Switching Ripple Artifact

Switching ripple originates from the pulse-width modulation (PWM) action of the supply modulator's power stage. The inductor-capacitor output filter cannot completely eliminate the switching frequency component, leaving a sawtooth-like voltage ripple superimposed on the desired DC or envelope-tracking voltage. The ripple's fundamental frequency equals the modulator's switching frequency (typically 10-150 MHz for modern ET modulators), with harmonics extending into the RF band.

When switching ripple is present on the PA's drain/collector supply, it multiplies with the RF carrier through the PA's intrinsic nonlinearity. This creates spurious sidebands offset from the carrier by the switching frequency and its harmonics. Key consequences include:

• Spectral mask violations at specific frequency offsets
• Receive band noise in FDD systems when ripple harmonics fall into the uplink band
• Adjacent channel leakage ratio (ACLR) degradation that cannot be corrected by conventional DPD alone

The PA's PSRR defines how effectively it rejects supply voltage variations. PSRR typically rolls off at higher frequencies, making the PA increasingly susceptible to switching ripple as the modulator's switching frequency increases. For GaN PAs operating in envelope tracking, PSRR at the switching frequency may be as low as 10-20 dB, meaning a 50 mV ripple can directly modulate the RF output with significant amplitude.

Switching ripple does not only cause amplitude modulation of the RF carrier. Through the PA's nonlinear input capacitance and supply-dependent phase shift, the ripple voltage is converted to unintended phase modulation. This AM-to-PM conversion creates asymmetric sidebands around the carrier that are particularly problematic for high-order QAM constellations, increasing error vector magnitude (EVM) and degrading demodulation performance.

Compensating for switching ripple artifacts requires extending the DPD model to include supply voltage as an explicit input. A dual-input behavioral model captures the PA's response to both RF input and supply ripple, enabling the predistorter to pre-cancel the intermodulation products. Alternatively, augmented Volterra models with supply-dependent kernels can model the ripple-to-RF coupling path, though this increases model complexity and coefficient count.

Characterizing switching ripple artifacts requires wideband observation receivers capable of capturing spectral content at offsets equal to the switching frequency (often 50-150 MHz from the carrier). Key measurement challenges include:

• Dynamic range: Ripple-induced spurs may be 60-80 dB below the carrier
• Synchronization: Correlating time-domain ripple waveforms with RF output distortion
• Isolation: Separating ripple effects from other nonlinear mechanisms like thermal memory

A passive mitigation strategy employing cascaded inductor-capacitor (LC) filter stages at the supply modulator output to attenuate ripple amplitude before it reaches the PA drain. Higher-order filters (4th-order and above) provide steeper roll-off beyond the switching fundamental frequency. Design trade-offs include filter Q-factor to avoid resonant peaking, component parasitics (ESR, ESL) that degrade high-frequency attenuation, and group delay that must remain constant across the tracking bandwidth to avoid envelope distortion. For wideband ET systems, a damped Cauer (elliptic) topology often provides the optimal balance between stopband rejection and passband flatness.

An active technique that dithers the switching frequency of the supply modulator to distribute ripple energy across a wider bandwidth, reducing peak spectral density at any single frequency. Frequency modulation index and dithering profile (triangular, pseudorandom, or sinusoidal) determine the degree of spreading. This approach reduces the probability of ripple components intermodulating with narrowband RF carriers to create spurs that violate spectral masks. The trade-off is a slight increase in the noise floor across the spreading bandwidth, which must be managed to avoid degrading wideband EVM.

Incorporating the switching ripple as a known deterministic input to the digital predistortion model, enabling the DPD to generate a pre-distorted cancellation signal that nullifies the ripple-induced intermodulation products at the PA output. This requires an augmented behavioral model that includes supply voltage terms with memory, such as a dual-input Volterra series or a neural network with supply voltage as an auxiliary input. The DPD effectively acts as a feedforward cancellation path, compensating for ripple artifacts that cannot be fully removed by passive filtering alone.

A hardware-in-the-loop technique where the residual ripple waveform is sampled at the PA drain through a high-bandwidth coupler, phase-aligned, and injected as an anti-phase cancellation signal into the RF path via an auxiliary modulator. This approach directly cancels the ripple-induced AM sidebands before they reach the antenna. Critical implementation challenges include delay matching between the sense and cancellation paths to within picoseconds, and maintaining cancellation depth across temperature and load variations. Often used in conjunction with DPD for maximum spectral purity.

A supply modulator architecture using multiple switching cells operating with phase-shifted clock signals (e.g., 4 phases at 90° intervals) to achieve ripple cancellation at the summed output. The interleaving causes ripple currents from individual phases to partially cancel, reducing the net ripple amplitude and effectively multiplying the ripple frequency by the number of phases. This pushes the fundamental ripple component to a higher frequency where output filter attenuation is greater and where it is less likely to fall within critical RF bands. The trade-off is increased component count and control complexity.

A tunable active filter placed in the supply voltage path that creates a frequency-agile notch precisely at the switching fundamental and its harmonics. The notch frequency is continuously adjusted based on the modulator's instantaneous switching frequency, which may vary with load or intentional spread-spectrum dithering. Implementation typically uses a gyrator-based active inductor or a switched-capacitor filter to achieve high-Q notching without bulky passive components. This technique is particularly effective when the switching frequency is relatively fixed and known a priori.