Constellation Diagram

The constellation diagram maps the in-phase (I) component on the x-axis and the quadrature (Q) component on the y-axis. Each discrete symbol is represented as a complex-valued point s(t) = I(t) + jQ(t), where the amplitude is the distance from the origin and the phase is the angle from the positive I-axis. This Cartesian representation allows any linear digital modulation scheme to be visualized as a geometric arrangement of points, making it the universal language for analyzing PSK, QAM, and APSK formats.

Each constellation point is surrounded by a Voronoi region—a convex polygon containing all locations in the IQ plane closer to that point than to any other. These regions define the optimal decision boundaries for minimum distance decoding in additive white Gaussian noise (AWGN). When a received symbol falls within a point's Voronoi region, the receiver assigns it to that symbol. The shape and area of these regions directly determine the symbol error probability for a given signal-to-noise ratio.

In a real receiver, noise and impairments cause received symbols to form Gaussian-distributed clusters around each ideal constellation point rather than landing exactly on them. The Error Vector Magnitude (EVM) quantifies the Euclidean distance between the ideal reference point and each received symbol. Key dispersion metrics include:

• Phase noise: Causes angular spreading of clusters, particularly affecting outer points
• Amplifier non-linearity: Compresses outer constellation rings inward
• IQ imbalance: Stretches the constellation into an elliptical shape
• Carrier frequency offset (CFO): Causes continuous rotation of the entire diagram

Two advanced techniques optimize constellation geometry beyond regular lattices:

Geometric shaping repositions constellation points in the continuous IQ plane to maximize mutual information for a specific channel model, creating non-uniform lattice structures.

Probabilistic shaping keeps a regular constellation but assigns non-uniform transmission probabilities—outer high-energy points are transmitted less frequently than inner low-energy points. This approaches the Shannon capacity limit by making the transmitted signal distribution appear Gaussian, offering shaping gains of up to 1.53 dB over uniform QAM.

Automatic modulation classification systems often reconstruct constellation diagrams from raw IQ samples without prior knowledge of the modulation format. K-means clustering partitions received samples into k clusters by minimizing within-cluster variance, enabling blind estimation of constellation points. Gaussian Mixture Models (GMMs) extend this by modeling each cluster as a Gaussian distribution, optimized via the Expectation-Maximization (EM) algorithm. The number of recovered centroids, their geometric arrangement, and the ring ratio in multi-amplitude formats serve as discriminative features for hierarchical modulation identification.

Beyond static geometry, modulated signals exhibit cyclostationary properties—periodic statistical variations caused by symbol rates, carrier frequencies, and guard intervals. The Spectral Correlation Density (SCD) function reveals these hidden periodicities as distinct patterns in the cyclic frequency domain. Different constellation formats produce unique cyclostationary signatures that are robust to stationary noise and interference, making SCD analysis a powerful complement to geometric constellation inspection for blind modulation recognition in low-SNR environments.

The orthogonal basis vectors of the constellation diagram. The in-phase (I) component modulates a cosine carrier, while the quadrature (Q) component modulates a sine carrier. Together, they form a complex number I + jQ, allowing any signal state to be represented as a single point in the complex plane. The precise mapping of digital bits to these two orthogonal components is the foundation of all vector modulation schemes.

A critical modulation quality metric quantifying the Euclidean distance between the ideal reference constellation point and the actual measured signal point. EVM captures the combined impact of all transmitter impairments—including IQ imbalance, phase noise, and compression—on the constellation diagram. It is typically expressed as a percentage of the peak constellation magnitude or in decibels, providing a single figure of merit for transmitter linearity and signal fidelity.

The geometric partitioning of the IQ plane that defines how a receiver interprets a noisy symbol. A Voronoi region is the convex polygon containing all points closer to a specific constellation point than to any other. The lines separating these regions are decision boundaries. In minimum distance decoding, a received symbol is assigned to the constellation point at the center of its Voronoi region, minimizing the symbol error probability in additive white Gaussian noise (AWGN).

Two advanced techniques to close the gap to the Shannon capacity limit by optimizing the constellation diagram. Geometric shaping moves points off a regular lattice in the continuous IQ plane to maximize mutual information. Probabilistic shaping keeps a regular QAM lattice but transmits outer, high-energy points less frequently than inner points. Both methods reshape the effective distribution of the constellation to better match the channel's characteristics.

Techniques to recover a distorted constellation without a training sequence. Blind equalization relies on statistical properties of the signal. The Constant Modulus Algorithm (CMA) is a classic method that penalizes deviations of the output signal's magnitude from a constant radius, effectively restoring the circular shape of PSK constellations or the ring structure of high-order QAM. This allows a receiver to 'de-rotate' and focus a smeared constellation diagram autonomously.

Hardware non-idealities that visibly distort the constellation diagram. Carrier Frequency Offset (CFO) causes the entire constellation to rotate continuously. IQ imbalance (gain or phase mismatch between I/Q branches) stretches the square lattice into an ellipse. Phase noise smears each point into an arc. Diagnosing these specific geometric deformations in a scatter plot is the primary method for troubleshooting direct-conversion receiver front-ends.