I was looking at your sweeps in REW, and noticed they were over 95dB. That may not be an issue, but you can sweep at a much lower SPL, if in a quiet environment.
I did notice a peak in 2nd order distortion on the mid around 1.5k. It shows both on and off axis. The tweeter measured considerably better at 1.5k. Sometimes a loose mounting screw can cause such a thing. I think an overly tight screw can as well. My driver baskets are thick aluminum. I usually tighten the screws with a minimal amount of torque. The ideal tightness may vary with several factors.
I did notice a peak in 2nd order distortion on the mid around 1.5k. It shows both on and off axis. The tweeter measured considerably better at 1.5k. Sometimes a loose mounting screw can cause such a thing. I think an overly tight screw can as well. My driver baskets are thick aluminum. I usually tighten the screws with a minimal amount of torque. The ideal tightness may vary with several factors.
I was indeed aiming to be near 95dB as I thought this is the common level for checking distortion. Also, I was wary of noise (wind, birds, passing cars and planes), so avoided lower levels, rightly or wrongly. I was getting about 45-51dB headroom readings during REW sweeps.
I was cautious when bolting on the drivers. I am sure they are well attached but not overtightened.
Give it a shot.
It depends on how you setup the optimizer. The way I showed, the output is a blend of on and off axis (in-room), so the optimizer will move the crossover frequencies a little bit, but not much. When I've run it then put the frequencies back it doesn't do much with your design.
If you heavily weight on-axis (or only optimize for it, maybe since you only have an on axis measurement) then it may move the x-over frequencies around and end up screwing up the off axis response.
It depends on how you setup the optimizer. The way I showed, the output is a blend of on and off axis (in-room), so the optimizer will move the crossover frequencies a little bit, but not much. When I've run it then put the frequencies back it doesn't do much with your design.
If you heavily weight on-axis (or only optimize for it, maybe since you only have an on axis measurement) then it may move the x-over frequencies around and end up screwing up the off axis response.
So what's the thinking on using EQ to smooth out narrow peaks and dips in the mid-range? Is the processing sufficient to allow very much? Does it negatively affect sound quality if over done? The reason I ask, is because I was looking at the polars for my sim, and I see a spot or two where I think adding 3dB or more at a Q of around 3 would improve both on, and off axis. Does the optimizer do any of this, if allowed in its setup? Would this just be overkill? I kind of think it would be, but it is interesting.
For example, look at 1.6k and 2k. These two dips are consistent as you move off axis. I assume that if you EQ them on axis, they will also smooth out off axis. These are coming from the mid. I know they are very minor. If the tweeter was crossed below 2k, they would likely be filled by the tweeters response. This is mostly a hypothetical question. I do not usually look at polars, but I think I'm beginning to see why others are fascinated by them.
For example, look at 1.6k and 2k. These two dips are consistent as you move off axis. I assume that if you EQ them on axis, they will also smooth out off axis. These are coming from the mid. I know they are very minor. If the tweeter was crossed below 2k, they would likely be filled by the tweeters response. This is mostly a hypothetical question. I do not usually look at polars, but I think I'm beginning to see why others are fascinated by them.
Last edited:
If you like playing, try this filter. My suspicion is that the polar will be smoother.
Woofer LR4 @ 400
Mid BW4 @ 400 and BW3 @ 2300
Tweeter BW3 @ 2500
EQ
W -4dB @ 250 Q=3
M +1dB @ 550 Q=4
M +3dB @ 725 Q=4
M +3dB @ 1600 Q=4
M +3dB @ 2000 Q=4
M -10dB @ 4500 Q=4
T No EQ
Levels
W +3dB
M -8dB
T -7dB
Woofer LR4 @ 400
Mid BW4 @ 400 and BW3 @ 2300
Tweeter BW3 @ 2500
EQ
W -4dB @ 250 Q=3
M +1dB @ 550 Q=4
M +3dB @ 725 Q=4
M +3dB @ 1600 Q=4
M +3dB @ 2000 Q=4
M -10dB @ 4500 Q=4
T No EQ
Levels
W +3dB
M -8dB
T -7dB
The dips seem consistent on and off axis, so that would lead me to believe that they are not diffraction related. Neither woofer, nor tweeter show dips at the same frequencies, so that leads me to think that it's not a measurement error, like a mic response error. It still could be an illusion. I've encountered some REW measurement issues with my sub that defy logic, so mostly I go with what sounds good to me.
Could I hear it? Probably not, but the frequencies involved are in the range where my hearing is likely pretty sensitive, so maybe it would be a subtle placebo type improvement.
I'm mostly looking so closely at this as a theoretical exercise, and to lean how to interpret the polar graph. My speakers are not usually designed using a sim at all, and they are passive designs. I sometimes use a sim to calculate a notch, or sometimes shape a roll-off that's problematic. I use trial and error, and lots of measurements, and listening.
Last edited:
You're thinking polars in the sense of smooth. Which is not what they are useful for in my opinion. You could after all hammer the response flat as a board with a FIR filter. So what.
Look at your x-over design, at the polar from 2khz to 5khz. You should be able to easily see that the dispersion narrows at the cross over and then gets wide again at 5khz. This is what causes the estimated in-room response to dip at the cross over and then bump up at 5khz.
You can either sacrifice on axis behavior to counter act this. Or, change your crossover design to not get off axis cancellation in the crossover region.
Look at your x-over design, at the polar from 2khz to 5khz. You should be able to easily see that the dispersion narrows at the cross over and then gets wide again at 5khz. This is what causes the estimated in-room response to dip at the cross over and then bump up at 5khz.
You can either sacrifice on axis behavior to counter act this. Or, change your crossover design to not get off axis cancellation in the crossover region.
I am still to try the suggested designs. This needs a lot of free time with the room available, which is not easy to get.
However, there is something new I tried yesterday - I learned to export the IR from REW and load the IR response into HFD. This makes designing the filters in HFD very intuitive and within 20min of fiddling I had this curve. Pretty flat and with driver delays set to deepest nulls. On loading to amp the speakers sounded great, but a bit bight. So I added a gentle downwards slope. Then added three notches at the worst room peaks and I have great sound!
Clean, powerful, with no boom. Nice!
However, there is something new I tried yesterday - I learned to export the IR from REW and load the IR response into HFD. This makes designing the filters in HFD very intuitive and within 20min of fiddling I had this curve. Pretty flat and with driver delays set to deepest nulls. On loading to amp the speakers sounded great, but a bit bight. So I added a gentle downwards slope. Then added three notches at the worst room peaks and I have great sound!
Clean, powerful, with no boom. Nice!
So you in your sim, the tweeter and mid are actually working against each other off axis, and have cancellations between 3k-5k, and that is what narrows the graph in that frequency range. This is the prediction anyway for the horizontal listening axis only. I assume this polar tells nothing about how the drivers add 20 to 30 degrees above this axis. Am I understanding what this shows, or does the software do more estimating vertically, even though no vertical measurements have been entered?
Last edited:
Does this use just the on axis IR for each driver, and generate the response for that one axis?
A question, but I assume that it is a correct assumption. I started to add a ? but it felt like an odd way to post. I'm trying to grasp just what is being shown. Please correct me where I'm mistaken if I have it wrong.
It is the only way I can see that the directivity of the tweeter can change at 5k. My sim crossed at 2k. The entire 5k frequency is solely handled the tweeter. Your sim crossed at 3.2k, and on axis, the tweeter is producing virtually all the 5k frequency. The only way I can logically envision less off axis at 5k in your sim, is if the mid is somehow canceling part of the tweeters response. In other words, a dip off axis. Actually, I don't really see how that can happen at 5k when the mid is down 18dB.
If the polar is taking into account a dip that would show up vertically on axis at the x-over frequency, and then altering the left / right horizontal off axis due to that vertical null, that would make more sense. On the other hand, If the software is strictly simulating what would be measured horizontally then I would come back to thinking the tweeter and mid fight each other off axis.
What would cause your sim to drop quickly between 4k and 5k? Maybe that's the question I should have asked.
In your sim and my sim the tweeter measures one dB difference at 5k! 88dB vs 89dB
It is the only way I can see that the directivity of the tweeter can change at 5k. My sim crossed at 2k. The entire 5k frequency is solely handled the tweeter. Your sim crossed at 3.2k, and on axis, the tweeter is producing virtually all the 5k frequency. The only way I can logically envision less off axis at 5k in your sim, is if the mid is somehow canceling part of the tweeters response. In other words, a dip off axis. Actually, I don't really see how that can happen at 5k when the mid is down 18dB.
If the polar is taking into account a dip that would show up vertically on axis at the x-over frequency, and then altering the left / right horizontal off axis due to that vertical null, that would make more sense. On the other hand, If the software is strictly simulating what would be measured horizontally then I would come back to thinking the tweeter and mid fight each other off axis.
What would cause your sim to drop quickly between 4k and 5k? Maybe that's the question I should have asked.
In your sim and my sim the tweeter measures one dB difference at 5k! 88dB vs 89dB
Last edited:
Ok, so that's pretty much like how my sims are done with PCD. We also have the option to enter data from a chosen off axis, and enter it. That's what I did a few posts back. I think I used 60 degrees.
For my speakers, I simply measure off axis, rather than use a sim to predict. Easier to make measurements on a small 2-way of course.
For my speakers, I simply measure off axis, rather than use a sim to predict. Easier to make measurements on a small 2-way of course.
Last edited:
Ok, I'm tracking now.
The two circular swirl things at 3500hz at 60 degrees off axis on my design are cancellations due to tweeter/midrange spacing, delay and phase. Changing the vertical spacing alters that behavior, as does altering the delay that I put on the midrange.
That said, I mentioned these earlier in the thread. I'm not sure they're real. If I don't export measured phase, and instead rely on VCad calculating phase, they go away.
If I were building these, I'd likely want to put it together and then measure at 60 degrees off axis and see if that cancellation actually shows up.
Here's my original idea:
Here's the same design without the measured phase:
This is if I remove the delay on the midrange (with measured phase):