Practical Ambisonic Clock Outputs Decoder with Special Crosstalk Cancellation

Here is a practical analogue Ambisonic decoder design that's better in many ways than the first one that I built.  If you're that way inclined, please feel free to go down the sensible route and do it all in DSP, or buy such a thing off the shelf.  It is for horizontal surround only.  

Front Panel Controls

Here we have the finished article sitting on top of the amplifiers on a bookshelf, with a suitable spacing device to avoid blocking the vents.  The switcher on top selects output from the TV, or the computer output elsewhere in the room.  That simply connects into the stereo UHJ input on the back.



On the left is an overall gain control for convenience.  The first toggle switch selects between the L R stereo UHJ inputs or the W X Y B-format inputs on the rear panel.  The second toggle switch changes the speaker feed generator from a conventional amplitude matrix to an experimental phased version.  The two 3.5mm mono jack sockets can be connected to remote closing contact buttons which achieve the same thing as moving the toggle switches into the lower position.  This allows A-B style comparisons of the different systems without having to get out of your chair.  The mains power switch is on the right.  This arrangement is much simpler than the V1 decoder with its 24-plus knobs. 

Rear Panel Connections

The L R stereo UHJ inputs are on the right.  The B format WXY inputs are next.  The 12 outputs correspond to speaker positions as seen on a clock face.  That gives you  a possible ideal speaker position every 30 degrees.  The instrument fuse and IEC mains inlet are on the left.

Internal Boards

The whole of the insides.

Front Board

Rear Board

PSU Board

Gosh, axial capacitors.  How posh!

Ambisonic Decoder V2 Schematics

Inputs and Phase Shifters ambi_clock_decoder01-01.pdf



Here's sheet 1, where L+R and L-R is created then given relative phase shifts over a wide frequency range to create (L+R), j(L+R), (L-R), and j(L-R).  Those components are then summed in the appropriate amounts to give the standard decode to produce the B' format signals W',X', and Y'.  The dash is there for a reason.  This isn't true B-format because there's still some phase-amplitude encoding left in it.  It does, however, decode to the correct speaker feed when you put it into a standard B-format to speaker feed amplitude matrix.  What's interesting?  Not much, other than to note that in some classic analogue designs, you'll see such things as 0.1% metal film resistors and 1% weird value custom capacitors.  There is no need for such extravagance.  It's the RC time constant that's important, so you choose a standard capacitor value and tweak the RC by creating a non-standard resistance.  I measure my 5% tolerance capacitors on a meter before choosing the smaller of the resistors.  You might not want to do that in full production, but at least you don't need custom capacitor values.  Anyway, measuring capacitors is what automation, or the summer student is for.

B-Format Phase Shifters ambi_clock_decoder01-02.pdf



This is what you won't see in a normal ambisonic decoder.  I've created a fully flexible set of W,X,and Y, so that you can create a speaker feed with an arbitrary phase.

Phase Shift Coeffs ambi_clock_decoder01-03.pdf



You won't see this in a conventional ambisonic decoder either.  Here, speaker feeds are created in such a way that they conform to the usual amplitudes expected, but opposite speakers are in anti-phase. 

Norm Coeff Weights ambi_clock_decoder01-04.pdf



This is the more conventional B-format to speaker feed decoding matrix which is the default option.

Speaker Sums ambi_clock_decoder01-05.pdf



After going through some switching, all the various resistor matrices are summed here and buffered to the phono output sockets. 

Switching and PSU ambi_clock_decoder01-06.pdf



Here's the standard linear PSU, an overall volume control, and switching for the UHJ / B-format input and the standard / freaky speaker feed generating schemes.  What's interesting?  Nothing, other than to note that I've used 74HC4053 chips to do all the switching.  Wafer switches are awful, especially after about the first two weeks of use, and it's better to keep all the signals on the PCB.  But, aren't 74HC4053s awful as well?  Not if you do it right.  Those cheap CMOS transmission gates do have quite high distortion when the voltage going through them gets closer to the analogue rails, but what did we have on the previous sheet?  Ah-hah!  Virtual ground summing amplifiers.  The CMOS transmission gates are held at zero volts so show low distortion.  They are switching currents.

UHJ Decoder

The UHJ decoder uses standard design practices and equations as detailed in the original design papers on the subject. The phase shifters are implemented with standard 2-pole per op-amp circuits rather than the phase sequence networks used in my previous decoder. This has the advantage of flat frequency response and use of fewer capacitors in their reactive region in the signal chain. Recovered B-format is presented externally on the standard 5-pin male XLR output and an external lead feeds this back into the decoder for normal straight through operation. This can be disconnected and used to feed in B-format sources from the outside world if desired.

B-Format Decoder- General

In general the B-format to speaker feed decoder presented here is substantially simplified from the usual concepts. It is assumed that the user has gain controls on each of the external amplifiers so that they can use recorded ambisonic material to set the gain of each speaker to give good results in a listening test. this removes the need for gain controls on this unit. It is further assumed that if the user requires more front gain then the external controls can be adjusted for this too. This decoder has outputs for twelve speaker feeds which have angles fixed internally, spread at 30 degree intervals around the soundstage. Each of the outputs labelled 1 to 12 creates a speaker feed designed for a speaker placed at that clock position, with the 12 O'Clock output being centre forward etc. It is expected that the user can find speaker positions close enough to these 12 possible positions to make errors practically undetectable in a home listening environment. In an experimental environment the speakers can be positioned exactly in any case. The O'Clock nomenclature gives an extremely simple and internationally understood method of expressing approximate angular direction and is seen as a significant advantage of this design. No complex front panel switching or programming is required to assign speaker feeds: The amplifier for a certain speaker is simply plugged into the appropriate output socket. The absence of front panel controls brings down cost and reduces both the perceived and actual complexity of operation. By making some reasonable compromises we have reduced front panel controls to a mains switch and a neon power indicator, which are combined in a rocker switch.

Standard B-Format Decoder Matrix

An internal 12-way jumper can be plugged to give completely standard B-format to speaker feed decoding coefficients for the twelve outputs. In common with the previous design, no shelf filtering options are included. The decoding coefficients produce a cardiod shape around the speaker circle, hence zero output from the speaker directly opposite the maximum direction of a panned sound.

Experimental Phase Shifted Speaker Feed Matrix

The W, X and Y signals are first passed through all-pass phase shift networks. These are the familiar imperfect phase-shift filters as used in UHJ encoding and decoding. This creates 0 and 90 degree versions of W, X and Y signals and 180 and 270 degree versions can then be generated by inversion. Now, using appropriate sine weighted sums of these signals will allow any arbitray phase of W X and Y to be generated. As these phase shifts are wideband and relative, any speaker feed with the correct ambisonic amplitude output can be created, only now it can be created with an arbitrary phase relative to the other speakers. The proposal is to apply a phase angle to a speaker feed which is numerically identical to its angle from centre front. So for speakers at the twelve positions of the clock, the relative phase shift will be as follows.

Clock Position Phase Shift

1 +330
2 +300
3 +270
4 +240
5 +210
6 +180
7 +150
8 +120
9 +90
10 +60
11 +30
12 0

From this you can see that each pair of opposite speakers are now driven in antiphase.

Mono signals with purely W content will tend to cancel at a point in the exact centre of a perfectly circular and balanced speaker array. Looking at the system slightly differently you can also see that a mono sine wave will spin around the array at its own frequency, with each speaker cone receiving the signal phase-shifted by the angle of it's physical position; A mono sine wave will create a spiralling wavefront. (A dual spiralling wavefront?)

A sine signal presented at position 12 for example, will have equal amplitude cancelling signals coming from positions 3 and 9, and less equal cancelling signals from other opposing speakers. Only the centre front signal from position 12 will have no interfering signal, as its opposite feed position 6 has no signal due to normal ambisonic amplitude based speaker feed generation.

It is hoped that this might provide an enhancement to the usually rather vague directional imaging of first-order ambisonic decoding. The normal amplitudes of a non-shelf filtered speaker feed decoder are maintained with just the phases being altered, so it is difficult to see how it can be made worse. As a cancellation strategy this method, to me, seems more logical than shelf filters as it at least avoids allowing antiphase components to emanate from the speaker opposite the desired maximum sound direction.

So, Tell Me Already.  Is your clever speaker feed phase cancellation scheme any good?

Nope.  Everything checks out on the 'scope when I've tested it using outputs from standard classic ambisonic equipment like the A+D Pan-Rotate and UHJ encoder, but while actually listening to that, or standard UHJ CDs including the Nimbus walkround, the standard decode gives better angular performance.  The "NORMAL - SPECIAL" toggle switch spends its entire life in the NORMAL position.  It's either a rubbish idea, or I've done something wrong.  Or both.

Final Thoughts

Q: How many speakers do you need in a first order horizontal ambisonic surround-sound set up?
A: This is not a variation on the light-bulb joke.  A very long time ago, the answer was, "As many as possible."  That's not true.  The answer from a mere long time ago is, "about four, maybe five, definitely not eight."  If you have too many, a phenomenon called spatial aliasing occurs which actually reduces the surround effect.  I can attest to this in practice.  My 'clever' system was an attempt to get around this.  If you have a higher order ambisonic system, more speakers is a good thing, but again there will be a limit.  You'll have to ask the academics for the maths.  I just change the light bulbs:)

Henry's email address:

Navigate Up

Recent Edit History

23-FEB-2006: first publication of clock outputs and phase shifted crosstalk cancellation concept
12-JAN-2026: major update, incantations, and I did actually make it