Problem with integrating a Flex Balanced into a 2.1 System. Need Advice!

[KEM]

I bought a Flex Balanced to see if it could do better than manual tweaking a 31 band Graphic Equalizer. I use REW and a miniDSP mic to make my measurements. No Dirac. This is my first foray into DSP. The learning curve was steep, but I was able to generate PEQ files in REW and use them with the Flex, but I am not an experienced user. I am, however, quite experienced and comfortable using REW.
The problem I have is that after successfully EQing both channels + sub separately using the Flex, I have a big dip with both speakers active at 50 Hz (Red trace). Please see attached graph and note that the traces have been separated to see the problem better. On this graph I also plotted the REW calculation (Blue trace) and my best effort at manual GEQ tweaking (Green trace). Please note that 50 Hz comes from the sub only as the crossover is at 120 Hz (24 dB/octave LR both speakers and sub).
I very carefully integrated the sub and then very carefully used the DSP to equalize both channels separately and to the same amplitude. The PEQ filters were very different since the non-equalized responses are quite different (not shown in graph). The actual EQ is identical to the dB Average calculation except for the big, wide measured dip at 50 Hz. I also carefully time aligned the main speakers (by moving one back and forth until the impulse response for the system looked the same as the response from one speaker.
My first thought was that the dip is caused by interference between the direct and reflected sound or a time mismatch.
Note that the manual tweaking with a Graphic Equalizer creates a very competitive FR.
I have removed the Flex. The original analog GEQ and crossover were both placed back into the system.
Neither manual GEQ nor DSP PEQ affected the time domain at all, not even a bit. Eventually, I would like to firm up the impulse response.

Please help me understand what happened when I integrated both speakers. What caused this gigantic dip?

 

[NTK]

The dip may be caused by a null at the listening/measuring position. At low frequencies where FR is greatly influenced by room modes, the response at the listening position is a combination (product) of the room modes at the speaker (source location) and at the listening position (receiver location). If you are sitting in a null, you can't solve it with EQ. You may be able to use acoustic treatments to alter the low frequency sound reflections behavior of the room, use double bass array if you have a rectangular room, or near field subs where you can increase the share of the total sound from direct sound vs reverberated sound. However, you may also just ignore it. Narrow high Q dips are usually not very audible.

 

[KEM]

The dip may be caused by a null at the listening/measuring position. At low frequencies where FR is greatly influenced by room modes, the response at the listening position is a combination (product) of the room modes at the speaker (source location) and at the listening position (receiver location). If you are sitting in a null, you can't solve it with EQ. You may be able to use acoustic treatments to alter the low frequency sound reflections behavior of the room, use double bass array if you have a rectangular room, or near field subs where you can increase the share of the total sound from direct sound vs reverberated sound. However, you may also just ignore it. Narrow high Q dips are usually not very audible.

Hey, thanks for your suggestions!
I have about 1/3 of the wall space covered in traps, especially in front and back of the listening position. There is a room mode near the 50 Hz dip. It is a ceiling or floor bounce. I do not have traps there. As I went through the procedure, I used the data for each channel separately. That is, I took the raw, unfiltered sound from one channel (including sub), performed the EQ in REW, imported into Device Console in the Flex. Then I did the next channel. Spectra from each channel showed that the EQ result for both channels separately was very close to the target. The target was the same for both channels. The exact PEQ filters were different since the raw speaker outputs were very different. The dip came when I ran the flex with both channels active.
The problem is confined to the region around 50 Hz, which is exclusively in the sub domain. Perhaps some error occurred because of my procedure. Maybe I should EQ the sub (only one) separately from the main speakers, then EQ the main speakers separately as I did before. Then put all of this together (two PEQ files for the mains and one for the sub).

 

[NTK]

I now see that there was no 50 Hz dip for the L or R speakers, meaning that the problem is probably with the sub sitting at a 50 Hz null. Deep nulls usually can't be EQ'd away because they are caused by destructive interferences (the reflections from different walls come back with the opposite phase and cancel each other).

May be you can try moving the sub. The 1/4 wavelength of 50 Hz is 1.7 m (= 0.25*343/50), which is the distance of a null to a peak. So you'll probably need to move 0.25 m or more (in the right direction) to start seeing significant effects.

Or add more sub(s).

 

[KEM]

I now see that there was no 50 Hz dip for the L or R speakers, meaning that the problem is probably with the sub sitting at a 50 Hz null. Deep nulls usually can't be EQ'd away because they are caused by destructive interferences (the reflections from different walls come back with the opposite phase and cancel each other).

May be you can try moving the sub. The 1/4 wavelength of 50 Hz is 1.7 m (= 0.25*343/50), which is the distance of a null to a peak. So you'll probably need to move 0.5 m or more (in the right direction) to start seeing significant effects.

Or add more sub(s).

The sub is in a really good spot in my room.

I've attached the spectrum of the Left Ch after PEQ in REW. You can see that the bass response is quite flat. For some reason, I seem to have erased the Right Ch result, but it was quite similar to the Left. Only when the two channels were combined did that dip occur.

When I performed the EQ in REW, I did it on the Left+Sub for one PEQ file and then Right+Sub for the other PEQ file. I'm thinking that this method is flawed. It is not the one that is described in the miniDSP tutorial. I'll have to re-do everything using that method.

 

[NTK]

The standard recommendation is to only EQ the frequency below the Schroeder frequency (~200-300 Hz) for room modes, and above to EQ the listening window (or on-axis) response to flat if you have access to high quality anechoic measurements of the main speakers.

So it is reasonable to have different EQ's for left channel, right channel, and sub below the Schroeder frequency with the goal of getting the combined response to match your target curve*. Above Schroeder, the left and right channels should have the same EQ. Otherwise the tonal balance of the direct sound (which is the most important sound) of the left and right speakers will be different, and likely won't sound very good.

It seems strange to me that the 50 Hz dip only showed up in the combined response, since the left and right speakers shouldn't be producing any significant output when your crossover is 120 Hz with 24 dB/octave slope (50 Hz is 1.26 = log2(120/50) octaves from 120 Hz).

[Edit] *Note:
It is not easy to optimize multiple different channels of EQ for the combined output to meet a target response. Since the sum is a complex sum that must include amplitude and phase, more complex techniques will be required (but should still be simpler than MSO as I assume we are optimizing for one listening position only).

 

[KEM]

The standard recommendation is to only EQ the frequency below the Schroeder frequency (~200-300 Hz) for room modes, and above to EQ the listening window (or on-axis) response to flat if you have access to high quality anechoic measurements of the main speakers.

So it is reasonable to have different EQ's for left channel, right channel, and sub below the Schroeder frequency with the goal of getting the combined response to match your target curve*. Above Schroeder, the left and right channels should have the same EQ. Otherwise the tonal balance of the direct sound (which is the most important sound) of the left and right speakers will be different, and likely won't sound very good.

It seems strange to me that the 50 Hz dip only showed up in the combined response, since the left and right speakers shouldn't be producing any significant output when your crossover is 120 Hz with 24 dB/octave slope (50 Hz is 1.26 = log2(120/50) octaves from 120 Hz).

[Edit] *Note:
It is not easy to optimize multiple different channels of EQ for the combined output to meet a target response. Since the sum is a complex sum that must include amplitude and phase, more complex techniques will be required (but should still be simpler than MSO as I assume we are optimizing for one listening position only).

Yes, I've come to the same conclusion that you have in your Edit note. I'll plan on following the miniDSP instructions next time. Thanks!

 

[DWPress]

Also give the MMM approach a try. You won't get the time domain information but it'll give you a better idea if that 50Hz null really a problem or just mic position.