Using Phase to remove room modes - single speaker system

[dweeeeb2]

The other day I saw something that I do not understand, and I am hoping someone here can enlighten me.

I was in a chat group talking about using phase to correct for room modes and speaker interference. Someone stated that in a single speaker system phase can be used to correct some room modes.
They showed some data for one of his speakers.

The group delay plot had several troughs/peaks, and the FR plot had troughs corresponding with the irregularities on the GD plot.

He then showed the after-correction plots. I believe he had used an allpass filter on one of the GD troughs. The corrected GD and FR graphs had been smoothed, ie it looked like a room mode had been corrected.

I do not understand how this has occurred. To my thinking any phase changes at the speaker just propagate out through the reflections and you end up with the same interference.

I'm sorry I don't have the plots but does anyone know how this has occurred? Have I missed something?

Thanks

 

[Keith_W]

OK let us assume you have one speaker producing a sine wave. At the listening position, two waves arrive (well, more than 2 but let's keep it simple for now):

1. The direct sound from the speaker
2. The reflected sound from the wall. This is delayed by several ms, depending on the extra distance the sound has to travel.

If the delay from the reflected sound is sufficient to create a 180deg phase misalignment, the two waves will cancel at the listening position and produce a null. If we want to correct this null, we can not simply boost the amplitude at the null because you are feeding energy into an infinitely deep hole and it will cancel. The energy is injected at the wrong point in time.

In the real world, true nulls rarely happen. Something like a 150deg - 170deg phase misalignment will still produce a null, and these can be partially filled by boosting the amplitude because the null is not infinitely deep. But let us put that aside for now and only look at a simplified model where there is true phase cancellation. This stuff is hard enough without the real world muddying up the picture.

In DSP terms, the null is a region of maximum phase. As we know, maximum phase has a zero outside the Z-plane unit circle and it can not be inverted, because inversion turns the zero into a pole outside the unit circle. A pole means that infinite energy is required to fill the hole.

The solution is to deal with the null before it happens. So, we get the speaker to send out two sounds:

1. Send out sound with the phase adjusted, so that when it reflects and arrives at the listening position, it will be in phase with the direct sound.
2. Delay the main sound for an appropriate amount so that it arrives at the listening position at the same time as the reflected sound.

If you do this, the energy is injected at the correct point in time, and the null will disappear. The ability to do this is a property of linear-phase FIR filters only.

 

[antcollinet]

1. Send out sound with the phase adjusted, so that when it reflects and arrives at the listening position, it will be in phase with the direct sound.
2. Delay the main sound for an appropriate amount so that it arrives at the listening position at the same time as the reflected sound.

But won't both these sounds be reflected in exactly the same way as the original sound, arriving at the listening position in anti phase with its direct counterpart, so that they each cancel themselves out? (Assuming they are both the same frequency as the original)

Or am I missing something?

 

[voodooless]

Or am I missing something?

Nope, you are not: if I have a tone from a single source, and I delay it, it's still the same tone. For the room, it doesn't matter when I play the tone.

If you have multiple physical sources, then it does matter because the sources can interact with each other. Now you can have one source act as multiple, but you cannot have it separate direct and reflected sound. It will always be a mix of both. So you can indeed fix part of the reflected sound by baking in the compensation into the signal with FIR, but it only works for a single listening position, and still won't fix your nulls.

 

[KSTR]

You cannot correct the physical room mode itself but you can change the way how it is excited.

The only successful case where one and the same speaker can be used as the main source and the correction source is the so-called virtual bass array. When a speaker, located on the front wall, emits a bass wave, the wave is reflected at the rear wall and bounces back to the front wall, where the speaker then can inject the proper cancelling wave so that it doesn't get reflected again. This turns a bouncing room mode into a single reflection.

The gotcha is that you have to do the same with the correction signal, and so on. This only works if the reflection has lowered its level when it reaches the speaker again and/or if you are OK with partial cancellation (still much better than doing nothing), and preferably the front wall speaker is covered with a distributed array and playing mono signal so that you get sort of a one-dimensional plane wave, thus this is a technique for subwoofers.

But of course you can never fix any deep null (deep means more than 10dB or so) as you just pump more useless energy into the room boosting the mode outside of the null even further. And you can do nothing against the basic specular single reflection which gives you a so-called Allison null when the total path length is is half a wave-length (or any of its odd multiples).

That said, proper digital "room correction" can still go a long way to even out the magnitude and phase response except for the nulls one is experiencing at the listening position.