A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Digital signal processing applies mathematical operations to discrete-time signals using digital hardware or software. It encompasses filtering, spectral estimation, modulation, and audio/image processing. DSP is enabled by fast sampling rates, the FFT algorithm, and efficient computational structures.
From your study of the DFT and FFT, you know how to transform a discrete-time signal into its frequency-domain representation and back. From your work on aliasing and reconstruction, you understand that a continuous signal must be sampled above twice its highest frequency to avoid aliasing, and that recovery requires a low-pass reconstruction filter. These two concepts — the sampling theorem and spectral analysis — are the twin foundations of digital signal processing. DSP is the discipline of performing useful signal manipulation computationally: filtering, detecting, transforming, compressing, and modulating signals after they have been converted to sequences of numbers by an analog-to-digital converter.
The central advantage of DSP over analog processing is precision, repeatability, and programmability. An analog filter is a physical circuit whose characteristics drift with component aging and temperature. A digital filter is an algorithm: it can be replicated exactly, updated by changing coefficients in software, and run on the same hardware to achieve completely different responses without touching any components. The same DSP processor that demodulates a radio signal today can implement a voice-cancellation algorithm tomorrow, simply by loading new code. This separates the physical constraints of the hardware from the signal-processing functionality — a separation impossible in purely analog systems.
The FFT is what makes real-time DSP computationally feasible. Computing an N-point DFT directly requires O(N²) multiply-accumulate operations; the FFT reduces this to O(N log N). For N = 1024, that is roughly 1,000,000 vs. 10,000 operations — a 100× speedup that translates directly into power consumption and latency. Modern DSP processors are architected around this: dedicated multiply-accumulate (MAC) units execute the butterfly operations at the FFT's core in a single clock cycle, and memory layouts are optimized for the bit-reversed access patterns the algorithm requires. Without the FFT, real-time spectral analysis, audio compression (MP3, AAC), OFDM wireless communication (Wi-Fi, LTE, 5G), and medical imaging would not be practical on affordable hardware.
Every DSP system follows the same structural pipeline: analog anti-aliasing filter → ADC → digital processor → DAC → analog reconstruction filter. The anti-aliasing filter is not optional — it removes frequency content above the Nyquist frequency before sampling, because aliased components fold into the signal band and are mathematically indistinguishable from legitimate signal once sampling has occurred. They cannot be removed digitally after the fact. The reconstruction filter smooths the staircase output of the DAC back into a continuous waveform. Between ADC and DAC, the digital processor has complete freedom to apply any transformation — linear or nonlinear, time-invariant or adaptive — to the sample stream. This fixed analog boundary around a flexible digital core is the architectural principle that has made DSP the dominant paradigm for signal processing in communications, audio, imaging, radar, and instrumentation.