Crest Factor Reduction (CFR)

The most fundamental CFR technique, where signal peaks exceeding a defined threshold are hard-limited, followed by filtering to suppress out-of-band spectral regrowth. While computationally simple, raw clipping introduces significant in-band distortion and Error Vector Magnitude (EVM) degradation. Iterative clipping and filtering refines this by repeating the process to better control the trade-off between PAPR reduction and Adjacent Channel Leakage Ratio (ACLR) compliance.

Instead of a hard clip, detected signal peaks are multiplied by a smooth windowing function—such as Gaussian, Kaiser, or raised-cosine—to soften the transition. This reduces the spectral splatter associated with hard clipping at the cost of a slightly wider peak suppression pulse. The technique is particularly effective in OFDM systems where maintaining subcarrier orthogonality is critical.

A sophisticated approach where a cancellation pulse, shaped to match the target emission mask, is subtracted from the signal at each detected peak location. This technique offers superior control over spectral regrowth compared to clipping. Key implementations include:

• Pulse Injection: Pre-computed cancellation pulses are scaled and subtracted at peak locations
• Noise Shaping: Quantization noise from the cancellation is shaped into unused spectrum or guard bands
• Multi-Stage Architectures: Cascaded stages progressively reduce peaks while managing EVM

A distortionless CFR method that reserves a subset of subcarriers within an OFDM symbol specifically for peak reduction. A carefully computed signal is transmitted on these reserved tones that destructively interferes with the time-domain peaks, reducing PAPR without introducing any in-band distortion or EVM on the data-bearing subcarriers. The trade-off is a reduction in net data throughput due to the reserved tones.

A technique that projects outer constellation points outward within their decision boundaries to reduce signal peaks without altering the symbol's demodulated value. By intelligently moving points in the complex baseband signal to the edges of their allowable regions, the aggregate time-domain peak can be reduced. This method preserves data rate and introduces no EVM, making it ideal for high-order QAM modulation schemes.

A non-linear transformation that compresses the dynamic range of a signal at the transmitter and expands it at the receiver, similar to audio noise reduction techniques. The μ-law and A-law companding algorithms amplify low-amplitude signals while compressing high-amplitude peaks. While effective at reducing PAPR, companding introduces distortion that increases with compression ratio and requires coordinated receiver-side expansion.

The fundamental metric that CFR seeks to minimize. PAPR is the ratio of the instantaneous peak power to the average power of a transmitted signal, typically expressed in dB.

• OFDM signals exhibit high PAPR (8–13 dB) due to the summation of many independent subcarriers.
• High PAPR forces power amplifiers to operate at significant back-off from their compression point, drastically reducing efficiency.
• CFR algorithms directly target PAPR reduction to enable higher average output power without violating linearity or emission masks.

EVM quantifies the in-band distortion introduced by CFR and other transmitter impairments. It measures the deviation of received constellation points from their ideal reference positions.

• Aggressive CFR clipping introduces in-band noise that degrades EVM, creating a direct trade-off with PAPR reduction.
• 3GPP specifications define strict EVM limits (e.g., 3.5% for 64-QAM in 5G NR) that constrain CFR algorithm design.
• Modern CFR techniques use peak windowing and filtered clipping to confine distortion energy outside the occupied bandwidth, preserving EVM.

ACLR is the primary out-of-band emission metric that CFR must respect. It measures the ratio of transmitted power in the assigned channel to power leaking into adjacent channels.

• CFR clipping generates spectral regrowth that directly degrades ACLR if not properly filtered.
• The peak-cancellation approach injects cancellation pulses shaped to match the transmit filter response, minimizing ACLR impact.
• Regulatory bodies (FCC, ETSI) mandate ACLR limits typically below -45 dBc, creating a hard constraint for CFR parameter tuning.

DPD is the inseparable counterpart to CFR in the transmit chain. While CFR reduces the signal's dynamic range, DPD compensates for the nonlinear distortion introduced by the power amplifier.

• CFR and DPD are jointly optimized: CFR reduces peaks to improve PA efficiency, while DPD linearizes the PA's remaining nonlinear response.
• The CFR-DPD cascade must be carefully sequenced—CFR typically precedes DPD to prevent DPD from attempting to correct clipping artifacts.
• In wideband systems, CFR reduces the bandwidth expansion factor that DPD must handle, simplifying the predistorter implementation.

A sophisticated CFR technique that multiplies detected signal peaks with a smooth window function (e.g., Gaussian, Kaiser, raised-cosine) rather than hard-clipping.

• The window shape is convolved with the peak, producing a spectrally contained cancellation pulse that minimizes out-of-band regrowth.
• Window length trades off PAPR reduction against EVM degradation—longer windows provide better spectral containment but spread distortion over more samples.
• Implemented efficiently in FPGAs using coordinate rotation digital computer (CORDIC) algorithms for magnitude detection and window lookup.

A frequency-domain CFR approach that intentionally shapes the clipping distortion to fall into unused or less-critical spectral regions.

• Uses filtered error subtraction: the difference between the original and clipped signal is filtered to remove in-band components before subtraction.
• Particularly effective in multi-carrier scenarios where guard bands between carriers can absorb shaped clipping noise.
• Enables more aggressive PAPR reduction without violating EVM limits by exploiting frequency-selective emission mask allowances.