AM-PM Distortion

The defining characteristic of AM-PM distortion is that the phase shift through the amplifier is not constant but changes with the input signal's envelope amplitude. As the instantaneous power level fluctuates—especially near the amplifier's compression point—the device's internal capacitances and transit times vary, causing a dynamic phase rotation. This creates a non-linear phase trajectory that distorts the constellation diagram of modulated signals, rotating outer constellation points differently than inner ones.

AM-PM distortion originates from several semiconductor physics effects within the transistor:

• Input capacitance variation: The gate-to-source capacitance (Cgs) in FETs changes non-linearly with the applied voltage swing, altering the input impedance and phase response.
• Drain-to-gate feedback: The Miller capacitance couples the output voltage back to the input, and its non-linear nature introduces amplitude-dependent phase modulation.
• Transit time modulation: In bipolar devices, the electron transit time through the base-collector depletion region varies with instantaneous collector current, directly shifting the phase of the output signal.
• Self-heating effects: Rapid temperature fluctuations within the die during high-power operation alter semiconductor mobility, contributing to slow-varying phase memory effects.

AM-PM distortion severely degrades Error Vector Magnitude (EVM) in modern communication systems:

• Constellation warping: Higher-order QAM constellations (64-QAM, 256-QAM) exhibit a spiral-like distortion where outer symbols experience greater phase rotation than inner symbols, collapsing the noise margin.
• Spectral regrowth: The phase non-linearity generates intermodulation products that spread energy into adjacent channels, increasing Adjacent Channel Leakage Ratio (ACLR) and violating regulatory emission masks.
• OFDM vulnerability: Orthogonal frequency-division multiplexing signals with high Peak-to-Average Power Ratio (PAPR) are particularly susceptible, as the frequent amplitude peaks repeatedly drive the amplifier into regions of severe phase distortion.

While both are manifestations of power amplifier non-linearity, they affect different signal properties:

• AM-AM distortion (amplitude-to-amplitude): The output amplitude is a non-linear function of the input amplitude, causing gain compression or expansion. This distorts the magnitude of constellation points.
• AM-PM distortion (amplitude-to-phase): The output phase shift varies with input amplitude, causing phase rotation that depends on instantaneous power.
• Joint impact: In real amplifiers, AM-AM and AM-PM occur simultaneously and are often correlated. Modern Digital Pre-Distortion (DPD) systems must compensate for both using complex-valued correction functions that address magnitude and phase errors jointly.

AM-PM distortion is quantified through specialized measurement techniques:

• Complex gain measurement: Using a vector network analyzer or vector signal analyzer, the amplifier's complex gain (magnitude and phase) is measured across a sweep of input power levels, producing AM-AM and AM-PM transfer curves.
• AM-PM coefficient: Expressed in degrees per dB, this metric quantifies how many degrees of phase shift occur for each decibel change in input power. Values above 1°/dB typically indicate significant distortion requiring linearization.
• Dynamic characterization: For amplifiers with memory effects, static single-tone measurements are insufficient. Two-tone or modulated signal tests with envelope tracking capture the dynamic phase behavior under realistic operating conditions.

Neural network and polynomial DPD models address AM-PM distortion through inverse phase pre-distortion:

• Complex-valued predistortion: The DPD actuator applies a complex gain correction that is a function of the instantaneous input magnitude, pre-rotating the phase in the opposite direction to cancel the amplifier's AM-PM characteristic.
• Memory polynomial phase terms: In Generalized Memory Polynomial (GMP) models, cross-terms between the current signal and lagging envelope values capture the phase memory effects caused by thermal and bias network dynamics.
• Neural network phase modeling: Architectures like the Real-Valued Time-Delay Neural Network (RVTDNN) process I and Q components separately with tapped delay lines, learning the non-linear mapping between amplitude history and required phase correction without explicit analytical formulation.

The complementary non-linearity to AM-PM, describing the amplitude-to-amplitude transfer function deviation. While AM-PM distorts phase, AM-AM causes gain compression or expansion as the input envelope varies. Together, they form the complex non-linear gain response of a power amplifier. Modern DPD systems must correct both simultaneously, as they are inherently coupled in the transistor's physics. AM-AM is typically characterized by the 1 dB compression point (P1dB), where gain drops by 1 dB from its small-signal value.

A comprehensive signal quality metric that directly captures the impact of AM-PM distortion on a transmitted constellation. EVM measures the vector difference between the ideal and actual symbol locations in the I/Q plane. AM-PM causes a rotational spreading of constellation points, increasing EVM even when amplitude errors are minimal. For high-order QAM schemes like 1024-QAM, tight EVM requirements (<1%) demand aggressive AM-PM correction, as phase errors become the dominant impairment limiting data throughput.

The phenomenon where a power amplifier's current output depends on past input values, causing AM-PM distortion to become frequency-dependent. Thermal memory arises from die heating over microsecond timescales, while electrical memory stems from bias network impedances and trapping effects in GaN transistors. These effects cause the phase shift to vary not just with instantaneous amplitude but with the signal's envelope history. Generalized Memory Polynomial (GMP) models explicitly include cross-terms between the signal and its lagging envelope to capture this dynamic AM-PM behavior.

A high-efficiency architecture that exhibits particularly severe AM-PM distortion due to its load modulation mechanism. As the peaking amplifier turns on and off, the impedance seen by the main amplifier changes dynamically, causing a pronounced phase shift with input power. This non-linear phase characteristic is a major design challenge, requiring advanced DPD with strong AM-PM correction capability. The Doherty's efficiency advantage—achieving high PAE over a wide power back-off range—makes it ubiquitous in modern base stations despite its complex distortion profile.

An advanced linearization approach where deep neural networks learn the inverse complex transfer function of the power amplifier, including its AM-PM characteristic. Unlike polynomial models, neural networks can capture highly non-linear phase responses without requiring explicit basis function engineering. Architectures like the Real-Valued Time-Delay Neural Network (RVTDNN) process I and Q components separately with tapped delay lines to model both static AM-PM and memory effects. These models excel in scenarios where traditional Volterra-based approaches struggle with model order explosion.

A critical regulatory metric that quantifies spectral regrowth caused by intermodulation distortion, of which AM-PM is a primary contributor. When a modulated signal passes through a non-linear amplifier, phase distortion generates intermodulation products that spill into adjacent frequency channels. ACLR measures the ratio of power in the main channel to power in the adjacent channels, typically requiring values below -45 dBc. AM-PM correction in DPD directly suppresses this out-of-band emission, ensuring compliance with 3GPP and FCC spectral masks.