Whatever Happened to Ambisonics?

The Ambisonic surround sound system was developed inBritain during the 1970s, hard on the heels of the so-called 'quadraphonic'techniques--and became tarred with the same brush. For a number of reasons,more political than technical, it has up to now only received limitedacceptance in the consumer and professional audio market places. But now allthat is changing. Major interest from Japanese hi-fi manufacturers, and thecurrent interest in encode-only stereo enhancement are bringing the system backinto the limelight. But according to Richard Elen who has been working withthe system for over 15 years, it never went away.

"Just as the accurate reproduction of performed music is the crucial test of audiofidelity, so the ability to reproduce correctly the directionality of naturalsounds is the crucial test of a surround sound system. Unless it can do this,there will not be the correct disposition of indirect sound which provides theacoustic ambience of the performance and gives the position-dependent labellingof direct sounds by their wall reflections, which is an important aspect of theappreciation of music.

"If a system can cope with this difficult task, it shouldgo without saying that it can easily deal with the relatively simple problemsof synthetic source material. A system of surround sound which is able toreproduce the directionality of indirect reverberant sounds, as well as ofdirect sources, is termed 'Ambisonic'"

Ambisonics was the brainchild of a small group of British academics, notably Michael Gerzonof the Mathematical Institute in Oxford, andProfessor P B Fellgett of the University of Reading. From the beginning, itwas designed as a surround sound system that would overcome the major problemsof the so-called 'quadraphonic' systems that were its predecessors -- the mainone being that they simply didn't work very well. Research rapidly indicated,however, that in addition to providing full surround sound in an encode/decodeenvironment (where the original recording is encoded into astereo/mono-compatible form for transmission and later decoded by the listenerinto multiple speaker feeds), Ambisonics could also offer a significant 'superstereo' capability without decoding. With current interest in single-endedstereo enhancement techniques like RSS and QSound [see Audio Media April andAug/Sept 1991 respectively], it's interesting to note that Ambisonic processingequipment has been used as a single-ended stereo enhancement device by radiostations, now especially AM stereo stations in the States, for almost adecade.

Ambisonics built on the astonishing work on stereo recording and reproductionperformed by Britain's early audio genius, Alan Dower Blumlein. Blumlein wasworking on stereo recording and disc-cutting in the Twenties, and as well asdeveloping the stereo cutting system introduced over 30 years later formicrogroove stereo LPs, he also invented what is at once the simplest and mostaccurate of all stereo recording systems: M-S coincident pair recording.

At a time when Bell Laboratories in the States were also investigating stereo,but with omnidirectional spaced microphones (which left a hole in the middlethat required a third, centre channel to fill it in - a precursor of Dolbysurround?), Blumlein realised that there was more to the ear/braincombination's ability to position sound source in space than merely thedifference in level between the ears. The principle is easily illustrated byconsidering a conventional mixing console panpot being used to pan a monosignal between two speakers -- an illustration that indicates, too, how littleBlumlein's work is now remembered in the audio industry.

Panpotted Mono

Just imagine that you're listening to a sound that's centre-stage. It has equallevels on both channels, and your meters will read identical values. But nowmove to the left and what happens? The sound follows you to the left, becausenow there's more energy reaching you from the left than the right. That's themain drawback of this system--because it's not 'stereo' at all: it's 'panpottedmono'. Send all the signal to the left speaker and it comes from the left. Send equallevels to both speakers and it's in the middle--or is it?

The standard stereo listening position, with the speakers at 60 degrees toeach other. Further apart than this and stereo begins to develop a 'hole in themiddle.' Normal panpots operate with level only, so the listener in thisexample hears the sound coming from the left, because there is more levelarriving from the left-hand speaker.

Now the panpot is central, but the listener has moved to the left. The soundstill appears to be coming from the left, because there is still more levelarriving from the left-hand speaker.

Phase-Shift Panning

If a sound is off to one side, we still hear it with both ears, but there is adifference between the signals arriving at the two ears. Apart from differencesin level (and high frequency content for that matter), there's another factor:phase. The wavefronts from the sound source don't reach the ears at exactly thesame time, and we interpret that phase difference as localisation information.It's a very impressive effect if you try it yourself in the studio.

A simple method of experimenting with phase-based localisation is to set up apair of delay lines, one variable and the other fixed. Send the same monosignal to both of them and pan the output of one hard left and the other hardright. Make sure that only the delayed signal is delivered to the output--noneof the input signal should be heard--and that the levels from both DDLs isidentical (eg. by setting them up on your console metering). Set the delay onthe fixed DDL to, say, 100 milliseconds. Set up the other delay to a basiclength of 100ms too, but with a knob to vary the delay equally above and belowthis figure--say 100ms +/-50ms. Now vary the delay back and forth either sideof the 100ms position and either side of the 100ms position and you'll hearthat, without changing the levels at all, you can create a remarkable panningeffect. You'll notice that at some settings, sounds can even seem to go waybeyond the speakers...

Switch your monitoring into mono while you do this, by the way, and you'll heara familiar sound--the 'swooshing' effect of true 'tape phasing' or 'flanging'.This was how, using tape machine record-play head delays instead of DDLs,George Chkiantz at Olympic Studios produced the original sound of 'phasing' onthe Small Faces hit, Itchycoo Park--possibly its first controlled use (theeffect had been used on the soundtrack of a Fifties movie, The Big Hurt, butthis was done by running two identical copies of a piece of music together andchanging the speed of one of them to bring it into sync--a rather haphazard wayof creating the effect).

Setup for phase-shift panning. Use this in mono to obtain flanging.

Echoes And Delays

Taking a signal, splitting it and delaying one path, then positioning the two signals hard left and hard right gives a useful effect. However, for the level onboth sides to sound the same, the delayed channel must be louder to overcome HaasEffect, which is trying to tell you that the delayed sound is just anecho.

The Simple Secrets Of Stereo

Blumlein's approach, on the other hand, utilised a pair of microphones at thesame point -- a coincident pair. One mic was an omnidirectional type, and thuspicked up everything--in stereo terms, it picked up left plus right (L+R). Atright angles to it, but as physically close to the omni as possible, was asecond microphone, with a figure-of-eight response, pointing to the left. Afigure-of-eight polar diagram means that sound waves hitting one side cause apositive displacement with respect to the other side, and so the signal pickedup is actually the difference between left and right (L-R).

The Blumlein coincident pair--an omni mic crossed with a figure-of-eightpointing left.

You'll notice that this 'stereo' is a bit odd. Instead of a left channel and aright channel, you have a 'sum' and a 'difference' channel. You can't listen tothem directly: they have to be decoded into the more usual left and rightchannels. This is done by a simple matrix. The sum of the two channels givesyou (L+R) + (L-R) = 2L --the left channel. Meanwhile, subtract one signal fromthe other (or simply mix them together, reversing the polarity of thedifference signal) and you get (L+R) - (L-R) = 2R --the right channel.Interestingly, you can simulate this effect without using a sum-and-differencetechnique. Just take two microphones with cardioid polar diagram and crosstheir capsules horizontally at 90 degrees. The effect is virtually identical toBlumlein's technique, and needs no matrix decoding.

Coincident-pair stereo is a remarkable technique. It is perhaps the simplestmicrophone technique that approaches our own hearing in its ability toreproduce spatial information. On speakers at a true 60 degrees to each other(as stereo speakers are meant to be) the sound has remarkable depth--it isn'tjust a straight line between the speakers--and the image is also incredibly stable,sounding more or less the same wherever you are between the speakers, unlikepanpotted mono. On headphones you can actually seem to hear things behind youand, occasionally, even above you. This is not as unlikely as it sounds-- do weneed ears in the back of our head to hear things going on behind us? No, thefront-back asymmetry of our ears changes the characteristics of sounds heardfrom behind us as compared to those in front.

Three-Dimensional Stereo

The Soundfield Mic--an omni crossed with three figure-of-eights atright-angles.

Soon after the development of the Soundfield microphone, developments began tobe made in the field of simulating soundfields as well as simply capturingthem. The result is that today there are comprehensive mixing systems thatallow individual multitrack signals to be panpotted into an Ambisonicpicture--an area we'll look at later in this article.

If a 'traditional' Blumlein M-S coincident pair gives you two signals whichneed to be decoded to derive the left and right speaker feeds, it's fairlyobvious that a three-dimensional Blumlein system will give you more of thesame. In fact, the 'studio format' for Ambisonics, generally known as B-Format,is exactly this: a mono (sum) signal from the omnidirectional component (Left +Right + Front + Back + Up + Down), known as the 'W' component, plus threedifference signals: Front - Back (known as the 'X' component), Left - Right(the 'Y' component), and Up - Down (the 'Z' component). Notice that only fourchannels are needed to encode not only surround information, but also height(Ambisonics with height is generally called 'Periphony'--"sound around theedge"). So why did the old 'Quad' systems need four channels to encode simplehorizontal surround?

Two Quadraphonic Fallacies

At the root of Quad's problems were several misconceptions. The idea was toreproduce a soundfield--which of course exists all around the listener--but theidea that this could be represented by recording four channels and replayingthem through four speakers at 90 degrees to each other around the listener wassimply incorrect. You can obtain some impressive effects, but in terms ofaccuracy, the results are disappointing. One reason is that stereo simply doesnot work with speakers at 90 degrees-- you get holes between them.

At the root of Quad was the idea of using panpotted mono in two dimensions withfour channels, and some of the (so-called 'Discrete' or '4-4-4') systems did nomore than this: utilising sum and difference systems in the same way as theyare used in FM stereo--with subcarriers on vinyl discs! --to get the monocompatible sum signals in the normal groove and the difference signalsmodulated on subcarriers. The listener without a decoder simply heard thebaseband signals--Left Front plus Left Rear on one side and Right Front plusRight Rear on the other--and missed out on the difference signals encoded onthe high frequency subcarriers.

To offer stereo compatibility without subcarriers, many of the several systemsavailable attempted to matrix the original four 'Discrete' channels down totwo, using phase relationships to encode the surround positions, and thensomehow recover the original four signals in the decoding process. Thesesystems were often referred to as '4-2-4' systems--four original signals,matrixed into two transmission channels, and then decoded into the originalfour again. Unfortunately, this is mathematically impossible, and '4-2-2.5'would have been a better name for them. Instead of a sound panned around theroom in a circle actually going around the room in a circle, it would dosomething else. In one case it went around a shallow ellipse, with littlefront-back definition. In another the front stage was fine but the rear was avery odd shape, with centre-rear being in the centre of the listening area.

Quadraphony: discrete 4-channel recording were distributed on 4-track tape,encoded as subcarriers on to disc, or matrixed into 2-channel and decoded

The spatial inaccuracies of the quad systems was one of their majorshortcomings as these attempts to pan a sound in a circle indlcate. Only UD-4got close.

The solution to poor localisation was a system called 'logic decoding'. Theprinciple here was that if there was, say, one sound source playing, the systemcould work out where it was supposed to be and turn the other speakers down.That's fine as long as there's only one thing going on. Of course, there seldomis.

The original Quad systems died out, but two developments of them were leftbehind. One was Dolby Surround, which is now widely used in the film industry.It owes a lot to two of the commonest Quadraphonic systems, CBS's SQ andSansui's QS systems. As its heritage might suggest, it is excellent forimpressive sound effects and ambience but is not highly accurate in itsrepresentations of localisation (it is not intended to be), and it is sometimesquite difficult to work with--logic decoding means that when severalwidely-spaced sounds are present, the sound stage tends to collapse as thelogic decoding is rendered less effective by the multiple sources.

A Working Matrix

The big challenge for Ambisonics was how to get the four sum-and-differencesignal components into a form that was stereo- and mono-compatible, so that thesystem was able to interface successfully with existing systems. This was thechallenge that Quad had failed, both with the expense and difficulty ofsubcarrier systems--with their special styli and loss of subcarrier informationdue to record wear--and with the inability of matrix systems to recover all thesurround information successfully. The answer was a phase encoding matrix thatbrought together work carried out by the Ambisonic team, the BBC, and some ofthe original designers of the UD-4 system. The Ambisonic team had developed amatrix called '45J', and the BBC were doing test transmissions with 'Matrix H'.Adding a dash of UD-4, the UHJ system was born.

Multi-Channel Compatibility

The theoretical path from B-format to the various stereo/mono-compatible UHJvariants. In fact many mixing applications will go straight from multitrack to2-channel UHJ at present.

Although there are some compromises as far as accuracy of localisation isconcerned in the 2-channel UHJ system, it is currently the encoding method ofchoice. UHJ recordings can be transmitted via all normal stereo channels andany of the normal media can be used with no alteration. Compact Disc has thecapability of carrying two additional audio channels over and above the twoused for stereo: these would be ideal for 4-channel UHJ but have as yet to beused for this purpose (there are of course no players with this capability atpresent either). [The emerging DVD standards may allow for multichannel Ambisonicsignals - for details of the most appropriate proposals, see the Acoustic Renaissance in Audio Web site --RE]

At Home With Ambisonics: The Decoder

The solution was to design the Ambisonic decoder in such a way that rather thaneach speaker receiving a single channel feed destined for it from thebeginning, as in Quad--where you had to place the speakers at home in the samerelative positions as they had been in the studio control room--you insteadpositioned the speakers in 'sensible' places, then told the decoder where theywere. The 'layout control' on an Ambisonic decoder, therefore, causes thedecoder to output the correct speaker feeds for the speaker positions you wouldlike (or are obliged to have). One result of this feature is that you can haveyour front speakers in a normal stereo position.

An extension of this principle is the ability to design Ambisonic decoders forany number of speakers. Four is the minimum for planar surround, and six forperiphony, but in a large environment such as a cinema, it may be a good ideato have a dozen speakers or more. There is no theoretical limit to the numberof speakers. Similarly, there are few limits to speaker positions, either. Yourfour speakers at home or in the control room can be placed in any rectangle,wide or narrow, as long as the ratio of the sides doesn't exceed 2:1. Andbecause Ambisonics tries to recreate the original soundfield, speakers tend towork together and thus smaller speakers are often more effective for Ambisonicreplay--they give more accurate localisation across the frequency range becausethe drivers are closer together, and they tend to exhibit better bass responsethan when the same speakers are used in stereo. A typical monitoring setup in acontrol room, therefore, is to use the main speakers for checking the stereoand four nearfield monitors connected to a decoder for Ambisonic monitoring.Most decoders have a bypass facility too, which enables the input signal to bemonitored on the front pair of speakers only, so stereo and mono nearfieldmonitoring can also be carried out.

Simplified block diagram of a planar-surround Ambisonic decoder

Part II of this article discusses "what happened?" to Ambisonics, and how it is being used in the real world.

For more on Ambisonics, see The Ambisonic Index at http://www.ambisonic.net