AM-PM Distortion

The fundamental mechanism of AM-PM distortion is a phase shift that changes with input power level. As the instantaneous envelope magnitude increases, the amplifier's insertion phase deviates from its small-signal value. This creates a nonlinear phase trajectory where different amplitude levels experience different phase delays, directly corrupting the phase information encoded in complex modulation schemes like QAM and OFDM. The result is constellation rotation that varies symbol-by-symbol based on instantaneous power.

A primary physical origin of AM-PM distortion in FET-based amplifiers is the nonlinear gate-to-source capacitance (Cgs) and gate-to-drain capacitance (Cgd) . These capacitances behave as voltage-dependent varactors whose value changes with the instantaneous gate and drain voltage swings. As the input drive level increases, the effective input capacitance modulates, altering the input matching network's phase response and introducing an amplitude-dependent phase shift through the device.

The AM-PM conversion factor quantifies the sensitivity of phase shift to amplitude variations, typically expressed in degrees per decibel (deg/dB) . A lower kp indicates better phase linearity. Key characteristics:

• Measured by applying a small-amplitude modulation tone and observing the resulting phase modulation sidebands
• Varies across the amplifier's operating range, often worsening near compression
• Critical for cascaded systems where AM-PM from driver stages compounds with final-stage distortion

AM-PM distortion directly degrades Error Vector Magnitude (EVM) by introducing phase errors that depend on the transmitted symbol's amplitude. In high-order QAM constellations (64-QAM, 256-QAM), outer constellation points experience greater phase rotation than inner points, creating a non-uniform phase dispersion. This amplitude-dependent phase error cannot be corrected by a simple static phase rotation and requires dynamic compensation through digital predistortion to restore modulation accuracy.

Unlike AM-AM distortion which produces symmetric spectral regrowth, AM-PM distortion generates asymmetric intermodulation sidebands around the carrier. This asymmetry is a telltale signature of phase nonlinearity in spectrum analyzer measurements. The upper and lower adjacent channel power levels become unbalanced, complicating ACLR compliance. Memory effects in AM-PM further distort this asymmetry, making it frequency-dependent and requiring wideband linearization solutions.

In Doherty amplifiers, AM-PM distortion is exacerbated by the load modulation mechanism itself. As the peaking amplifier activates and injects current into the combiner, the impedance seen by the carrier amplifier changes dynamically. This impedance trajectory traverses non-constant-phase contours on the Smith chart, introducing an additional amplitude-dependent phase shift beyond the transistor's intrinsic AM-PM. The combined effect creates a complex, power-dependent phase profile that demands sophisticated memory-polynomial or neural network DPD models.

The complementary nonlinear effect to AM-PM, representing the amplitude-to-amplitude transfer characteristic deviation. While AM-PM distorts phase, AM-AM compresses or expands the output envelope magnitude as a function of instantaneous input power. Both distortions originate from the same nonlinear device physics—primarily the input capacitance variation with gate voltage and the transconductance nonlinearity of the transistor. In GaN HEMTs, AM-AM often manifests as soft compression, while AM-PM shows a characteristic phase expansion followed by compression. Digital predistortion must correct both simultaneously using a complex baseband model where the predistorter's in-phase component addresses AM-AM and the quadrature component addresses AM-PM.

Dynamic nonlinear phenomena that cause the amplifier's AM-PM characteristic to depend not only on the instantaneous envelope but also on past signal history. Memory effects introduce hysteresis into the AM-PM curve, creating a scattered, frequency-dependent distortion pattern rather than a static single-line transfer function.

• Thermal memory: Self-heating changes junction temperature over microsecond-to-millisecond timescales, altering phase shift dynamically.
• Electrical memory: Bias network impedance at the envelope frequency modulates the drain voltage, causing phase variations tied to the signal envelope bandwidth.
• Trapping effects: Charge capture and release in semiconductor surface states create slow dispersion with sub-hertz to kilohertz time constants.

Compensating memory effects requires a predistorter with memory, such as a memory polynomial or Volterra series model, not a simple static lookup table.

A critical modulation quality metric directly degraded by AM-PM distortion. EVM measures the vector difference between the ideal constellation point and the actual transmitted symbol. AM-PM causes a phase rotation of the received symbol that varies with the symbol's amplitude, creating a phase-dependent error that cannot be corrected by a simple gain adjustment.

For high-order QAM schemes like 1024-QAM or 4096-QAM used in 5G and microwave backhaul, the phase error budget is extremely tight—often less than 1 degree RMS. Uncompensated AM-PM in a Doherty amplifier can introduce 5–15 degrees of phase shift across the dynamic range, making linearization essential. EVM specifications in 3GPP require values below 3.5% for 256-QAM, which demands AM-PM correction to within fractions of a degree.

In many transistor technologies, particularly GaN HEMTs and LDMOS, AM-AM and AM-PM exhibit a characteristic coupled behavior: as the input drive approaches compression, the gain begins to decrease (AM-AM compression) while the phase simultaneously advances (AM-PM expansion). This phase expansion results from the nonlinear input capacitance—the gate-to-source capacitance Cgs decreases as the depletion region widens under large-signal swing, reducing the input time constant and causing a phase lead.

At deeper saturation, the phase may reverse and begin to lag as the transconductance collapses. This non-monotonic AM-PM shape creates a challenging predistortion target, requiring models with sufficient nonlinear order (typically 7th to 11th order polynomials) to capture the inflection point accurately.

A systematic measurement methodology used to characterize AM-PM as a function of load impedance presented to the transistor. By varying the load reflection coefficient across the Smith chart under large-signal drive, engineers map contours of constant AM-PM conversion. This reveals the sensitivity of phase distortion to impedance—critical for Doherty amplifier design where the carrier amplifier's load impedance is actively modulated by the peaking amplifier's current injection.

Load-pull data shows that AM-PM is typically minimized near the maximum efficiency impedance but may be compromised at impedances chosen for bandwidth or gain flatness. Modern vector load-pull systems with wideband modulation capability can extract dynamic AM-PM surfaces, providing the training data for neural network predistortion models that must operate across varying VSWR conditions.

The primary technique for compensating AM-PM distortion in modern transmitters. DPD applies an inverse nonlinearity to the baseband signal before the power amplifier, such that the cascade of predistorter and amplifier yields a linear overall response. For AM-PM correction, the predistorter must generate a phase rotation opposite to the amplifier's phase distortion at each instantaneous envelope level.

Implementation approaches include:

• Lookup table (LUT): Maps envelope magnitude to complex gain correction values, effective for static AM-PM but limited for memory effects.
• Memory polynomial: Extends the LUT concept with FIR filter taps to capture frequency-dependent AM-PM.
• Neural network: Deep learning models that learn the inverse AM-PM characteristic from training data, excelling at complex memory and multi-band scenarios.

Adaptive DPD systems continuously update coefficients using an observation receiver that captures the amplifier's output, comparing it to the desired linear response.