Linearity-Efficiency Trade-off

The amplifier's conduction angle—the portion of the RF cycle the transistor is active—is the primary control lever. Class-A biasing (360° conduction) offers the highest linearity but a theoretical maximum efficiency of only 50%, often dropping below 20% in practice. Class-B (180°) improves efficiency at the cost of crossover distortion. Class-AB is a compromise, while Class-C used in Doherty peaking amplifiers maximizes efficiency but is inherently highly nonlinear. The choice of bias class directly determines the severity of the AM-AM distortion that DPD must later correct.

Modern communication signals like OFDM (used in 5G and Wi-Fi) exhibit high PAPR, often exceeding 10 dB. To avoid clipping and spectral regrowth, the power amplifier must operate at a significant Output Back-Off (OBO) from its saturated power, where efficiency is poor. This forces a direct trade-off: high linearity at a low average power point sacrifices DC power efficiency. The higher the signal PAPR, the more severe the efficiency penalty, making linearization techniques like DPD critical for recovering efficiency without violating ACLR limits.

The trade-off is not purely instantaneous; it has a temporal dimension. Memory effects mean the amplifier's nonlinear behavior depends on previous signal states, not just the current envelope. These arise from:

• Thermal memory: Transistor self-heating changes gain and phase over microseconds.
• Electrical memory: Bias network impedance variations and envelope frequency-dependent matching.
• Trapping effects: Slow charge capture/release in GaN HEMT devices. These effects create a frequency-dependent nonlinearity that complicates the trade-off, requiring DPD models with memory, such as Memory Polynomials, to achieve both linearity and efficiency.

In a Doherty Power Amplifier, the linearity-efficiency trade-off is managed through active load modulation. As the Peaking Amplifier turns on, it injects current into the Doherty Combiner, dynamically reducing the impedance seen by the Carrier Amplifier. This allows the carrier to remain in saturation (high efficiency) over a wider OBO range. However, the transition region where the peaking device activates introduces significant AM-PM distortion and a sharp nonlinearity 'kink' in the AM-AM characteristic, placing a heavy burden on the DPD system to linearize this complex, modulated behavior.

The underlying transistor technology sets the physical boundaries of the trade-off. GaN HEMT devices offer high power density and low knee voltage, enabling higher efficiency at a given linearity compared to LDMOS or GaAs. GaN's soft compression characteristic is more gradual and thus more amenable to DPD correction than the hard saturation of other technologies. However, GaN also exhibits pronounced trap effects and self-heating, introducing complex memory that expands the dimensionality of the linearization problem.

Wider signal bandwidths (e.g., 100 MHz for 5G NR carrier aggregation) exacerbate the trade-off. The amplifier's frequency response becomes non-flat across the band, and memory effects become more pronounced. Simultaneously, a high crest factor (PAPR) forces deeper back-off. The combination of wide bandwidth and high PAPR demands DPD with extremely wide linearization bandwidth (typically 3-5x the signal bandwidth) and deep memory, pushing the limits of FPGA-Based DPD Implementation and Coefficient Estimation Algorithms to maintain both linearity and efficiency.

The power-added efficiency (PAE) of an amplifier when operating at an average output power level significantly below its saturated maximum. This is the critical metric for amplifying signals with high peak-to-average power ratios (PAPR).

• Class-A amplifiers maintain high linearity but suffer from poor back-off efficiency, often below 10%.
• Doherty architectures specifically target high back-off efficiency by using load modulation to maintain voltage swing.
• Typical back-off levels for 5G NR signals range from 6 to 12 dB, where conventional amplifiers waste significant DC power as heat.

Amplitude-to-amplitude modulation distortion represents the nonlinear relationship between the input signal envelope magnitude and the output signal envelope magnitude of a power amplifier.

• Manifests as gain compression at high drive levels, where incremental gain decreases.
• Quantified by the 1-dB compression point (P1dB), marking the onset of significant nonlinear behavior.
• In Doherty amplifiers, the transition between carrier-only and carrier-plus-peaking operation creates a characteristic AM-AM kink that digital predistortion must precisely invert.

Amplitude-to-phase modulation distortion represents the nonlinear phase shift introduced by a power amplifier that varies as a function of the instantaneous input signal envelope magnitude.

• Caused by voltage-dependent parasitic capacitances in the transistor, particularly gate-to-source and drain-to-gate capacitances.
• In GaN HEMT-based Doherty amplifiers, AM-PM can exceed 10 degrees across the dynamic range.
• AM-PM is often more challenging to linearize than AM-AM because phase distortion creates asymmetric spectral regrowth that degrades error vector magnitude (EVM).

Dynamic nonlinear distortions where the current output depends not only on the instantaneous input envelope but also on past signal values due to thermal, electrical, and trapping time constants.

• Short-term memory effects arise from bias network impedance and envelope frequency-dependent matching.
• Long-term memory effects stem from self-heating and charge trapping in GaN HEMT devices.
• Memory effects break the static AM-AM/AM-PM assumption, requiring Volterra series or memory polynomial models for accurate predistortion.

A regulatory compliance metric measuring the ratio of transmitted power within an assigned channel to the power that leaks into adjacent frequency channels due to spectral regrowth from amplifier nonlinearity.

• 3GPP specifications typically require ACLR better than -45 dBc for base station transmitters.
• Operating closer to saturation improves efficiency but degrades ACLR, creating the core trade-off.
• Digital predistortion can improve ACLR by 15-25 dB, enabling operation at higher efficiency points while maintaining spectral compliance.

A modulation quality metric quantifying the vector difference between the ideal reference constellation point and the actual measured transmitted symbol, degraded by amplifier nonlinearity and phase distortion.

• For 256-QAM and 1024-QAM modulation in 5G NR, EVM requirements are stringent, often below 3.5%.
• AM-PM distortion is a dominant contributor to EVM degradation in Doherty amplifiers.
• Linearization techniques must simultaneously optimize both ACLR and EVM, as improving one metric can sometimes degrade the other.