Hard Clipping

Hard clipping operates by comparing the instantaneous signal envelope to a fixed clipping threshold. Any sample exceeding this limit is forced to the threshold amplitude while preserving the original phase. This is a memoryless operation—the output depends only on the current input sample, with no consideration of past or future signal values. The transfer function is a perfect linear response up to the threshold, followed by a flat saturation region.

The abrupt truncation of signal peaks creates sharp corners in the time-domain waveform. These discontinuities represent high-frequency content that did not exist in the original signal. Mathematically, the clipping operation is equivalent to multiplying the original signal by a time-varying gain factor that drops instantaneously from 1.0 to a lower value at each peak excursion, introducing broadband distortion products.

The time-domain discontinuities directly cause out-of-band spectral splatter. The clipping operation is a nonlinear process that spreads energy into adjacent frequency channels, dramatically degrading the Adjacent Channel Leakage Ratio (ACLR). Unlike windowed or filtered approaches, hard clipping offers no inherent spectral containment. The regrowth spectrum rolls off slowly, often violating regulatory spectral mask requirements without additional filtering stages.

While clipping reduces PAPR, it simultaneously corrupts the in-band signal. The amplitude truncation distorts the constellation points, increasing the Error Vector Magnitude (EVM). This in-band distortion cannot be filtered out without regenerating the clipped peaks. The EVM penalty is directly proportional to the Clipping Ratio (CR)—more aggressive clipping yields higher PAPR reduction but introduces greater modulation inaccuracy and potential bit error rate degradation.

Hard clipping is rarely used in isolation. When the clipped signal passes through a band-limiting filter to suppress spectral regrowth, the filtering operation smooths the sharp time-domain discontinuities. This smoothing causes peak regrowth—previously clipped peaks reappear at amplitudes exceeding the original threshold. This necessitates iterative Clipping and Filtering stages, where each iteration clips the regrown peaks and re-filters, converging toward the target PAPR.

The primary advantage of hard clipping is its computational simplicity. The operation requires only a magnitude calculation, a comparison against the threshold, and a complex scaling multiplication per sample. It introduces zero processing latency since no filtering or windowing is involved. This makes it attractive for hardware implementations where logic resources and power consumption are tightly constrained, though the spectral penalty usually demands subsequent filtering stages.

A direct alternative to hard clipping that multiplies detected peaks by a smooth time-domain window (e.g., Gaussian, Kaiser, or raised-cosine). Unlike the sharp truncation of hard clipping, windowing creates a gradual amplitude transition that dramatically reduces spectral splatter. The trade-off is a slight reduction in PAPR suppression efficiency, as the window's finite duration causes some peak regrowth and may affect adjacent samples. Commonly implemented with coefficient-scalable windows to balance ACLR compliance against EVM degradation.

An iterative process that addresses the primary failure mode of hard clipping: uncontrolled spectral regrowth. The workflow applies hard clipping to suppress peaks, then passes the distorted signal through a low-pass filter to remove out-of-band emissions. However, filtering causes peak regrowth—previously clipped peaks reappear above the threshold. This necessitates multiple stages of clipping and filtering with progressively tighter thresholds. Modern implementations use frequency-domain filtering for computational efficiency in OFDM systems.

A more sophisticated approach that avoids the sharp discontinuities of hard clipping. Instead of truncating the signal, a pre-designed cancellation pulse is subtracted at each detected peak location. These pulses are spectrally shaped to confine energy within the transmission band, achieving PAPR reduction with controlled ACLR. Key design parameters include:

• Pulse shape: Determines spectral confinement
• Pulse scaling: Matches cancellation magnitude to peak excess
• Pulse alignment: Centers cancellation on peak location This method is widely used in 5G NR base stations for multi-carrier signals.

Replaces the hard saturation function of hard clipping with a smooth nonlinearity, such as a hyperbolic tangent or polynomial limiter. The smooth transition region reduces high-frequency spectral components generated by the abrupt discontinuity. While soft clipping produces less ACLR degradation, it achieves less aggressive PAPR reduction for a given threshold because the transition region allows some peak energy to pass. Often implemented via look-up tables in FPGA-based CFR pipelines for real-time execution.

The primary in-band distortion metric directly impacted by hard clipping. EVM measures the deviation of received constellation points from their ideal reference positions. Hard clipping introduces nonlinear distortion that scatters constellation points, increasing EVM and degrading bit error rate (BER). Regulatory standards like 3GPP TS 38.104 specify maximum EVM limits per modulation scheme:

• QPSK: 17.5% EVM
• 64QAM: 8% EVM
• 256QAM: 3.5% EVM Aggressive clipping ratios force a direct trade-off between PAPR reduction gain and EVM budget.

The critical out-of-band emission metric that hard clipping severely degrades. ACLR quantifies power leaking into adjacent frequency channels relative to in-band power. Hard clipping's sharp amplitude discontinuities generate broadband spectral splatter that directly increases ACLR. Regulatory bodies mandate strict limits:

• 3GPP: Typically -45 dBc for adjacent channel
• FCC: Specified in spectral mask requirements Without subsequent filtering, hard clipping alone cannot meet these requirements, making it a component rather than a complete CFR solution.