Digital Pre-Distortion (DPD)

The core principle of DPD is to cascade a non-linear element—the predistorter—with the power amplifier (PA) such that the combined transfer function is linear. The predistorter implements an inverse model of the PA's AM-AM and AM-PM distortion characteristics.

• Gain Expansion: The predistorter applies gain expansion where the PA compresses, and vice versa.
• Phase Rotation: It introduces a complementary phase shift to cancel the PA's AM-PM conversion.
• Cascaded Response: The goal is a straight-line output vs. input relationship up to the PA's saturation point.

Modern wideband signals (e.g., 100 MHz for 5G) excite memory effects in PAs, where the current output depends on past inputs. DPD must model these temporal dependencies to be effective.

• Thermal Memory: Slow changes in transistor junction temperature affecting gain.
• Electrical Memory: Fast effects from bias network impedance and trapping phenomena.
• Tapped Delay Lines: DPD models use delayed versions of the input envelope to capture these dynamics.
• Volterra Series: The mathematical foundation, though often pruned to a Generalized Memory Polynomial (GMP) for practical implementation.

DPD systems estimate the predistorter coefficients using one of two primary learning architectures, each with distinct trade-offs.

• Indirect Learning Architecture (ILA): Swaps the PA's input and output to identify a post-distorter, then copies it to the predistorter. It is computationally simple but sensitive to measurement noise.
• Direct Learning Architecture (DLA): Iteratively minimizes the error between the desired linear output and the actual PA output. It is more robust but requires a real-time PA model.
• Neural Network DPD: Replaces polynomial models with feed-forward or recurrent neural networks to learn a more generalized inverse function, often outperforming classical models for complex Doherty PAs.

A deployed DPD system is not static. It must continuously adapt to track changes in the PA's behavior over time and environmental conditions.

• Online Training: Coefficients are updated during live transmission without interrupting service, using the transmit signal and a feedback observation path.
• Offline Training: Initial model identification is performed in a lab using dedicated training sequences before deployment.
• Drift Compensation: Tracks slow changes due to temperature drift, component aging, and antenna load mismatch (VSWR changes).
• Model Order Reduction: Techniques like principal component analysis (PCA) are used to reduce the number of adaptive coefficients, lowering computational load.

The primary economic and engineering driver for DPD is to maximize Power-Added Efficiency (PAE) while meeting strict regulatory emission masks.

• Back-Off Reduction: Without DPD, a PA must operate with a large output back-off (OBO) from its saturation point to maintain linearity, severely reducing efficiency. DPD allows operation closer to the 1 dB compression point (P1dB).
• Spectral Regrowth Suppression: Non-linearity causes intermodulation distortion (IMD) that spreads the signal's bandwidth, leaking power into adjacent channels. DPD suppresses this regrowth.
• ACLR Compliance: DPD is the primary technique to meet Adjacent Channel Leakage Ratio (ACLR) specifications mandated by 3GPP for 4G and 5G base stations.
• Crest Factor Reduction (CFR): Often paired with DPD, CFR reduces the Peak-to-Average Power Ratio (PAPR) before the predistorter, further improving efficiency.

Next-generation radio systems introduce complexities that require DPD to evolve beyond single-antenna, single-band correction.

• Massive MIMO DPD: In large antenna arrays, each beam experiences a different composite PA distortion. Beam-dependent DPD linearizes the signal in the direction of the beam, not just at the antenna element.
• Over-the-Air (OTA) DPD: Uses a remote observation receiver to capture the radiated signal, compensating for antenna mutual coupling and impedance mismatches that internal feedback paths miss.
• Dual-Band DPD: Linearizes concurrent multi-band transmission, where intermodulation products fall across widely separated frequencies.
• Envelope Tracking (ET) PA DPD: When the PA's supply voltage is dynamically modulated, the distortion becomes a complex function of both the input envelope and the supply voltage, requiring a multi-dimensional DPD model.