When are multiple side surrounds necessary?

I haven't been able to find much about this, so I thought I'd ask here. When is it recommended to use multiple side surrounds (i.e. a side surround pair for each row)?

I assume that's true even when rows are at different levels such as each row being on a different level riser?

That certainly makes sense from an intuitive standpoint. Hard to believe it's that simple Thanks, Dennis.

You have to be careful with how you wire multiples like that. You need to present the receiver with a load that it wants to see. For example, if you wire two 8 ohm speakers in parallel, the impedance becomes 4 ohms. OTOH, if you wire them in series, the impedance becomes 16 ohms. Back to what Dennis mentioned above, IIRC from what I've read in Toole's book, you also want to decorrelate the two sets of surrounds. I've also read that this can be accomplished by adding the correct signal delay for each speaker (hence the pricey DSP required).

On another note, I'm not sure why adding a time delay will "decorrelate" the two signals. It would seem that the two signals would still have a strong correlation, only shifted. Perhaps the term is used in a different context here, or maybe I just don't remember my signal processing as well as I would like

Assuming a uniform phase shift, wouldn't that be the same as a time delay (e.g. an 90 degree phase shift of a sine wave is a cosine)? If you are doing some sort of manipulation of the phase across the spectrum, I don't know what that would do.

IIRC in my signal processing studies, decorrelation is also referred to as "whitening." Which is a reference to what you are doing to the frequency content of the signal. When you decorelate it, you pass the signal through a matched filter, and the result is a signal with a uniform frequency spectrum. Like you would see with white noise, hence the term "whitening." However, I don't think white noise from the second row surrounds would get the job done

Keep in mind that I'm referring to things I haven't studied in years. So I'm afraid there are little truths spread amongst erroneous information. I'd probably be better off keeping my hands taped up and away from my keyboard

I've been thinking about this more, probably more than I should, and this is what I've come up with. If you look at the Fourier transform (or really just it's properties), a time delay results in a phase shift. The amount of that phase shift depends on the frequency. That does not affect the correlation, and the correlation coefficient is still 1 between the two signals. Now, what happens if you apply a 180 degree phase shift to ALL the frequencies? Think reversing the polarity on your speakers. You get a negative correlation coefficient of -1 (same signal just opposite polarity). Finally, if you want to perfectly decorrelate these signals, you essentially randomly change the phase of the signal without manipulating the magnitude of the frequencies in it. Ths will give a correlation coefficient near zero.

How do you add all that random phase to your signal? Apparently that's called reverb.

Edit: I didn't get to finish my thought earlier. My guess is that calibrating a setup like this would add a delay to get the signals aligned in time at the LP, and then a little reverb to decorrelate the signals. How much reverb is probably where all the calibration experience comes into play. From what I've read, this is the same process they use to create a stereo signal from a mono signal and to simulate concert hall effects.

Here's an interesting paper that discusses most of this.
Edited by J_P_A - 1/31/13 at 7:07pm

You may be right regarding the comb filtering. However, I think the phenomena that we are trying to replicate is much more complicated. The paper I linked to gives a good example of this. In a large concert hall, we hear a direct sound from the stage that is nearly perfectly correlated. The sound originates from the same spot and traveled the same distance to our ears on either side of our head. However, there is also sound reflecting from other surfaces in the concert hall, and some of those arrive at our ears a little later. The sounds arriving at the left ear have interacted with different stuff as well as traveling a different distance than the sounds arriving at the right ear. So both of these waves have a different "character" (or internal phase structure) than one another, and have a near 0 correlation. We've learned over the centuries to interpret this information and decide we are in a large reverberant space. Meaning if you were blindfolded and dropped in an concert hall, you would be able to identify that space as large and reverberant based on the sound signature alone.

For our small listening rooms (relative to that concert hall) we don't have the benefit of of all that space to allow our sound to interact with stuff as well as be delayed, so we try to mimic that with our discrete sources. The original mixer will create a track for that single side surround by adding the sound effects that he/she likes, and by adding the appropriate amount of reverb to get the spaciousness that they want (I'm inferring that at this point). We want to essentially mix our own discrete channel, so we add a little delay to account for the path difference as well as add some reverb to further decorrelate our sounds. Again, we're trying to reproduce or predict that "character" or internal phase structure that would have occurred in the concert hall.

It may very well boil down to comb filtering, but that paper suggests that decorrelation reduces the perception of comb filtering. It may be semantics, however.

BTW, I'm using the term reverb, but that may not be the right thing to call this randomization of the phase structure. I'm fairly confident that's what we're trying to emulate, though.

Wow. Talk about diarrhea of the hands. I just can't stop myself today I agree that some HAA classes would be awesome. Maybe we could get a group discount

I'm not sure adding white noise is the ticket. That would certainly help to decorrelate the signals, but you would change the spectral content. The idea is to keep the same spectral content but to change the phase content.

Dennis, very informative paper. Thanks for the reference. I haven't made it to the patents yet, but I'm interested to see what's in there.

BIG, hopefully Dennis will elaborate on that, but from looking at the spec sheets for the DSP it has extensive filtering capabilities, one of which is an all-pass filter. Presumably, this filter will allow the phase variations necessary to decorrelate the signals. That really seems to be the key here.

Quite helpful! Can this sort of calibration be done with REW, or is more sophisticated analysis equipment required?