I/Q Constellation Statistical Moments

The second central moment of a constellation point cluster, measuring the spread or dispersion of measured symbols around their ideal centroid. In I/Q analysis, variance is computed independently for the in-phase and quadrature dimensions, revealing the noise power contributed by each signal path. A transmitter with a noisy I-channel amplifier will exhibit asymmetric variance between the I and Q axes, creating a measurable signature. The square root of variance—standard deviation—directly relates to the Error Vector Magnitude (EVM) of that specific symbol cluster. Engineers track variance across multiple constellation points to build a noise profile that captures the device's unique additive impairment characteristics.

The third standardized moment, quantifying the asymmetry of a constellation point distribution around its mean. A skewness of zero indicates a perfectly symmetric Gaussian cluster, while non-zero values reveal directional bias in the impairment. Positive skewness in the I-dimension suggests the point cloud has a longer tail toward higher in-phase values, often caused by non-linear amplifier compression that clips one side of the waveform more than the other. Skewness is particularly sensitive to even-order distortion products in the transmitter chain. By computing skewness for each symbol cluster across both I and Q axes, a multi-dimensional vector emerges that captures subtle, device-specific non-linearities invisible to variance alone.

The fourth standardized moment, measuring the tailedness or propensity for outliers in a constellation point distribution. A Gaussian distribution has a kurtosis of 3 (excess kurtosis of 0). Distributions with excess kurtosis > 0 (leptokurtic) have heavier tails, indicating intermittent burst noise or phase hits that scatter points far from the centroid. Distributions with excess kurtosis < 0 (platykurtic) suggest a more uniform, bounded impairment, such as quantization noise from a low-resolution DAC. Kurtosis is exceptionally effective at distinguishing transmitters with different phase noise profiles, as oscillators with sporadic phase jumps produce characteristically heavy-tailed constellation clusters.

The joint second moment between the I and Q dimensions, capturing the linear dependency between in-phase and quadrature errors. In an ideal transmitter, I and Q impairments are independent, yielding a covariance near zero and a circularly symmetric point cloud. Non-zero covariance indicates I/Q crosstalk or coupling, where an error in one channel predicts an error in the other. This manifests visually as a tilted elliptical cluster in the constellation diagram. The Pearson correlation coefficient normalizes covariance to a -1 to +1 range, providing a scale-invariant measure of I/Q dependency. This feature is highly stable over time and robust to channel conditions, making it a cornerstone of many fingerprinting feature vectors.

Statistical measures beyond the fourth moment that capture non-Gaussian signal structure while being theoretically immune to Gaussian noise. Unlike moments, cumulants of order greater than two are zero for Gaussian distributions, making them ideal for isolating the deterministic impairment signature from additive thermal noise. The third-order cumulant (related to skewness) and fourth-order cumulant (related to kurtosis minus 3) are commonly used. These features are particularly powerful in low-SNR environments where variance-based metrics are overwhelmed by noise. Cumulant-based fingerprinting exploits the fact that transmitter impairments are inherently non-Gaussian processes, allowing the extraction of a noise-robust device signature.

The systematic process of assembling statistical moments into a fixed-length numerical vector suitable for machine learning classifiers. A typical feature vector concatenates the mean, variance, skewness, and kurtosis computed for each symbol cluster in the constellation, across both I and Q dimensions. For a 16-QAM signal, this yields 16 symbols × 2 dimensions × 4 moments = 128 features. Dimensionality reduction via Principal Component Analysis (PCA) is often applied to retain only the most discriminative moment combinations. This approach transforms the raw, high-dimensional constellation point cloud into a compact, computationally efficient representation that feeds directly into support vector machines, random forests, or neural network classifiers for real-time emitter identification.

A two-dimensional scatter plot visualizing the in-phase (I) and quadrature (Q) components of a digitally modulated signal. Each point represents a received symbol, and the systematic distortion of these points from their ideal grid locations reveals the transmitter's unique hardware signature. The statistical moments are calculated directly on the distribution of these points to quantify the shape of the distortion.

The statistical dispersion of measured signal points around an ideal constellation locus. This cloud is caused by additive noise, phase noise, and inter-symbol interference. The variance of this cloud is a primary statistical moment, and its higher-order shape—quantified by skewness and kurtosis—provides a unique noise signature that is distinct from the deterministic I/Q imbalance signature.

The comprehensive study of the shape, symmetry, and statistical structure of constellation point clusters. This analysis goes beyond simple error vector magnitude to extract a multi-dimensional feature vector. Key morphological features include:

• Cluster ellipticity and tilt angle
• Centroid offset from the ideal symbol location
• Higher-order moments like cluster kurtosis, which measures the 'tailedness' of the distribution

The use of cumulants, bispectrum, and trispectrum processing to characterize non-Gaussian signal behavior. While variance is a second-order moment, skewness (third-order) and kurtosis (fourth-order) capture the asymmetry and peakedness of a constellation cluster. These higher-order moments are particularly robust against Gaussian noise, making them powerful features for emitter identification in low signal-to-noise ratio environments.

A multi-parameter characterization of a transmitter's unique impairment fingerprint. This profile maps the specific statistical moments—including variance, skewness, and kurtosis—for each constellation point across different power levels and frequencies. The complete profile forms a high-dimensional feature vector used to train a one-class classifier for device authentication.

The degree to which a transmitter's I/Q impairment signature and its associated statistical moments remain constant over short time intervals under fixed environmental conditions. High stability is a critical requirement for reliable fingerprinting. If the kurtosis or skewness of a constellation cluster drifts significantly without a change in channel conditions, it may indicate a component failure or a spoofing attempt.