From Analogue to Digital: The Sampling Revolution
Claude Shannon's 1948 sampling theorem established the theoretical foundation: a band-limited signal with highest frequency component f_max can be perfectly reconstructed from discrete samples taken at a rate f_s ≥ 2·f_max. This is the Nyquist criterion. Below this rate, aliasing occurs — high-frequency components fold back into the sampled spectrum and appear at wrong frequencies, producing artefacts that cannot be removed.
The Z-Transform: DSP's Laplace Transform
For discrete-time systems, the Z-transform plays the role that the Laplace transform plays for continuous systems. It converts a difference equation (the digital equivalent of a differential equation) into an algebraic equation, making filter design tractable:
FIR vs IIR Filters: The Fundamental Trade-off
Digital filters fall into two fundamental classes, each with distinct advantages and limitations:
DSP Everywhere: Applications Across Engineering
The pervasiveness of DSP in modern engineering is difficult to overstate. In communications, digital modulation schemes (OFDM, QAM, CDMA) process signals entirely in the digital domain, enabling the spectral efficiency that supports 4G/5G. In audio, noise-cancelling headphones sample the ambient noise, compute an inverted waveform, and play it back to cancel the original — all within microseconds. In medical devices, pacemakers filter sensed cardiac signals to distinguish real beats from noise artefacts before deciding whether to pace. In structural health monitoring, accelerometers on bridges and buildings are sampled continuously, and DSP algorithms detect shifts in natural frequencies that indicate structural change. The common thread is that DSP enables functionality that would be impossible or impractical with analogue components: filters with precisely defined cut-off frequencies that don't drift with temperature; adaptive algorithms that learn the characteristics of a specific environment; complex spectral analysis performed in real time; and numerically stable implementations using floating-point arithmetic that analogue circuits cannot match.
Quantisation and Finite Precision Effects
Digital filters operate with finite-precision arithmetic — 16-bit, 24-bit, or 32-bit numbers rather than exact real values. Every multiplication and addition rounds the result, introducing quantisation noise. For simple filters this is inconsequential. For high-order IIR filters with poles close to the unit circle (sharp narrow-band filters), quantisation of the filter coefficients can shift poles outside the unit circle — making a theoretically stable filter unstable in implementation. This is why signal processing engineers implement high-order IIR filters as cascades of second-order sections (SOS) rather than direct-form structures. Build a multi-component signal, add noise, apply FFT analysis, and visualise the frequency spectrum. EngForge's signal analyser helps you understand filtering requirements before writing a single line of DSP code.