I/Q Constellation Distortion Modeling

The core linear transformation representing I/Q imbalance. This 2x2 matrix captures amplitude mismatch (gain ratio α) and phase deviation from orthogonality (quadrature skew φ).

• Diagonal elements represent the gain applied to each branch
• Off-diagonal elements model the crosstalk caused by phase error
• The matrix is typically parameterized as a function of α and φ
• A perfectly balanced modulator has an identity matrix

In practice, even a 1% gain error and 1-degree phase error create a unique, measurable distortion signature that distinguishes one transmitter from another.

A constant additive term [c_I, c_Q]ᵀ that displaces the entire constellation from the origin. This vector models local oscillator leakage and baseband amplifier offsets.

• c_I component: Shifts the constellation horizontally along the in-phase axis
• c_Q component: Shifts the constellation vertically along the quadrature axis
• The combined effect creates an origin point offset visible as carrier feedthrough
• DC offset magnitude is typically expressed in dBc relative to the signal power

This offset is highly device-specific, as it depends on microscopic manufacturing variances in the mixer and DAC stages.

The complete mathematical model combines the imbalance matrix and offset vector into a single affine transformation:

x_measured = M(α, φ) · x_ideal + c

Where:

• x_ideal is the intended complex baseband symbol [I, Q]ᵀ
• M(α, φ) is the 2x2 gain/phase imbalance matrix
• c is the DC offset vector [c_I, c_Q]ᵀ
• x_measured is the observed distorted symbol

This compact representation allows distortion parameters to be estimated via least-squares fitting on measured constellation data, forming the basis for both compensation and fingerprint extraction.

The distortion parameters (α, φ, c_I, c_Q) are extracted by analyzing the deviation of measured symbol clusters from their ideal reference positions.

• Gain imbalance α is estimated from the ratio of I-axis to Q-axis variance across all symbol clusters
• Phase skew φ is derived from the cross-correlation between I and Q error components
• DC offset is computed as the centroid of the entire constellation relative to the origin
• EVM serves as a composite quality metric aggregating all impairment sources

Modern systems use blind estimation algorithms that operate without known pilot symbols, enabling passive fingerprinting of unknown emitters.

The basic narrowband model assumes impairments are constant across the signal bandwidth. For wideband signals, the model extends to include frequency-selective I/Q imbalance.

• The scalar gain α and phase φ become frequency-dependent functions α(f) and φ(f)
• Modeled using FIR filter structures with asymmetric tap coefficients
• Frequency-dependent imbalance creates inter-carrier interference in OFDM systems
• The distortion profile across subcarriers provides additional fingerprinting dimensions

This extension is critical for modern wideband protocols like Wi-Fi 6 and 5G NR, where a single impairment value cannot capture the full hardware signature.

The accuracy of a distortion model is quantified by how well it predicts observed constellation errors after parameter fitting.

• Residual EVM: The EVM remaining after applying the modeled distortion to ideal symbols and comparing against measurements
• Coefficient of determination (R²): Measures the proportion of variance in measured errors explained by the model
• Prediction stability: The consistency of estimated parameters across multiple captures under identical conditions

A well-fitted model typically achieves residual EVM below 0.5%, indicating that the linear impairment model captures the dominant hardware signature. Non-linear residuals point to power amplifier compression or phase noise effects requiring separate modeling.

The mathematical heart of distortion modeling. A 2x2 matrix captures gain imbalance (α) and phase imbalance (θ) to transform ideal symbols into distorted ones. The model is expressed as:

[I' Q']^T = [[α·cos(θ), sin(θ)], [α·sin(θ), cos(θ)]] · [I Q]^T

• Gain Imbalance (α): Ratio of Q-channel amplitude to I-channel amplitude. A value deviating from 1.0 causes constellation scaling error.
• Phase Imbalance (θ): Deviation from the ideal 90-degree separation, known as quadrature skew, causing axis non-orthogonality.
• This matrix is the primary source of constellation warping into a parallelogram shape.

A static displacement added to the entire constellation, modeled as a simple additive vector [D_I, D_Q]. This is distinct from the multiplicative matrix distortion.

• Origin Point Offset: The primary visual symptom, shifting the (0,0) center of the diagram.
• Causes: Primarily local oscillator leakage in zero-IF architectures and DAC offset error.
• Modeling: The complete model is [I' Q']^T = M_impair · [I Q]^T + [D_I D_Q]^T, combining the impairment matrix and the offset vector.
• This offset is a highly stable, device-specific parameter used for identification.

The observable geometric consequences of the mathematical model, used as input features for AI-based fingerprinting.

• Constellation Ellipticity: A gain/phase imbalance ratio stretches circular symbol clusters into ellipses.
• Constellation Tilt Angle: The orientation of the ellipse's major axis, a sensitive measure of quadrature skew.
• Constellation Centroid: The calculated center of a symbol's point cloud, directly quantifying the static offset for that specific symbol.
• Statistical Moments: Variance, skewness, and kurtosis of point clusters provide robust, multi-dimensional features for distortion profiling.

A distortion profile is a multi-parameter vector mapping the specific gain error, phase error, and DC offset across different operating conditions (power, frequency).

• Distortion Uniqueness: The property that a transmitter's specific combination of impairments is sufficiently distinct from all others, enabling reliable identification.
• Distortion Stability: The degree to which the signature remains constant under fixed environmental conditions, a critical requirement for a viable biometric.
• Distortion Drift: The slow, temporal variation of the profile due to temperature and aging, requiring adaptive tracking algorithms for long-term authentication.

Digital signal processing techniques that dynamically estimate and invert the impairment model. While designed to clean up a signal, the correction coefficients themselves reveal the hardware signature.

• Blind Estimation: Algorithms that estimate I/Q imbalance without a known training sequence, analyzing statistical properties of the received signal.
• Feedback Loops: Systems that continuously track and nullify constellation rotation and origin point offset.
• Fingerprinting Application: The converged values of the correction filter taps (gain, phase, DC) serve as a direct, low-dimensional feature vector for device identification.

Direct-conversion (zero-IF) transceivers are particularly rich sources of constellation distortion, making them ideal for fingerprinting.

• Severe DC Offset: A dominant impairment caused by local oscillator self-mixing, far larger than in superheterodyne architectures.
• Flicker Noise (1/f): Low-frequency noise that adds a time-varying component to the DC offset, creating a unique noise signature.
• I/Q Mismatch: The integration of I and Q paths on a single die still results in measurable gain and phase imbalances.
• These combined, architecture-specific flaws create a highly unique and unclonable I/Q distortion signature.