I/Q Imbalance Modeling

Represents the amplitude mismatch between the I and Q branches, typically expressed in decibels (dB) or as a fractional ratio. A non-zero gain error causes the ideal square constellation to stretch into a rectangle.

• Typical Range: 0.1 dB to 2.0 dB for commercial transceivers
• Effect: Amplifies one signal component relative to the other
• Modeling: Applied as a multiplicative factor (1+ε) to the Q branch before combining
• Fingerprint Utility: Highly stable over temperature, making it a reliable long-term identifier

Represents the deviation from perfect 90-degree orthogonality between the I and Q local oscillator signals. This error causes cross-talk where the I component leaks into the Q component and vice versa.

• Typical Range: 1 to 10 degrees for integrated transceivers
• Effect: Rotates and skews the constellation, transforming a square into a parallelogram
• Modeling: Expressed as a phase error matrix applied to the complex baseband signal
• Signature: The specific angle of skew is unique per device due to manufacturing variances in the quadrature mixer

A holistic metric quantifying the combined effect of gain and phase imbalance. IRR measures the power difference between the desired signal and the unwanted mirror-frequency image generated by the imbalance.

• Calculation: IRR (dB) = 10 * log10( (1 + γ² + 2γ cos(φ)) / (1 + γ² - 2γ cos(φ)) ) where γ is the gain ratio
• Typical Values: 25-40 dB for uncorrected consumer hardware
• Fingerprint Role: The specific IRR value across frequency forms a spectral signature that is difficult to clone

Unlike static imbalance, this model captures mismatches that vary across the signal bandwidth, caused by unequal low-pass filter responses in the I and Q paths.

• Source: Component tolerances in analog baseband filters and trace length differences on the PCB
• Modeling: Implemented using asymmetric FIR or IIR filters on the I and Q branches
• Impact: Causes frequency-selective constellation warping, where subcarriers at band edges exhibit different distortion than the center
• Fingerprint Depth: Provides a richer, higher-dimensional feature set for deep learning classifiers compared to static imbalance alone

A secondary impairment often modeled alongside I/Q imbalance. DC offset is a constant voltage added to the I or Q signal, while LO leakage is the unintended radiation of the unmodulated carrier.

• Constellation Effect: Shifts the entire constellation away from the origin
• Spectral Effect: Produces a visible tone at the carrier frequency, even without modulation
• Modeling: Added as a complex constant c = c_I + j*c_Q to the baseband signal
• Synergy: The interaction between DC offset and I/Q imbalance creates a unique composite distortion that enhances fingerprint distinctiveness

The complexity trade-off in I/Q imbalance simulation. A single-tap model applies a static gain and phase error, while a multi-tap model captures memory effects and frequency selectivity.

• Single-Tap: y(t) = x(t) + α * conj(x(t)) where α is a complex imbalance coefficient. Fast to compute, suitable for narrowband signals
• Multi-Tap: Uses a widely-linear filter with multiple coefficients to model frequency-dependent effects. Required for wideband signals like OFDM
• Selection Criteria: Narrowband IoT (single-tap) vs. Wi-Fi/LTE (multi-tap)
• Training Impact: Multi-tap models generate more realistic synthetic data, improving neural network generalization to real-world captures

The direct visual and mathematical manifestation of I/Q imbalance on a signal's constellation diagram. Gain mismatch causes the ideal square constellation to become rectangular, while phase error skews the axes, warping the symbol points. This distortion is a primary, stable feature used for transmitter fingerprinting because it creates a unique, device-specific constellation signature that is independent of the transmitted data.

A related impairment often modeled alongside I/Q imbalance. It represents the unintended radiation of the mixer's unmodulated carrier signal, which manifests as a DC offset in the baseband I/Q constellation. This shifts the entire constellation away from the origin. The combination of LO leakage and I/Q imbalance creates a compound, highly unique hardware signature that is difficult to clone.

The holistic metric used to quantify the severity of I/Q imbalance and other impairments. EVM measures the Euclidean distance between the ideal constellation point and the actual transmitted point. By synthetically injecting a specific I/Q gain and phase mismatch, you can precisely degrade the EVM to a target value, creating a labeled dataset where the impairment severity is a known, controllable parameter for training robust fingerprinting models.

While DPD is designed to linearize a power amplifier, the correction process itself leaves behind unique, low-level residual distortions. These artifacts are influenced by the initial I/Q imbalance of the transmitter's modulator. Modeling the interaction between the I/Q mismatch and the DPD correction loop generates a subtle, second-order fingerprint that is extremely difficult to mask or spoof, providing a deep hardware identity marker.

I/Q imbalance is fundamentally an analog impairment, but it interacts with data converter errors. DAC quantization error and ADC jitter can exacerbate or mask the gain and phase mismatches. A complete synthetic model must co-simulate these effects to understand how the I/Q imbalance signature survives the conversion to and from the digital domain, which is critical for generating realistic training data for receiver-side fingerprinting algorithms.

The primary challenge in using I/Q imbalance for fingerprinting is that multipath fading and Doppler shift can distort the received constellation, obscuring the hardware impairment. This term refers to the machine learning techniques, such as domain randomization and contrastive learning, that force a neural network to isolate the stable I/Q imbalance signature from the time-varying channel effects, ensuring reliable authentication in dynamic environments.