Spectral Regrowth

Memory effects mean the PA's current output depends not only on the present input but also on past signal values. This dynamic nonlinearity is a critical barrier to linearization. Short-term (electrical) memory is caused by bias network impedance and trapping effects, while long-term (thermal) memory results from die temperature changes with signal envelope variations.

• Impact: Causes an asymmetric spectral regrowth profile that simple memoryless DPD cannot correct.
• Mitigation: Requires Volterra series or memory polynomial-based predistorters.

Modern wideband signals like OFDM have a high PAPR, meaning they contain infrequent but extreme amplitude peaks. To avoid clipping, the PA must operate with a large back-off from its saturation point, but even momentary excursions into the nonlinear region during a peak generate severe spectral regrowth.

• Challenge: A 10-12 dB PAPR in a 5G signal forces the PA to operate at very low average efficiency.
• Solution: Crest Factor Reduction (CFR) is often applied before DPD to limit peaks, trading off a controlled amount of in-band distortion (EVM) for a dramatic reduction in out-of-band emissions.

As signal bandwidths increase to 100 MHz and beyond for 5G, the PA's frequency response is no longer flat. The nonlinear behavior becomes frequency-selective, meaning the distortion characteristics change across the channel. This is exacerbated by the bandwidth expansion factor, where the predistorted signal must be several times wider than the original to cancel out-of-band distortion.

• Consequence: A single, wideband DPD model becomes insufficient.
• Approach: Requires multi-rate DPD or frequency-selective predistortion architectures to handle the varying nonlinear profile across the spectrum.

In high-efficiency architectures like Envelope Tracking (ET), the PA's drain voltage is dynamically modulated to follow the signal envelope. This intentional modulation creates a complex, dynamic interaction where the PA's gain and phase response become a function of both the input signal and the instantaneous supply voltage.

• Complexity: The DPD system must linearize a 2D nonlinear surface (input power vs. supply voltage).
• Requirement: A joint ET-DPD characterization and linearization strategy is essential to prevent catastrophic spectral regrowth while maintaining efficiency gains.

The antenna's impedance is not perfectly constant; it changes with frequency and environmental conditions. A varying load impedance presented to the PA output alters its nonlinear characteristics, a phenomenon quantified by load-pull contours. This mismatch can instantly degrade the performance of a fixed DPD model.

• Real-world impact: A handset's antenna detuning can cause a sudden spike in ACLR.
• Countermeasure: Adaptive DPD algorithms that can track and compensate for load variations in real-time are necessary for robust mobile operation.

The primary regulatory metric for quantifying spectral regrowth. ACLR measures the ratio of filtered mean power centered on the assigned channel frequency to the filtered mean power centered on an adjacent channel. It is typically specified in dBc.

• 3GPP specifications mandate ACLR limits (e.g., -45 dBc for base stations) to prevent interference.
• Directly impacted by the order of nonlinearity; third-order intermodulation products are the dominant contributors.
• Improving ACLR is the explicit optimization target for Digital Pre-Distortion (DPD) systems.

The physical mechanism causing spectral regrowth. When multiple frequency components pass through a nonlinear device, they mix to generate spurious products at sums and differences of integer multiples of the original frequencies.

• Third-order IMD (IMD3) falls directly adjacent to the original carriers, causing immediate spectral regrowth.
• Fifth-order IMD (IMD5) affects channels further out but is typically lower in power.
• The amplitude of IMD products grows nonlinearly with input power; a 1 dB increase in fundamental power can cause a 3 dB increase in IMD3 power.

The ratio of the predistorted signal bandwidth to the original signal bandwidth. To cancel third-order distortion, the predistorter must generate inverse distortion products that occupy up to 3x to 5x the original signal bandwidth.

• A 100 MHz 5G NR signal requires a DPD feedback path with at least 300-500 MHz of linear observation bandwidth.
• This expansion drives the requirement for high-speed Analog-to-Digital Converters (ADCs) in the observation receiver.
• Insufficient bandwidth leads to aliasing distortion and incomplete cancellation of spectral regrowth.

Power amplifier nonlinearity is not static; it depends on past signal values. Memory effects cause asymmetry in the spectral regrowth profile, making the upper and lower sidebands unequal.

• Electrical memory effects are caused by bias network impedance variations at the envelope frequency.
• Thermal memory effects result from dynamic transistor junction heating and cooling.
• Memory effects require Volterra series or Memory Polynomial models for accurate DPD compensation.

Regulatory bodies like the FCC and ITU define strict spectral masks that limit out-of-band emissions. Spectral regrowth is the primary cause of mask violations in transmitters using efficient but nonlinear power amplifiers.

• Spectrum Emission Mask (SEM) defines absolute power limits across a wide frequency range.
• Spurious emission limits apply far from the carrier, often requiring additional bandpass filtering.
• DPD is the primary active technique to ensure compliance without sacrificing power efficiency.

Spectral regrowth is modeled entirely in the complex baseband domain, representing the signal as I/Q components centered at zero frequency. This simplifies simulation and DPD algorithm design.

• Nonlinearity is applied to the complex envelope magnitude, generating baseband distortion products.
• The AM-AM and AM-PM characteristics of the PA fully describe the nonlinearity.
• Baseband modeling avoids simulating the carrier frequency, reducing computational load by orders of magnitude.