Open Wave Field Synthesis

After a couple of test iterations, I settled on modular 32-driver arrays of Peerless TA6FD00-04-ND full-range drivers, in two rows (both rows evenly spaced but with different spacings to counteract the comb filtering which comes from a regularly-spaced single row). Each driver was built into an individually sealed cell as part of a metre-wide module, and these were designed to be able to array together into one arbitrarily long array. Simulation showed that the frequency and phase response from a relatively small (0.3 litre) enclosure would be acceptable and allow for a crossover at 210Hz without exceeding the maximum excursion of the drivers.

The enclosures were all cut by Tom Cecil at his workshop in East London on a CNC machine, so the drivers would be millimetre-aligned, and assembled and soldered by hand (with many thanks to James Stevenson and Jake Tyler).

We wanted to keep the processing and amplification cost down as much as possible, but wanted to allow for at least 256 drivers in an array, at a relatively low power of 10-20W per driver. Existing solutions were prohibitively expensive, and either required multiple interfaces in an aggregate device and/or proprietary interconnection protocols.

For the initial iteration, we wanted to be able to feed each driver individually from a computer to allow for rapid iteration of different algorithms. The timing of each driver and between arrays needs to be sub-sample aligned for the wavefield effect to work (each source might only be delayed by a fraction of a sample to position it with millimetre accuracy). My target was for microsecond-aligned latency for all drivers.

I settled on using 802.1 AVB as an audio transport protocol. This is natively supported by macOS and allows for 256 output channels, sample-synchronous, from the Ethernet port of a Mac studio, and through a TSN network switch.

I built custom 32-channel amplifier modules based on a Texas Instruments class-D amplifier architecture, and custom modules with an XMOS XEF216 chip for AVB/TDM conversion. These provide 30W/channel of amplification for each speaker and can be scaled up arbitrarily. I chose to sit the chip on a module to allow for replacement with an FPGA/SoC for a later revision of the project where wavefield processing will be on each module.

The amplifier board connects to each speaker module with a 72-pin card edge connector, and allows for AVB input, or USB (as a 32-channel UAC 2 class compliant audio device), with support for external word-clock I/O if needed. (In AVB mode, the clocking is all via gPTP so no external clocking is needed – just power and Ethernet to each module).

Software and algorithms for this project were built by Daniel Jones, based on his open-source Signalflow library. Running across all cores of a Mac Studio, we had more than enough power to spatialise dozens of sources into 256 output channels, with an interactive editor and visualiser, and/or MIDI control of source position.

The installation was commissioned and generously supported by Electromagnetic Field Festival – who provided us with a shipping container to house the installation for its initial outing. We opted for four arrays, each with 64 drivers, placed to the left, right, in front and above the listener position.

Daniel composed a new spatial sound piece, specifically for the installation, On The Surface Of The Ocean. The installation ran for three days, with wonderful feedback from attendees; future installations are in the works.

Roles

Year