Digit-Serial Computation
The sampling rates for digital signal processing (DSP) in such applications as speech, telephony, mobile radio, video, radar and sonar, ranges from 10 kHz to 100 MHz. Real-time implementation of such systems requires design of hardware that can process signal samples as these are received from the source, rather than storing them in buffers for batch-mode processing. Efficient implementation of DSP hardware demands a study of families of architectures and styles, selecting an appropriate architecture for a specific application. Digit-Serial Computation is proposed as an appropriate design methodology when bit-serial systems cannot meet sampling rate requirements, and where bit-parallel systems require excessive hardware. A family of implementations can be obtained by changing the digit size parameter, allowing an optimum trade-off between throughput and size. Digit-Serial Computation describes the architecture, and the design and layout methods used in Parsifal: the silicon compiler developed at GEC's Corporate R & D Laboratory. The structures architecture of digit-serial designs lends itself to automatic compilation from algorithmic descriptions. The book also goes on to discuss wider-ranging issues in digit-serial design in chapters on `folding' and `unfolding', as well as in chapters on systolic arrays, canonic-signed-digit number representation and carry-save arithmetic. The book is an excellent source of reference and may be used as a text for an advanced course on the subject.
