I/Q Imbalance Compensation

Gain imbalance occurs when the I and Q branches have unequal amplification, causing the constellation to stretch into a rectangle rather than a perfect square. Compensation applies a complex scaling factor to equalize the amplitudes.

• Origin: Component tolerances in mixers and DACs
• Impact: Asymmetric constellation distortion
• Correction: Adaptive amplitude scaling on one branch
• Metric: Gain error expressed in dB

Phase imbalance arises when the local oscillator signals are not exactly 90 degrees apart, causing the I and Q axes to lose orthogonality. This results in cross-talk between the I and Q channels and constellation rotation.

• Origin: Imperfect quadrature splitting in the LO path
• Impact: Inter-symbol interference and EVM degradation
• Correction: Complex rotation matrix applied to the baseband signal
• Metric: Phase error expressed in degrees

Unlike static imbalance, frequency-dependent I/Q mismatch varies across the signal bandwidth due to differing low-pass filter responses in the I and Q paths. Compensation requires adaptive filtering rather than a single complex coefficient.

• Origin: Mismatched analog baseband filters
• Impact: Frequency-selective image interference
• Correction: Complex FIR filter with asymmetric taps
• Joint consideration: Often combined with channel equalization

Modern transmitters integrate I/Q imbalance compensation directly into the DPD model because the two impairments cascade non-linearly. A power amplifier amplifies an already-imbalanced signal, creating cross-modulation products that a separate compensator cannot fully correct.

• Architecture: Augmented DPD coefficient vector
• Model: Generalized memory polynomial with I/Q cross-terms
• Benefit: Single adaptive loop corrects both impairments
• Application: Critical for high-order QAM (256-QAM, 1024-QAM)

Blind I/Q imbalance estimation derives correction parameters directly from the transmitted signal statistics without requiring dedicated training sequences. This enables online adaptation during live transmission.

• Method: Circularity-based algorithms assume proper complex signals
• Assumption: Ideal constellation is circularly symmetric
• Advantage: No spectral overhead for pilot tones
• Limitation: Sensitive to other impairments like DC offset

The image rejection ratio quantifies the effectiveness of I/Q imbalance compensation by measuring the suppression of the unwanted image signal. A high IRR indicates excellent quadrature balance.

• Definition: Ratio of desired signal power to image signal power
• Uncompensated: Typically 25-35 dB
• Compensated: Can exceed 60 dB with digital correction
• Measurement: Single-tone or multi-tone test signals

The root physical impairment where the in-phase (I) and quadrature (Q) branches of a modulator exhibit unequal amplitudes and a phase difference deviating from the ideal 90 degrees. This creates a mirror-frequency image that interferes with the desired signal, degrading the Error Vector Magnitude (EVM). In direct-conversion transmitters, these mismatches are caused by component tolerances in the local oscillator path and baseband amplifiers.

A unified linearization architecture that corrects power amplifier non-linearity and I/Q modulator impairments simultaneously using a single model. This is critical because the cascaded non-idealities interact; correcting them sequentially is suboptimal. A dual-input neural network or a Generalized Memory Polynomial (GMP) with augmented basis functions can model the combined non-linear dynamic response of the PA and the static frequency-independent I/Q imbalance in a single forward path.

A more complex impairment where the gain and phase mismatch vary across the signal's modulation bandwidth. This is typically caused by mismatched anti-aliasing filters or unequal trace lengths in the I and Q baseband paths. Compensation requires a complex-valued filter structure, such as a widely linear filter, rather than a simple scalar correction. Neural network models with tapped delay lines inherently capture this frequency-selective behavior.

The primary metric for quantifying I/Q imbalance severity, defined as the power ratio of the desired signal to the unwanted mirror-frequency image. A high IRR indicates good balance. Mathematically, it is derived from the complex gain imbalance factor. Compensation algorithms aim to maximize IRR, often targeting values above 50 dB for high-order QAM transmissions to prevent the image from corrupting the intended constellation points.

The mathematical framework essential for modeling I/Q imbalance. A real-valued bandpass signal is represented as a complex envelope centered at zero frequency. I/Q imbalance is modeled as a linear combination of the signal and its complex conjugate, where the conjugate term represents the distortion. This widely linear model is the foundation for both analytical compensator design and the construction of input feature vectors for neural network-based correction.

A class of algorithms that correct I/Q imbalance without requiring a dedicated training sequence, operating on the live payload signal. These methods exploit statistical properties of the transmitted signal, such as circularity or properness—the condition that a complex signal is uncorrelated with its complex conjugate. A blind adaptive filter iteratively minimizes this correlation to force the signal back to a proper state, effectively removing the image.