Probabilistic Constellation Shaping

28 concepts15 papers1 proposal

In digital communications, maximizing the rate of data transmission over a noisy channel is governed by the Shannon limit. Information theory dictates that to achieve maximum capacity over an Additive White Gaussian Noise (AWGN) channel, the transmitted signals must follow a continuous Gaussian distribution.

However, real-world hardware relies on discrete alphabets, such as Pulse Amplitude Modulation (PAM) or Quadrature Amplitude Modulation (QAM). Traditionally, systems transmit these discrete constellation points with uniform probability—meaning every symbol is equally likely to be sent. As noted by Lin Zhou et al. in An Improved Design of Gallager Mapping for LDPC-coded BICM-ID System and Ahmed Elzanaty et al. in Adaptive Coded Modulation for IM/DD Free-Space Optical Backhauling: A Probabilistic Shaping Approach, this uniform assumption creates an inherent efficiency loss known as the shaping gap, which caps out at 1.53 dB in standard AWGN channels.

To bridge this gap, researchers use Probabilistic Constellation Shaping (PCS). Instead of changing the physical coordinates of the constellation points (geometric shaping), PCS alters the frequency with which existing points are used. By transmitting lower-energy or lower-cost symbols more frequently, the system mimics a discrete Gaussian or Maxwell-Boltzmann distribution, lowering average power consumption and pushing data rates closer to the theoretical Shannon limit (Lin Zhou et al. et al.; Jie Feng et al. in MIMO Transmission Under Discrete Input Signal Constraints).

The Core Mechanism: Distribution Matching

How do you transform a standard, uniform stream of data bits (0s and 1s) into a non-uniform sequence of symbols without losing information? The answer lies in an architecture called a Distribution Matcher (DM).

A DM acts as an invertible pre-encoder before the channel coding stage. It takes a block of uniform bits and maps them to a sequence of symbols whose frequencies match a target non-uniform distribution. This architecture ensures that the transmitted signal achieves the precise entropy required for the channel, as detailed by Alexandre Graell i Amat et al. in Probabilistically-Shaped Coded Modulation with Hard Decision Decoding for Coherent Optical Systems.

By biasing symbol selection toward lower intensities, the system reduces the overall energy footprint (Ahmed Elzanaty et al.). In extreme low-signal-to-noise ratio (SNR) environments, this can even mean explicitly placing a high-probability symbol exactly at the origin of the constellation, allowing more power to be preserved for outer points when they are required, as shown by Rajet Krishnan et al. in Constellation Optimization in the Presence of Strong Phase Noise.

To implement this efficiently at high speeds, Mohamad Dia et al. in A Compressed Sensing Approach for Distribution Matching framed distribution matching as a compressed sensing problem. By utilizing the Generalized Approximate Message Passing (GAMP) algorithm, the authors achieved a low-complexity, joint channel coding and matching solution that scales effectively.

Balancing Blocklength and Complexity

While theoretical PCS assumes infinitely long sequences, practical hardware operates under strict latency and computational constraints.

In the short blocklength regime (where packets contain 1,000 bits or fewer), traditional distribution matchers suffer from severe rate loss and divergence from the target distribution. To recover these lost shaping gains, specialized algorithms like shell mapping are used. As reviewed by Mustafa Cemil Coskun et al. in Efficient Error-Correcting Codes in the Short Blocklength Regime, shell mapping successfully preserves shaping gains even when channel uses are highly restricted. Building on this, Felipe Areces in Achieving Short-Blocklength RCU Bound via CRC List Decoding of TCM with Probabilistic Shaping demonstrated that pairing shell-mapping matchers with trellis-coded modulation (TCM) and cyclic redundancy check (CRC) list decoding allows short-block systems to closely approach fundamental finite-length capacity bounds.

Receiver complexity presents another bottleneck. Decoding non-uniform symbols typically requires intensive soft-decision computations. To bypass this, Alexandre Graell i Amat et al. integrated PCS with hard-decision decoding (HDD) via high-speed staircase codes, bridging the capacity gap while maintaining low hardware complexity.

Conversely, in multi-antenna (MIMO) systems, Kirill Ivanov et al. in Probabilistic Shaping in MIMO: Going Beyond 1.53dB AWGN Gain With the Non-Linear Demapper showed that a non-linear sphere decoder designed to explicitly account for the non-uniform symbol distribution can actually reduce receiver complexity. Simultaneously, it breaks past the traditional 1.53 dB AWGN limit, unlocking performance gains of up to 3.5 dB.

Custom Shapes for Complex Channels

The true power of PCS lies in its flexibility: the optimal "shape" depends entirely on the physical constraints and noise profiles of the underlying channel.

Optical and Multi-User Wireless Networks

In Visible Light Communication (VLC) and hybrid LiFi-WiFi networks, signals must be unipolar (non-negative) and are heavily constrained by illumination requirements and hardware distortions. By adaptively adjusting the likelihood of individual signal points rather than treating them uniformly, systems can dynamically adapt to local noise and multi-user interference, as shown by Shuai Ma et al. in Optimal Discrete Constellation Inputs for Aggregated LiFi-WiFi Networks and Amanat Kafizov et al. in Probabilistic Constellation Shaping for Enhancing Spectral Efficiency in NOMA VLC Systems.

This power-shaping capability can even serve environmental and safety goals: Sidrah Javed et al. in EMF-Aware Probabilistic Shaping Design for Hardware-Distorted Communication Systems designed a shaping strategy that prioritizes low-power symbols specifically to reduce human electromagnetic field (EMF) exposure by up to 98% while maintaining target data rates.

Nonlinear Channels

In optical fiber communications utilizing the Nonlinear Fourier Transform (NFT), different spectral components (eigenvalues) incur vastly different costs in pulse duration and noise sensitivity. To maximize efficiency, Alexandre Graell i Amat et al. introduced Probabilistic Eigenvalue Shaping (PES) in Probabilistic Eigenvalue Shaping for Nonlinear Fourier Transform Transmission, nearly doubling data rates by tailoring symbol probabilities to account for these non-linear resource costs.

Data-Driven and Model-Free Shaping

When channels are too complex or mathematically intractable to model analytically, researchers leverage machine learning. Dor Tsur et al. in Optimizing Estimated Directed Information over Discrete Alphabets and Data-Driven Optimization of Directed Information over Discrete Alphabets reframed input distribution optimization as a reinforcement learning problem. By using deep generative models to interact directly with channel samples, their framework automatically discovers optimal non-uniform signaling patterns without requiring any prior theoretical channel models, dynamically interpolating between distribution types based on the operating SNR.

Key Foundational Work

An Improved Design of Gallager Mapping for LDPC-coded BICM-ID System (2016) — contributed 3 concept elements to this direction.
Constellation Optimization in the Presence of Strong Phase Noise (2013) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_probabilistic_constellation_shaping_d186d9cf,
  title = {probabilistic constellation shaping},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/probabilistic-constellation-shaping/rdr_d186d9cf63bb4caa850d20e90bb366cc},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_probabilistic_constellation_shaping_d186d9cf_v1,
  title = {probabilistic constellation shaping},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/probabilistic-constellation-shaping/rdr_d186d9cf63bb4caa850d20e90bb366cc/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake