Frequency Band

Description

In 3GPP specifications 26.402 and 26.404, which cover the Enhanced Low Delay Codec (ELD) and other audio codecs, a frequency band is defined as an interval in frequency that consists of a group of consecutive Quadrature Mirror Filter (QMF) subbands. The QMF bank decomposes the audio signal into multiple subbands, each covering a specific frequency range. By grouping these subbands, the codec can process the signal in perceptually relevant frequency bands, such as critical bands of human hearing. This approach is common in parametric and transform-based audio codecs, where the signal is analyzed in the frequency domain and parameters are extracted per band.

The QMF filterbank used in these codecs typically has 32 or 64 subbands, depending on the sampling rate and the specific codec configuration. Each subband has a bandwidth equal to the sampling frequency divided by twice the number of subbands (for complex QMF) or the sampling frequency divided by the number of subbands (for real QMF). For example, with a 16 kHz sampling rate and a 32-band complex QMF, each subband covers 250 Hz. A frequency band may consist of one or more adjacent subbands, and the grouping is often non-uniform to match the frequency resolution of the human auditory system. Low-frequency bands may consist of a single subband, while high-frequency bands may group several subbands to reduce the number of parameters to transmit.

Frequency bands are used in various coding tools, such as bandwidth extension (BWE), noise filling, and spatial audio coding. In BWE, the high-frequency content is generated based on parameters extracted from lower frequency bands, and the frequency band structure defines the mapping between the low and high bands. In noise filling, the codec fills spectral gaps with noise in specific frequency bands to improve perceptual quality. The concept is also used in the MPEG-D Unified Speech and Audio Coding (USAC) standard, which shares many principles with 3GPP codecs.

The definition of frequency band has remained consistent across 3GPP releases from Rel-8 to Rel-19, with only minor editorial changes. The specifications 26.402 and 26.404 provide the normative details for the codecs that use this concept, including the exact QMF filterbank design and the band grouping tables. Implementers must adhere to these tables to ensure interoperability and correct decoding of the audio signal. The frequency band concept is fundamental to the operation of these codecs, as it determines how the audio spectrum is partitioned and processed.

In summary, a frequency band in 3GPP is a grouping of consecutive QMF subbands that defines a spectral interval for coding purposes. It enables efficient parametric representation of audio signals by concentrating processing on perceptually relevant frequency regions. The term is part of the technical vocabulary of audio codec engineers and is essential for understanding the architecture of codecs like ELD and USAC.

Purpose & Motivation

The concept of frequency bands exists to provide a structured way to process audio signals in the frequency domain, aligning with the human auditory system's non-uniform frequency resolution. Human hearing is more sensitive to changes in low frequencies than high frequencies, and the critical bands of hearing are narrower at low frequencies and wider at high frequencies. By grouping QMF subbands into frequency bands, codecs can allocate bits and processing power more efficiently, focusing on perceptually important spectral regions. This is particularly important in low-bit-rate audio coding, where every bit counts.

Before the adoption of frequency bands in 3GPP codecs, early audio codecs often processed the entire spectrum uniformly or used fixed transform windows. This approach was less efficient because it did not exploit the perceptual properties of hearing. The introduction of QMF-based frequency bands allowed for flexible, perceptually motivated processing. For example, in bandwidth extension, the high-frequency bands can be represented by a small set of parameters derived from low-frequency bands, significantly reducing the bit rate required to transmit high-frequency content. Similarly, in noise filling, the codec can selectively add noise to frequency bands where the signal is masked, improving the subjective quality without increasing the bit rate.

The frequency band concept was standardized in 3GPP Release 8 as part of the Enhanced Low Delay Codec (ELD) and later extended in other codecs. It addresses the limitations of earlier approaches by providing a scalable and interoperable framework for spectral processing. The specifications 26.402 and 26.404 define the exact band structures and QMF configurations, ensuring that encoders and decoders from different vendors can interoperate. Over successive releases, the definition has remained stable, reflecting the maturity of the technology. The concept continues to be relevant in modern codecs, including those used for immersive audio and 5G voice services.