Advanced Signal Processing Unit 5 Review: Multirate Processing & Filter Banks

Fundamentals of Multirate Systems

• Multirate systems involve processing signals at different sampling rates within a single system
• Enables efficient processing and analysis of signals by adapting the sampling rate to the signal characteristics
• Key concepts include upsampling (increasing the sampling rate) and downsampling (reducing the sampling rate)
• Upsampling involves inserting zeros between samples to increase the sampling rate by an integer factor
• Downsampling involves discarding samples to reduce the sampling rate by an integer factor
• Aliasing can occur during downsampling if the signal is not properly bandlimited
  • Aliasing introduces distortion and can cause frequency components to overlap
• Anti-aliasing filters are used before downsampling to prevent aliasing by removing high-frequency components

Decimation and Interpolation Techniques

• Decimation reduces the sampling rate of a signal by an integer factor $M$
• Involves two steps: low-pass filtering followed by downsampling
  • Low-pass filtering removes high-frequency components to prevent aliasing
  • Downsampling discards $M-1$ samples for every $M$ samples, reducing the sampling rate
• Interpolation increases the sampling rate of a signal by an integer factor $L$
• Involves two steps: upsampling followed by low-pass filtering
  • Upsampling inserts $L-1$ zeros between each sample, increasing the sampling rate
  • Low-pass filtering removes the high-frequency replicas introduced by upsampling
• Decimation and interpolation can be cascaded to achieve non-integer sampling rate changes
• Efficient implementation techniques, such as polyphase decomposition, are used to reduce computational complexity

Polyphase Decomposition

• Polyphase decomposition is a technique for efficiently implementing multirate systems
• Decomposes a filter into a set of parallel subfilters called polyphase components
• Enables efficient implementation of decimation and interpolation by reducing computational complexity
• For decimation, the input signal is split into $M$ polyphase components, filtered, and then downsampled
  • Each polyphase component operates at the reduced sampling rate, reducing the overall computational cost
• For interpolation, the input signal is upsampled, split into $L$ polyphase components, and then filtered
  • The polyphase components are combined to form the interpolated output signal
• Polyphase decomposition allows for the realization of multirate systems using parallel processing architectures
• Facilitates the design of efficient filter banks and wavelet transforms

Filter Bank Theory and Design

• Filter banks are a collection of filters used to analyze and synthesize signals in multirate systems
• Consist of an analysis bank that decomposes the signal into subbands and a synthesis bank that reconstructs the signal
• Perfect reconstruction filter banks allow for the exact recovery of the original signal from the subband signals
  • Requires the analysis and synthesis filters to satisfy certain conditions
• Maximally decimated filter banks have a decimation factor equal to the number of subbands
  • Provide the most efficient representation but require careful design to avoid aliasing
• Oversampled filter banks have a decimation factor less than the number of subbands
  • Provide increased robustness and flexibility at the cost of increased computational complexity
• Design techniques for filter banks include the use of polyphase decomposition and lattice structures
• Filter banks find applications in signal compression, feature extraction, and multiresolution analysis

Quadrature Mirror Filters (QMF)

• Quadrature Mirror Filters (QMF) are a special class of two-channel perfect reconstruction filter banks
• Consist of an analysis filter pair and a synthesis filter pair
• The analysis filters are designed to be mirror images of each other in the frequency domain
  • One filter captures the low-frequency content, while the other captures the high-frequency content
• The synthesis filters are also mirror images and are used to reconstruct the original signal
• QMF banks have the property of perfect reconstruction, allowing for the exact recovery of the input signal
• Commonly used in subband coding and wavelet transform applications
• Design techniques for QMF banks include the use of half-band filters and lattice structures

Wavelet Transform and Filter Banks

• Wavelet transforms provide a multiresolution analysis of signals using a set of basis functions called wavelets
• Wavelets are localized in both time and frequency, allowing for the capture of both temporal and spectral information
• The discrete wavelet transform (DWT) can be implemented using a tree-structured filter bank
  • The signal is recursively decomposed into low-frequency (approximation) and high-frequency (detail) subbands
• The inverse discrete wavelet transform (IDWT) reconstructs the original signal from the wavelet coefficients
• Wavelet filter banks can be designed using various wavelet families (Haar, Daubechies, Symlets, etc.)
• Wavelet transforms find applications in signal denoising, compression, and feature extraction
• The choice of wavelet family and decomposition level depends on the specific application and signal characteristics

Applications in Signal Compression

• Multirate systems and filter banks are widely used in signal compression applications
• Subband coding is a compression technique that exploits the frequency-dependent characteristics of signals
  • The signal is decomposed into subbands using a filter bank
  • Each subband is encoded independently based on its perceptual importance and statistical properties
• Wavelet-based compression schemes, such as JPEG 2000, use wavelet transforms for image compression
  • The image is decomposed using a wavelet filter bank, and the coefficients are quantized and encoded
• Adaptive filter banks can be used to optimize the compression performance based on the signal characteristics
• Perceptual audio coding techniques, such as MP3 and AAC, use filter banks to exploit the psychoacoustic properties of the human auditory system
• Multirate techniques are also used in speech coding and video compression standards (H.264, HEVC)

Advanced Topics and Current Research

• Multirate signal processing continues to be an active area of research with various advanced topics and applications
• Nonuniform filter banks involve the use of different decimation factors for each subband
  • Provide increased flexibility and adaptability to signal characteristics
• Multidimensional filter banks extend the concepts of multirate processing to higher-dimensional signals (images, videos)
• Adaptive filter banks dynamically adjust their parameters based on the signal characteristics or external factors
• Multirate techniques are being explored in the context of graph signal processing for analyzing signals defined on graphs
• Compressed sensing leverages the sparsity of signals in certain domains to enable efficient acquisition and reconstruction
• Machine learning techniques, such as deep learning, are being integrated with multirate systems for improved performance and adaptability
• Multirate processing finds applications in emerging areas such as wireless communications, sensor networks, and biomedical signal analysis