I have been away from the room correction rabbit hole for many years now (since my AV32R died) but now its time again. It is fantastic to see the how you have developed the old TMReq application over the years. Fantastic work.
I have an idea to create and add filters in the PC application JRIVER (feeds an USB DAC) to correct my Calliopes with Bassmodules. I wonder (a long shot) if you happens to have old measurements from these in an anechoic chamber? Or would this not make sense?
I have been away from the room correction rabbit hole for many years now (since my AV32R died) but now its time again. It is fantastic to see the how you have developed the old TMReq application over the years. Fantastic work.
I have an idea to create and add filters in the PC application JRIVER (feeds an USB DAC) to correct my Calliopes with Bassmodules. I wonder (a long shot) if you happens to have old measurements from these in an anechoic chamber? Or would this not make sense?
If John has that, it would potentially be helpful, but you would want to make sure he has the full polar response. Additionally, if the speaker has any directivity anomalies, you may find that REW’s built in eq facilities are not well suited to those adjustments. I actually develop my eq filters for speakers using a different software that is specifically designed for that purpose. It’s a crossover software with optimizer that allows you to enter in full polar data. Critical to making meaningful correction above 300-500hz.
Thank you for good insights Matthew. It would be very interesting to know more about your approach, tools, challenges, achievements and result. I quess there is a large chance that you have written about it here in the forum. Is there a thread that you recommend?
One question though - is a polar response a 2D measurement of the frequency curve or is it some kind of plot of the wave nodes or something else?
Thank you for good insights Matthew. It would be very interesting to know more about your approach, tools, challenges, achievements and result. I quess there is a large chance that you have written about it here in the forum. Is there a thread that you recommend?
One question though - is a polar response a 2D measurement of the frequency curve or is it some kind of plot of the wave nodes or something else?
a Polar plot is actually a 3D plot as it displays the amplitude by frequency at different angles. This is critical for applying EQ to a speaker because otherwise you may make the response worse at some angles than others or attempt to EQ stuff like diffraction. You should never take in room measurements at a listening position and apply EQ to that above say 300-500hz, the main reason being that what you have measured is a mix of the direct sound and all the reflected sound along with all the room's comb filtering. What you EQ is captured by a mic that doesn't hear like you hear and so the correction is highly likely to just make things worse. However, if you characterize the speaker in an anechoic like environment, then you remove the effect of the room. If you do this at various radiation angles, you then know what the direct sound, listening window sound, and reflected sound all looks like and can be sure that the EQ you apply doesn't make any of that worse/bad.
Because the data can be tricky to visualize, I am sharing some different examples.
Dayton Audio MK402 Polar graph over 90 degrees. Note that the amplitude window of this graph makes some serious response issues a little hard to see. What is going on between 500hz and 10khz is actually pretty bad. This speaker is pretty good for the money. Add some EQ and its quiet listenable for a very cheap speaker. However, in asbolute terms, it's a very poor measuring speaker with classic dispersion flaws. The rate of change in the response across various angles is different at different frequencies, it isn't constant, so the smoothness of the overall response at different angles is different. That means, for example, that the tonality of the speaker changes as you move your head. Worse, it also means the tonal balance of the room reflections won't match the tonal balance of the direct sound. This is a bad thing, and also happens to be partly why absorbing first reflections can be bad (they don't fully absorb and they absorb selectively, so there is still a reflection, but it's response is far from that of the direct sound).
Same speaker visualized differently. Now we are including the directivity index (the ratio of off-axis sound to direct sound. Basically, if it is flat, the response shape doesn't change as radiation angle changes and this is a good thing always. Again, this speaker isn't great, but also isn't terrible. The tweeter beams and there is some weird behavior in the upper end of the midbass.
Similar to the above graphs, but for a very good waveguide. What you see in this graph is that the response starts to get rough as you move off-axis. This is caused by reflections in the waveguide along with some diffraction. At 10khz notice a peak that shows up in every response, but gets worse at certain angles. This is probably a cavity resonance or early reflection in the mouth, and as such, is something you could EQ. Notice a big peak at just over 20khz, that is the resonance of the breakup mode in the dome of the CD. That is also constant with angle, so again, you could apply eq to that. Notice that at 2.8khz or so there is a roughness and peak that develops off-axis, but its precise placement actually shifts a little with angle and it gets worse. That is likely a diffraction effect, you can't and shouldn't EQ that. An in room measurement could potentially show that problem and you might attempt to EQ it, but as you can see here, if you did, it would ruin a perfectly good response at most of the other radiation angles.
Ok one last method of showing polar data. Because you want to have a lot of polar data and because that becomes near impossible to visualize and read, this approach has become popular. It's more of a graphic than a graph.
This particular approach is too hard to achieve by folks like you and I, Earl Geddes did this using a very specialized scanning approach and a lot of fancy math I largely do not understand. That is less important that the graphic however, looking at this graphic you can see that the color gradation shows a great deal of detail in the response. This makes it a lot easier to see the effect of EQ on the overall response. Any of those circles you see in the response, small areas either directly on axis or paired on either side (that is a 180 degree plot) is something that is not constant across the radiation angle, and so if you applied EQ to fix that, it would cause problems. Make things worse.
As for where I've shown how this is done, if you look at my videos over in Techtalk on REW, it very lightly covers it. It's a topic I didn't feel most people would care much about. I actually added it to REW to show that it's possible but also as a means of giving a cautionary tale. I wanted to point out that you shouldn't apply EQ across a wide bandwidth because at mid and high frequencies the EQ is likely to be wrong. I wanted to show that there is a way to do it, but it's a lot of work that most people won't bother with.
To do this I use VituixCAD 2. I take a series of measurements at 5 degree increments under semi-anechoic conditions. You only need the response to be accurate down to about 500hz, so what you do is figure out the best means to measure the speaker such that the first reflection path length is longer that 500hz. Based purely on calculations this would seem easy to achieve. 500hz is 2.28 feet which is a 2ms window. However, you will find that if you apply a 2ms window the response is too smoothed between 500hz and 1khz to be useful for eq. What that means is that you really need to avoid reflections down to half of that, 250hz, closer to 5-6 feet, and 4ms. For me, I can only do this outside as my house as 10 foot ceilings and relatively small rooms. There is no space in my house in which I can place a speaker and get clean measurements down that low. That means if I need to measure inside (like I need low noise or I'm feeling lazy) I will accept a clean response to a higher frequency and shorten the gate window.
Once I have the rig setup and I get clean measurements down to my target lower limit, I just rotate the speaker in 1 degree, 2.5 degree, or 5 degree increments. The resolution is up to what I want/need. For EQ, you can actually get away with being fairly course. I might choose 2.5 degrees over the first 30 degrees and switch to 5 degrees over the remaining 60. I then label these carefully and save them out as FRD files. I then import these to VituixCAD and it generates the plot for me. I create a response curve for optimization including a DI curve and Power response curve. I then add the desired number of biquads (that is the digital EQ bands you will apply) and set them up for what I need, Shelf, PEQ, etc. I then apply values that look close to what I want. I run the optimizer and look at the final results. It often gives overly precise results so I will sometimes reset everything to match acceptable results for my EQ tools. Change the Q to the next closest whole number, same with frequency and amplitude.
As with everything EQ related, I don't recommend adding a lot of EQ to a speaker, even using this approach. I just don't think that level of flattness is audible and it runs the risk of other problems. The dayton speaker, as noted, doesn't measure very good, yet I managed a very listenable response with just a handful if biquads.
My polar of that speaker
and the VituixCAD 2 output
I want to say I achieved this with just 3-4 biquads.
Wow, thank you for a fantastic thorough explanation of DSP correction of speakers in practice. Here is a lot to digest. I will think twice before (if ever) I go this path. If I do so, the scope will definitively be the lower frequencies.
It is still annoying knowing that there in some cases are areas for improvement but that it is very difficult to achieve it.
Somehow I cannot understand why the speaker manufacturers not design and publish some filters based on data gathered with their perfect measurement facilities. It is quite many people today that have some kind of possibility to add transferfunctions/filters in their audio chains (JRiver, miniDSP etc).The speaker manufacturers could even provide it as a standardized industry cloud solution where the amplifier reads the parameters automatically.
Wow, thank you for a fantastic thorough explanation of DSP correction of speakers in practice. Here is a lot to digest. I will think twice before (if ever) I go this path. If I do so, the scope will definitively be the lower frequencies.
It is still annoying knowing that there in some cases are areas for improvement but that it is very difficult to achieve it.
Somehow I cannot understand why the speaker manufacturers not design and publish some filters based on data gathered with their perfect measurement facilities. It is quite many people today that have some kind of possibility to add transferfunctions/filters in their audio chains (JRiver, miniDSP etc).The speaker manufacturers could even provide it as a standardized industry cloud solution where the amplifier reads the parameters automatically.
I agree, the changes needed are immense. Manufacturers would need to be willing to provide detailed measurement data on their speakers. If you think about it, the marketing of this would be troubling. "Our speakers are great, they sound better than everyone else, but we built in flaws that you can correct with these filters." That is the reality, but...all it would take is for another manufacturer to provide less filters and advertise their speaker as superior, even if it's not. On top of that, it would require a standardization of measurement and sharing, something that isn't true of domestic speaker manufacturing. A number of speaker manufacturers don't have anechoic chambers and don't design based around those. In other words, they don't have perfect measurement facilities at all.
I admit to sometimes providing too much information to act as a kind of lesson. One of my worries is that the ease with which we can measure and implement EQ has lead to over-reliance and under-analyzing the system before applying this eq. The reality is that it is very hard and very time consuming to properly apply EQ to a speaker, and because of the workload to do it, easy to screw up and introduce errors. It's also critical that the microphone have a characterized and linear response out to as far as you plan to apply EQ. Many of the cheaper mics have some inaccuracies, especially if no correction filter is provided. A lot of people are better off using a good automatic room correction system that builds in the necessary analytic capabilities to develop good EQ. DIRAC is a good example of this and if used correctly, I find it works well.
In my review of the MK402 I applied DIRAC vs manual EQ and did some listening tests. I found both sounded similar and better than no EQ, but one was much easier.
No problem it was a long shot. I guess that things were regulated by some kind of Non Disclosure Agreement went TAG was sold? Nevertheless, my Calliopes still rocks. But I can admit that I not took them to my heart until I begun to room corrected them. First then showed the bass modules what they were capable of (without correction are they boomy).
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
