Signal Envelope Clipping

Hard clipping operates as a zero-memory nonlinearity, applying the amplitude limit sample-by-sample without regard to past or future signal values. The transfer function is defined as:

• If |x(n)| ≤ A_clip: output = x(n)
• If |x(n)| > A_clip: output = A_clip × e^(j∠x(n))

This sample-level decision preserves phase information while saturating magnitude. Unlike peak windowing or cancellation, there is no temporal smoothing, making it the fastest possible CFR method with zero computational latency.

The abrupt truncation of waveform peaks creates mathematical discontinuities in the time-domain signal. By Fourier theory, these sharp corners correspond to broadband spectral energy extending far beyond the occupied channel.

• Produces sinc-function spectral side-lobes that decay slowly (~6 dB/octave)
• Causes severe ACLR degradation, often exceeding regulatory spectral masks
• Generates third-order and fifth-order intermodulation products that fall into adjacent channels

This spectral regrowth is the primary drawback of hard clipping and necessitates subsequent filtering stages.

The Clipping Ratio (CR) defines the aggressiveness of peak suppression:

CR = A_clip / σ_x

where A_clip is the clipping threshold amplitude and σ_x is the RMS level of the unclipped signal. Typical values range from 3 dB to 7 dB.

• Lower CR (e.g., 3 dB): Aggressive clipping, high PAPR reduction, severe distortion
• Higher CR (e.g., 7 dB): Gentle clipping, modest PAPR reduction, minimal distortion

At a CCDF probability of 10⁻⁴, a CR of 4 dB can achieve 3-5 dB of PAPR reduction for OFDM signals.

Hard clipping introduces nonlinear distortion within the occupied channel bandwidth, directly degrading modulation accuracy:

• Constellation point scattering: Clipped peaks cause demodulated symbols to deviate from ideal reference positions
• EVM floor: Even after filtering, residual in-band distortion sets a minimum achievable EVM
• Error vector distribution becomes non-Gaussian, with heavy tails corresponding to clipped peaks

For a given CR, the EVM degradation follows a predictable relationship that can be analytically derived from the Bussgang theorem for Gaussian input signals.

When a hard-clipped signal passes through a band-limiting filter to suppress out-of-band emissions, the filtering operation causes peak regrowth:

• Filtering smooths the sharp clipping discontinuities, partially reconstructing the original peaks
• The regrown peaks can exceed the original clipping threshold by 0.5-1.5 dB
• This necessitates iterative clipping and filtering (ICF) with multiple passes

Each iteration further reduces PAPR but adds computational cost and incremental distortion. Typically 3-5 iterations achieve convergence for most waveforms.

Hard clipping is implemented as a simple magnitude comparison and saturation in the digital baseband I/Q sample stream:

• Computes instantaneous magnitude: |x| = √(I² + Q²)
• Compares against clipping threshold A_clip
• If exceeded, scales I and Q by (A_clip / |x|)
• Requires only a multiplier, CORDIC, or lookup table for magnitude computation

In FPGA implementations, a CORDIC vectoring mode efficiently computes magnitude and phase, consuming minimal logic resources. The entire operation fits within a single clock cycle.

A refined alternative to hard clipping that multiplies detected peaks by a smooth time-domain window function (e.g., Gaussian, Kaiser, or Hamming). Peak windowing spectrally confines the distortion by avoiding the sharp discontinuities of hard clipping, dramatically reducing out-of-band emissions at the cost of slightly less aggressive PAPR reduction. The window length controls the trade-off between spectral containment and peak regrowth.

An iterative process that alternates between hard clipping to suppress peaks and low-pass filtering to remove out-of-band spectral regrowth. Each filtering stage causes peak regrowth, where previously clipped peaks partially reappear due to Gibbs phenomenon. Multiple iterations with progressively tighter clipping thresholds are typically required to converge on the target PAPR while meeting the spectral mask.

The primary metric for quantifying in-band distortion caused by signal envelope clipping. EVM measures the Euclidean distance between received constellation points and their ideal reference positions, expressed as a percentage of the reference amplitude. Aggressive clipping increases EVM, degrading modulation accuracy and bit error rate. Standards like 3GPP specify maximum EVM limits (e.g., 3.5% for 256-QAM) that constrain CFR aggressiveness.

A CFR technique that subtracts a pre-designed cancellation pulse from the original signal at each detected peak location. Unlike clipping, which is memoryless, peak cancellation uses spectrally shaped pulses that concentrate distortion energy within the signal bandwidth while strictly controlling ACLR. The cancellation pulse is typically derived from the impulse response of the target channel filter, ensuring spectral compliance without iterative filtering.

A nonlinear transformation inspired by audio noise reduction that compresses high-amplitude peaks and expands low-amplitude valleys. The companding function (e.g., μ-law or A-law) reduces PAPR while preserving the signal's zero-crossing structure. At the receiver, an inverse expansion restores the original dynamic range, though quantization noise enhancement is a trade-off. Less common in modern wireless than clipping-based methods due to receiver complexity.

A cascaded architecture that chains multiple CFR stages with progressively tighter clipping thresholds and narrower filter bandwidths. Early stages perform aggressive hard clipping for coarse PAPR reduction; later stages apply peak windowing or peak cancellation for fine spectral shaping. Multi-stage CFR achieves higher overall PAPR reduction gain while maintaining EVM and ACLR within specification, at the cost of increased latency and hardware complexity.