Immersive audio is the buzzword across the industry right now. Anthony Grimani looks at what it means for integrators moving forward.

Immersive audio is everywhere, including at CEDIA 2023, where CEDIA and the Consumer Technology Association released its immersive audio recommended practices. The 148-page PDF is called CEDIA-CTA-RP22 and may be downloaded free from the CTA’s website. Learn it, live it, love it.

There are lots of good nuggets in there, but I want to focus on one which is kind of the icing on the cake – the thing that makes the whole long process come together, the final link between engineering and ears.

Ultimately, the goal of an immersive system is to make listeners believe that sounds can emanate from any point in space around them. We can’t put speakers literally everywhere, so we must rely on phantom imaging to create virtual speakers between the physical speakers. RP22 tells you all about how many you need and where to put them to accomplish this in various size rooms and at several different performance levels; that’s great, but how do you verify that all those phantom-imaged virtual speakers are actually doing what they are supposed to be doing? Can you measure it? Can you plot it? Yes… and not really.

Before we get into that, let’s cover a few basics from RP22 Section 6 for those who, inevitably, will have somehow managed not to download and read a free document that tells them how to do their job properly.

To make phantom imaging work, all the speakers need to have the same frequency response, the same phase characteristics, the same SPL and the same distance/delay. Not that frequency response and SPL aren’t important, because they are, but I want to focus specifically on delay and phase for a minute.

You may think delay, which commonly masquerades as speaker distance, is easy. You measure the distance to each speaker and plug that number into the surround processor. It’s not that simple; keep a few things in mind. First, the acoustic center of the speaker isn’t the grille, the front baffle or even the driver surfaces. In most cases, you can’t actually measure to the acoustic center of the speaker, because it’s inside the speaker. Your measuring tape will get you close, but not all the way there… literally. Furthermore, DSP active speakers, including subwoofers or DSP inserted in the signal path in the form of EQs may introduce latency, which effectively moves the speakers farther from you than you actually measure. If you’re lucky enough that the manufacturer of the DSP knows and provides the latency (typically in milliseconds), you could convert that to distance and add it to what you measure with your meter and plug that into the surround processor, but you still can’t reach the acoustic center of the speaker.

Enter acoustic detection of distances using impulse response. This makes a lot of sense, as it does get you to the acoustic center of the speaker and takes into account any latency in DSP. However, it does require a good impulse response, clean acquisition and good computation. And there’s the rub – especially with active subwoofers, where the intentionally limited bandwidth and other factors usually scrub the impulse response and send the algorithm into fits trying to find the arrival.

Fortunately, there are other ways to deal with subwoofer delays, which I won’t go into here, because that’s actually under the bass management topic. Even if the system computes the correct distance from the impulse response, there’s always the possibility that the conversion from distance to actual delay (which is what matters) will not be calculated properly by the processor. This seems like a pretty basic concept, but there was a case in recent history where some enterprising individuals figured out that a well-known processor company was, apparently, using the wrong value for the speed of sound.

No, I’m not making this up. I wish I were. I hope I am.

Anyway, this very long paragraph is to tell you to get as close as you can with the distances/delays, but you must still verify them.  And we haven’t even gotten to phase yet!

Fortunately, we’re only really concerned with phases from 500Hz to 3kHz. Above and below that, hearing doesn’t really interpret sound arriving from multiple sources as “imaging”. If you have phase problems at high or low frequencies, that’s probably ok – at least for our purposes here.

Phase can be affected by both the speaker (crossovers, etc.) and the room, so using the same speaker model everywhere is not a legitimate shortcut – even if you could swing it, practically. That being said, some companies take great care to make sure the phase rotations of their speakers are the same in the critical midrange imaging frequencies. All companies should, really, but the ones that do tend not to advertise it, because it’s not a typical selling point.

(Yes, that is our fault, because we don’t ask. Be honest, did you even think this was a thing you needed to ask about before reading this?)

Even if all the speakers have consistent phase, though, the room can still rotate it all around. Here, you need to pull out REW and do a phase analysis of the speakers from 500 Hz to 3kHz. You can correct any phase misalignments in several ways – often at the same time you are correcting magnitude errors. Handy.

Use a minimum-phase EQ method (like a good ole parametric EQ) to correct phase errors that have a corresponding magnitude error. Minimum-phase is a term that, at least in my opinion, comes up a little short in describing what it actually means – which is, basically, that the magnitude and phase error are associated or linked in such as way that correcting one corrects the other as well. If you measure a phase error that’s not associated with a magnitude error (i.e., non-minimum-phase), then you can use an all-pass network to change just the phase. Once you’ve got the phase all lined up, if you still need to shape the curve some, you can use a fancy linear-phase EQ to correct magnitude without affecting phase – with the caveat that the EQ may phase shift the entire range a few degrees (also known as, you know, delay), requiring a touch-up of the distance/delay settings to compensate. Yay. That again. This is especially true any time you apply correction to one (or several) speakers but not all of them. Some get delayed by the filter processing, some don’t; you have to sort it out. Really earning that money now, aren’t we?

And now we’re finally around to where we started, which is verification that you have done all that work properly. For this, you use your ears. The good news is that it’s fun and easy to actually do.  The bad news I’ll get to in a minute. You use a special test signal for this. It’s gated (meaning it pulses on/off), narrowband pink noise (500Hz to 2kHz). You play this through the pair of speakers you want to test and listen to see if they make a phantom image – a virtual speaker – between them, when you sit at the main listening position or in an average location of all the seats. If you’ve done everything properly, the gated noise bursts will come from a position magically floating in the air between the two speakers. If not, try adjusting the delay, either by incrementing milliseconds in the DSP processing stage or changing the distance setting on one or the other speakers in the pair.

One pair down… well, lots to go. It turns into a bit of a Rubik’s Cube 3D puzzle as you go around the room listening to various phantoms and tweaking the distance/delay settings so all the phantoms land in the right spot.

This can take some time – a lot of time.

You need to plan for at least an hour within your total of two days for audio calibration. Not all of that will be on this one test. Which brings us to the bad news. Over 20 years ago, I made a DVD called The 5.1 Audio Toolkit that included the gated noise signal for all the speakers in a (somewhat contradictory given the name of the disc) 6.1 system using Surround EX matrix processing for the sixth channel. Since that time, I’ve been super lax about updating the disc (or files, as they are these days) for 7.1 and now immersive. My apologies. I’ve been busy. Yes, for 20 years. Anyway, the Toolkit is fantastic for knocking out the most important phantom image tests between – what remain even today – the five main channels. You have to get a little creative for the others.

But wait – there’s more bad news. The Toolkit is out of print; you probably can’t find one new. Yes, I’ll admit that the situation is pretty dire; however, I have heard that there may be some others out there making discs or files with similar test signals. I really hope there are. Your homework assignment is to go find them and tell me about it. Eventually (I hope) I will create a set of files with the gated pink noise rendered to all speaker pairs. It would really help if all of you pestered me about this continually. I give you permission to spam me about it.

In the meantime, there are several things you can do with the existing Toolkit to light up the rest of the speakers. The first is to patch the processor’s outputs to the appropriate amplifier inputs by swapping the cables. I take short patch cables with me in my calibration kit for just such a purpose.  Remember, if you do this, transfer all the relevant settings in the processor from the channels you are testing to the channels you are using for the test.

What’s that, you say? Here’s an example. The Toolkit has the gated half-point noise for the left and side left channels. But you want to test the side left and back left. So, you patch the side left speaker to the front left output and the back left speaker to the side left output. You also need to transfer all the settings (distance, level trims, EQ etc.) for the side left speaker to the front left and from back left to side left. Oh, no. Now that’s a lot of work. Just think about how many speaker pairs there are to test.

Recall that I did apologise for not updating the Toolkit…

Some processors allow you to change the routing inside the processor through a matrix process.  This greatly simplifies the process, but you still must keep your head on straight to make sure all the settings are correct for the speaker pair you’re testing. Oh, and remember to record them and put them all back when you’re finished.

At this point, assuming you’re still reading, I should mention that this is not just important for the main listening position, or Reference Seating Position as RP22 calls it. I won’t try to describe it here, but all the time synchronisation work you’re doing affects how the soundstage falls off away from the main seat as you move to the outside seats. This is not, counter intuitively, something that only affects one person. It actually affects every seat in the house. Otherwise, I might just tell you to forget about it.

To wrap this all up, even after you finish checking all the phantoms, it’s best to listen to a wide range of program material to make sure the time synchronisation of the speakers has truly improved the width, height and depth of the sound field and the ability of the system to spatially resolve all details to their proper places. That is the goal after all.

Yes, I know you think this is a giant waste of time. You’re already saying you don’t have that time to waste, it’s not worth it or your clients will never hear such nonsense anyway. Sorry; that’s not true. Trust me, I hear these excuses from integrators a lot – mainly because they just really don’t want to run all the imaging verification tests.

I probably can’t convince you that any of this matters if you’re already inclined to lean that way, but I’ll leave you with this. Time synchronisation of the speakers is one of the last things I do – one of the final tweaks. I have done tests where I bring the clients in to listen before the final timing checks, then again after. It blows their minds. It’s like the missing piece falls into place. The light bulb goes on. Sound mixing, sound design finally make sense. The fog clears and they understand why they paid big money. I’d like to challenge you to do the same thing. Properly time synchronise one system – even if it’s your own – and listen to the difference. Then maybe try it just once for a client with their system. See if I’m crazy.