Peak Signal-to-Noise Ratio

Description

Peak Signal-to-Noise Ratio (PSNR) is a mathematical, objective quality assessment metric expressed in decibels (dB). In the context of 3GPP multimedia specifications, it is used to quantify the fidelity of a reconstructed video frame or image after it has undergone processing such as encoding, decoding, transmission, and error concealment. The 'signal' in this case is the original, reference data (e.g., a raw YUV video frame), and the 'noise' is the error or distortion introduced by the processing chain, calculated as the difference between the original and the processed signal. A higher PSNR value indicates a smaller difference and thus higher reconstruction quality, though it does not always perfectly correlate with subjective human perception.

The calculation of PSNR for video typically operates on a per-frame basis, often in the luminance (Y) component, as the human visual system is more sensitive to luminance distortions. The process begins by computing the Mean Squared Error (MSE) between the original and the processed frame. MSE is the average of the squared differences of each pixel's intensity value. PSNR is then derived from MSE using the formula: PSNR = 10 * log10( (MAX_I)^2 / MSE ), where MAX_I is the maximum possible pixel value (e.g., 255 for 8-bit samples). This formula highlights that PSNR is a logarithmic measure of the ratio between the peak power of the signal (the square of the maximum value) and the power of the noise (the MSE). For color video, PSNR may be calculated separately for Y, U, and V components and sometimes averaged.

Within 3GPP's Multimedia Broadcast Multicast Service (MBMS) and Media Streaming working groups, PSNR serves as a key performance indicator (KPI) in codec characterization and selection. Its role is to provide a reproducible, quantitative benchmark for comparing different video coding standards (like H.264/AVC, H.265/HEVC, H.266/VVC), encoding parameters (bitrate, quantization), and error resilience tools. Test plans in specifications like 26.904 define rigorous methodologies for calculating PSNR under various network conditions (e.g., packet loss). While instrumental for engineering comparisons, it is often used alongside subjective testing methods (like Mean Opinion Score - MOS) because PSNR can fail to capture certain types of perceptual artifacts that are more annoying to viewers but contribute less to the numerical MSE.

Purpose & Motivation

PSNR exists as a fundamental engineering tool to objectively and automatically evaluate the performance of video compression algorithms and systems. The primary problem it solves is the need for a fast, consistent, and cost-effective method to compare the output quality of different codecs or encoding settings during development, standardization, and network planning. Subjective testing with human viewers, while the ultimate benchmark, is time-consuming, expensive, and difficult to scale or automate. PSNR provides an immediate, numerical score that can be integrated into automated test benches and regression testing.

The motivation for its adoption in 3GPP stems from the need to select and specify video codecs for mobile services like packet-switched streaming (PSS) and MBMS. During competitive codec selection processes, proponents of different technologies must demonstrate the efficiency of their proposals. PSNR-rate curves, which plot achieved PSNR against required bitrate, became a standard way to compare the rate-distortion performance of codecs. This allowed for data-driven decisions on which codec delivers the best quality for a given bandwidth constraint—a critical consideration for mobile networks with limited and shared radio resources.

However, the creation and continued use of PSNR also acknowledge its limitations, which subsequent research and standards work aim to address. PSNR is a pixel-based difference metric that does not align perfectly with human vision; it weights all errors equally, whereas the human eye is more sensitive to errors in smooth areas than in textured ones. This limitation motivated the development and investigation of more advanced perceptual quality metrics (like SSIM, VMAF) within later 3GPP studies. Nevertheless, PSNR remains a widely understood and historically important benchmark due to its simplicity, reproducibility, and the vast body of existing research and results that rely on it, providing a common baseline for performance evaluation across the industry.