Peak Windowing

Peak windowing operates by multiplying the original signal by a carefully designed window function centered at each detected peak. Unlike clipping, which truncates the signal envelope, windowing applies a smooth amplitude taper. The window is typically a raised-cosine, Kaiser, or Gaussian function whose width determines the spectral containment trade-off. The operation can be expressed as:

y(n) = x(n) * w(n - n_peak)

where w(n) is the window function and n_peak is the peak location. This multiplicative approach avoids the sharp discontinuities that cause spectral splatter in hard clipping.

The primary advantage of peak windowing is its ability to confine distortion energy within the transmit channel. By selecting a window with a narrow frequency-domain response, out-of-band emissions are dramatically reduced compared to hard clipping. The window's Fourier transform directly shapes the distortion spectrum:

• Wider windows (in time) produce narrower spectral regrowth but affect more samples per peak
• Narrower windows affect fewer samples but cause broader spectral leakage

This predictable relationship allows engineers to design window functions that meet specific spectral mask requirements without iterative filtering.

The choice of window function directly determines the trade-off between PAPR reduction and EVM degradation. Common window families include:

• Raised-Cosine: Offers a smooth roll-off with a tunable roll-off factor α controlling the excess bandwidth of the distortion
• Kaiser: Provides a flexible parameter β to adjust the main lobe width versus sidelobe level, enabling precise spectral mask compliance
• Gaussian: Minimizes the time-bandwidth product, offering the best joint localization in time and frequency
• Taylor: Designed for specified sidelobe levels with minimal main lobe broadening

Each window is characterized by its peak reduction capability and spectral spreading factor.

In dense peak environments, multiple peaks may occur within a single window duration. The system must handle overlapping windows intelligently. The standard approach is to apply the maximum attenuation at each sample point across all active windows:

a(n) = max(a_1(n), a_2(n), ..., a_k(n))

where a_k(n) is the attenuation factor from the k-th window at sample n. This peak-preserving combination ensures that all peaks are suppressed to the target threshold. More sophisticated implementations use recursive windowing where the residual signal after one window is re-evaluated for remaining peaks.

Peak windowing is well-suited for FPGA and ASIC implementation due to its regular structure. The core operations include:

• Peak detection: A magnitude comparator identifying samples exceeding the clipping threshold
• Window LUT: A pre-computed look-up table storing the window coefficients
• Multiplier array: Parallel multipliers applying window weights to signal samples
• Overlap management: Logic for handling concurrent window applications

The computational complexity scales with the window length and peak arrival rate. Modern implementations use pipelined architectures to maintain throughput at multi-GHz sample rates required for 5G and wideband signals.

Peak windowing occupies a middle ground between hard clipping and peak cancellation in the CFR design space:

Peak windowing is preferred when spectral mask compliance is critical and moderate EVM degradation is acceptable, making it ideal for base-station transmitters where ACLR is tightly regulated.

The parent category of signal conditioning techniques that deliberately limit peak amplitude to improve power amplifier efficiency. Peak windowing is a specific CFR method that multiplies detected peaks by a smooth time-domain window function, offering a superior trade-off between PAPR reduction and spectral regrowth compared to hard clipping. CFR is essential in modern OFDM-based systems like 5G NR and Wi-Fi where high PAPR forces inefficient amplifier back-off.

Hard clipping applies a memoryless amplitude threshold that saturates the signal envelope at a fixed limit, creating sharp discontinuities that cause severe spectral splatter and high ACLR. Soft clipping uses a smooth saturation function (e.g., hyperbolic tangent) to reduce spectral regrowth at the expense of less aggressive PAPR reduction. Peak windowing extends the soft clipping concept by applying a time-domain window (e.g., Hanning, Kaiser, Gaussian) centered at each detected peak, providing superior control over the spectral mask compliance.

The phenomenon where filtering a clipped signal causes previously suppressed amplitude peaks to reappear above the target threshold. This occurs because the filtering operation, necessary to remove out-of-band distortion, effectively interpolates between samples and can reconstruct peak amplitudes. Peak windowing mitigates regrowth by using spectrally confined window functions that inherently limit out-of-band energy, reducing the need for aggressive post-filtering and minimizing the number of iterative clipping stages required.

A critical metric quantifying in-band distortion introduced by CFR algorithms. EVM measures the deviation of received constellation points from their ideal reference positions, expressed as a percentage or in dB. Peak windowing introduces controlled EVM degradation because the window function modifies the signal envelope within the occupied bandwidth. System designers must balance:

The ratio of transmitted power within the assigned channel to power leaking into adjacent frequency channels, measured in dBc. Regulatory bodies like 3GPP and ETSI mandate strict ACLR limits to prevent interference. Peak windowing's primary advantage over hard clipping is its ability to achieve aggressive PAPR reduction while maintaining ACLR compliance. The window function's spectral properties directly determine the out-of-band emission profile, making window selection a critical design parameter.

A cascaded approach applying successive stages of peak windowing with progressively tighter clipping thresholds. Each stage uses a window function optimized for a specific peak magnitude range, allowing aggressive overall PAPR reduction while distributing EVM degradation across stages. Typical implementations use 2-3 stages: