I think you're going to find that those 9 steps have a lot of overlap when working with a DSP. Or are at least simplified to the point where they are all done simultaneously. For example, 1 and 2 takes just a few minutes in HFD. At that point, you could start doing sweeps on a given channel with REW to make sure things are generally doing what you expected while looking for any obvious issues.
A lot of the rest of the items you've called out can probably be done in one step once it's in the final listening position.
Phasing might be the exception if you plan to use FIR, that would require rephase or something like that.
A lot of the rest of the items you've called out can probably be done in one step once it's in the final listening position.
Phasing might be the exception if you plan to use FIR, that would require rephase or something like that.
I don't know why, but I would like to read about FIR in this topic, I may be wrong Hypex FA253 allows you to do this.
Here's how I would do it. This might not work for you, but if I was working with a passive x-over, this is one way I might approach it. The woofer portion is more complicated. No need to talk about that, if the mid / tweeter x-over fails to work.
1 Measure the mid on axis starting with a 300hz LR2 high-pass filter. Set the level at 90dB at 1k.
2 Import targets for the high, and low filters. Set targets for 85dB. LR4 at 300 and 3k.
Keep the high-pass 2nd order at 300hz, and set a low-pass LR4 at 3k.
Alter the frequency, gain, and slope, as needed to match the 300 to 3k target.
3 Use additional EQ to shape the response to match the target as close as possible. Try not to boost anything by more than a dB or two.
Reduce the gain for the tweeter by 6dB, or whatever would be needed to match the 85dB target.
1 Measure the tweeter on axis with a 1k LR2 high-pass. (Use a suitable cap initially to protect the tweeter too.) Adjust to 85dB at around 5k.
2 Import a 3k LR4 high-pass target for the high-pass. Set the target at 85dB.
3 Set a high-pass 4th order filter at 3000hz, and make a sweep. Remove the cap if you are sure you won't somehow sweep it full range by accident .
Alter the frequency, and slope, as needed to match the 3k target. You might only need a 2nd order filter, and a notch.
4 Use additional EQ to shape the response to match the target as close as possible.
Now, find out if the tweeter, and mid add correctly.
Sweep the mid, then sweep the tweeter, and have them on the screen at the same time. Gated measurements only. They should both be at 85dB, and cross each other at 3k, and be 6dB down. (79dB)
Now sweep them at the same time, and observe how they combine at 3k. They can't add to more than 85dB, but they can add to less. If they add below 79dB at 3k , flip the polarity of the tweeter.
Add some delay to the tweeter in .005ms increments, and sweep again. At some delay, the sum should look flat.
Flip the tweeter polarity, and see if there's a nice null. Then remember to flip it back.
This would be a good time to measure off axis to the side. Measure at 30, 45, and 60 degrees and see if anything looks lumpy.
Measure distortion.
Here's an example of matching a target.
1 Measure the mid on axis starting with a 300hz LR2 high-pass filter. Set the level at 90dB at 1k.
2 Import targets for the high, and low filters. Set targets for 85dB. LR4 at 300 and 3k.
Keep the high-pass 2nd order at 300hz, and set a low-pass LR4 at 3k.
Alter the frequency, gain, and slope, as needed to match the 300 to 3k target.
3 Use additional EQ to shape the response to match the target as close as possible. Try not to boost anything by more than a dB or two.
Reduce the gain for the tweeter by 6dB, or whatever would be needed to match the 85dB target.
1 Measure the tweeter on axis with a 1k LR2 high-pass. (Use a suitable cap initially to protect the tweeter too.) Adjust to 85dB at around 5k.
2 Import a 3k LR4 high-pass target for the high-pass. Set the target at 85dB.
3 Set a high-pass 4th order filter at 3000hz, and make a sweep. Remove the cap if you are sure you won't somehow sweep it full range by accident .
Alter the frequency, and slope, as needed to match the 3k target. You might only need a 2nd order filter, and a notch.
4 Use additional EQ to shape the response to match the target as close as possible.
Now, find out if the tweeter, and mid add correctly.
Sweep the mid, then sweep the tweeter, and have them on the screen at the same time. Gated measurements only. They should both be at 85dB, and cross each other at 3k, and be 6dB down. (79dB)
Now sweep them at the same time, and observe how they combine at 3k. They can't add to more than 85dB, but they can add to less. If they add below 79dB at 3k , flip the polarity of the tweeter.
Add some delay to the tweeter in .005ms increments, and sweep again. At some delay, the sum should look flat.
Flip the tweeter polarity, and see if there's a nice null. Then remember to flip it back.
This would be a good time to measure off axis to the side. Measure at 30, 45, and 60 degrees and see if anything looks lumpy.
Measure distortion.
Here's an example of matching a target.
Last edited:
Thank you, I will study this. I forgot polarity and delays!
Not sure what a target is, where do I get it and where I import it to!
In PCD software, they have targets, so you can match them there, but what would be more convenient is to have them in REW, so that you can see them right away, after a sweep. Short of making one using an frd file, I don't really know. The one in the picture is part of REW, but it is not handy like what I suggest. I may try to make one in PCD.
I'll write up something when I get home tonight and can get some screen grabs and whatnot.
The procedure that I have used with Hypex FA: (place speaker far from walls and mic at 1-1,5m at listening height and menorize this setup!) No simulation needed!
- check that HFD has downloaded flat presets to the FA board!
Don't expect all this to happen in one day! Gradually you will learn to detect measurement artefacts, mistakes and bad choices
1 play sweeps and measure each way with moderate fixed spl (70dB is ok) setting of REW, dont preset anything except channelling in HFD --> raw response in-box
- on-axis and at least 15,30,45,60deg horizontally With the tweeter you can start the sweep from 500Hz up
- save each way's raw off-axis set of measurements as .mdat for futher use
- study the measurement's quality - impulse response and find reflections to set as long gating as possible The first reflection is most likely the floor at 2-3ms, walls and ceiling at 5-8ms
- try to improve measurement quality by adding pillows, moving speaker, mic, furniture etc. Reduce ambient noise!
- study spl response irregularities = interferences and resonances, directivity and distortion of each way
- take also nearfield sweeps of each driver and save them (lower spl to prevent mic/ADC clipping
2 Equalize each way's on-axis response flat at least 1,5 oct beyond expected xo manually/parametric (FIR). Don't add gain more than 2-3dB!
- BSC compensation uses mainly negative gain highpass, fine adjust with peak correction
- remember to load save settings and measurements regularly!
- use appropriate gating of measurements for each way (freq band)
- this step greatly helps you afterwards, and Hypex and minidsp have eq and xo functions separated just because of this!
3 set preliminary low-and highpass slopes (filter presets give them!)
- measure all ways individually in same.mdat and use "All spl" view of rew to see how they match
- reduce gain of high spl ways to match them
- check distortion and try different xo types - use higher spl for this
- when using LR2 switch the mid's polarity
- reduce mid's gain with LR2
4 start adding delay to improve summation - several iterations needed, this is hard!
- 3-4 way keep woofer delay zero, then tune mid's delay to it without the tweeter. Then same for mid-tweeter without woofer
5 Measure full system's on- and off-axis responses in one -mdat
- study directivity and distortion
- try different xo freq and slopes
6 Go back to stages 2-5 etc. fine tuning..... measure and listen
7 put the speaker on intended location and measure room response around listening spot (MMM)
- do some room correction eq and tonality changes with low q-factor or each way's gain
When you master all stages you can redo single stages, changes and measurements to the system afterwards.
Have fun! No good luck needed, just hard work and patience...
- check that HFD has downloaded flat presets to the FA board!
Don't expect all this to happen in one day! Gradually you will learn to detect measurement artefacts, mistakes and bad choices
1 play sweeps and measure each way with moderate fixed spl (70dB is ok) setting of REW, dont preset anything except channelling in HFD --> raw response in-box
- on-axis and at least 15,30,45,60deg horizontally With the tweeter you can start the sweep from 500Hz up
- save each way's raw off-axis set of measurements as .mdat for futher use
- study the measurement's quality - impulse response and find reflections to set as long gating as possible The first reflection is most likely the floor at 2-3ms, walls and ceiling at 5-8ms
- try to improve measurement quality by adding pillows, moving speaker, mic, furniture etc. Reduce ambient noise!
- study spl response irregularities = interferences and resonances, directivity and distortion of each way
- take also nearfield sweeps of each driver and save them (lower spl to prevent mic/ADC clipping
2 Equalize each way's on-axis response flat at least 1,5 oct beyond expected xo manually/parametric (FIR). Don't add gain more than 2-3dB!
- BSC compensation uses mainly negative gain highpass, fine adjust with peak correction
- remember to load save settings and measurements regularly!
- use appropriate gating of measurements for each way (freq band)
- this step greatly helps you afterwards, and Hypex and minidsp have eq and xo functions separated just because of this!
3 set preliminary low-and highpass slopes (filter presets give them!)
- measure all ways individually in same.mdat and use "All spl" view of rew to see how they match
- reduce gain of high spl ways to match them
- check distortion and try different xo types - use higher spl for this
- when using LR2 switch the mid's polarity
- reduce mid's gain with LR2
4 start adding delay to improve summation - several iterations needed, this is hard!
- 3-4 way keep woofer delay zero, then tune mid's delay to it without the tweeter. Then same for mid-tweeter without woofer
5 Measure full system's on- and off-axis responses in one -mdat
- study directivity and distortion
- try different xo freq and slopes
6 Go back to stages 2-5 etc. fine tuning..... measure and listen
7 put the speaker on intended location and measure room response around listening spot (MMM)
- do some room correction eq and tonality changes with low q-factor or each way's gain
When you master all stages you can redo single stages, changes and measurements to the system afterwards.
Have fun! No good luck needed, just hard work and patience...
I agree with Juhazi's approach broadly.
If you can get the speaker outside on a step ladder, you'll increase the usability of the response.
I use REW to measure, but VituixCAD to actually plan out the filters.
Measure on axis and off for each driver/way individually, from the tweeter level. So at tweeter level, tweeter only, 0,10,20 etc off axis. Then from tweeter level measure the midrange only, then woofer section. All that in REW.
Making sure they're labeled correctly (reference vituixcad help file), export them to a .frd file and import into VituixCAD. Then build your speaker. Below is an example with a single measurement for each driver. If you had all the off axis measurements, they'd be listed out under each driver.
If you actually have all the off axis measurements, you can simulate the entire response to include in room response and all sorts of goodness. I didn't do this when I was building/figuring out my speakers. It would have saved me a bunch of time though.
If you can get the speaker outside on a step ladder, you'll increase the usability of the response.
I use REW to measure, but VituixCAD to actually plan out the filters.
Measure on axis and off for each driver/way individually, from the tweeter level. So at tweeter level, tweeter only, 0,10,20 etc off axis. Then from tweeter level measure the midrange only, then woofer section. All that in REW.
Making sure they're labeled correctly (reference vituixcad help file), export them to a .frd file and import into VituixCAD. Then build your speaker. Below is an example with a single measurement for each driver. If you had all the off axis measurements, they'd be listed out under each driver.
If you actually have all the off axis measurements, you can simulate the entire response to include in room response and all sorts of goodness. I didn't do this when I was building/figuring out my speakers. It would have saved me a bunch of time though.
Also, be very deliberate with HFD, it's a finicky mess. Select the amp section you want to measure, then mute the channels not being played. The measure. Once you're done with a channel, unmute everything, select the next channel, mute the un-used channels, repeat.
You can jump between amp channels with all sorts of things muted/not muted and it gets confusing quick. You don't want to try playing the tweeter full range at high volume. So make the selections with your source at low volume, verify the correct drivers are working/muted then set the volume back to where you're measuring and measure.
You can jump between amp channels with all sorts of things muted/not muted and it gets confusing quick. You don't want to try playing the tweeter full range at high volume. So make the selections with your source at low volume, verify the correct drivers are working/muted then set the volume back to where you're measuring and measure.
Ok. I have dual woofers (on one amp channel). Do I sweep them individually or together?
Re HFD, I found it manageable. Danner’s tutorial on this this was extremely helpful.
VituixCad is a bit intimidating at the moment. But I will press on.
I don't use VCAD for design/settings because I am too lazy and dumb to learn it. Too many important parameters to set right... too little lifetime left. I understand the value and versatility of VCAD and recommend learning it!
A by-product of this project/thread could be a step-by-step guide on how to set up a 3-way using REW and Hypex HFD.
I will attempt to put it together based on what knowledge and experience I pick up. I will incorporate advice from others to my best ability.
I would very much welcome as much input as possible from more experienced members here.
Here is my starter file (just started, work in progress). Feel free to comment!
I will attempt to put it together based on what knowledge and experience I pick up. I will incorporate advice from others to my best ability.
I would very much welcome as much input as possible from more experienced members here.
Here is my starter file (just started, work in progress). Feel free to comment!
Measure the dual woofers together. Measure at a fixed distance (1 meter is good) on your listening axis, the BSC gets taken care of naturally.
The main reason I recommend you figure out VCAD initially is the off axis and estimated in room responses.
You can hammer the chosen axis flat as a billiard table, but that doesn't mean the response off that axis will look good.
When I built my main speakers, I measured a single axis (that's the data above) and then flattened the response on that axis. The vocals sounded recessed and I got headaches listening to them. It wasn't until I did a full spin to check it out that I saw the predicted in room response had a massive peak in the upper treble and a hollowed out midrange.
For reference on actual VCAD complexity. The snippets I posted above started as an .mdat from REW. I exported to .frd, imported to VCAD and built that example filter design in about 15 minutes, just to take that screen grab to post.
So steps:
-Put speaker on the tallest thing you're comfortable (outside), as far away from hard objects as possible, with the microphone on your listening axis (usually tweeter level)
-Full spinorama (every 10 or 15 degrees out to 90 or so) on each "way" or driver section (tweeter, mid, woofers) label and save in accordance with VCAD naming convention ("hor_0_tweeter" is horizontal 0 degrees, "hor_45_tweeter" would be horizontal 45 degrees off axis)
--If you want to do vertical spin, you'll need to lay the speaker on its side
-Export the REW data to .frd format
-Import into VCAD on the drivers tab (this will be the entire list of "hor_0_xxxxx" files
--SET THE SMOOTHING TO 1/3. Your measurements will have a lot of garbage in them, you don't want to end up trying to correct that.
-Move to crossover tab, drop in your three drivers
-Drop in individual filter functions and see what it looks like.
--Goal is flat frequency response on your chosen listening axis (start on axis) and a constant downward in-room/power response
--If the speaker dispersion angle gets wider at the crossover from midrange to tweeter, you would see a bump up in in-room response. This would necessitate a crossover move, since if you tried to fix it with an EQ filter, you'd get a dip in your on-axis response.
Just to give you an idea of what I'm talking about. Below is my main speakers. These are coaxials, and needed a bunch of compromises to sound good. The first picture shows non-smoothed responses, on axis in gray, power response in dark blue. You can see the directivity weirdness in the tweeter.
The ripples from 100hz up through 2khz is due to reflections in the measurement. Using REW you can gate out the reflections to verify what is real and what isn't. Don't try to fix things that aren't real. Also, you will get a dip at the floor bounce freq, the higher your step ladder, the lower the frequency that will occur. https://mehlau.net/audio/floorbounce/ Check that website to see where you expect that to occur.
This is 25 degrees off axis (where I listen) and smoothed at 1/3 octave.
You shouldn't have to make bananas compromises like this with your chosen drivers. Coaxials are their own weird thing. But voicing changes are a thing. The picture below is my speakers, final filters done, but not voicing. This has recessed sounding vocals still. What I ended up on was two filters of only 0.7db in magnitude to push the treble down a bit and lift the upper midrange a bit. The picture above is the final setup. You can see the listening axis response doesn't look much different. But it sounds different. And you can see it in the power response.