Processeurs de traitement numérique du signal (DSP)

The use of digital techniques has clearly overtaken analog techniques in many fields, including, in no particular order, the following:

• conventional signal processing (filtering, fast transforms, signal generation, etc.);

• telecommunications (coding-decoding, modulation-demodulation, adaptive equalization, echo cancellation, encryption, etc.);

• speech processing (coding-compression, analysis, recognition, synthesis, etc.) and image processing (coding-compression, pattern recognition, etc.);

• radar (multi-mode tracking, anti-reverberation processing, target identification, etc.);

• medical applications: EEG, EMG, EOG signal processing... biomedical imaging in NMR...);

• controls (industrial, avionics, etc.), etc.

The advent of standard microprocessors, followed by microcontrollers and then DSPs (Digital Signal Processors), microprocessors dedicated to signal processing, rapidly ensured the advantage of programmed over wired solutions. These techniques offer many interesting features, including :

• reproducibility of treatments to facilitate testing;

• perform functions that have no analog equivalent, or at least only with great difficulty;

• easy modification of the algorithms and/or parameters required;

• the availability of simulation tools.

From an economic point of view, we can be sure of significant growth in the DSP market, given the equipment in which they are used:

• cell phones in the form of specialized circuits with DSP cores;

• modems in a similar form;

• DSL (Digital Subscriber Line), HDSL, ADSL terminals, etc., where they provide equalization, scrambling, echo cancellation, crosstalk suppression, etc.

• high-definition television (HDTV) (sound and image coding/decoding);

• digital broadcasting (DAB Direct Broadcast Audio) ;

• DVDs (Digital Video Disks) for Dolby AC-3 sound decoding and MPEG-2 video decoding;

• AC motor control ;

• hard disk control: use of maximum likelihood techniques (PRML Partial Response, Maximum Likehood) to increase recording density, etc.

We'll talk about what really identifies signal processors in terms of the processing they need to perform. We will then look at the techniques used in the standard processors used in microcomputers and other workstations. These techniques - Harvard structure, cache memory and pipeline - are systematically used in DSPs. The concepts presented will be illustrated by examples taken from the DSPs of three manufacturers: AT&T, Motorola and Texas Instruments. Other players in the DSP market, whose names also...