Peak Windowing

When a signal peak exceeds the clipping threshold, peak windowing multiplies the signal by a smooth window function (e.g., Gaussian, Kaiser, Hanning, or raised-cosine) centered at the peak location. Unlike hard clipping which abruptly truncates the waveform, the window gradually attenuates the signal envelope around the peak, creating a smoother transition that concentrates distortion energy within the signal bandwidth rather than spreading it into adjacent channels. The window duration is typically 1-3 chip periods for WCDMA or a few OFDM symbol samples for LTE/5G NR, balancing peak reduction against EVM degradation.

The primary advantage of peak windowing over hard clipping is superior spectral regrowth control. Hard clipping creates sharp discontinuities in the time-domain waveform, which correspond to broadband spectral splatter in the frequency domain. Peak windowing replaces these discontinuities with smooth transitions, confining the resulting distortion products to frequencies near the carrier. This yields 10-15 dB better ACLR performance at equivalent PAPR reduction levels. The window shape directly determines the spectral roll-off of the clipping noise, with raised-cosine and Kaiser windows offering excellent stopband attenuation for stringent 3GPP spectral mask compliance.

Peak windowing introduces in-band distortion that degrades Error Vector Magnitude (EVM), creating a fundamental trade-off between PAPR reduction and modulation quality. Key parameters controlling this balance:

• Clipping threshold: Lower thresholds clip more peaks, increasing PAPR reduction but degrading EVM
• Window length: Longer windows improve spectral containment but spread distortion over more samples, increasing EVM
• Window shape: Sharper windows (e.g., rectangular-like) preserve EVM at the cost of spectral regrowth

Typical implementations target 0.5-2% EVM degradation for 3-6 dB PAPR reduction, keeping the transmitter within 3GPP modulation accuracy requirements (e.g., 17.5% EVM for 64QAM in 5G NR).

Modern base station transmitters often employ multiple peak windowing stages in cascade to achieve aggressive PAPR reduction while maintaining spectral compliance. A typical two-stage architecture:

• Stage 1: Aggressive clipping with a short window to capture and reduce large, infrequent peaks
• Stage 2: Gentle clipping with a longer, spectrally-optimized window to clean up residual peaks and shape the final spectrum

This cascaded approach can achieve 8-10 dB total PAPR reduction with better ACLR than a single aggressive stage. Some implementations combine peak windowing with pulse injection or peak cancellation techniques, where a pre-computed cancellation pulse is subtracted at each peak location rather than multiplying the signal.

Peak windowing is widely deployed in FPGA and ASIC baseband processors due to its deterministic latency and moderate computational complexity. Key implementation aspects:

• Peak detection: Requires magnitude computation (CORDIC or sqrt approximation) and threshold comparison on every sample
• Window storage: Pre-computed window coefficients stored in look-up tables (LUTs) for real-time multiplication
• Overlapping peaks: Requires peak sorting and arbitration logic when multiple peaks occur within one window duration
• Latency: Typically 1-3 µs for a single stage, critical for TDD systems with strict timing requirements

Compared to iterative clipping and filtering (ICF), peak windowing avoids FFT/IFFT operations, reducing gate count by 40-60% in FPGA implementations while achieving comparable spectral performance.

Peak windowing occupies a middle ground in the CFR design space:

• vs. Hard Clipping: 10-15 dB better ACLR at same PAPR reduction, but higher computational cost
• vs. Tone Reservation (TR): No reserved subcarrier overhead (preserves throughput), but introduces in-band EVM distortion
• vs. Active Constellation Extension (ACE): Works with any modulation format, not limited to QAM constellations, but degrades EVM rather than being distortionless
• vs. Iterative Clipping and Filtering (ICF): Lower latency and gate count (no FFT), but slightly worse spectral containment per iteration
• vs. Pulse Injection: Simpler peak detection (no alignment required), but less flexible cancellation pulse shaping

Peak windowing is the dominant CFR method in 4G/5G base station transmitters due to its excellent balance of performance, complexity, and latency.

The primary nonlinear impairment that peak windowing seeks to avoid. Clipping occurs when a power amplifier is driven beyond its saturation point, abruptly truncating waveform peaks.

• Hard Clipping: Generates severe out-of-band spectral components and high Adjacent Channel Leakage Ratio (ACLR).
• Soft Clipping/Peak Windowing: Replaces the abrupt truncation with a smooth, filtered cancellation pulse, concentrating distortion energy within the signal's own bandwidth rather than spreading it to adjacent channels.
• Result: A controlled trade-off between in-band distortion (EVM) and out-of-band emissions.

The fundamental signal characteristic that necessitates peak windowing. PAPR is the ratio of a signal's instantaneous peak power to its average power, expressed in dB.

• High PAPR Signals: Modern modulation schemes like OFDM exhibit high PAPR, forcing power amplifiers to operate with significant power back-off to avoid nonlinear distortion.
• Efficiency Impact: High back-off drastically reduces amplifier efficiency, generating excess heat and operational cost.
• Peak Windowing's Role: Reduces PAPR before the amplifier, allowing operation closer to the 1dB compression point (P1dB) for higher efficiency without violating the spectral mask.

The primary regulatory compliance metric that peak windowing is designed to improve. ACLR quantifies the ratio of transmitted power within an assigned channel to the power leaking into adjacent frequency channels.

• Measurement: Typically measured as the power in a 3.84 MHz channel offset by 5 MHz from the carrier.
• Spectral Regrowth: Nonlinear amplification causes spectral regrowth, which directly degrades ACLR.
• Peak Windowing Benefit: By smoothing signal peaks, peak windowing produces far less spectral spreading than hard clipping, resulting in superior ACLR performance and easier compliance with 3GPP and other regulatory spectral masks.

The in-band modulation quality metric that trades off against ACLR when applying peak windowing. EVM measures the vector difference between ideal reference constellation points and actual transmitted symbols.

• The Fundamental Trade-off: Aggressive peak windowing reduces PAPR and improves ACLR but distorts the signal's constellation, increasing EVM.
• Window Design: The shape and width of the windowing function directly control this trade-off. A wider, smoother window minimizes spectral regrowth but causes more symbol distortion.
• System Budget: System designers must allocate a specific EVM budget for CFR/peak windowing within the overall transmitter impairment budget.

A related signal processing technique that can be combined with peak windowing to further optimize spectral containment. Noise shaping intentionally redistributes quantization or clipping noise energy away from critical frequencies.

• Mechanism: Uses feedback loops in a delta-sigma modulator structure to push noise from the in-band region to out-of-band frequencies where it is less harmful.
• Synergy with Peak Windowing: The peak-canceling pulses generated by a peak windowing block can be spectrally shaped to concentrate any resulting distortion in frequencies that are already allocated for filtering or are less sensitive.
• Application: Used in advanced CFR algorithms to achieve aggressive PAPR reduction while maintaining very low in-band EVM.