Home Theater Forum and Systems banner

New function required : implementation of Low pass and High pass filters to perfectly simulate response curve with EQs

17K views 88 replies 5 participants last post by  Philippe75 
#1 ·
Hi everyone,

I am using REW for several months now, and I am slowly improving my measurement and optimization method. Quite time consuming, but when we like it… :innocent:

My subject of interest is a 3 voices multi amplified system with a Xilica crossover. I measured each speaker individually on an appropriate frequency range, near field method for Low voice and 1 meter distance for Medium and Tweeter voices. Once measures are OK, I ran several EQ wizards on each speaker curve to obtain an acceptable linear response curve using limited EQ gain values. Each speaker is also affected by High pass and Low pass filters implemented in the crossover (model is JMLC “almost perfect filter” based on 3rd order Butterworth filter). Therefore, I ran EQ wizard outside the -3dB filtering range to optimize the overall curve (i.e. within the filtering range, at crossing frequencies and outside).

The Generic equalizer can simulate High pass and Low pass filters. These filters are not implemented in the Xilica, so I have to swap between Generic and Xilica which is not perfect but acceptable. I found that I have to manually adjust EQ to better match the LP & HP filtered curve. However, I have no idea of what the characteristics of these LP and HP filters are (2nd order?) and no possibility to setup a filter of my own (Butterworth, Linkwitz–Riley or Bessel). No additional options in the Target Settings section… My manual EQs are maybe not that appropriate.

That’s why I suggest to create a new function that would enable to setup a Butterworth, Bessel or Linkwitz-Riley LP and HP filters. EQ would be defined as they are currently done, but simulating LP and HP filters would give an overview of the very final result. That would be perfect!!!

Hope to see this in the coming weeks…

Regards
Philippe
 
See less See more
#4 ·
Thanks for the link, I now know what are the specifications of the filters. Unfortunately, these are not 3rd order Butterworth, so I can not simulate the digital filters applied to each voice to then optimize SPL curve with PEQ.

As I mentioned, but however my method is probably incorrect, my objective is to record each voice independantly (done), to identify optimal frequency cutoffs (done) for each voice thanks to manufacturers data and REW measures (done, but never good enough :)), to apply filters to each voice and to optimize the curve.

I am applying this method because my system is 3 voices multi amplified, and simulations prevent me from doing live testing all the time. Any better idea to achieve my objective ?

Regards,
 
#5 ·
Philippe75,
I am a little unclear on your objective, your current approach and any imposed constraints your may have, but will make the following comments in case they are helpful.

If the objective is just to find the best settings for a 3-way speaker using an Xilica speaker management box then I am be able to offer an alternative approach. We can take the approach that what is important is the phase tracking through the 2 XO ranges.

If 2 drivers (voices) phase track closely it does not cause erratic SPL behavior in that range. The worse the phase tracking the greater the SPL is impacted both on and off axis.

Driver phase tracking alignment is measured on the acoustic XO not on the electrical XO. It is not important if the Acoustic XO follows a particular textbook path like for instance a LR-24, or but-6, or JMLC “almost perfect filter”. If the drivers acoustical phase tracks well then any combination of electrical XO settings is optimized. That may be achieved for example with a But-24 LPF at 2k with a BES-12 HPF at 2.5k and 0.036ms Tweeter (TW) delay between the 2 drivers. In practice we can usually find several combinations of setting that meet this criteria when using a speaker management box like the Xilica.

The JMLC “almost perfect filter” was designed to minimize the phase tracking/SPL issue given the constraints of a passive XO implementation. In that case there is very limited ability to "adjust" the delay. It is "locked-in" by the filter chosen and the driver offset. Active XOs are much more flexible as the delay can be set at any value.

Once the XO settings are determined via the filters and delays chosen then the final level settings and EQ can be done. If the EQ is done on the input of each channel the phase of all 3 drivers are impacted in the same way and the good phase tracking that was achieved in the XO ranges is not changed.
 
#6 ·
Thanks for your feedback, but I think we are talking of 2 different subjects. Even though, they are highly related.

In regards to JMLC filter, Jean-Marie Le Cleah managed to design a XO filter that focused on a flat SPL response at XO frequency, optimized group delay propagation at XO frequency and limited impact on phase tracking. This is the reason why it is called "almost perfect" as those 3 criteria are to be taken into account for best sound quality. To achieve this objective, the filter requires to respect physical positions of drivers using a given formula at XO frequency. This filter is easy to implement in a Xilica box as you mentioned. Not perfect anyway, but the best filter I tested so far...

From my understanding and readings, there are no perfect 12db or more XO as they all impact these 3 criteria, except maybe FIR filters... If you have any suggestion, I would be grateful to read from you if I can implement it in the Xilica Box. In regards to your comments, I haven't yet measured the whole XO result (3 voices) as I was still focusing on optimizing XO points and individual speakers' SPL curve. But from your comment, I understand that the SPL curve might be impacted by unpleasant phase tracking.

Below is a detailed presentation of my approach:
1) I measured each voice individually on the whole range of frequency (10 Hz - 22 500 Hz) respecting the signal path. I.e. PC -> Preamp -> Xilica (flat) -> voice Amp.
2) Based on measures and manufacturer information, I define ideal XO frequencies taking into account SPL curve flatness (no EQ), distortions levels...
3) Once XO are defined (long and iterative approach, hum), I simulate XO (JMLC is based on Butterworth 18 dB (3rd order) => this is the missing function in REW, my request.
4) I optimize resulting curve with EQ to get the best result, but I try to limit the effect of an EQ to where it is needed.
------ This is where I am
5) I will have to optimize physical position of drivers to have aligned impulse from the 3 drivers.
6) I will have to measure the whole result and see if the SPL curve remains flat at XO frequencies and if phase is not too erratic. I must admit this is not the criteria I understand the best…….

Any suggestions?
 
#7 ·
There is nothing wrong with pursuing your path. I have tried all sorts of XO/EQ setup to decide for myself what makes a significant difference.

I was just pointing out that the JMLC filter is designed within the constraints of a passive XO network. The active XO setup with the Xilica box removes those constraints. That would allow a more perfect XO to be created. It also allows several different types of perfect XO to be created for comparisons.

I think I now understand your question however. You would like REW to create a But-18 House Curve for you. In the meantime you are asking how to import a But-18 target curve into REW so that you can try to EQ the "Voice" to match it.

Any House Curve can be imported into REW. The REW "Help" explains the method. The part that is probably the issue for you is the creation of the data to import.
1. Possibly someone will create the file for you to import. I did a spreadsheet for But-12 and LR-24. but it does not include But-18. There is no doubt a program available somewhere. I know it can be done easily in Matlab and probably many others so someone could provide you a trace, or better yet, the coordinates.
2. If you find a graph of the But-18 response you are wanting (possibly at the discussion site of the JMLC filter or in his presentation?) there is a program that will allow the coordinate file to be created:
SPL Trace
 
#8 ·
From my understanding, passive or active filters have the same effects on SPL, group delay and so on. Am I wrong ? If this is not correct, active can get ride off passive drawbacks, then indeed I should test different XO structure.

In regards to room curve, I think this would be a too hard task because I need as much house curve as I have speakers and XO points... I don't think this would be a good approach. I still think the best would be to implement, in addition of the existing HP and LP filters, 2 parameters : filter type (Link, Butt, Cheb...) and order (6dB, 12 dB, 18 dB...). This being done, we could all optimize PEQ to the final curve. Not a too big deal, isn't it ?

I now plan to measure all speakers with XO and PEQ to see if time overall volume is linear and if adjustment is OK as you described.
 
#9 ·
From my understanding, passive or active filters have the same effects on SPL, group delay and so on. Am I wrong ?
This is basically correct in one sense, but it is an over simplification.

If this is not correct, active can get ride off passive drawbacks, then indeed I should test different XO structure.
The active box can create the classic electrical filters listed in its documentation accurately. A passive design is typically more limiting. They can approach the classical electrical filters for the low order (1st and 2nd order) reasonably well. Their accuracy is not as good because of; component tolerances, and the speaker impedance is often not uniform in the XO range. This make the higher orders problematic to implement. Regardless, the electrical XO accuracy is not the important characteristic anyway. It is the acoustic XO result that is important not the electrical settings that were used to get there.

The biggest difference is the active filter will allow a delay to be set to align the phase tracking through the XO. This is one of the important consideration in XO design. With a passive design this design consideration is not possible unless the horn is moved forward such that the compression driver is much closer to the woofer than the normal situation with the horn mouth at/near the baffle.

The JMLC design is for a passive XO and does not align the Phase tracking, hence the "“almost perfect filter” designation. It allows the phase to cross at the XO frequency at a high slope and thus the SPL is erratic in that range.

See the chart in this post for an example of a non aligned horn setup. The chart is windowed to remove the room effects so we can easily see the effect on the direct signal, oops, I mean the effect on the "Direct Sound".

There is nothing "wrong" with this type of XO alignment. It is common for all large horn speakers using passive XO filters and with the horn mouth at the baffle. I was only pointing out that it is not necessary to make this XO compromise when using an active XO box where delays can be set.

In regards to room curve, I think this would be a too hard task because I need as much house curve as I have speakers and XO points... I don't think this would be a good approach. I still think the best would be to implement, in addition of the existing HP and LP filters, 2 parameters : filter type (Link, Butt, Cheb...) and order (6dB, 12 dB, 18 dB...). This being done, we could all optimize PEQ to the final curve. Not a too big deal, isn't it ?
I thought you wanted a target curve in REW that allows you adjust the Xilica XO filters and EQ to match that target. REW does not currently do that for us but allows us to input any target we want in the EQ panel. REW calls it a "House Curve", but it can be used to set any target curve we like, for any reason including, e.g., But-18 HPF at 1kHz. The horn voice can then be adjusted to match it. The target curve for the LPF can then be entered and the midrange voice adjusted to match that, etc. Once all the filters/voices are set then the overall house curve can be entered and the room EQ applied. Is that what you want to do? There are many DIY's that do it that way (even if I personally don't think it is efficient or advantageous).

I now plan to measure all speakers with XO and PEQ to see if time overall volume is linear and if adjustment is OK as you described.
The XO alignment / phase tracking cannot be easily evaluated with an overall measurement. This must be done by measuring the voices independently with "loopback timing" turned on in REW. This thread has more info on what is involved in case you ever decide to try it that way.

That method is likely more complicated that you are interested in now, so you may just want to start with something easier. Just select LR-24 XO filters and adjust the delays using the REW RTA feature until the SPL is as smooth as possible in the XO range. That "easy" method will provide a similar result to the JMLC passive design.
 
#10 ·
I took some times to read carefully your post and links. There are very interesting and I think I was not clear enough in my first posts (my English is not that good)...

In regard to passive and active XO, you are perfectly right, time alignment is one of the key for a good signal response. My system is made of three speakers, and the medium is a horn which is "far behind" the woofer and the tweeter (40 cm). At the moment, I did a measurement of each individual speaker positions to align them all using delays. I still need to confirm measurements with REW as physical measurements are not precise enough... These delays are implemented in all XO I parameterized in the Xilica.


For the sake of the discussion, I don't understand your point when you say JMLC filter was designed for passive XO. Jean-Marie Le Cleah spent a lot of time seeking for the perfect XO that met all these criteria : SPL, minimum effect on group delay and phase rotation. His final proposal is (the tweeter is the reference) :
- LP and HP are Butterworth of the 3rd order, XO is designed at -5 dB
- LP XO frequency = 0,8729 * XO frequency (to match -5 dB)
- HP XO frequency = 1,1456 * XO frequency (to match -5 dB)
- Medium of Woofer are to be moved ahead = 0,22 * wave length at XO
- Medium phase is to be inversed
The result is a quite linear response curve, group delay and phase rotation is in between -150 and -20 degree.

I am not sayin it is the best as there are no best XO except maybe FIR, but it is a good approach to manage all criteria. The filter you are using in your link looks like Samuel Harsch proposal : LP=Butterworth 4th order, HP=Bessel of the 2nd order and a constant delay is to be set for HP=(1/fc)/0,5. I have not tested it so far, I will in coming weeks. For your information, JMLC also desiged a good Excel simulation sheet to test various kind of XO. you will find it on that link (French) : http://nicolas.davidenko.perso.sfr.fr/filtragejmlc/filtragejmmlc.html


In regards to REW new function required, I will test your proposal but I need to generate 3 house curve, one for each individual speaker. Just for your understanding, I found out that when I aligned one speaker curves using PEQ, some PEQ were not well adapted when applying the HP and LP. This is why I would like to simulate filters in REW to manually fine tune REW PEQ proposal directly.

Philippe
 
#11 ·
Please understand that I recognize a JMLC recommended XO as a very good theoretical design. I would expect the sonic results to be excellent.

My understanding of the difference between JMLC approach and the alternate one I mentioned is that JMLC provides a analytical tool to predict the acoustic XO performance. It is a theoretical design. The alternate approach is to measure the actual acoustic XO response and adjust the Filters, Delays and EQ as needed to achieve the best measured results.

If I understood correctly, your plan was to take a recommend JMLC XO design and implement that using the Xilica. I see no issue with that.

It also sounded like you intended to try to EQ to the voices to match the theoretical/calculated response path of the individual voices. I do not see an advantage to that that approach. You could instead just implement the JMLC filters and delays; then fine tune the delays based on the actual measurements as needed, and then EQ to the overall response, i.e., the response of all 3 voices working together. I believe this approach negates the need import a But-18 target into REW. I probably should have ended my recommendation with this suggestion.

I went a little further and tried to point out that it is possible to just approach the whole problem using an empirical approach rather than analytic approach. That is, we can select the filters; measure and select the best delay for those filters (to achieve the best phase tracking), and then EQ to the overall response. If we are looking for optimized phase tracking, it may be necessary to change one or both filters in the XO to optimize the tracking. I have not studied the JMLC info to be sure, but understood that the phase tracking was not optimized in the particular design I saw. It appeared the phase of the voices crossed at the XO freq and diverged from each other in the rest of the XO range. This compromise was made to accommodate the limited offsets possible in a passive XO system. This may, or may not, be the case for the particular alignment you are considering.

For the sake of the discussion, I don't understand your point when you say JMLC filter was designed for passive XO. Jean-Marie Le Cleah spent a lot of time seeking for the perfect XO that met all these criteria : SPL, minimum effect on group delay and phase rotation. His final proposal is (the tweeter is the reference) :
- LP and HP are Butterworth of the 3rd order, XO is designed at -5 dB
- LP XO frequency = 0,8729 * XO frequency (to match -5 dB)
- HP XO frequency = 1,1456 * XO frequency (to match -5 dB)
- Medium of Woofer are to be moved ahead = 0,22 * wave length at XO
- Medium phase is to be inversed
The result is a quite linear response curve, group delay and phase rotation is in between -150 and -20 degree.
Possibly this particular alignment creates close phase tracking throughout the XO range, but the one I saw crossed at the XO freq as it was intended as the best solution for a passive XO. The example in the link your posted also may not be optimized for phase tracking. It is impossible for me to tell for sure as there is no phase tracking chart shown and I do not read French. It appears to be compromise design from what I can glean, but again, I am not sure.

I am not sayin it is the best as there are no best XO except maybe FIR, but it is a good approach to manage all criteria. The filter you are using in your link looks like Samuel Harsch proposal : LP=Butterworth 4th order, HP=Bessel of the 2nd order and a constant delay is to be set for HP=(1/fc)/0,5. I have not tested it so far, I will in coming weeks. For your information, JMLC also desiged a good Excel simulation sheet to test various kind of XO. you will find it on that link
It is, no doubt, a good approach and since you are measuring, it will be easy to see just how ideal it is.

The example in my link followed no theoretical approach or any recommendation. It was empirically determined to be a good solution for my particular setup of voices and listening axis. It was one good solution of several that I have found empirically.

In regards to REW new function required, I will test your proposal but I need to generate 3 house curve, one for each individual speaker. Just for your understanding, I found out that when I aligned one speaker curves using PEQ, some PEQ were not well adapted when applying the HP and LP. This is why I would like to simulate filters in REW to manually fine tune REW PEQ proposal directly.
This is main point I was trying to help with. I was trying to provide you an alternate approaches that will provide the same or better result. One that allows you continue on with your setup with out the need to figure out how to enter these target curves into REW. After all, you did ask for options in Post 2. :)
 
#12 ·
Don't make me wrong : I highly appreciate your feedbacks, sharings, and very detailed explanations. :bigsmile:

I will test your approach once I will be back and able to install the measurement tools again. I will keep in touch in that thread to let you know in coming days.
 
#13 · (Edited)
Hi,

Following the method described in given Thread, I tried today to measure "raw" delays in between individual speakers, i.e. without any filters. I wanted to be trained with the method by measuring delays generated by the design of the speaker (the Supravox is roughly +40 cm in front of the Beyma and +8 cm in front of the Fostex). I activated the "Use Loopback as Timing Reference" setting in REW, but I get 3 impulses almost at 0. Same results if I add a 1000 mm delay in Xilica as proposed in thread...

Following is the my measurement setp :
- MiniDSP UMIK1 USB mic attached to a laptop
- USB link in between the laptop and the DAC of the PreAmp - the soundcard is not used.
No physical loopback... and I don't see how to design one.

Any idea on how to setup a loopback of force REW to not align impulses on 0 ?

Thanks in advance.

------
Up date : I just see 2 posts below which deal with this issue (dealing with UMIK). I will have to find a way with my desktop I think...
 
#14 ·
Yes, the popular USB mics are not compatible with loopback timing. This makes it impossible to use them with the current version of REW to fine tune the phase handoff between voices/drivers. They are fine for all, or at least most all, other acoustic measurements. This is also a limitation if fine tuning of the delay/distance settings for a SW to main speaker XO is needed. However in the SW to main speaker case, it is reasonably easy to find good settings without using loopback timing. That's because the wavelength is so large at low freqs a ±0.5 m error in distance does not have major impact on SPL or sound quality. The midrange is not a forgiving.

It is too bad that many of the DIY speaker builders using a speaker management box like your Xilica or my DCX do not realize this limitation when deciding on the mic to purchase.

I expect it is possible to use a typical internal soundcard loopback with a typical "cheap computer mic" for this type of phase alignment work. Once the delays are established then the USB mic can be used for EQ and any other analysis that is needed.
 
#15 ·
Yesterday, I purchased a Taskam US-125M external USB sound card with 2 Line- out and 2 Line-in slots. I still have my old Ecm8000 mic, so I should be able to achieve the delay measure in coming days…


Taking a deep look at Ecm8000 and Umik-1 measures, I found-out that, if SPL are comparable, there are big discrepancies in GD, Impulse and Phase. Globally speaking, Umik-1 measures are weird compared to Ecm8000 ones. For instance, Supravox impulse starts normally with Ecm8000 but reversed with Umik-1; Beyma phase is weird below 600 Hz with Ecm8000 but still weird at 2K Hz with Umik-1; GD is almost flat with Ecm8000 but weird again with Umik-1 below 2K Hz.

To be complete, Umik-1 measures were done through a USB-to-SPDIF bridge connected to the Accuphase DAC-20 extension board when Ecm8000 measures were done through the PC sound card (calibrated) connected to one analog input of the Accuphase. I will have to make some measures with Umik-1 and the PC sound card to see which component is generating such results…

Any idea?
 
#16 ·
I wouldn't expect any significant difference in the Phase/GD results of the 2 setups, but...

At the LP even a very minor change in the mic position can have a large apparent impact on the charts as the room reflections may be very different. With the mic at 1 m a small position difference should be negligible because the direct signal is much strong than the reflections for the mid to upper frequency ranges. The lower freqs can still be problematic.

With proper scaling, windowing and filtering of the measurements, we can often see the phase trend of the direct sound even with the mic at the LP. It is the direct sound phase that we need to accurately measure in order to properly align the XO handoff.

I would need to see the .mdat file and understand your test conditions fully to provide a response specific to your situation.
 
#17 ·
I found a mistake in my analysis yesterday: I forgot to adjust the IR Window... :R Once done (depending on measures spec), results are much better for Phase and GD but Phase remains very different even if curves have now similar shapes.

Following graphs are of the Beyma horn speaker. In GREEN, Ecm8000 measure. In BLUE and RED, 2 different Umik-1 measures done in January and May but with similar audio path and physical mic location. For the 3 measures, mic distance to the floor is 1 meter and mic distance to the horn speaker mouth is 1m for Umik-1 and 70 cm for Ecm8000 (old measure). The major difference (to me) is that Ecm8000 measure was done through the PC Soundcard and Preamp analog input whereas Umik-1 measures were done through the USB to SPDIF bridge and Preamp DAC input. Frequency range is 200 Hz to 22 050 Hz.

SPL
Text Line Plot Pattern Design


Phase
Line Text Slope Plot Pattern


Group delay
Text Line Pattern Design Parallel



However, Umik-1 impulses remain very different compared to Ecm8000. As if there was a 180° applied to the speaker which can also be seen in Phase graph.
Text Line Plot Diagram Font
 
#18 ·
...The major difference (to me) is that Ecm8000 measure was done through the PC Soundcard and Preamp analog input whereas Umik-1 measures were done through the USB to SPDIF bridge and Preamp DAC input.
Just to clarify/confirm: The Tascam/ECM setup should bypasses the internal PC soundcard. The Tascam output should go directly to the Preamp input.

Also:
You are correct that the polarity is reversed for one of the 2 measurement setups. Using the Tascam/ECM setup this is easy to determine and correct as needed during the Tascam loopback calibration process. Neither of these setup should result in inverted polarity however. If the internal PC soundcard was actually used then it would not be surprising if it inverts the polarity.

There are a couple of other minor observations that are of no concern at this point.
> The SPL calibration for the 2 mics is significantly different at high freqs. Possibly the ECM is not calibrated or the appropriate mic calibration files were not loaded. This has no impact on phase/GD charts.
> The red UMIK-1 phase at the high freqs has an issue. Without looking at the .mdat file it is impossible to tell if that is just related to the REW settings applied or possibly to an issue with that particular measurement. Since we will not be using that particular measurement, it is irrelevant.
> REW allows the position of the IR to be offset as needed to align with other measurements. We could invert the ECM IR to match the polarity of the two UMIK-1 measurements and then offset the 3 IRs as needed to line them up. That way the phase chart overlay would track very similarly. They would overlap until the red trace falls away a 12kHz. This is just a convenient way to better depict how closely they track.

The Tascam/ECM setup is ideal for the process of adjusting the delays and XO setting to achieve close phase tracking throughout the XO range.
 
#19 ·
Yes, you are right. When using the Tascam card, I will have to generate a calibration file and then use the board directly connected to the Pre and Ecm8000 (back to analog measure). I will not use the laptop internal soundcard.

In regards to mic, you are also right : the Ecm8000 is using the generic calibration file proposed with REW as it was not calibrated when I purchased it. This probably explains the SPL curve at high frequencies. To be honnest, I had no idea of calibration when I purchased it 3 years ago... This is one of the reason why I purchased the Umik-1 mic early this year (in addition to the simpler USB interface but with limitation).

As far as polarity is concerned, the GREEN curve was measured with the internal PC sound card and the Ecm8000. I have no idea if polarity was reversed by the card or if the measure was done using a specific parameter set by error. I can e-mail you the .mdat file if you don't mind.

Regarding high frequencies SPL measure, I have this issue at 13 200 Hz with both mics. I thought this could be an effect of the horn... I will anyway set a LowPass filter at 9 000Hz max. You may find some information about the horn on this french site : http://www.guigue-locca.com/pavillons.html - reference is 300 C1.4.


By the way, one question : as you can see, the Beyma is naturally having a strong curve below 800 Hz. I always wondered if it was good to setup a HighPass filter which will reinforce the curve or let it as it is with a HP set low (at 300 Hz for instance) to protect the speaker...

In regards to the topic initial subject, I recently found that I can generate a HP or LP (Link) impulse using rePhase and import it as a .wav file to simulate the effect of the filter on a SPL curve. I just have to generate the different HP and LP required and use the A*B calculation provided within REW. Simple solution for RAW curve, but I have to find how to save an parametric equalized curve first as I can not apply the function on a target SPL curve...
 
#20 ·
As far as polarity is concerned, the GREEN curve was measured with the internal PC sound card and the Ecm8000. I have no idea if polarity was reversed by the card or if the measure was done using a specific parameter set by error. I can e-mail you the .mdat file if you don't mind.
You can post .mdat files on here. No need for e-mail. There is no need to post this file though as there is no open question that I am aware of. It is not unusual for an internal soundcard to invert the signal. The Tascam will not do that. I don't see that these old measurements need any more analysis as they cannot be used.

Regarding high frequencies SPL measure, I have this issue at 13 200 Hz with both mics. I thought this could be an effect of the horn... I will anyway set a LowPass filter at 9 000Hz max. You may find some information about the horn on this french site : http://www.guigue-locca.com/pavillons.html - reference is 300 C1.4.
That is a beautiful horn!

By the way, one question : as you can see, the Beyma is naturally having a strong curve below 800 Hz. I always wondered if it was good to setup a HighPass filter which will reinforce the curve or let it as it is with a HP set low (at 300 Hz for instance) to protect the speaker...
It is normally recommended to set the XO an octave or more above where the SPL response starts rolling off. It looks like the manufacturer suggests 400Hz as a minimum useful lower limit. If you are intending a 3rd or higher order filter then any freq higher than that will be okay from a safety perspective. I would have expected the best XO to be higher than that however. Normally the lower limit is a function of the horn mouth size. Below that starts to cause issues, but I am not really skilled in best horn practices. It does take a pretty large horn to reach 400 Hz comfortably.

A lot also depends on the capability of the lower voice. It's best to stay in the comfort range of both voices. You can experiment with various XO points if you like or just pick one that you think is a good compromise. I tend to try to split the difference in the overlap capabilities of the 2 drivers. A good 15" will go to 400-500 and a good 12" to 800 or so.

In regards to the topic initial subject, I recently found that I can generate a HP or LP (Link) impulse using rePhase and import it as a .wav file to simulate the effect of the filter on a SPL curve. I just have to generate the different HP and LP required and use the A*B calculation provided within REW.
I hadn't thought about using RePhase that way - interesting.

Simple solution for RAW curve, but I have to find how to save an parametric equalized curve first as I can not apply the function on a target SPL curve...
I suppose there are programs that allows IIR filter settings to be input and will then create and save the IR, but I am not aware of them as I have not had the need for it.
 
#21 ·
It just occurred to me that your above question regarding the 300 HPF filter is maybe in reference to a process where the EQ is applied to make the voice flat before applying the XO. If you do that process then keep the volume low and set the REW sweep to start at 300Hz or your idea of what is safe and required. That can be done in the measurement pop-up box.

I would avoid any process that results in any significant EQ boost in the cut off region of a voice due to both safety and distortion concerns for the voice.
 
#22 ·
No, my point was about the "natural" slope of the speaker below 700 Hz which is close to a 18db HP filter. I was wondering if there was an interest in not implementing a HPF to take the benefit of this natural slope. But I was also keeping in mind that there should be a problem passing too low frequencies in the speaker. So a solution could be to implement a lower HPF to protect the speaker only. But I don't know if this makes sense...

In regards to PEQ, I am only willing to use PEQ to linear the curve in between HPF and LPF XO points. Not above or below. But based on all your feedback and learning a of the past days, I don't know if I should spend too much time on this since phase and then SPL is the basis.

So far, thanks again for all informations shared with me !!!
 
#24 ·
No, my point was about the "natural" slope of the speaker below 700 Hz which is close to a 18db HP filter. I was wondering if there was an interest in not implementing a HPF to take the benefit of this natural slope. But I was also keeping in mind that there should be a problem passing too low frequencies in the speaker. So a solution could be to implement a lower HPF to protect the speaker only. But I don't know if this makes sense...
Then my original comments regarding XO selection are appropriate (post 20).
 
#25 ·
Interesting.

I am confident the REW phase chart is correct, but, for me, it has an inflection at about the 3kHz range. I believe it is the real characteristic of this particular horn and driver. I just haven’t seen one measure like this before. Possibly it is common as I have only worked with a limited number of horns.

Other comments:
> The polarity appears to be negative as the initial rise of the IR is negative. The UNIK-1 and REW is not the source of the inversion so it must be elsewhere. Very possibly the unusual phase response of the horn causes this effect? This is not really an issue though because the tweeter (TW) polarity is normally set to a positive polarity. The midrange (MR) polarity and delay are selected to provide the closest phase tracking with the TW. Then the woofer (W) polarity and delay are selected to provide the closest phase tracking with the MR. It is not required to start with the TW as positive, but it is convention and it helps a little in order to simplify the analysis. The polarity of the MR and W that create the best phase tracking throughout the XO ranges will be impacted by the filter slopes chosen and the driver characteristics so there is no way to simply state which polarity is “correct”. If the acoustic phase slopes are very near LR-24 then of course all the drivers will be the same polarity. If we select XO settings to best accommodate the 2 voices natural characteristics then it is good to be open minded and select whichever settings provide the best measured results.

> My frame of reference in the bullet point above is with my process of setting up an XO empirically in mind. If you want me to help work through an example of that process using your measurements, I can do that. Just advise me that this is your intent and I will provide the first steps. If you instead intend to use the initial process you mentioned, or some other one, then let me know that also. That way I will try to avoid confusing you with more comments like this that do not pertain to your approach.
 
#26 ·
All measures of the Beyma speaker + horn give the same result at 1m. As you mentioned, I suppose this is the normal behavior of this association as manufacturers measures are not equivalent (but usually, manufacturers measures always look better, no?).

Based on your findings, REW and Umik-1 are OK. Good! So, I suppose that the inverted phase may be caused by the USB Bridge -> Accuphase internal DAC audio path. I will do a new batch of measures once the Tascam audio card is there and calibrated. The audio path will be Tascam -> Accuphase analog input -> Xilica -> Amp -> Umik-1. I then will be able to see if phase is back to normal or no. It was when I measured with audio path : Internal audio card -> Accuphase analog input -> Xilica -> Amp -> Ecm8000.

I would be very happy if you could drive me in this attempt. Don’t worry, I am very open-minded! I just want to understand what is happening as most of my “limited” knowledge is based on readings and theory. Thanks very much in advance!
 
#28 ·
All measures of the Beyma speaker + horn give the same result at 1m. As you mentioned, I suppose this is the normal behavior of this association as manufacturers measures are not equivalent (but usually, manufacturers measures always look better, no?).
:)

I would be very happy if you could drive me in this attempt. Don’t worry, I am very open-minded! I just want to understand what is happening as most of my “limited” knowledge is based on readings and theory. Thanks very much in advance!
Good. :) I think you will find this process results in closer phase tracking and SPL flatness than can be obtained using analytical process alone. While the empirical method it is complicated and confusing at first, I find it to be much easier to learn and implement than the analytical approach.

> I suggest we first do the MR-TW XO together as it is easier because there are no room modes to muddle the charts. That will establish the process. Then maybe you will want to try the W-MR XO yourself. I will still help as needed.

> Let me know when the loopback calibration of the Tascam setup is completed and you are ready to start measuring voices. I will then provide setup/test instructions.

> It will be helpful if you provide (for example): W and TW ID/Specs?, Box type for the W (vented, or ??)?, Amps used?, Room Size?, Speaker and LP locations, Speakers toed in? , Music system only?, Music server in use?
This basic type of info will help me stay on course and avoid problems stemming from assumptions about what your setup is.
 
#27 ·
I'd like to add a vote for the request in this thread (if I understand it right) that it would be huge for a xo designer to be able to generate target xo filters in REW. I've done the loopback and import house curve method in the past but it can be tedious if you are trying out different filter topologies, corner freqs, etc. ARTA has this ability and it's my go to in these situations, otherwise I usually prefer REW.

In my case I use JRiver for xo and eq. It only has the ability to do Butterworth filters (and LR types with cascaded 2nd order BW) so if I want to try different topologies I'm SOL without knowing what they actually look like.
 
#29 · (Edited)
To answer to natehansen66, yes my initial request was to introduce XO filters simulation in REW so that we can simulate impacts on SPL, phase, etc. I still think this would be a great improvement for esigning and testing XO.


> Let me know when the loopback calibration of the Tascam setup is completed and you are ready to start measuring voices. I will then provide setup/test instructions.
Unfortunately, the Tascam US-125M is not the good external card... It is a mixer that combines multiple inputs (line, mic and instrument) and sends back the mixed signal to line output or the computer. Therefore, I cannot monitor the mic individually as it is mixed with REW SPL. In result, I get a loud interesting larsen... Nice :)! I will change for the US-122M or US-322. If you have any idea before I send it back for an exchange...


> It will be helpful if you provide (for example): W and TW ID/Specs?, Box type for the W (vented, or ??)?, Amps used?, Room Size?, Speaker and LP locations, Speakers toed in? , Music system only?, Music server in use?
Amps
- Accuphase E350 + DAC20: used as a PRE and as an external AMP for the Beyma voice (middle)
- Microméga PW400: used for the Supravox voice (low)
- Myriad Z62: used for the Fostex voice (tweet)

Digital crossover
- Xilica XP3060 : audiopath is Accuphase -> Xilica => 3 amps

Speaker
- The system is based on the Acanthe speaker you can find in the link "Guigue & Locca". It is a vented box for low voice + horn for the middle + tweeter. All voices are independent. The overall level is close to 98 dB.
- Woofer: Supravox 285 GMF
- Middle: Beyma CP755Nd + Guigue Horn 300 C1.4
- Tweeter: Fostex T90A

Room
Room dimension: 8,04m * 5,53 m, height is standard 2,5m (appartment).

Usage
The system is for music mainly, I also linked the TV box on it but it is for convinience. Sources are a laptop with an old version of JRiver Media Center and a CD player. The laptop is linked to the Accuphase DAC using a USB to SPDIF bridge (DAC20 option card is only SPDI/F).

Here it is, I have to wait for the new external sound card now.
Keep in touch.


----- Update
I anyway did some measures using the Tascam and Umik-1. Phase is still inversed for all speakers on impulse. I was anable to test back with Ecm8000 because of the Mixer issue with the Tascam...
 
#30 ·
To answer to natehansen66, yes my initial request was to introduce XO filters simulation in REW so that we can simulate impacts on SPL, phase, etc. I still think this would be a great improvement for esigning and testing XO.
There are several here at HTS interested in this feature and maybe JohnM will implement it someday. For anyone that needs a workaround, I just confirmed Philippe75's comment that RePhase can create a wide variety of filter shapes that can be converted for use in REW. It takes a few steps to do, but it is not too difficult. See the chart below as an example. I created 200Hz Low-cut 18 dB/octave and 6kHz High-cut 24 dB/octave filters in RePhase. I then converted the file for use as a target curve (house curve) in REW. I then loaded a random full range measurement and used the REW EQ window to calc EQ filters from 150-7k Hz. It worked fine. RePhase can create any freq, any slope rate, and can even stack filters so this is a good workaround if needed.
Text Blue Line Green Pattern


Unfortunately, the Tascam US-125M is not the good external card... It is a mixer that combines multiple inputs (line, mic and instrument) and sends back the mixed signal to line output or the computer. Therefore, I cannot monitor the mic individually as it is mixed with REW SPL. In result, I get a loud interesting larsen... Nice :)! I will change for the US-122M or US-322. If you have any idea before I send it back for an exchange...
Do your have the Tascam "Loop Mix" switch set in the "Off" position? If it is set "On" feedback will be created. All of these USB audio interfaces have a feature like this using various names that will create this problem for REW. This interface will work as well as any other if the settings of the computer and the USB interface are appropriate. Someone here will be able to help you troubleshoot any problem you have.

Thanks for the info on your system. I will look up your voices and get my idea for XO points. Let me know if you have particular XO idea for the MR-TW. We just need a freq and the filter slopes as a starting point. We may later decide to adjust these if that is needed to achieve good phase tracking.
 
#31 ·
Just received the new Tascam box, I should now be able to achieve testing in good conditions. But unfortunately not before next Wednesday as I will not have the ability to save time for installing and testing the box before that date.

I was working on a MR-TW XO at 8 kHz or a little higher. Indeed, the Beyma speaker (if the measure was correct) becomes slightly erratic above 8 kHz and more beyond 10 kHz, and Fostex recommended XO frequency is 7 kHz or higher. In addition, I wanted to rely as much as possible on the Beyma to limit XO effects on key frequencies, and ideally respect the MR-TW XO > 10 * SW-MR XO rule (theory again).

Any advice so far?
 
#32 ·
8kHz sounds fine to me.

I took a quick look at the published responses of the 3 units. There appears to be significant overlap available so there is good flexibility in the XO point selections and the filter orders for both XOs.

I would suspect that the MR starts to narrow its high freq dispersion rapidly above 9.6k (depends on the design of the throat of the horn) and would think the horizontal dispersion of the MR and TW may match a little better at less than 8k. I was thinking that 5k-6k may be a better guess for the XO. I couldn't find any horizontal dispersion info for the MR however so I really have no evidence to support this. Even if the best match does happen to be a bit lower than 8k I would not expect it to be a very significant sonic impact in practice.

The Fostex 7kHz min XO recommendation has in mind a first order XO as shown on their schematic. They also have in mind the 50 W max power it's rated for. With such a high sensitivity I would doubt that you would even put 1 W to it. 106 dB above 5k Hz in a typical room is huge. If you are intending a higher order filter or not intending to push the limits of its output then it will be perfectly safe to XO as low as 5kHz in my opinion.

My thoughts above are all just guesswork so let's do the 8kHz. That is as good a starting point as any. You can always try something lower or higher later if you like. At this XO you can use whatever filter slopes you like to start. As we review the measurements we may decide to adjust a filter slope slightly to achieve the best phase tracking.

Whenever you are ready the initial 2 measurements needed are:
> TW (One channel, L or R)
> MR (same channel)

Conditions:
> If you have distance settings in the Pre-Pro just set them all to equal distance. We will not be using that feature.
> Set 8k Hz XO filters in the Xilica Use whatever filter slopes you want to start.
> Set 400 Hz? XO (Use your initial desired target freq and filter slopes) I would suggest a 12 to 24 dB/octave HPF for the MR if the XO freq is this low. If you go higher, to maybe 600, or more then a But-6 could be possible also. I'm just trying to avoid significant output from the horn below where it unloads.
> Set all Xilica delays at 0 ms.
> I suggest we start with no EQ activated. EQ can be left on for delay adjustments once we have it set reasonably.
> REW Loopback Timing engaged
> Full range sweeps, Approx, 20-20k Hz for all this work. We can run this 8k XO effort (2k-20k Hz) if you prefer. I usually run full range for everything as it has a couple minor advantages. Your choice. Just be sure all XO filters are set properly and active. This means the lower XO needs to be in place also and set to reasonable values.
> Mic at LP and speaker in its normal position. The mic orientation is not critical for phase response, but the impact of reflections at high freq will be reduced slightly if the mic is pointed foreword (at or in the general direction of the speaker).
> Approx 75 dB is okay for the measurement level.

I hope I remembered of all the important bits!

Post the .mdat file. I will review, and adjust polarity and delay for the best phase tracking. You can do the same if you followed and understood the link I posted earlier. I will then explain; what I did, where we stand, and recommendations for better tuning. If your just looking for good results rather than all the confusing detail, I will save effort and skip the explanations. :)
 
#33 ·
Done!

The new TASCAM is the good one. I also take the opportunity to buy a better mic with its calibration file (Audix Tm1 plus). All speakers have been measured with following audio path: Tascam -> Accuphase -> Xilica -> Voice amp -> TM1. Impulses look correct now.

I followed your instructions to measure Tweeter and MR with REW Loopback function activated. I can now clearly see delays in between TW and MR.

I have done several measures with different XO types and slopes (Link, Butt and Bess for 12 dB to 36 dB) to visualize effects of each filter on the phase. Following parameters are standard to all measures:
- MR: Highpass XO=500 Hz, Lowpass XO=8000 Hz (a starting point as you mentioned)
- Tweeter: Highpass XO=8000 Hz
- Mic: distance to the box=100 cm, pointing the middle of Woofer-Tweeter distance
- No delays
- Normal (positive) phase
- No PEQ at that step indeed

MDAT files (one for each XO type and slope) are huge, even zipped. I have to find a way to post one here. Tell me if you have a preference in terms of filter type and slope.

I have to analyses and understand results now. I would be very grateful if you could explain steps, findings and recommendations as I am interesting in understanding things.
 
#35 ·
The new TASCAM is the good one. I also take the opportunity to buy a better mic with its calibration file (Audix Tm1 plus). All speakers have been measured with following audio path: Tascam -> Accuphase -> Xilica -> Voice amp -> TM1. Impulses look correct now.
Looks good!

I followed your instructions to measure Tweeter and MR with REW Loopback function activated. I can now clearly see delays in between TW and MR.
Yes, as we expected, the MR is significantly delayed as shown by the overlay of the 2 IRs. The approximate excess delay of the MR in this measurement is about 1.16 ms.

I have done several measures with different XO types and slopes (Link, Butt and Bess for 12 dB to 36 dB) to visualize effects of each filter on the phase. Following parameters are standard to all measures:
- MR: Highpass XO=500 Hz, Lowpass XO=8000 Hz (a starting point as you mentioned)
- Tweeter: Highpass XO=8000 Hz
- Mic: distance to the box=100 cm, pointing the middle of Woofer-Tweeter distance
- No delays
- Normal (positive) phase
- No PEQ at that step indeed
All good, except mic position. I requested LP mic position and your intended speaker position. Maybe you are not testing with the room already completed or setup. This is no problem, but ideally the mic should be located on your intended "listening axis" we could also think of it as the line-of-sight axis. if we sit at the LP and look at a midpoint point between the MR and TW (vertically), that will be the listening axis.

For example, If we intend to use a standard equilateral triangle setup and to face the speakers straight forward, then the listening axis is not directly in front of the speaker. it would be 30° off the speaker horizontally. If the mic was position within 20° or maybe 30° of the intended listening axis horizontally then that is probably an insignificant difference. We should be much more careful vertically. We want the forward lobe to be faced directly at the LP, not tilted up or down. That lobe will be very narrow so the sound will change significantly with vertical position changes, particularly so because of the large center-to-center distances of the voices and the high XO point. I would target being within ±3° of the intended vertical listening axis.

Also, 1 m is very close for a speaker this size. We should consider something nearer 2 m (if the LP is not convenient or available). This will help reduce error in locating the mic on the vertical listening axis.

MDAT files (one for each XO type and slope) are huge, even zipped. I have to find a way to post one here. Tell me if you have a preference in terms of filter type and slope.
The slopes you provided appear to be pretty steep. I think that is a good idea in this case. It will to keep the XO range relatively small. The drivers are large and the XO is very high so this choice will minimize the overall disruption to the SPL response by limiting it to a narrow range. LR-24's are popular and have some benefits. Since we have wide overlap, I would have suggested we start there. Others may work out just as well so if you prefer to try LR-48 or something else that is okay.

Minor points:
> The REW SPL meter doesn't appear to be calibrated correctly unless it was extremely loud when the measurements were taken. I would suggest either setting it with an SLM or just calibrate the REW meter at 75 dB using a comfortably loud PinkPN signal. Testing should then be done to something near 75dB.
> It would be best to use the Xilica to decrease the MR level a little so it is closer to the TW level. We probably will want some HF decrease above 8k though so MR level can be a little higher than the TW. Maybe a MR decrease of 4 or 5 dB would be about right.
> Please set REW to measure using only 1 sweep for the measurements. There is no issue using 2 sweeps, but the file size is 2x as large with no benefit. Multiple sweeps are only important for certain types of measurements and phase timing is not one of them.

Path forward:
> I can use these measurements for timing alignment as we planned, but please advise me if you want me to do so. These are not suitable for best results unless the mic is at least close to the intended listening axis as defined above. Please advise.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top