Over-the-Air DPD

OTA DPD uses a spatially separated observation receiver—typically a probe antenna placed in the far-field or near-field of the array—to capture the actual radiated waveform. This feedback path inherently includes the effects of antenna impedance mismatch, mutual coupling between array elements, and frequency-selective fading in the channel. By closing the loop over the air, the predistorter learns a composite inverse model that linearizes the entire transmit chain, not just the power amplifier in isolation.

In phased-array and massive MIMO systems, each beamforming configuration presents a different effective load impedance to each power amplifier element. This causes the non-linear distortion to become beam-dependent—a single DPD model trained for one beam angle will fail when the beam is steered elsewhere. OTA DPD addresses this by either:

• Training beam-indexed predistorter banks where each beam direction has a dedicated coefficient set
• Using neural network models that accept beamforming weights as auxiliary inputs to generalize across steering angles

When antenna elements are placed in close proximity—as in modern massive MIMO arrays with half-wavelength spacing—the radiated field from one element induces currents in neighboring elements. This mutual coupling creates a non-linear interaction that varies with frequency, beam angle, and element position. OTA DPD captures this coupling in the feedback signal, allowing the predistorter to pre-compensate for the cross-element distortion that conducted DPD systems cannot observe.

OTA DPD naturally integrates with array calibration procedures. The remote observation receiver simultaneously provides data for:

• I/Q imbalance correction across array branches
• Gain and phase alignment between transmit paths
• Non-linear predistorter coefficient extraction This unified approach eliminates the need for separate calibration and linearization workflows, reducing system complexity and improving overall error vector magnitude (EVM) performance at the radiated output.

A critical challenge in OTA DPD is decoupling the power amplifier non-linearity from the over-the-air channel response in the feedback path. Advanced techniques include:

• Training during quiet periods with minimal multipath
• Channel estimation and equalization applied to the observation receiver before DPD coefficient extraction
• Anechoic chamber pre-characterization to establish baseline models that are then adapted in the field Without this decoupling, the DPD model may inadvertently attempt to invert the channel rather than the amplifier.

OTA DPD imposes stringent latency and throughput requirements on the adaptation loop. The observation receiver must capture, digitize, and align the feedback signal with the transmitted reference within the coherence time of the amplifier's thermal dynamics. Typical implementations use:

• High-speed ADCs with sampling rates exceeding 5× the signal bandwidth
• FPGA-based time alignment using cross-correlation
• Reduced-complexity model architectures such as pruned neural networks or compact memory polynomials to meet real-time coefficient update deadlines

The deviation from the ideal 50-ohm impedance at the antenna port, causing signal reflections that alter the load seen by the power amplifier. In active antenna arrays, this mismatch varies dynamically with beam steering angle and mutual coupling between elements. Over-the-air DPD must capture the radiated signal to linearize the amplifier under these non-ideal, time-varying load conditions that conventional conducted DPD cannot observe.

The electromagnetic interaction between adjacent antenna elements in a phased array, where current induced in one element affects the radiation pattern and impedance of its neighbors. This coupling creates an element-specific non-linear load for each power amplifier that changes with scan angle and frequency. Over-the-air DPD uniquely captures the composite radiated field, inherently compensating for these coupling-induced distortions that conducted feedback loops miss.

A dedicated receiver placed in the far-field of the antenna array to capture the over-the-air transmitted signal for DPD coefficient extraction. The ROR must have a wider bandwidth and higher linearity than the main transmitter to faithfully digitize the distorted waveform, including spectral regrowth products. Key design challenges include maintaining phase coherence with the transmit reference and achieving sufficient signal-to-noise ratio at the observation point.

A phenomenon in active phased arrays where the composite distortion of the radiated signal changes as the beam is electronically steered. This occurs because each beam angle presents a different effective load impedance to each power amplifier due to varying mutual coupling and reflection coefficients. Over-the-air DPD must either learn a beam-indexed model or adapt in real-time as the beam scans, making it fundamentally more complex than per-element conducted DPD.

The critical process of characterizing and compensating for the transfer function of the observation receiver and propagation channel to isolate the true power amplifier distortion. This involves:

• Cable and connector de-embedding for conducted reference paths
• Free-space path loss compensation for wireless observation
• Phase synchronization between the transmit reference and observation receiver Without precise calibration, the DPD model learns the combined channel and amplifier response, degrading linearization performance.

An advanced over-the-air DPD architecture that combines conducted feedback from individual power amplifiers with radiated observation from a remote receiver. The conducted paths provide per-element non-linearity estimates, while the over-the-air path captures array-level effects like mutual coupling and beam-dependent distortion. A fusion algorithm reconciles both data streams to build a comprehensive predistorter model that linearizes the entire array as a unified system.