Pulse Injection

The algorithm continuously monitors the complex baseband signal envelope to identify samples exceeding a defined clipping threshold. Once a peak is detected, the system aligns a pre-computed cancellation pulse precisely in time with the peak location. The pulse is scaled to match the peak magnitude exceeding the threshold and then subtracted from the original signal. This process is inherently signal-dependent, activating only when and where peaks occur, which preserves signal quality during low-PAPR periods.

The defining advantage of pulse injection over hard clipping is spectral control. The cancellation pulse is designed using a band-limited window function (e.g., a Kaiser, raised-cosine, or Sinc window) whose frequency response is strictly confined to the transmit channel bandwidth.

• No Out-of-Band Splatter: The subtracted pulse contains no energy in adjacent channels.
• ACLR Preservation: Adjacent channel leakage ratio remains theoretically unchanged.
• Trade-off: The pulse has finite time-domain extent, causing minor in-band distortion (EVM) but zero spectral regrowth.

A single pass of pulse injection cannot cancel all peaks because subtracting a pulse at one location may create new peaks or cause peak regrowth at nearby samples. Practical implementations use iterative processing:

• Sequential Cancellation: Process peaks in descending magnitude order, updating the signal after each subtraction.
• Multi-Stage Architectures: Cascade multiple pulse injection stages with progressively tighter thresholds.
• Convergence: Typically 3–5 iterations achieve target PAPR with diminishing returns beyond that.

This iterative approach balances PAPR reduction gain against computational latency.

The shaped cancellation pulse is pre-computed offline and stored as a set of complex FIR filter coefficients in a look-up table (LUT). Key design parameters include:

• Pulse Duration: Longer pulses provide sharper spectral confinement but increase overlap between adjacent peak cancellations.
• Oversampling Ratio: The pulse must be designed at a sample rate sufficient to capture peak locations accurately (typically 4×–8× the signal bandwidth).
• Quantization: Coefficient bit-width affects EVM floor and hardware resource utilization in FPGA implementations.

This LUT-based approach enables deterministic, low-latency real-time operation.

Pulse injection introduces in-band distortion because the subtracted cancellation pulse corrupts the data symbols within its time span. The relationship is governed by:

• Pulse Energy: Larger-amplitude peaks require more energetic cancellation pulses, increasing EVM.
• Clipping Threshold: Lower thresholds (more aggressive CFR) increase the frequency of pulse injections and cumulative distortion.
• Modulation Order Sensitivity: Higher-order QAM constellations (64-QAM, 256-QAM) have tighter EVM budgets, limiting the acceptable PAPR reduction.

System designers must balance power amplifier efficiency gains against modulation accuracy requirements specified in 3GPP standards.

Pulse injection maps efficiently to FPGA and ASIC architectures due to its feed-forward structure:

• Peak Search Logic: Magnitude comparators identify samples exceeding the threshold in real time.
• Complex Multipliers: Scale the stored pulse coefficients by the detected peak's excess magnitude and phase.
• Accumulator Buffering: Overlapping pulse subtractions require managing a window of pending corrections.
• Latency: Typical implementation latency is 1–3 µs, suitable for 5G NR numerologies.

The deterministic data flow avoids the feedback loops present in some adaptive CFR algorithms, simplifying timing closure.

The foundational CFR architecture that pulse injection implements. Rather than simply saturating the signal, peak cancellation subtracts a spectrally shaped cancellation pulse from the original waveform at each detected peak location. This preserves in-band signal quality while aggressively suppressing amplitude excursions. The key advantage over clipping is the ability to control ACLR independently of PAPR reduction by designing the cancellation pulse spectrum.

The real-time signal processing block that identifies amplitude excursions exceeding the target threshold. Detection must operate at sample-level granularity to catch all peaks, including those between sampling instants. Advanced implementations use interpolation filters to estimate true peak locations with fractional-sample accuracy. The detection threshold is typically set relative to the RMS signal level, defining the target clipping ratio.

A critical challenge in pulse injection systems where filtering after cancellation causes previously suppressed peaks to reappear. This occurs because the cancellation pulse itself has finite bandwidth, and subsequent channel filtering alters the composite waveform. Solutions include:

• Iterative pulse injection: Multiple passes with progressively refined cancellation
• Over-cancellation: Injecting slightly larger pulses to compensate for regrowth
• Predictive filtering: Accounting for filter response in the pulse design stage

The primary regulatory metric that pulse injection must control. ACLR measures the ratio of in-channel power to power leaking into adjacent frequency bands, typically specified at 5 MHz and 10 MHz offsets for 3GPP systems. Pulse injection achieves superior ACLR compared to hard clipping because the cancellation pulse spectrum is pre-designed to be band-limited, minimizing spectral regrowth into neighboring channels. Typical targets are -45 dBc or better.

Pulse injection introduces in-band distortion that degrades modulation accuracy. The EVM budget allocates how much signal degradation is acceptable—typically 2-5% for 64-QAM and < 1% for 256-QAM. The cancellation pulse amplitude directly trades off against EVM: larger pulses reduce PAPR more aggressively but add more distortion to the constellation points. System designers must balance PAPR reduction gain against the EVM floor required for the target modulation scheme.