What Is Maximum Likelihood Detection?

Maximum likelihood detection is a statistical decision procedure that selects, from a finite set of hypotheses, the one that maximizes the probability of the observed data given that hypothesis. In communications and signal processing, it is applied at the receiver to determine which symbol or signal was transmitted based on the noisy waveform collected at the input. The technique belongs to the intersection of detection theory and statistical inference, drawing on probability theory, filtering theory, and optimization. It provides the theoretically optimal decision rule for minimizing detection error probability when all hypotheses are equally likely, making it the performance baseline for practical detector designs.

Maximum likelihood detection differs from maximum a posteriori detection in that it does not incorporate prior probabilities over the hypotheses. When the symbols are transmitted with unequal probabilities, MAP detection outperforms ML detection in terms of average error rate, but ML detection has the advantage of requiring no knowledge of the source distribution, which is often unavailable or unreliable in practice. The two criteria coincide under uniform priors.

Matched Filtering and Optimal Receivers

The matched filter is the linear receiver that maximizes the output signal-to-noise ratio for a deterministic signal corrupted by additive white Gaussian noise, and it is the core building block of the ML receiver for digital modulation. For binary signaling, the ML receiver computes the inner product of the received waveform with each of the two candidate signal waveforms and selects the hypothesis yielding the larger value. The matched filter output at the sampling instant is a sufficient statistic for the ML decision: all the information about the hypothesis contained in the received waveform is captured in this single scalar. Extensions to M-ary modulation replace the binary comparison with a nearest-neighbor search in the signal space, selecting the constellation point whose Euclidean distance to the received vector is smallest. A detailed treatment of ML receiver design for AWGN channels, including derivations of error probability bounds, is available through ScienceDirect topics on maximum-likelihood decoding.

Filtering Theory and Sequential Detection

In sequential and dynamic settings, maximum likelihood detection integrates with filtering theory to track the evolution of a hidden state over time. When the transmitted signal sequence passes through a dispersive channel with intersymbol interference, the optimal ML sequence detector processes the entire received sequence jointly rather than symbol by symbol. The Viterbi algorithm achieves this by operating on the trellis of possible transmitted sequences, exactly as in convolutional decoding. For linear Gaussian state-space models, the Kalman filter computes the maximum likelihood estimate of the state trajectory given the observation sequence, connecting ML detection to classical linear filtering theory. The interplay between ML estimation and Kalman-type filtering is developed in the context of noisy parameter identification in IEEE Xplore work on maximum likelihood parameter estimation from incomplete data.

MIMO and Multi-Hypothesis Detection

In multiple-input, multiple-output wireless systems, the receiver observes a linear superposition of simultaneously transmitted streams, each independently modulated. The ML MIMO detector finds the vector of transmitted symbols that minimizes the Frobenius distance between the received vector and the product of the channel matrix with the candidate symbol vector. This is an integer least-squares problem whose exact solution requires exponential search in the number of transmit antennas and constellation size. Sphere decoding and semidefinite relaxation offer approximate solutions with polynomial average complexity and near-ML performance at practical operating signal-to-noise ratios. Research on ML-based detection in broadband systems is discussed across multiple publications accessible through IEEE Xplore on maximum likelihood parameter estimation.

Applications

Maximum likelihood detection has applications in a wide range of sensing and communication systems, including:

• Digital wireless receivers, including 4G LTE and 5G NR base stations, for symbol detection under multipath fading
• Radar and sonar systems, where ML detectors resolve target presence and range from received echoes
• Optical fiber communications, where the ML sequence detector compensates for chromatic dispersion and nonlinear distortion
• Medical diagnostics, including MRI reconstruction and ultrasound imaging, where signals must be resolved from noisy measurements
• Cognitive radio and spectrum sensing, where ML detectors determine channel occupancy from received power observations