Pipelined Special Purpose Computer

System architecture and circuit topology.

Hardware

The vOICe hardware. Photography: P. Meijer.

• This approach increases the flexibility for the implementation, via programming, of alternative image-to-sound mappings. The frequency distribution of the 64 oscillators is freely programmable, which may be used to deal with individual hearing problems by offering a personalized frequency spectrum. In addition, the image to sound conversion time is user selectable, presently offering a choice between 1.05 and 2.10 seconds, but still longer or shorter conversion times are easily implemented in the design. Relatively long conversion times help in analyzing complicated or detailed images at a high resolution, while shorter ones can help in tracking rapidly changing scenes containing moving objects. Making the conversion time user selectable allows the user to adapt to any given situation.

• Expensive hardware multipliers or sine evaluators were avoided for a low-cost design, and only commercially available components of standard speed were used. Apart from CMOS memory components, most of the system logic was implemented using standard LS TTL technology. By not using any new or fancy technology at the component level, excellent reliability has been achieved, with an expected MTBF (mean time between failure) of many years. This is a clear advantage over most tactile imaging devices. Even the prototype has already been operating for many hundreds of hours without failing once since its creation.

Real-time 64-channel sound synthesis
Using a system clock of only 2 MHz, a pipelined serial computation of samples of the superposition of 64 oscillator signals takes place at a frequency of 31.25 kHz, which is sufficiently high for very good audio quality (cf. CD players using 44.1 kHz). The superposition samples are represented by 16-bit values, again to achieve high audio quality.

The pipeline for parallel processing consists of 5 stages, in which alternatingly 2 and 3 oscillator samples are simultaneously processed during each phase of the system clock. Every 64 clock cycles the system churns out a newly synthesized 16-bit sound sample of the superposition of the 64 oscillators.

Pipeline flow animation.

Single-board prototype
The resulting special purpose computer was optimized towards the DSP processing for image-to-sound conversion. The whole conversion system, including 20 ms frame grabbing and 16 grey-tone digitization hardware for input, and the analog output stages for the headphones, has been implemented on a single 236 × 160 mm circuit board, dissipating a measured low-power 4.4 Watt. Even while this is already comparable to the power consumption of small portable (battery powered!) video cameras in the consumer market, a further drastic reduction of power consumption could easily be obtained by replacing the LS TTL components by CMOS equivalents.