A PRACTICAL SYSTEM FOR THREE-DIMENSIONAL SOUND PROJECTION (Vennonen, Cont'd)
6. CRITICAL EVALUATION
I used the above architecture for the channel card based on MDAC IC's available at the time (1991). The circuit delivers high static audio performance with no aurally perceptible noise or distortion. However, transitions between attenuation values are in 1.5dB steps, a cause for concern. This was very much the case in the pre-B-format version of the card, when these steps were clearly audible. In B-format Ambisonics, this effect is greatly reduced to the point of imperceptibility with most program material. This is explained by the fact that individual speaker outputs (D-format) are not isolated unique signals, in fact, all four B-format channels contribute to the total amplitude of any D-format output so we gain some extra resolution. The Ambisonic team at the University of York had a similar observation and noted
" This may be illustrated by considering the addition of four digital numbers of equal length where the sum has two more bits of resolution than the original numbers."[34]
The lack of resolution is most audible as mild square wave modulation on pure low frequency tones when using high MDAC update rates. The maximum attainable speed of movement of a sound depends on this update rate, however the channel card hardware is not the limiting factor. With the low speed 1616 running optimised Forth code, it is possible to achieve in excess of fifteen thousand updates per second. This corresponds to a pair of sources being moved over one thousand times per second. To avoid digital noise caused by lack of resolution, it is advisable to make sure that there are minimal occurrences of steps greater than 1.5dB. For instance, a circular movement pattern using several hundred updates per channel card per revolution is quite acceptable. For optimum results one must find a compromise between maximum speed and best resolution.
With this in mind, it was judged that it would be worthwhile to completely redesign the channel card using the latest MDAC chips. The Analog Devices AD7111/7112 and Crystal Semiconductor CS3310 are probably the best. The 1616 could also be replaced by a faster computer. A Macintosh running at over 20Mhz with NuBus slots is one possibility as a dedicated host for the channel cards. Alternatively, the 1616 and channel cards could be emulated entirely in software by a fast (and expensive) RISC computer like a high end SGI. While not discounting the latter entirely, it was decided that one could live with the resolution problem a little longer, remembering after all that it is only an evaluation system.
Overall, the Ambisonic approach certainly is a great improvement on previous methods. There is a pleasant precision to the quality of imaging and the ability to control a three dimensional soundfield with just four channels is very elegant and effective. The playback environment needs to have reverberation minimised for best results, and the optimum listening position is central. In the dome acoustic reverberation tends to diffuse imaging and acoustic treatment would be the best solution. Otherwise, effective if not completely accurate imaging is perceived by the off-centre listeners, who are after all in the majority. This is true not just for frontal images, but all around, with vertical localization being generally the most difficult. Still, the non-central listeners get a much more convincing localization than with conventional quad, with the added benefit of a sense of psychoacoustical verity due to correspondence of Ambisonics to how we actually localize in real life. In the ACAT quad studio the acoustic is not the problem, but the side and rear imaging is considerably more delicate than in the dome. This is a well known limitation of the square speaker configuration. Ambisonic circular pans around the speakers are much smoother than with conventional pair-wise panpot techniques and the reproduction of Soundfield microphone recordings is excellent. These observations attest to the correct calculation and calibration of the decoder matrix parameters, as well as the channel card design and accompanying software.