Features and characteristics of DSP

A digital signal processor (DSP) is an application-specific instruction-set processor optimised for repetitive, numeric-intensive signal-processing tasks such as filtering, FFT and convolution, on streams of real-time data (audio, video, radar, telecom).

Characteristics and advantages

• Hardware multiply-accumulate (MAC) unit: completes multiply and add in one cycle, the core of FIR/IIR filters (y = Σ a·x).
• Harvard architecture: separate program and data memories (often two data memories) allow instruction and operand fetch at once.
• Pipelining and parallel execution: several instructions overlap; VLIW or SIMD in modern DSPs.
• Zero-overhead looping: hardware loop counters avoid branch time.
• Special addressing modes: circular (modulo) addressing for buffers and bit-reversed addressing for FFT.
• Fast on-chip memory and DMA for high data throughput.
• Fixed-point/floating-point and saturation arithmetic to prevent overflow wrap-around.
• Low power, low cost, and real-time response with predictable timing.
• Good for processing continuous data streams quickly, giving better performance than a general-purpose processor at lower cost.

Common features of ASIPs for DSP

• A DSP is an ASIP: it has a datapath tailored to the domain (MAC, shifters, accumulators) and a smaller/specialised instruction set.
• Extra on-chip peripherals: ADC/DAC, timers, serial ports.
• Reduced program size and faster execution than a GPP for the same task, yet still programmable (flexible).