AM-AM Distortion

As the input amplitude increases, the amplifier's gain deviates from its linear small-signal value. The output amplitude stops increasing proportionally and compresses toward a maximum saturated output power (Psat).

• The 1 dB compression point (P1dB) marks where the actual gain drops 1 dB below the ideal linear gain.
• Beyond P1dB, the amplifier operates in severe non-linearity, causing clipping of signal peaks.
• This compression is the dominant mechanism behind spectral regrowth into adjacent channels.

The AM-AM characteristic is mathematically represented as a memoryless non-linear transfer function, mapping instantaneous input envelope voltage to output envelope voltage.

• Common models include the Rapp model for solid-state amplifiers and the Saleh model for traveling-wave tube amplifiers.
• The transfer function is typically expressed as a polynomial or rational function: Vout = f(Vin).
• Memoryless models assume the output depends only on the current input, ignoring thermal and trapping effects.

AM-AM non-linearity causes intermodulation distortion that spreads the signal's bandwidth into adjacent frequency channels.

• When a band-limited signal passes through a non-linear amplifier, the amplitude distortion generates third-order and fifth-order intermodulation products.
• This regrowth is quantified by the Adjacent Channel Leakage Ratio (ACLR).
• The severity of regrowth depends on the signal's Peak-to-Average Power Ratio (PAPR) and the amplifier's back-off from P1dB.

AM-AM distortion affects only the magnitude of the output signal, while AM-PM distortion introduces an unwanted phase shift dependent on input amplitude.

• Real amplifiers exhibit both effects simultaneously, forming a complex gain expansion/compression profile.
• AM-AM is measured by comparing input vs. output envelope voltages.
• AM-PM is measured as the phase difference between input and output as a function of instantaneous power.
• Together they form the complex gain characteristic used in memory polynomial digital pre-distortion models.

AM-AM distortion directly degrades the Error Vector Magnitude (EVM) of digitally modulated signals.

• Compression causes the outer constellation points to shrink inward, reducing the Euclidean distance between symbols.
• This increases the Bit Error Rate (BER) at the receiver, especially for high-order QAM schemes like 64-QAM and 256-QAM.
• In OFDM systems, the non-linear distortion also destroys subcarrier orthogonality, causing inter-carrier interference (ICI).

Each physical power amplifier has a unique AM-AM transfer function due to microscopic manufacturing variances in transistor doping, gate oxide thickness, and matching networks.

• These subtle differences create a hardware-specific distortion signature that can be extracted and used for transmitter identification.
• Synthetic impairment generators replicate these curves using parameterized models to create digital twins of specific devices.
• The compression knee sharpness, saturation level, and small-signal gain slope form a multi-dimensional feature vector for deep learning classifiers.

The companion non-linearity to AM-AM distortion, where the input signal amplitude causes an unwanted phase shift in the output signal. While AM-AM distorts magnitude, AM-PM introduces phase modulation that varies with instantaneous power. Together, they fully characterize a memoryless power amplifier's non-linear behavior. In synthetic impairment generation, AM-PM is modeled alongside AM-AM using complex gain polynomials or Saleh models to produce realistic constellation warping and spectral regrowth.

A critical figure of merit defining the input power level at which the amplifier's gain drops by 1 dB from its ideal linear value. P1dB marks the transition from linear operation to gain compression. In synthetic impairment modeling, P1dB serves as a key parameter to scale the AM-AM distortion curve, determining how aggressively the amplifier saturates. Signals operating near or above P1dB experience significant non-linear distortion, making this metric essential for realistic digital twin calibration.

A theoretical metric extrapolated from the slopes of the fundamental and third-order intermodulation products. IP3 quantifies an amplifier's linearity performance and predicts the severity of intermodulation distortion. Higher IP3 values indicate better linearity. In synthetic impairment generation, IP3 is used to parameterize the polynomial coefficients of the AM-AM transfer function, ensuring that the simulated spectral regrowth and adjacent channel interference match real hardware behavior.

The unintended broadening of a signal's occupied bandwidth caused by amplifier non-linearity. When a modulated signal with a non-constant envelope passes through an AM-AM distorted amplifier, intermodulation products generate out-of-band emissions in adjacent channels. Spectral regrowth is the primary observable consequence of AM-AM distortion in the frequency domain. Synthetic impairment simulators must accurately reproduce this regrowth to train deep learning models that can identify devices by their unique spectral leakage patterns.

A widely-used two-parameter mathematical model for representing traveling-wave tube amplifier (TWTA) non-linearity. The Saleh model provides closed-form expressions for both AM-AM conversion (amplitude distortion) and AM-PM conversion (phase distortion) using only four scalar parameters. Its simplicity and accuracy make it a standard building block in synthetic impairment generation pipelines, allowing engineers to rapidly generate diverse, labeled datasets of non-linear amplifier behavior for training robust fingerprinting classifiers.

A behavioral model specifically designed for solid-state power amplifiers (SSPAs) that exhibit a sharper transition from linear to saturation regions compared to TWTAs. The Rapp model uses a smoothness factor to control the knee sharpness of the AM-AM curve. In synthetic impairment generation, varying the Rapp smoothness parameter across simulated devices creates distinct, device-specific compression characteristics that serve as highly discriminative features for RF fingerprinting neural networks.