What Is IEEE Transactions on Signal Processing?

IEEE Transactions on Signal Processing is a peer-reviewed biweekly journal that publishes original research on theory, algorithms, and applications for the processing and analysis of signals. It is one of the flagship publications of the IEEE Signal Processing Society and covers a broad class of signals, including audio, video, speech, images, communications waveforms, geophysical measurements, radar, sonar, medical data, and financial time series. The journal traces its origins to 1953, when it was established as the IRE Transactions on Audio; it was renamed several times as the field expanded, passing through IEEE Transactions on Audio and Electroacoustics (1966) and IEEE Transactions on Acoustics, Speech, and Signal Processing (1974) before reaching its current title in 1992.

The journal draws on mathematics, statistics, and electrical engineering to address problems of information extraction from measured data. Its scope encompasses both the theoretical foundations of signal processing, including sampling theory, spectral estimation, and optimal filtering, and the computational methods that implement these foundations in practical systems.

Statistical Signal Processing and Estimation

A foundational area of the journal covers statistical signal processing: methods for estimating signal parameters, detecting signals in noise, and tracking time-varying phenomena using probabilistic frameworks. Classical approaches include Wiener filtering, Kalman filtering, and maximum likelihood estimation. The journal has consistently published advances in Bayesian inference applied to signal models, particle filtering for nonlinear state estimation, and sparse signal recovery methods including compressed sensing and the LASSO.

Array signal processing, which extracts spatial information from signals measured at multiple sensors, forms a major thread covering direction-of-arrival estimation, beamforming, and source separation. The IEEE Signal Processing Society, which sponsors the journal, organizes its submission categories using an Editors' Information Classification Scheme (EDICS) that reflects these research areas and their relationships.

Adaptive and Multidimensional Signal Processing

Adaptive filtering, in which filter coefficients are updated continuously to track nonstationary signal environments, is a recurring topic. The least mean squares (LMS) and recursive least squares (RLS) algorithms, along with their many variants, appear throughout the journal's history in new applications and improved formulations. Adaptive methods for channel equalization in communications, acoustic echo cancellation, noise reduction, and system identification have produced extensive literature in the Transactions.

Multidimensional signal processing extends one-dimensional filtering and spectral analysis to images and volumetric data. Research in this area covers two-dimensional filter design, wavelet transforms and their generalizations, image denoising and restoration, and the analysis of spatiotemporal data. The journal's coverage of time-frequency analysis, including the short-time Fourier transform, wavelet packets, and empirical mode decomposition, addresses signals whose spectral content changes over time, a situation common in speech, biomedical, and seismic applications. Compressed sensing, formalized in influential early papers by Candes, Romberg, Tao, and Donoho, found one of its primary publication venues in the Transactions; the IEEE Signal Processing Magazine has published accessible treatments of many of these foundational advances.

Machine Learning and Data-Driven Methods

Beginning in the early 2000s, the journal absorbed a growing body of work connecting signal processing to machine learning, a trend that has accelerated significantly. Deep neural networks for audio and speech processing, graph signal processing for data on irregular domains, and reinforcement learning methods for adaptive sensing systems have all appeared in the Transactions. This expansion reflects the field's recognition that many signal processing problems, particularly those involving complex natural signals, benefit from learned representations rather than analytically designed ones.

Tensor decompositions, which generalize matrix methods to higher-order data, and federated learning methods for distributed signal analysis connect the journal to contemporary directions in both data science and communications systems. Tensor decompositions, which generalize matrix methods to higher-order data, and federated learning methods for distributed signal analysis connect the journal to contemporary directions in both data science and communications systems. The full archive, accessible through IEEE Xplore, spans over seventy years of published research and provides an unbroken record of the field's theoretical and computational development.

Applications

IEEE Transactions on Signal Processing publishes research with applications in:

• Wireless communications, including channel estimation and multicarrier modulation
• Medical imaging and physiological signal analysis
• Radar and sonar target detection and tracking
• Audio and speech enhancement for telecommunications and hearing aids
• Seismic data processing for geophysical exploration