• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Meridian inversion PEQ method - comments or caveats?

klettermann

Active Member
Joined
Mar 2, 2022
Messages
134
Likes
93
Location
Rocky Mountains
In this thread @LDG_ recommends the EQ method described by Meridian Soundworks as described HERE. The write-up is very detailed and informative and involves various manipulations of REW measurements to create EQ filters for DSP. In a nutshell the process is:
  1. Take room measurements - single L and R measurements at MLP.
  2. Convert measurements to smoothed minimum phase copies
  3. Design the target curve and make a minimum-phase copy of that
  4. Generate inverted versions of MP measurements from step 2.
  5. Export the inverted curves as a stereo .wav file at desired sampling rate(s). These are filters.
  6. Export filters to convolver (MiniDSP Studio in my case)
Can anybody comment on this method? First, it just uses MLP speaker measurements, which would be a limitation for many I suppose. Is this somehow better than just using REW's EQ funtionality to generate filters? And what's the purpose of the minimum-phase conversion? Finally, it seems rather involved as opposed to automated methods like Dirac. Regardless, I guess I'll try it and see what happens. In the meantime, thanks in advance for any remarks or suggestions. Cheers,
 
@UliBru might want to comment and share some of his knowledge as well
 
If the correction is made only to the measurement converted to minimum phase (i.e. we are not correcting phase beyond minimum phase which PEQ's will automatically correct), what benefits are we getting over using PEQ's which are much more computationally economical? It is not like we need to correct every little wrinkles in the frequency response, and 10 PEQ's should be more than enough.
 
Last edited:
that's @OCA's area of expertise :D
The method described in that link is based directly on one of my older video tutorials. While I didn’t have the chance to read the entire post, I didn’t notice any reference. That said, it’s a solid approach, though the added complexity of generating a minimum-phase version of the target curve is unnecessary and best avoided.

I’ve also recently moved away from using direct inversion because the sharp edges of inverted filters can introduce minor phase and group delay issues—even in their minimum-phase versions. Instead, I recommend using REW’s minimum-phase auto EQ filters. These filters are much easier to implement in a MiniDSP and more practical overall. Exporting a convolution filter as described in the link isn’t suitable for MiniDSP due to its limited tap count; that approach works better with platforms like Roon or EQ APO, which can handle unlimited taps, as shown in my video.

In summary, working on the minimum-phase version of the speaker response is absolutely the correct approach since it represents the only truly correctable part of the steady-state response. However, sticking to REW’s EQ filters, applied to the bass frequencies of an "unsmoothed" min phase speaker response, will yield better results.
 
The method described in that link is based directly on one of my older video tutorials. While I didn’t have the chance to read the entire post, I didn’t notice any reference. That said, it’s a solid approach, though the added complexity of generating a minimum-phase version of the target curve is unnecessary and best avoided.

I’ve also recently moved away from using direct inversion because the sharp edges of inverted filters can introduce minor phase and group delay issues—even in their minimum-phase versions. Instead, I recommend using REW’s minimum-phase auto EQ filters. These filters are much easier to implement in a MiniDSP and more practical overall. Exporting a convolution filter as described in the link isn’t suitable for MiniDSP due to its limited tap count; that approach works better with platforms like Roon or EQ APO, which can handle unlimited taps, as shown in my video.

In summary, working on the minimum-phase version of the speaker response is absolutely the correct approach since it represents the only truly correctable part of the steady-state response. However, sticking to REW’s EQ filters, applied to the bass frequencies of an "unsmoothed" min phase speaker response, will yield better results.
Hi @OCA , what is the "REW’s minimum-phase auto EQ filters"?

"However, sticking to REW’s EQ filters, applied to the bass frequencies of an "unsmoothed" min phase speaker response, will yield better results."
CMIIMW, the step that you mean is: measure speaker FR -> generate min. phase copy -> EQ in REW to the target curve -> and apply the filter in miniDSP or equalizerAPO?
 
Hi @OCA , what is the "REW’s minimum-phase auto EQ filters"?

"However, sticking to REW’s EQ filters, applied to the bass frequencies of an "unsmoothed" min phase speaker response, will yield better results."
CMIIMW, the step that you mean is: measure speaker FR -> generate min. phase copy -> EQ in REW to the target curve -> and apply the filter in miniDSP or equalizerAPO?
Yes, untick "drop filters if small" default option in REW/Preferences while you are at it. But do not eq above 200-250Hz.
 
  • Like
Reactions: Tio
In summary, working on the minimum-phase version of the speaker response is absolutely the correct approach since it represents the only truly correctable part of the steady-state response
If I'm not mistaken the majority of tutorials/guides/ASR doesn't mention working on minimum phase though, any idea why?

edit probably because you're talking about EQing speaker itself here, not room?
 
Last edited:
If I'm not mistaken the majority of tutorials/guides/ASR doesn't mention working on minimum phase though, any idea why?

edit probably because you're talking about EQing speaker itself here, not room?
AFAIK, nobody used the technique before. It's actually based on a very basic and hard to argue with idea so it's interesting that it's undocumented.
 
The method described in that link is based directly on one of my older video tutorials. While I didn’t have the chance to read the entire post, I didn’t notice any reference. That said, it’s a solid approach, though the added complexity of generating a minimum-phase version of the target curve is unnecessary and best avoided.

I’ve also recently moved away from using direct inversion because the sharp edges of inverted filters can introduce minor phase and group delay issues—even in their minimum-phase versions. Instead, I recommend using REW’s minimum-phase auto EQ filters. These filters are much easier to implement in a MiniDSP and more practical overall. Exporting a convolution filter as described in the link isn’t suitable for MiniDSP due to its limited tap count; that approach works better with platforms like Roon or EQ APO, which can handle unlimited taps, as shown in my video.

In summary, working on the minimum-phase version of the speaker response is absolutely the correct approach since it represents the only truly correctable part of the steady-state response. However, sticking to REW’s EQ filters, applied to the bass frequencies of an "unsmoothed" min phase speaker response, will yield better results.
Huh. What I get from this is that the inversion/minimum phase approach is solid yet best avoided. And then the min phase is the absolute best yet using REWs filters yields better results. I certainly don't understand it well enough to grasp these apparent contradictions but, the way it's stated, it seems pretty nonsensical. So, at the end of the day I guess I'll stick with REW and MSO EQing. Maybe Dirac too but so far just REW and MSO seems to sound best. Cheers,
 
Huh. What I get from this is that the inversion/minimum phase approach is solid yet best avoided. And then the min phase is the absolute best yet using REWs filters yields better results. I certainly don't understand it well enough to grasp these apparent contradictions but, the way it's stated, it seems pretty nonsensical. So, at the end of the day I guess I'll stick with REW and MSO EQing. Maybe Dirac too but so far just REW and MSO seems to sound best. Cheers,
Inversion on target and REW auto EQ on target are pretty much the same operations. And what I said to be avoided and unnecessary is about creating a min phase version of the "target curve" not the "speaker" response.
 
Back
Top Bottom