Introduction
In the deployment of 5G and massive MIMO systems, phased arrays are critical for precise beam steering. To maximize energy efficiency, power amplifiers (PAs) within these arrays are often driven near their saturation points. This operational state introduces significant nonlinearity, leading to spectral regrowth and signal distortion. While Digital Predistortion (DPD) is the standard remedy, implementing individual feedback loops for every PA in a large array is hardware-intensive and cost-prohibitive.
This paper presents a streamlined DPD strategy utilizing a single "Over-the-Air" (OTA) observation antenna. By identifying an optimal steering angle for calibration, the researchers demonstrate that a single set of DPD coefficients can effectively linearize the entire array across all scanning directions.
Proposed Architecture
The authors propose a simplified DPD architecture that replaces complex per-branch feedback with a single observation antenna (OA) located in the far field. This antenna captures the combined signal of the array, which is then processed through an observation chain to update predistortion coefficients.
The system utilizes a joint nonlinear model based on Memory Polynomials (MP). Instead of characterizing each PA individually, the model describes the aggregate distortion of the array as perceived at the far-field observation point. An indirect learning architecture is then employed to find the DPD coefficients that invert this joint PA model.
Training Methodology and Configuration
A core challenge of OTA-DPD is its sensitivity to the placement of the observation antenna. If the main beam steers away from the calibration point, performance typically degrades. The researchers addressed this by analyzing the relationship between steering angles and DPD robustness.
The methodology was validated using simulations of a 1 × 4 Uniform Linear Array (ULA) and a 4 × 4 Uniform Rectangular Array (URA). The study found that at a steering angle of (60°, 60°), the beamwidth is widest and signal amplitude changes are minimized. By briefly suspending user transmission to point the array at this "optimum" angle for training, the extracted DPD coefficients become robust to placement mismatches of up to ±10°.
Key Results and Performance
The proposed strategy was evaluated using a 256-QAM modulated signal with a 20 MHz bandwidth. The results confirm that the method meets stringent industry standards:
- Spectral Purity: The Adjacent Channel Power Ratio (ACPR) was improved to less than -55 dBc, satisfying 3GPP requirements.
- Signal Integrity: The Error Vector Magnitude (EVM) was reduced to below 3.5%.
- Efficiency: The system successfully linearizes the array across its entire scanning range using only one feedback circuit and one set of universal coefficients.
Conclusion
By leveraging the inherent properties of phased array beam patterns, this research provides a practical path for implementing DPD in massive MIMO systems. Training at a robust "sweet spot" angle eliminates the need for massive coefficient libraries or complex hardware, ensuring high-quality signal transmission with minimal overhead. Future work will focus on validating these findings in dynamic real-world environments with hardware-in-the-loop testing.
Summary updated