Compute Intensive: Most DSP kernels are compute intensive with weighted-sum being the core computation. A programmable DSP hence incorporates a dedicated hardwired multiplier and its datapath supports single cycle multiply-accumulate (MAC) operation.
Data Intensive: In most DSP kernels, each multiply operation of the weight-sum computation is performed on a new set of coefficient and data values. A programmable DSP is hence pipelined with an operand read stage before the execute stage, has an address generator unit that operates in parallel with the execute datapath and uses a Harvard architecture with multiple busses to program and data memory.
Repetitive: DSP algorithms are repetitive both at micro-level (e.g. Multiply-accumulate operation repeated N times during an N-term weighted-sum computation) and at macro-level (e.g. kernels such as filtering repeated every time a new data sample is read). A DSP architecture hence uses special instructions (such as RPT MAC) and control mechanisms to support zero overhead looping.
KeywordsPower Dissipation Signal Flow Graph Stopband Attenuation Passband Ripple Adjacent Signal
Unable to display preview. Download preview PDF.