Constellation Diagram

The diagram maps symbols onto a complex plane where the I-axis (In-Phase) represents the cosine carrier component and the Q-axis (Quadrature) represents the sine component. Each point is a complex number I + jQ, encoding both amplitude (distance from origin) and phase (angle from positive I-axis). This orthogonal basis allows two independent data streams to modulate a single carrier without interference.

Each discrete point, or symbol, represents a unique binary sequence. The number of symbols M defines the modulation order (e.g., M=4 for QPSK, M=16 for 16-QAM). Decision boundaries partition the plane into regions; a receiver classifies a received noisy sample by finding the nearest ideal symbol point using minimum Euclidean distance detection. Closer symbol spacing increases spectral efficiency but reduces noise immunity.

The diagram serves as a powerful diagnostic tool. Additive White Gaussian Noise (AWGN) manifests as a circular cloud around each ideal symbol point. Phase noise causes an angular smearing or rotation of the entire constellation. Gain compression pulls outer points inward, while IQ imbalance distorts the diagram into an elliptical shape. The spread of these clouds directly visualizes the Error Vector Magnitude (EVM).

Higher-order constellations pack more bits into each symbol, increasing spectral efficiency (bps/Hz). Common schemes include:

• QPSK (4-QAM): 2 bits/symbol, robust against noise.
• 16-QAM: 4 bits/symbol, moderate efficiency.
• 64-QAM: 6 bits/symbol, used in Wi-Fi and LTE.
• 256-QAM: 8 bits/symbol, requires high SNR. The trade-off is density: as M increases, points crowd together, demanding higher Signal-to-Noise Ratio (SNR) for error-free reception.

In Deep Learning Modulation Recognition, the raw constellation image or its 2D histogram is fed directly into a Convolutional Neural Network (CNN). The network learns geometric features—such as the grid structure of QAM, the circular arrangement of PSK, or the cruciform shape of APSK—to automatically classify the modulation scheme. This bypasses manual feature engineering and is robust to moderate channel impairments when trained with data augmentation.

While a static diagram shows ideal symbol locations, plotting the signal trajectory between consecutive symbols reveals the pulse-shaping filter's effect. The path taken from one point to another shows the peak-to-average power ratio (PAPR) characteristics and filter ringing. Sharp, direct transitions indicate unfiltered signals with high spectral splatter, while smooth, curved paths indicate Nyquist pulse shaping like root-raised cosine filtering for bandwidth control.

A constellation diagram maps symbols onto the complex plane, where the horizontal axis represents the In-Phase (I) component and the vertical axis represents the Quadrature (Q) component. Each point corresponds to a unique combination of amplitude and phase that defines a specific digital symbol.

• The I-axis carries the cosine-modulated carrier component
• The Q-axis carries the sine-modulated carrier component
• Distance from origin represents instantaneous signal amplitude
• Angle from the positive I-axis represents instantaneous phase shift

This representation allows engineers to visualize how a modulation scheme encodes bits into distinct physical states, with the number of points directly corresponding to the modulation order (e.g., 4 points for QPSK, 16 for 16-QAM).

QPSK places four equally spaced points on a circle at phase angles of 45°, 135°, 225°, and 315°, encoding 2 bits per symbol. Each symbol has identical amplitude, with information carried solely in the phase.

• Symbol mapping: 00, 01, 11, 10 (Gray coding minimizes bit errors)
• Constant envelope simplifies power amplifier design
• Minimum Euclidean distance between symbols determines noise immunity
• Spectral efficiency: 2 bits/s/Hz before filtering

QPSK is widely used in satellite communications, LTE uplink, and WLAN standards due to its balance of spectral efficiency and robustness against noise and amplifier non-linearity.

16-QAM arranges 16 symbols in a 4×4 square grid, encoding 4 bits per symbol through simultaneous amplitude and phase modulation. Three distinct amplitude levels create the grid structure.

• Inner ring: 4 symbols at lowest amplitude
• Middle ring: 8 symbols at intermediate amplitude
• Outer ring: 4 symbols at highest amplitude
• Gray coding ensures adjacent symbols differ by exactly 1 bit

16-QAM doubles the spectral efficiency of QPSK but requires a higher Signal-to-Noise Ratio (SNR) to maintain reliable detection. It is a cornerstone of LTE downlink, DOCSIS cable modems, and digital microwave links.

64-QAM packs 64 symbols into an 8×8 grid, delivering 6 bits per symbol. Higher-order QAM variants (256-QAM, 1024-QAM) further increase spectral efficiency at the cost of noise resilience.

• 64-QAM: 6 bits/symbol, used in 802.11ac Wi-Fi and LTE-Advanced
• 256-QAM: 8 bits/symbol, introduced in 802.11ax (Wi-Fi 6) and 5G NR
• 1024-QAM: 10 bits/symbol, specified in Wi-Fi 6E and 802.11be
• Symbol spacing shrinks exponentially with modulation order

These dense constellations demand high SNR and sophisticated channel impairment compensation including equalization and phase noise correction. Modern systems dynamically switch between QAM orders based on link conditions.

Real-world channel effects distort the ideal constellation, causing symbol points to spread into clouds. Analyzing these distortions reveals specific impairment types.

• Additive White Gaussian Noise (AWGN): Points scatter isotropically around ideal locations
• Phase noise: Points rotate tangentially, forming arcs rather than clusters
• I/Q imbalance: Constellation becomes elliptical or skewed
• Non-linear distortion: Outer points compress inward while inner points expand
• Multipath fading: Creates rotated, scaled, and delayed copies that blur the constellation

Error Vector Magnitude (EVM) quantifies the deviation between measured and ideal symbol positions, serving as a key metric for transmitter quality and link diagnostics.

In Automatic Modulation Classification (AMC), constellation diagrams serve as powerful visual features for Convolutional Neural Networks (CNNs). The spatial arrangement of received symbols directly encodes the modulation scheme's geometric signature.

• CNNs treat constellation images as 2D input tensors
• Density-based representations (heatmaps) improve robustness at low SNR
• Residual Networks (ResNets) extract hierarchical features from constellation structure
• Graph Neural Networks (GNNs) model symbol relationships as graph edges

Unlike raw IQ sample processing, constellation-based classification is invariant to symbol timing offsets and provides a compact, interpretable representation. However, it discards temporal ordering information that may be critical for identifying schemes with memory, such as GMSK.

The raw, time-domain digital representation of a radio signal consisting of paired In-Phase (I) and Quadrature (Q) component values. These samples capture both the instantaneous amplitude and phase of the signal, forming the direct data source from which a constellation diagram is plotted. A constellation point is simply an IQ sample taken at the optimal symbol sampling instant.

A frequency-shifted representation of a bandpass signal centered at zero hertz. This representation preserves all modulation information in a complex-valued format (I+jQ), making it the native domain for constructing and analyzing constellation diagrams. It allows for efficient digital processing by eliminating the carrier frequency while retaining the signal's envelope and phase.

A measure comparing the power of a desired signal to the power of background noise. In a constellation diagram, a low SNR manifests as a 'cloud' of points around each ideal symbol location, increasing the probability of symbol errors. It is the primary metric for evaluating the robustness of a modulation classifier, as higher noise levels make the geometric separation between symbols ambiguous.

A deep learning architecture employing learnable filters that slide across input data to extract spatial hierarchies of features. When applied to a constellation diagram, a CNN treats the plot as a 2D image, learning to identify the geometric arrangement of symbol clusters. This approach bypasses manual feature engineering by automatically detecting the shape, density, and relative positions of the signal states.

A measure of modulation quality that quantifies the Euclidean distance between the measured symbol location on a constellation diagram and its ideal reference position. A high EVM indicates significant distortion from the transmitter, channel impairments, or noise. It is a critical diagnostic metric for hardware testing and a powerful discriminative feature for automatic modulation classification.

Statistical measures that characterize the shape of a probability distribution beyond simple variance. For a constellation diagram, the distribution of IQ points yields specific cumulant values that act as a fingerprint for different modulation types. For example, QPSK and 16-QAM have distinct fourth-order cumulant signatures, making them robust, hand-crafted features for classification algorithms that are resilient to phase and frequency offsets.