What Is Signal Representation?

Signal representation is the mathematical and computational framework by which a signal is expressed in a form suitable for analysis, processing, storage, or transmission. A signal, whether it originates from a microphone, an antenna, a medical sensor, or a camera, carries information embedded in its variations over time or space. The choice of representation determines which properties of the signal are made explicit and which are hidden, and it directly governs the efficiency and accuracy of every subsequent processing step.

The field draws on functional analysis, linear algebra, and harmonic analysis. Engineers and mathematicians working in telecommunications, audio, and image processing developed the principal representations over the twentieth century, building on the Fourier series introduced by Joseph Fourier in the early nineteenth century and extended through the development of digital signal processing theory in the 1960s and beyond.

Time and Frequency Domain Representations

The most fundamental distinction in signal representation is between the time domain and the frequency domain. A time-domain representation records how a signal's amplitude varies moment by moment, preserving the exact temporal structure of the waveform. The Fourier transform converts this representation into the frequency domain, expressing the signal as a superposition of sinusoidal components at different frequencies and phases. For discrete-time signals, the discrete Fourier transform (DFT) performs this conversion on sampled data, and the fast Fourier transform (FFT) algorithm reduces the computational cost from O(N²) to O(N log N), making spectral analysis practical for large datasets. Research published in an IEEE Signal Processing Magazine overview of time-frequency analysis surveys how short-time Fourier analysis and related methods address the fundamental trade-off: the Fourier transform gives precise frequency information but no temporal localization, which limits its usefulness for signals whose spectral content changes over time.

Wavelet Transforms

Wavelet transforms address the limitations of the Fourier transform by providing simultaneous localization in both time and frequency. A wavelet is a short oscillating function that serves as the analysis kernel; by scaling and translating the wavelet, the transform captures features at multiple resolutions. High-frequency, short-duration events are resolved with fine time precision, while low-frequency, slow variations are resolved with fine frequency precision. The paper by Daubechies on the wavelet transform, time-frequency localization, and signal analysis published in IEEE Transactions on Information Theory provides one of the foundational mathematical treatments of wavelet bases and their construction. Wavelet representations have become central to image compression standards, including the JPEG 2000 format, and to the analysis of biomedical signals such as electroencephalograms (EEGs) where transient events carry diagnostic significance.

Approximation Methods

Approximation methods produce compact representations of signals by expressing them as weighted sums of basis functions, retaining only the most significant components. Sparse representations seek to decompose a signal using a small number of elements drawn from an overcomplete dictionary, a set of more basis functions than the signal dimension requires. Matching pursuit and basis pursuit are algorithms that identify the most relevant dictionary atoms iteratively or through convex optimization. The NIST Digital Library of Mathematical Functions provides reference definitions for the orthogonal polynomial families, including Chebyshev and Legendre polynomials, that serve as basis functions in many approximation schemes. Compressed sensing, developed in the 2000s by Candes, Romberg, Tao, and Donoho, showed formally that signals with sparse representations can be accurately recovered from far fewer measurements than the Shannon-Nyquist theorem would conventionally require.

Applications

Signal representation has applications across a wide range of disciplines, including:

• Image and video compression, where wavelet and transform-based representations underpin standards such as JPEG 2000 and HEVC
• Wireless communications, where signal representations support channel estimation, equalization, and spectral efficiency analysis
• Biomedical signal processing, including EEG, ECG, and fMRI analysis where time-frequency representations reveal transient physiological events
• Radar and sonar, where range-Doppler representations separate targets by velocity and distance
• Audio coding and music information retrieval, where spectral representations capture timbre and pitch