Maximum Likelihood

Description

Maximum Likelihood (ML) is a principle and method of statistical estimation used throughout 3GPP specifications to optimize signal processing tasks in wireless communication systems. In essence, given a statistical model and observed data (e.g., a received radio signal corrupted by noise), the ML estimator finds the parameter values (e.g., transmitted symbol, channel coefficients, user position) that make the observed data most probable. Mathematically, it maximizes the likelihood function, which is the probability of the observed data given the parameters. In digital communications, this often translates to minimizing a distance metric between the received signal and all possible transmitted signals, making it the optimal detector in the presence of additive white Gaussian noise (AWGN).

In the physical layer of 3GPP radio access technologies (LTE, NR), ML algorithms are employed in several key areas. For channel estimation, ML techniques can be used to estimate the complex gains of the radio channel from reference signals, providing a more accurate picture of how the signal was distorted during propagation compared to simpler methods like Least Squares. In MIMO (Multiple-Input Multiple-Output) detection, ML detection (or approximations like ML-MIMO) is the optimal method for separating spatially multiplexed data streams at the receiver, though its complexity grows exponentially with the number of streams. For decoding, the Viterbi algorithm—an implementation of ML sequence detection—is used for convolutional codes, while ML principles underpin the decoding of other channel codes.

Beyond the physical layer, ML estimation is crucial for positioning techniques in 3GPP. For Observed Time Difference of Arrival (OTDOA) positioning in LTE and NR, the UE measures time differences of arrival from multiple base stations. The ML estimator can be used to compute the UE's location from these noisy measurements, providing higher accuracy than linearized methods, especially in non-line-of-sight conditions. Similarly, in angle-based positioning using massive MIMO, ML estimation helps resolve the Angle of Arrival (AoA) or Angle of Departure (AoD).

The implementation of ML in network equipment and UEs involves significant computational complexity, especially for high-order modulation or large MIMO systems. Therefore, 3GPP specifications often reference ML as a performance benchmark, while practical implementations may use sub-optimal but less complex approximations (like MMSE for MIMO detection). The role of ML in the network is to push the boundaries of performance—increasing data rates, improving coverage, enhancing positioning accuracy, and enabling more efficient use of spectrum—by providing the theoretical optimum against which real-world algorithms are measured and improved.

Purpose & Motivation

Maximum Likelihood estimation was incorporated into 3GPP standards as a foundational mathematical tool to achieve optimal or near-optimal performance in various signal processing tasks inherent to digital wireless communication. Early cellular systems used simpler, less optimal estimators and detectors due to limited computational power. However, as data rate demands increased and systems employed more complex techniques like MIMO and higher-order modulation, the performance gap between simple methods and the theoretical optimum (often ML) became a limiting factor for network capacity and user experience.

The adoption of ML-based techniques within 3GPP was motivated by the need to overcome these limitations. For instance, in MIMO-OFDM systems introduced in LTE, linear detectors like Zero-Forcing suffered from noise amplification, especially in ill-conditioned channels. ML detection offered significantly better bit-error-rate performance, enabling the full spatial multiplexing gain promised by MIMO theory. Similarly, for advanced positioning requirements mandated for emergency services and commercial location-based services, traditional geometric positioning methods were insufficient in multipath environments. ML estimation provided a robust statistical framework to handle measurement noise and non-line-of-sight errors, improving location accuracy.

Furthermore, ML serves as a common benchmark in 3GPP performance requirements and conformance testing. Receiver performance tests (e.g., for reference sensitivity) often assume an ideal ML receiver to define the theoretical limit, ensuring that real implementations achieve a performance close to this bound. By standardizing the use of ML principles in specifications for channel estimation, detection, decoding, and positioning, 3GPP ensures that equipment from different vendors is designed to meet a high, consistent performance standard, driving continuous improvement in wireless technology efficiency and capability.

Key Features

• Provides statistically optimal estimation/detection in the presence of Gaussian noise
• Used as a performance benchmark for receivers in 3GPP conformance testing
• Applied in MIMO detection to separate spatially multiplexed data streams
• Enhances accuracy of channel estimation from reference signals (e.g., DMRS, SRS)
• Fundamental to high-accuracy positioning algorithms (OTDOA, AoA/AoD)
• Underpins optimal decoding algorithms for various channel codes (e.g., Viterbi algorithm)

Evolution Across Releases

Defining Specifications