DSPs

Explainer

DSPs explained

A DSP, or digital signal processor, is a processor built to handle streams of real-world data such as audio, radio signals and sensor readings. It takes samples from an analog-to-digital converter and filters, compresses or analyses them in real time.

DSPs are fast at multiply-accumulate operations, the building block of filters and transforms. They typically have dedicated hardware for these operations, special memory access patterns, and predictable timing, so a result always arrives within a known deadline.

How it differs from the usual approach

A general CPU can run signal-processing code, but its timing varies and it uses more power. A DSP is designed for repetitive maths on data streams with fixed deadlines and low power. Compared with a GPU, a DSP works at smaller scale and lower power, usually inside embedded devices. Many modern chips include DSP cores next to CPUs and NPUs.

Where it is used

• Mobile phone modems and radio baseband processing.
• Audio processing in headphones, speakers and hearing aids.
• Radar and sonar systems.
• Motor control and power electronics.
• Camera image processing and sensor fusion.

What to watch

• Texas Instruments and Analog Devices are long-standing DSP suppliers, and Qualcomm's Hexagon is a DSP-derived core used in phones.
• DSP cores are increasingly combined with NPUs so that one block handles both signal processing and small neural networks.
• Wireless standards raise the processing load on basestations and devices with each generation.
• Licensable DSP designs from firms such as Cadence and CEVA are used inside custom chips.

Frequently asked questions