Filterbankar

Filterbankar are arrays of filters used to separate a signal into multiple components, each occupying a defined frequency band. They are widely used in digital signal processing to perform spectral analysis, subband coding, or feature extraction. The term filterbank combines 'filter' and 'bank', emphasizing parallel processing of frequency channels. In Swedish, filterbankar is the plural form of filterbank.

In digital implementations, filter banks can be uniform (equal bandwidth bands) or nonuniform (bands with different widths). They can be realized as polyphase filter banks, cosine-modulated banks, or FFT-based banks. Common filter types include FIR and IIR filters. A uniform polyphase bank can be more efficient by sharing computations across channels.

Applications include audio processing (equalization, compression, perceptual coding such as MFCC mel filter banks), speech and music analysis, and telecommunications (channelization, subcarrier allocation, and OFDM). Nonuniform banks are used when perceptual or channel constraints require uneven spacing.

Key design considerations include bandwidth, stop-band attenuation, passband ripple, and phase response. Aliasing and leakage between channels are concerns in critically sampled banks; oversampling, padding, or polyphase decomposition help mitigate. Real-time hardware implementations may use dedicated DSP processors or SIMD architectures.

In Swedish contexts, filterbankar is the plural form of filterbank. Filter banks have roots in analog and digital signal processing research of the 20th century. Classic families include uniform, nonuniform, Gammatone and Mel filter banks used in modeling human perception and speech processing.