn't make as much sense.
Instruments have more power in space, i.e., better defined, easily localized and I can follow them more easily. In general, it feels more relaxed as I can follow things more easily.