Quadrature Mirror Filter

Description

A Quadrature Mirror Filter (QMF) bank is a specific type of filter bank used extensively in signal processing for sub-band coding. Its primary function is to decompose a full-band signal (like an audio waveform) into a set of sub-band signals, each occupying a distinct portion of the original frequency spectrum. Conversely, a synthesis filter bank can reconstruct the original signal from these sub-bands. The 'quadrature' aspect relates to the filters being designed in pairs with specific phase relationships, while 'mirror' refers to the symmetrical frequency response of the analysis low-pass and high-pass filters around the quadrature frequency (π/2).

The QMF bank consists of two main components: an analysis filter bank and a synthesis filter bank. The analysis bank typically uses a pair of filters: a low-pass filter (H0) and a high-pass filter (H1). These filters are designed such that their frequency responses are mirror images of each other around the half-Nyquist frequency. The signal is passed through these filters, and the outputs are downsampled (decimated) by a factor of two, creating the sub-band signals. This process can be iterated on the low-pass output to create a hierarchical, multi-resolution decomposition (like in a wavelet transform).

For perfect reconstruction—where the output signal is a delayed, possibly scaled, version of the input—the synthesis filters (G0 and G1) must be carefully designed in conjunction with the analysis filters. The conditions involve constraints on the filter coefficients to eliminate aliasing and amplitude distortion. In 3GPP specifications, particularly those related to audio codecs (like the Enhanced Voice Services codecs), QMF banks are employed as a tool for spectral analysis and synthesis. They provide an efficient way to partition the audio signal into frequency bands for subsequent processing, such as perceptual coding where bits are allocated based on the importance of each sub-band to human hearing.

Purpose & Motivation

Quadrature Mirror Filter banks were developed to address the need for efficient, reversible signal decomposition for compression and processing. Before their widespread adoption, processing wideband signals directly was computationally expensive and inefficient for applications like audio coding, where different frequency components have different perceptual importance. Simple filtering and downsampling would introduce aliasing artifacts that corrupted the signal upon reconstruction.

The creation of QMF banks solved the critical problem of aliasing cancellation in two-channel filter banks. Their specific design property allows the aliasing components introduced by downsampling in the analysis stage to be canceled out by the upsampling and filtering in the synthesis stage. This enables perfect (or near-perfect) reconstruction, which is essential for lossless or high-quality lossy coding. In the context of 3GPP audio codecs, referenced from Rel-8 onwards, QMF banks provide a standardized, mathematically sound method for splitting the audio signal into sub-bands. This allows codecs to apply psychoacoustic models more effectively, allocating fewer bits to less perceptible frequency components, thereby achieving higher compression ratios without perceptible quality loss, which is vital for efficient voice and audio transmission over bandwidth-constrained mobile networks.

Key Features

• Decomposes a signal into sub-bands for multi-resolution analysis
• Enables perfect reconstruction of the original signal when conditions are met
• Provides inherent aliasing cancellation between analysis and synthesis stages
• Used as a building block for more complex filter banks and wavelet transforms
• Efficient implementation possible using polyphase structures
• Referenced in 3GPP for spectral processing in audio codec specifications

Evolution Across Releases

Defining Specifications