I/Q Imbalance MIMO DPD

Integrates I/Q imbalance correction and power amplifier linearization into a single unified model rather than cascading separate compensators. This approach captures the cross-interaction between modulator imperfections and PA nonlinearity, which cascade systems miss. The joint model uses an augmented basis function set that includes both the original signal terms and their conjugate mirror images to address frequency-dependent I/Q mismatch. By processing these terms through a shared nonlinear polynomial structure, the architecture corrects the composite distortion in one computational pass, reducing overall latency and coefficient count compared to sequential correction chains.

Unlike frequency-independent I/Q correction, this technique models frequency-dependent gain and phase imbalance across the signal bandwidth. The impairment is characterized by:

• Amplitude imbalance: α(ω) — frequency-varying gain difference between I and Q branches
• Phase imbalance: φ(ω) — frequency-dependent deviation from 90° orthogonality
• DC offset: Static and dynamic offsets introduced in the modulator

The model incorporates these parameters into the DPD coefficient estimation, ensuring that the predistorter simultaneously flattens the in-band response and suppresses the image frequency interference that frequency-selective mismatch generates.

Employs widely-linear (WL) processing to handle the improper nature of I/Q imbalanced signals. Standard linear models assume circular symmetry, which breaks down when quadrature errors are present. The WL framework augments the DPD basis with:

• Direct signal terms: x(n), x(n)|x(n)|², x(n)|x(n)|⁴
• Conjugate signal terms: x*(n), x*(n)|x(n)|², x*(n)|x(n)|⁴
• Cross-memory terms: Products of delayed direct and conjugate samples

This dual-basis structure enables the predistorter to independently control the upper and lower sidebands, correcting the asymmetric spectral regrowth characteristic of I/Q-impaired transmitters.

In MIMO arrays, each transmit branch exhibits unique I/Q imbalance signatures due to component tolerances, temperature gradients, and manufacturing variance. The DPD system maintains per-branch calibration coefficients that capture:

• Individual modulator frequency responses
• Branch-specific DC offset values
• Temperature-dependent drift parameters

These coefficients are stored in a calibration table and applied as pre-correction before the common DPD engine. During online operation, periodic recalibration using loopback paths or over-the-air measurements updates the per-branch parameters without interrupting transmission, ensuring consistent linearity across the entire array despite environmental changes.

A primary performance metric for I/Q imbalance DPD is image rejection ratio (IRR) — the power difference between the desired signal and its unwanted image. Joint correction typically achieves:

• 60-70 dB IRR for narrowband signals
• 45-55 dB IRR for wideband (>100 MHz) signals
• 10-15 dB improvement over separate I/Q correction followed by DPD

The enhanced rejection directly translates to improved error vector magnitude (EVM) and reduced adjacent channel leakage ratio (ACLR). This is particularly critical in massive MIMO systems where the aggregate image power from dozens of branches can create significant spatial interference patterns.

The parameter extraction process extends standard least-squares estimation to handle the augmented widely-linear model. The estimation framework:

• Constructs a composite regression matrix including both direct and conjugate basis functions
• Applies regularization to prevent overfitting from the expanded parameter set
• Uses recursive least squares (RLS) for online tracking of slowly varying I/Q parameters
• Incorporates conjugate symmetry constraints to reduce the effective parameter count

The computational complexity scales linearly with the number of MIMO branches when coefficient sharing is employed, making the approach feasible for massive MIMO arrays with 64+ transmit chains.

The foundational technique for correcting gain mismatch and phase orthogonality errors in quadrature modulators. Frequency-independent imbalance creates a mirror image of the desired signal, while frequency-dependent imbalance—caused by mismatched low-pass filters in the I and Q branches—produces a more complex, frequency-selective image. Compensation typically applies a widely-linear filter structure that processes both the original signal and its complex conjugate to cancel the unwanted image component.

A signal processing method that mitigates unintended electromagnetic interaction between adjacent antenna elements in a MIMO array. Crosstalk paths create a nonlinear mixing of signals from different branches before they reach the antennas, complicating the predistortion problem. Cancellation requires modeling the coupling matrix between elements and applying inverse filtering to decouple the branches before or during DPD processing.

A widely-used behavioral model structure that captures both static nonlinearity and memory effects in power amplifiers. The generalized memory polynomial (GMP) extends this by including cross-terms between delayed envelope samples and complex baseband samples. For I/Q imbalance MIMO DPD, the model is augmented with conjugate signal paths to represent the image components generated by quadrature errors, creating a widely-linear memory polynomial structure.

The mathematical techniques used to extract predistorter parameters from observed data. Common approaches include:

• Least Squares (LS): Batch estimation minimizing squared error
• Recursive Least Squares (RLS): Adaptive tracking for time-varying impairments
• Least Mean Squares (LMS): Low-complexity gradient-based adaptation Joint I/Q and nonlinearity estimation requires solving a widely-linear system that includes both direct and conjugate basis functions.

An adaptive predistortion architecture that iteratively minimizes the error between the desired linear output and the actual PA output to directly estimate predistorter coefficients. Unlike indirect learning, DLA does not assume the PA model is invertible. For I/Q imbalance MIMO DPD, the direct learning approach jointly optimizes the widely-linear predistorter by comparing the transmitted signal against a clean reference, accounting for both quadrature errors and nonlinearity in a single optimization loop.

The primary regulatory motivation for DPD—reducing adjacent channel leakage ratio (ACLR) caused by PA nonlinearity. I/Q imbalance introduces additional spectral asymmetry, where the image component spills into adjacent channels differently on the upper and lower sides. Joint correction must simultaneously suppress both symmetric spectral regrowth from AM-AM/AM-PM distortion and asymmetric regrowth from quadrature modulator impairments to meet emission mask requirements.