rePhase, loudspeaker phase linearization, EQ and FIR filtering tool

[dlr]

Originally Posted by pos

Another test with the same EQ at different frequencies, with a 512 taps impulse:
20Hz, 40Hz, 80Hz, 160Hz, 320Hz, 640Hz

I have a question for John on this. I find it curious that as one tries to lower the cuttoff, at some point the absolute output level at the low end drops.

I wonder if this effect is part of the issue I've had with my dipole system. Do you think that trying to EQ the low end will have some hard and fast limit as to cutoff? What I see in those graphs makes me think so, at least for any specific sample rate. Looks like in my case I'm better off using a 44.1 sample rate since I need woofer dipole correction.

dlr

p.s. Sorry for the odd look. I wanted to include the image, but a QUOTE of the post dropped it.

[john k...]

Hi dave,

Yes, it could be the problem. Once you get low in frequency the accuracy limits can take things in weird directions. The best thing to do is to measure the UE generated filter and see what it really look like compared to what is shown in the UE screen. Then you can manipulate the target (like exaggerate the dipole boost) and remeasure the actual filter. You should be able to get something that ultimately looks like what you really want but in the UE screen it may look very different.

For example you are trying to generate a Q boost with 8dB gain and Q = 1 but it comes out looking like a Q = 0.5 with gain = 5. So try Q = 2 with gain of 11dB and see what comes out. It's manually trying to correct for the inaccuracies.

[pos]

Originally Posted by john k...

Your manual approach is pretty much the way the UE works deep down inside. You start with a measurement. That measurement can be a lot of things: the on axis response, the smoothed axial response, a spatially averaged response,..... Then, the user specifies a frequency range over which minimum phase equalization is applied to flatten the response.

I have never used UE but it looks like a very nice tool. In comparison rephase is not an integrated framework, but just a small tool to be used together with other tools (measurement software, convolution engine, etc.), that the user has to can inside its design chain.

Still, I am more confortable with manual EQ when it comes to flattening a driver.
Automated corrections can be very good (if the measurement is good and representative, that is), but this requires a lot of care.
DRC-FIR does that very well (inverted response special tricks), as well as PORC (multiple EQ with controlled Q), and I am sure UE does that quite well also.
Still... I prefer manual EQ

[pos]

Originally Posted by john k...

For example you are trying to generate a Q boost with 8dB gain and Q = 1 but it comes out looking like a Q = 0.5 with gain = 5. So try Q = 2 with gain of 11dB and see what comes out. It's manually trying to correct for the inaccuracies.

That it what rePhase does during its automatic optimization steps: it compares the target curve (in blue) and the result curve (in red, which is a fft of the generated impulse) and internally modifies the target to correct the result.

Here is an illustration of the effect of the iterative optimization for a filter and an EQ with a window and a short number of taps:

[kgrlee]

pos, I haven't tried rePhase but I hope when you look at the effect of your short FIRs, you use a MUCH larger FFT block.

ie if looking at a 512 pt FIR, you are using an FFT block of 16K or more.

Small FFT blocks hide the info between the bins. Good windowing alleviates the nasty effects but info is still hidden or fudged. Bigger FFT blocks move the problem to frequencies where it is less important or at least let you see there is a problem.

A 512 pt FIR at 48kHz has a 'resolution' of 93.75Hz so any effect it has at 20Hz is purely due to 'windowing' artifacts.

[john k...]

Originally Posted by pos

That it what rePhase does during its automatic optimization steps: it compares the target curve (in blue) and the result curve (in red, which is a fft of the generated impulse) and internally modifies the target to correct the result.

Here is an illustration of the effect of the iterative optimization for a filter and an EQ with a window and a short number of taps:

Nice. But I don't think that is exactly what dlr and I are referring to. In theory the impulse response generated by the UE should exactly produce the desired filter. But the impulse is only 8192 samples long or 5.86 Hz resolution. When actually played through the convolution engine they lose accuracy because of the required windowing which has the greatest effect on the low frequency information (windowing is akin to smoothing in the frequency domain). To improve this a longer impulse is needed (not just a longer FFT block) so that a longer window can be used and more low frequency info is retained. I don't think you can conclude what you have without making the measurment of the output of the convolution engine.

[thierry38efd]

a capture of 512 taps 100 Hz +6dB eq.

sound card output after convolution,rectangular window and blackman in green.



and with the same eq with 1024 taps

[john k...]

Originally Posted by thierry38efd

a capture of 512 taps 100 Hz +6dB eq.

sound card output after convolution,rectangular window and blackman in green.

Exactly as expected.

[CopperTop]

Hi. A question regarding phase correction. I enclose a snapshot of a nearfield measurement of a tweeter's response taken using REW. I have used the Estimate IR Delay option to, in theory, remove any delay from the IR, yet the phase still shows several complete 360 degree 'wraps'. In your opinion, is there still a delay that has not yet been removed from the IR?

[thierry38efd]

after estimated delay,adjust sample by sample to get the correct minimum phase of requency response (i think tweeter has no electrical filter or wave guide.)

the tab on the right side below