Inferensys

Glossary

Companding

A non-uniform quantization technique that compresses high-amplitude signal components and expands low-amplitude ones, reducing PAPR at the cost of introduced distortion that must be managed to control spectral regrowth.
Stylish WeWork-like workspace with hot desks and document wall, professional searching through enterprise knowledge base on a mounted ultrawide display, warm industrial pendants overhead.
SIGNAL COMPRESSION

What is Companding?

Companding is a non-uniform quantization technique that compresses high-amplitude signal components and expands low-amplitude ones to reduce the peak-to-average power ratio (PAPR), introducing controlled distortion that must be carefully managed to prevent spectral regrowth.

Companding (a portmanteau of compressing and expanding) applies a nonlinear transfer function to a signal before transmission and the inverse function after reception. At the transmitter, high-amplitude peaks are logarithmically compressed while low-amplitude regions are amplified, effectively reducing the peak-to-average power ratio (PAPR) and allowing the power amplifier to operate closer to saturation with improved efficiency.

The compression stage introduces intentional AM-AM distortion that generates out-of-band spectral components, requiring careful compander design to balance PAPR reduction against adjacent channel leakage ratio (ACLR) degradation. The complementary expansion at the receiver restores the original signal dynamic range but also expands any noise or distortion accumulated in the channel, making companding most effective in high-SNR environments where the efficiency gains outweigh the fidelity penalty.

NON-UNIFORM QUANTIZATION

Key Characteristics of Companding

Companding is a signal processing technique that applies a nonlinear compression function to high-amplitude signals before transmission and an inverse expansion function at the receiver. This process reduces the peak-to-average power ratio (PAPR) and improves the signal-to-quantization noise ratio for low-level signals, but introduces deliberate distortion that must be carefully managed to control spectral regrowth.

01

Compression Characteristic: µ-Law vs. A-Law

The compression curve defines the nonlinear mapping between input and output amplitudes. The two dominant international standards are µ-law, used in North America and Japan, and A-law, used in Europe. µ-law provides a slightly wider dynamic range and is defined by the continuous logarithmic formula F(x) = sgn(x) * ln(1 + µ|x|) / ln(1 + µ) with µ typically set to 255. A-law is a piecewise approximation optimized for PCM voice telephony, offering superior small-signal performance with a reduced computational load. Both curves are designed to make the quantization step size proportional to the signal amplitude, effectively allocating more bits to quiet passages.

02

PAPR Reduction Mechanism

Companding directly reduces the peak-to-average power ratio (PAPR) by compressing the dynamic range of the signal before it enters the power amplifier. By attenuating high-amplitude peaks and boosting low-amplitude valleys, the signal's envelope becomes more uniform. This allows the power amplifier to operate closer to its 1dB compression point (P1dB) without clipping, improving power efficiency. The reduction in PAPR is a direct trade-off: the signal's crest factor is lowered, but the process introduces in-band distortion that degrades the error vector magnitude (EVM).

03

Spectral Regrowth and Out-of-Band Emissions

The nonlinear compression function generates intermodulation products that cause spectral regrowth into adjacent channels. This is the primary penalty of companding. The sharp transitions in the compression curve, particularly near the saturation point, create high-frequency spectral components that degrade the adjacent channel leakage ratio (ACLR). To mitigate this, companding is often paired with pulse shaping filters or peak windowing techniques. The design challenge is to balance the PAPR reduction benefit against the resulting spectral mask violations, ensuring compliance with regulatory spectral mask requirements.

04

Companding Distortion vs. Quantization Noise

Companding fundamentally trades quantization noise for companding distortion. In a uniform quantizer, low-amplitude signals suffer from poor signal-to-quantization noise ratio. Companding improves this by effectively expanding the quantization levels for small signals. However, the nonlinear mapping introduces harmonic and intermodulation distortion that is signal-dependent. This distortion is not random like quantization noise; it is deterministic and correlated with the signal envelope. For communication systems, this means the distortion can be partially compensated for at the receiver if the companding parameters are known, a technique known as decompanding.

05

Application in OFDM Systems

Orthogonal Frequency Division Multiplexing (OFDM) signals exhibit inherently high PAPR due to the summation of multiple independent subcarriers. Companding is a popular, low-complexity PAPR reduction technique for OFDM. The process is applied to the complex baseband time-domain samples after the IFFT operation. Key design parameters include the companding threshold and the compression ratio. A high compression ratio aggressively reduces PAPR but causes severe spectral regrowth. Adaptive companding schemes dynamically adjust these parameters based on the instantaneous signal statistics to maintain an acceptable balance between power efficiency and ACLR.

06

Synergy with Digital Predistortion

Companding and digital predistortion (DPD) are complementary linearization strategies. Companding addresses the signal's statistical properties before amplification, while DPD corrects the amplifier's nonlinear transfer characteristic. In a modern transmitter chain, companding is applied first to reduce PAPR, allowing the power amplifier to operate at a higher average power. The residual nonlinearity, including the distortion introduced by the compander itself, is then corrected by the DPD engine. This cascade requires the DPD model to be trained on the companded signal to effectively linearize the entire chain, including the AM-AM and AM-PM distortion introduced by both the compander and the PA.

COMPANDING EXPLAINED

Frequently Asked Questions

Clear, technically precise answers to the most common questions about companding, its role in PAPR reduction, and its impact on spectral regrowth in modern communication systems.

Companding is a non-uniform quantization technique that applies a compression function to high-amplitude signal components before transmission and a complementary expansion function upon reception. The process deliberately reduces the dynamic range of a signal to lower its Peak-to-Average Power Ratio (PAPR). During compression, large signal amplitudes are attenuated more than small ones according to a nonlinear curve—typically the μ-law (used in North America and Japan) or A-law (used in Europe). At the receiver, the expander applies the inverse curve to restore the original signal envelope. This signal conditioning allows a power amplifier to operate closer to its 1dB compression point without hard clipping, but it introduces intentional AM-AM distortion that generates spectral regrowth which must be carefully managed through filtering and Digital Pre-Distortion (DPD).

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.