Does Phase Distortion/Shift Matter in Audio? (no*)

[gberchin]

Please note that at only 0 and pi phase shift is this not a complex signal. The i is the 'imaginary number' representing the quadraphase signal.

This is why your head hurts so very much trying to do a 90 degree phase shift of a real signal. This is also why 'DC' makes your head ache, because the DC and Pi components of the FFT are STRICTLY REAL.

It might also be worth pointing-out that the "imaginary" part of a complex signal is not a signal that only exists in the "4th-dimension of hyperspace". If you take a "real" cosine wave signal ("real" in both senses: the real part of a complex signal, and also one that you can create with the signal generator on your bench), and phase-shift it by 90° so that it becomes "imaginary", that only means that it is changed from a cosine wave to a sine wave. It is still a signal that you can create with a signal generator. What's the difference between a cosine wave and a sine wave? The cosine wave has value "1" when time t=0. The sine wave has value "0" when time t=0.

What does it mean to phase shift DC by 90°? The equation for a discrete-time cosine wave is x[n]=cos(2*pi*n/N), where N is the period, in samples. At DC, N approaches infinity and thus (2*pi*n/N) approaches zero, so x[n]=cos(0)=1 for all n. Phase-shift that by 90° and x[n]=sin(0)=0 for all n.

Similarly, at half the sampling frequency, N=2 (two samples per cycle). Thus, x[n]=cos(2*pi*n/2)=cos(pi*n), and since n is an integer the value of x[n] alternates between +1 and -1. (You can think of this as sampling the cosine wave at its positive and negative peaks.) Phase-shift that by 90° and x[n]=sin(pi*n)=0 for all n. (You can think of this as sampling the sine wave at its zero-crossings.)

This is why the concept of using an FFT/DFT to create imaginary values at DC and pi makes no sense. There is no way to interpret the results.

 

[j_j]

It might also be worth pointing-out that the "imaginary" part of a complex signal is not a signal that only exists in the "4th-dimension of hyperspace". If you take a "real" cosine wave signal ("real" in both senses: the real part of a complex signal, and also one that you can create with the signal generator on your bench), and phase-shift it by 90° so that it becomes "imaginary", that only means that it is changed from a cosine wave to a sine wave. It is still a signal that you can create with a signal generator. What's the difference between a cosine wave and a sine wave? The cosine wave has value "1" when time t=0. The sine wave has value "0" when time t=0.

What does it mean to phase shift DC by 90°? The equation for a discrete-time cosine wave is x[n]=cos(2*pi*n/N), where N is the period, in samples. At DC, N approaches infinity and thus (2*pi*n/N) approaches zero, so x[n]=cos(0)=1 for all n. Phase-shift that by 90° and x[n]=sin(0)=0 for all n.

Similarly, at half the sampling frequency, N=2 (two samples per cycle). Thus, x[n]=cos(2*pi*n/2)=cos(pi*n), and since n is an integer the value of x[n] alternates between +1 and -1. (You can think of this as sampling the cosine wave at its positive and negative peaks.) Phase-shift that by 90° and x[n]=sin(pi*n)=0 for all n. (You can think of this as sampling the sine wave at its zero-crossings.)

This is why the concept of using an FFT/DFT to create imaginary values at DC and pi makes no sense. There is no way to interpret the results.

Indeed, "no way to interpret the results" because the result of '0' makes the transform singular (information is lost) unless you stick to complex domain.

Which is a whole other ball of wax. The issue at pi creates another issue, in that this creates patterning in half-band filters used for decimation/interpolation if one is not very careful (i.e. longer filters). I wasn't going to try to explain that here. (Same issue as Daubachies "smoothness" theorem.)

 

[gberchin]

Indeed, "no way to interpret the results" because the result of '0' makes the transform singular (information is lost) unless you stick to complex domain.

Yes, exactly. While complex math can accommodate a nonzero imaginary DC value, you could not create a nonzero imaginary DC value with a signal generator, battery, etc. Similarly, complex math can accommodate a nonzero imaginary value at pi (half the sampling frequency), but when you tried to create that sine wave with a signal generator, you'd have no way to know what the amplitude should be -- it is sampled at the zero-crossings so the amplitude is indeterminate. As you said; "information is lost".

EDIT: Earlier in this thread we discussed applying a Discrete Hilbert Transform twice to achieve 180° phase shift, and its equivalence to a polarity swap. It turn out that that equivalence can only hold if the amplitudes at DC and pi are maintained through both transformations, i.e., the first 90° phase shift converts the real parts at DC and pi into equivalent imaginary parts at DC and pi, and the second 90° phase shift converts the imaginary parts at DC and pi into real parts at DC and pi. Strictly speaking, the Discrete Hilbert Transform does not do this; at DC its response is zero, and its response at pi is also zero if there is a frequency bin there.

 

[Tim Link]

I have been playing with FIR filters for fun using REW. With a near field measurement of my tweeter horn I can do an inverse convolution of the resulting impulse and get a nearfield measurement that follows the target curve (flat with 24dB/ octave roll off at 800 Hz) with amazing precision, and the phase response is flat as a pancake all the way down through the crossover slope. At the listening position the phase is still very flat and the response follows the Harman Target very closely. What does it sound like? Nothing I can go "Wow!" about. More like "pretty nice." It sounds very smooth and even, but the frequency response is very smooth and even so no surprise there. Maybe I should try two settings, both getting the frequency response equally smooth, but not flattening out both phase responses.

 

[Tim Link]

One thing I like about using the FIR filters to flatten the phase of each driver is that it makes it easier to make sense of what's happening with the time alignment of the drivers, and eliminates unexpected issues at the crossover zone. Everything sums nicely. My speaker arrangement is unusual as my tweeters are not right next to the woofers, so time alignment at the listening position can get tricky. With the phase all over the place it makes it hard to understand what to do. With both the woofer's and tweeter's phase straightened out it's very easy to interpret the phase response at the listening position. If the tweeter is flat but the woofers are showing rolling phase with a downward tilt, I know the woofers are late and I need to put more delay on the tweeters. This is also easier to see as group delay or on the spectrogram. I can see a well defined kink standing out at the crossover frequency if it's off. As a result I'm getting better frequency response, and I can more easily match drivers. I'm running a two way now instead of a three way because I can now get the tweeters and woofers to directly shake hands very nicely. And it doesn't seem to a be a head in a vice situation either. I'm apparently far enough away from all the drivers that anywhere near the listening position things line up more nicely than usual. It' a matter of being reasonably close to aligned rather than perfectly aligned.

One surprise for me is how well the phase alignment remains even at the listening position. Sure, it's got some peaks and dips here and there because of room reflections, but with 1/6 smoothing it's easy to see that the phase relationships are averaging a basically flat line across the frequency spectrum rather than rolling in different directions in different zones. I don't see or hear any downsides to it, the audio lag time is totally acceptable for movies and games, I don't hear artifacts from the processing, and my Mac Mini can easily do it. Why not?

Does it sound better? I like what it's enabled me to do. I can't say it's because of the phase correction that I like the way it sounds, but I think I'm getting frequency response that's smoother and more natural sounding than I can easily get with just IIR PEQ. The FIR does very fine adjustment of levels, and the phase correction comes along for free and makes everything easier.

 

[Keith_W]

I have been playing with FIR filters for fun using REW. With a near field measurement of my tweeter horn I can do an inverse convolution of the resulting impulse and get a nearfield measurement that follows the target curve (flat with 24dB/ octave roll off at 800 Hz) with amazing precision, and the phase response is flat as a pancake all the way down through the crossover slope. At the listening position the phase is still very flat and the response follows the Harman Target very closely. What does it sound like? Nothing I can go "Wow!" about. More like "pretty nice." It sounds very smooth and even, but the frequency response is very smooth and even so no surprise there. Maybe I should try two settings, both getting the frequency response equally smooth, but not flattening out both phase responses.

I have been doing similar experiments to you. I take a nearfield measurement of all my drivers, and then I flatten the both amplitude and phase. My verification sweeps show that both are flat as a pancake. I then put the mic at MLP and do an overall room correction. The phase mostly remains flat, except at bass < 50Hz. I have also generated a set of filters without the nearfield phase linearization, with amplitude linearization only. Verification measurements of both filters at MLP shows that both comply with the target curve, both have the same step response, and only the phase curve is different. I even did mono measurements (i.e. both speakers playing together) to see if there was any interaction between the speakers in both cases, e.g. a band suckout. Nothing like that was present.

There is a huge difference between the two filters. I simply do not like the phase flattened version. It sounds throttled and smeared. Surprisingly, even the tonality is different. The filter without the phase linearization sounds much more natural, with a more open sound.

This has left me quite confused. I have no idea what a "good" phase curve is supposed to look like, or what I should be aiming for. At the moment my aim is for subjectively more pleasing sound, but aiming for that means countless iterations of DSP until I find the right formula.

 

[levimax]

I have been doing similar experiments to you. I take a nearfield measurement of all my drivers, and then I flatten the both amplitude and phase. My verification sweeps show that both are flat as a pancake. I then put the mic at MLP and do an overall room correction. The phase mostly remains flat, except at bass < 50Hz. I have also generated a set of filters without the nearfield phase linearization, with amplitude linearization only. Verification measurements of both filters at MLP shows that both comply with the target curve, both have the same step response, and only the phase curve is different. I even did mono measurements (i.e. both speakers playing together) to see if there was any interaction between the speakers in both cases, e.g. a band suckout. Nothing like that was present.

There is a huge difference between the two filters. I simply do not like the phase flattened version. It sounds throttled and smeared. Surprisingly, even the tonality is different. The filter without the phase linearization sounds much more natural, with a more open sound.

This has left me quite confused. I have no idea what a "good" phase curve is supposed to look like, or what I should be aiming for. At the moment my aim is for subjectively more pleasing sound, but aiming for that means countless iterations of DSP until I find the right formula.

I have never had great success trying to manipulate phase separate from amplitude. So far my best result have been taking MMM measurements of each driver at the LP and then generating crossover curves in REW (LR 4 in my case) and then overlaying my measuremens and the use use EQ filters to match the measurements to the LR 4 curves. The filters to acoustically match the LR4 curves are significantly different than an electronic LR4 filter. I allow for a little downward slope with frequency due to distance to LP. I then time align the drivers and then do room EQ below schroeder and not worry about phase. I am happy with results YMMV.

 

[KSTR]

Verification measurements of both filters at MLP shows that both comply with the target curve, both have the same step response, and only the phase curve is different.

Hhm, if the step response is the same then the phase (and magnitude) response must be the same, too.

There is a huge difference between the two filters. I simply do not like the phase flattened version. It sounds throttled and smeared. Surprisingly, even the tonality is different. The filter without the phase linearization sounds much more natural, with a more open sound.

This has left me quite confused. I have no idea what a "good" phase curve is supposed to look like, or what I should be aiming for. At the moment my aim is for subjectively more pleasing sound, but aiming for that means countless iterations of DSP until I find the right formula.

"Smeared" sound often indicates over-correction and/or issues with subtle pre-/post-ringing/-echos.

To avoid that, I prefer an approach as outlined in the Grimm LS white-paper. The core is to apply the phase correction a) globally to the input signal and b) derive the phase correction/unwarping kernel analytically. In the end it's a "regular" design with only analog/IIR minimum-phase filters for the drivers (and true time delays when using DSP) which has a "regular" allpass phase response for the sum. All drivers, after filtering, acoustically have the exact same constant phase offset at all frequencies (like 0deg phase offset for a Linkwitz-Riley). This gives the maximum coherence and least possible time-smear. The difference with the phase correction switched on/off is then only very subtle, a slightly different timbre and slightly different speed.

 

[Tim Link]

I have been doing similar experiments to you. I take a nearfield measurement of all my drivers, and then I flatten the both amplitude and phase. My verification sweeps show that both are flat as a pancake. I then put the mic at MLP and do an overall room correction. The phase mostly remains flat, except at bass < 50Hz. I have also generated a set of filters without the nearfield phase linearization, with amplitude linearization only. Verification measurements of both filters at MLP shows that both comply with the target curve, both have the same step response, and only the phase curve is different. I even did mono measurements (i.e. both speakers playing together) to see if there was any interaction between the speakers in both cases, e.g. a band suckout. Nothing like that was present.

There is a huge difference between the two filters. I simply do not like the phase flattened version. It sounds throttled and smeared. Surprisingly, even the tonality is different. The filter without the phase linearization sounds much more natural, with a more open sound.

This has left me quite confused. I have no idea what a "good" phase curve is supposed to look like, or what I should be aiming for. At the moment my aim is for subjectively more pleasing sound, but aiming for that means countless iterations of DSP until I find the right formula.

That's interesting. I've heard things I would describe similarly from various attempts at standard IIR equalization. Two results look very similar but I took different approaches and got a different sound. Throttled and smeared is a good description of what I hear sometimes. My first attempts at FIR correction were almost that bad. I'd describe it as overly soft sounding, but also something was very compelling about it. Next attempt I got it close enough to the non phase corrected version that I had to listen carefully for a little while if I forgot which mode I was in and didn't want to look. I always guessed it right, and always prefer the phase corrected version.

What kind of IR windows are you using when you measure the drivers up close? I'm setting about 9 cycles. There should be no sharp micro jaggies in the response that's being corrected with FIR it seems. It's just way too much to expect. I'm after the basic response of the driver as it's interacting with my horns, not any later interactions with nearby stuff. It seems pretty intuitive to me that those complex interactions are impossible to correct in any meaningful way. Everything that ends up sounding good for me also ends up being a very short little impulse if I look at the wave form, which I always do. If I get a long impulse I know it's not going to sound good. I've also learned to look at the resulting correction curve that I export the impulse from. If it's got weird little oscillations at either end I know it's going to be a dud. I just erase that response and try to make another with different settings until I see it come out perfectly smooth, and everything about it makes sense as a mirror opposite to the difference curve from the target response.

Hhm, if the step response is the same then the phase (and magnitude) response must be the same, too.


"Smeared" sound often indicates over-correction and/or issues with subtle pre-/post-ringing/-echos.

To avoid that, I prefer an approach as outlined in the Grimm LS white-paper. The core is to apply the phase correction a) globally to the input signal and b) derive the phase correction/unwarping kernel analytically. In the end it's a "regular" design with only analog/IIR minimum-phase filters for the drivers (and true time delays when using DSP) which has a "regular" allpass phase response for the sum. All drivers, after filtering, acoustically have the exact same constant phase offset at all frequencies (like 0deg phase offset for a Linkwitz-Riley). This gives the maximum coherence and least possible time-smear. The difference with the phase correction switched on/off is then only very subtle, a slightly different timbre and slightly different speed.

The Grimm paper is interesting, and points out to me that I forgot that some people are trying to correct little dome tweeters and little woofers on baffles with significant edge diffraction. I'm using horns but have come to see Grimm's way of doing things for the most part. The paper gives me a little more to think about. What about the diffraction off the mouth of my horns? How close to them should I be? My rule of thumb has been to get close enough that the little micro jaggies in the response are very close together and small in magnitude, and then close down the IR window length until they disappear, but not to the point of loosing too much detail in the response curve.

With my horns I've gotten my best sounding results with 48 dB/ Octave slopes at 700Hz with the FIR. Looking at Grimm's DSP diagram mine looks the same except that I run a separate FIR for each driver ahead of the IIR EQ, IIR crossover filters, and time delays. I guess that means I'm doing the steep slopes partially in IIR, and partially with the natural roll off of the drivers, so the FIR isn't really creating it, just fine tuning it. I run a room correction IIR EQ in front of everything. It's pretty minimal with just two small PEQ adjustments and doesn't seem to change the phase much.

 

[Keith_W]

Hhm, if the step response is the same then the phase (and magnitude) response must be the same, too.

Sorry. Perhaps I should have said "similar". If you take a look at my system thread, especially from this post onwards, I show some graphs and I am tearing my hair out trying to figure out why they sound different.

"Smeared" sound often indicates over-correction and/or issues with subtle pre-/post-ringing/-echos.

Hmm, that is an excellent insight. Pre-ringing is easy to see, I am not so sure about post-ringing because it's hidden in the impulse. How do you look for it?

To avoid that, I prefer an approach as outlined in the Grimm LS white-paper. The core is to apply the phase correction a) globally to the input signal and b) derive the phase correction/unwarping kernel analytically. In the end it's a "regular" design with only analog/IIR minimum-phase filters for the drivers (and true time delays when using DSP) which has a "regular" allpass phase response for the sum. All drivers, after filtering, acoustically have the exact same constant phase offset at all frequencies (like 0deg phase offset for a Linkwitz-Riley). This gives the maximum coherence and least possible time-smear. The difference with the phase correction switched on/off is then only very subtle, a slightly different timbre and slightly different speed.

Thanks. I am aware of that paper, and I did apply a global reverse all pass. As you say, it's quite subtle and I am not sure which I prefer. I have more listening to do.

 

[ernestcarl]

... I prefer an approach as outlined in the Grimm LS white-paper. The core is to apply the phase correction a) globally to the input signal and b) derive the phase correction/unwarping kernel analytically. In the end it's a "regular" design with only analog/IIR minimum-phase filters for the drivers (and true time delays when using DSP) which has a "regular" allpass phase response for the sum. All drivers, after filtering, acoustically have the exact same constant phase offset at all frequencies (like 0deg phase offset for a Linkwitz-Riley). This gives the maximum coherence and least possible time-smear. The difference with the phase correction switched on/off is then only very subtle, a slightly different timbre and slightly different speed.

That's not always so easy or optimal to do in a multichannel system where one is forced to pair very different speaker designs (e.g. LF extension and phase profiles) all to the same .1 LFE and bass managed sub(s).

 

[ernestcarl]

Result between two different xo EQ designs I've cooked up for my own 7.1c desk setup:

DESK 7.1 setup - exclusive min phase vs mixed phase xo EQ.mdat

drive.google.com

Raw measurements taken at the MLP:

DESK 7.1 setup - Unequalized Measured Response at MLP.mdat

drive.google.com

Which one do you think looks better?

From my own informal subjective evaluation the "mixed phase" (linearized) version sounds better.

What modifications -- if any -- would you suggest for the EQ created for this single listening position?

 

[gnarly]

One thing I like about using the FIR filters to flatten the phase of each driver is that it makes it easier to make sense of what's happening with the time alignment of the drivers, and eliminates unexpected issues at the crossover zone. Everything sums nicely.

Yes. I too like that....so much.

When each driver section has flat phase throughout its range, including the critical summation regions with neighboring driver sections, time alignment gets very easy.

What I've found from hundreds, thousands, of xovers put together for 4-5way multi's, is that when driver sections have plat phase,
the time alignment...the fixed delays needed between sections...ends up being the physical distance between acoustic centers.

If it's not very, very close, I know I've screwed the pooch somewhere.

 

[gnarly]

"Smeared" sound often indicates over-correction and/or issues with subtle pre-/post-ringing/-echos.

That's my take too....over-corrections.

To avoid that, I prefer an approach as outlined in the Grimm LS white-paper. The core is to apply the phase correction a) globally to the input signal and b) derive the phase correction/unwarping kernel analytically.

I think what the Grimm paper gets right, is that they reduced the degree of correction.
But otherwise I see it as much more of a marketing piece that a paper.

The train wreck example was the overly corrected part imo...solved by ditching FIR overcorrection for more gentle bi-quad IIR correction.
Well, the exact same IIR bi-quad corrections, with the same degree of gentleness, could have been used their FIR filter(s) instead of their train wreck example.

As for using global FIR to correct the phase rotation of the prefered IIR LR4...ok fine, provided all it is in an inverse all pass that does nothing but correct the IIR LR4.
If global correction does more, it's overstepped itself ime/imo.

But I say why not just use a linear phase LR4 to begin with? Only reason I can see is is takes two channels of FIR instead of one for global.

So, on the whole, I like the Grimm "paper" for casting light on overcorrections in general,........ but the implementations touted I put into the "marketing" bucket.

 

[KSTR]

But I say why not just use a linear phase LR4 to begin with? Only reason I can see is is takes two channels of FIR instead of one for global.

Technically it is of course equivalent as the transfer function multiplications follow the associative law.

The point is that the phase unwrapping is best based on a clean analytically computed phase response (the time-inverse of a simplified model of system allpass function's impulse response), not coming from a measurement+inversion process. This gives an approximate but benign phase compensation without any noise/artefact issues as we are only correcting the trend line but keep all the remaining small natural warts and wiggles.

Key point is close matching of the model to the real response (with well-executed min-phase filter versions to arrive at the acoustical min-phase targets for each driver). Best derived by curve-fitting using an optimizer. This should be based on quasi-free field measurements, later adding any actual "room correction" in a second step. From there, you can (conceptually) sample each final IR of the correction chain of each driver, yielding individual final IR kernels for a single direct convolution per driver and speaker.

That's at least what I personally found as consistanty good and "natural" sounding and being reliable/repeatable... back when I experimented with this, years ago. Today I'm mostly on headphones and use a very reduced speaker setup, so YMMV of course

 

[gnarly]

Technically it is of course equivalent as the transfer function multiplications follow the associative law.

The point is that the phase unwrapping is best based on a clean analytically computed phase response (the time-inverse of a simplified model of system allpass function's impulse response), not coming from a measurement+inversion process. This gives an approximate but benign phase compensation without any noise/artefact issues as we are only correcting the trend line but keep all the remaining small natural warts and wiggles.

Key point is close matching of the model to the real response (with well-executed min-phase filter versions to arrive at the acoustical min-phase targets for each driver). Best derived by curve-fitting using an optimizer. This should be based on quasi-free field measurements, later adding any actual "room correction" in a second step. From there, you can (conceptually) sample each final IR of the correction chain of each driver, yielding individual final IR kernels for a single direct convolution per driver and speaker.

That's at least what I personally found as consistanty good and "natural" sounding and being reliable/repeatable... back when I experimented with this, years ago. Today I'm mostly on headphones and use a very reduced speaker setup, so YMMV of course

All you say makes sense to me, and essentially reflects the process I use.

I start with each driver, applying both in-band and out-of-band minimum phase magnitude flattening, which as we know flattens phase too.
This is the only phase flattening I feel is fully valid.
I learned from POS of rephase, that if out-of-band min phase flattening is extended far enough, adding complementary linear phase xovers would create complementary acoustic speaker response with no phase rotations, even throughout the drivers summations ranges.

Since all the min-phase corrections were manual, all one had to do the assure overcorrections were not made, was simply not use any high Q EQ's.
And I've yet to hear or measure pre-ring, even off-axis, when the only linear-phase implementation is complementary xovers.

To reduce the amount on manual work with rePhase, I tried using REW to generate min-phase corrections for import into rePhase.
Also tried using REW's trace arithmetic to achieve impulse inversion.
But I've never felt either of those equaled the old manual process.

So I've come to I agree with you that measurement + an inversion process has potential perils.

That said, I've found a "manual and choose how automatic", measurement based FIR generator, that I think gets the job done and avoids over corrections.
FirDesigner matches the process I use with rePhase, by letting me select how closely or loosely to match min-phase driver corrections, to whatever linear-phase xover target I've chosen. The "close vs loose matching" shows the windowing error between ideal and realized filter, in real time as adjustments are made.

I honestly don't even think about over-corrections anymore.
Or pre-ring, for that matter .....
my 2c...and as you say, ymmv

 

[levimax]

All you say makes sense to me, and essentially reflects the process I use.

I start with each driver, applying both in-band and out-of-band minimum phase magnitude flattening, which as we know flattens phase too.
This is the only phase flattening I feel is fully valid.
I learned from POS of rephase, that if out-of-band min phase flattening is extended far enough, adding complementary linear phase xovers would create complementary acoustic speaker response with no phase rotations, even throughout the drivers summations ranges.

Since all the min-phase corrections were manual, all one had to do the assure overcorrections were not made, was simply not use any high Q EQ's.
And I've yet to hear or measure pre-ring, even off-axis, when the only linear-phase implementation is complementary xovers.

To reduce the amount on manual work with rePhase, I tried using REW to generate min-phase corrections for import into rePhase.
Also tried using REW's trace arithmetic to achieve impulse inversion.
But I've never felt either of those equaled the old manual process.

So I've come to I agree with you that measurement + an inversion process has potential perils.

That said, I've found a "manual and choose how automatic", measurement based FIR generator, that I think gets the job done and avoids over corrections.
FirDesigner matches the process I use with rePhase, by letting me select how closely or loosely to match min-phase driver corrections, to whatever linear-phase xover target I've chosen. The "close vs loose matching" shows the windowing error between ideal and realized filter, in real time as adjustments are made.

I honestly don't even think about over-corrections anymore.
Or pre-ring, for that matter .....
my 2c...and as you say, ymmv

Makes sense but how do you deal with Baffle Step compensation when "flattening" each driver since baffle width issues won't show up in near field measurements?

 

[gnarly]

Makes sense but how do you deal with Baffle Step compensation when "flattening" each driver since baffle width issues won't show up in near field measurements?

I use acoustic far-field measurements for the tuning process. ...... As quasi-anechoic as I can reasonably attempt.
Got an outdoor mic rig off a deck. The vertical mic array, working with a spinorama, lets me assess polars pretty easy..
......or I should say gives me more data than my mellon can take in sometimes

And baffle step, or baffle ramp as I like to think about it, is less of an issue with the type DIYs I build....largish unity/synergy horns like these below.
Where the horn loses its horizontal and vertical pattern control, is their analogous issue.

 

[levimax]

I use acoustic far-field measurements for the tuning process. ...... As quasi-anechoic as I can reasonably attempt.
Got an outdoor mic rig off a deck. The vertical mic array, working with a spinorama, lets me assess polars pretty easy..
......or I should say gives me more data than my mellon can take in sometimes

View attachment 356361


And baffle step, or baffle ramp as I like to think about it, is less of an issue with the type DIYs I build....largish unity/synergy horns like these below.
Where the horn loses its horizontal and vertical pattern control, is their analogous issue.

View attachment 356363

Shoot I don't have a set up like that.... looks like fun! Do you have any neighbors that wonder what all the strange sounds are about?

 

[gnarly]

Shoot I don't have a set up like that.... looks like fun! Do you have any neighbors that wonder what all the strange sounds are about?

Only when I run REW sweeps to check out harmonic or modulation distortion. A 130dB log sweep can sound like some kind dang warning alarm, or something... lol

But transfer functions, which comprise at least 95% of measurement time, are no problem at all.
Smaart allows the use of music as the stimulus signal, so folks often don't even know I'm testing.
And even when I use straight pink noise, nobody pays it any attention. Its sound that's kin to rushing water is very gentle compared to most lake noise. (boats, PWC, etc)
(Heck, the worst noise there is imo, is dang leaf blowers and such).

So other than occasional air-raid swoops from REW, everybody just shakes their heads, and say crazy ole Mark is at his speaker building again.