Polyphase Filter Bank

The entire channelization performance hinges on a single, carefully designed prototype low-pass filter. This FIR filter defines the sub-band shape, stop-band attenuation, and pass-band ripple. Key design parameters include:

• Stop-band attenuation: Determines adjacent channel isolation and alias rejection.
• Pass-band ripple: Controls amplitude flatness within each sub-band.
• Transition bandwidth: Dictates the guard band between channels. The prototype is typically designed using the Parks-McClellan algorithm or a windowed sinc function to meet a specific spectral mask.

The PFB's efficiency comes from applying the Noble Identity of multirate signal processing, which allows the filtering operation to be moved after the downsampler. Instead of filtering at the high input rate and then discarding samples, the PFB:

• Decimates first, then filters at the lower output rate.
• Partitions the prototype filter into M polyphase component filters.
• Computes a commutator model, where the input is sequentially switched to each polyphase branch. This reduces the required multiplications per second by a factor of M, where M is the number of channels.

A critically sampled PFB is mathematically equivalent to a windowed Discrete Fourier Transform (DFT). The architecture consists of:

• A polyphase decomposition of the prototype filter into M branches.
• An M-point Inverse FFT (IFFT) applied to the output of the polyphase branches. This structure, often called a DFT-modulated filter bank, uniformly stacks the translated low-pass responses across the spectrum. The IFFT efficiently computes the complex exponential modulation needed to shift each sub-band to baseband, making the architecture highly suitable for FPGA and ASIC implementation using standard FFT IP cores.

While a critically sampled PFB has decimation factor D equal to the number of channels M, an oversampled PFB uses D < M. This provides significant benefits:

• Reduced aliasing: Increased guard bands between sub-band spectra.
• Relaxed prototype filter constraints: Simpler filter design with gentler transition bands.
• Enhanced signal reconstruction: Less distortion when resynthesizing the wideband signal. The trade-off is a higher aggregate output sample rate. Oversampled PFBs are essential in applications like radio astronomy and spectrum monitoring where signal fidelity is paramount.

The Weighted Overlap-Add (WOLA) PFB is an advanced variant that applies a time-domain window and overlap before the FFT. This technique:

• Suppresses spectral leakage far beyond a standard DFT filter bank.
• Decouples the analysis window from the number of channels, allowing independent optimization.
• Provides superior dynamic range, making it ideal for detecting weak signals in the presence of strong adjacent interferers. WOLA is a cornerstone of modern wideband spectrum analyzers and signal intelligence receivers, where a high spurious-free dynamic range (SFDR) is non-negotiable.

A perfect reconstruction (PR) polyphase filter bank allows the original wideband signal to be exactly reconstructed from its sub-band components, minus a system delay. This requires a matched synthesis filter bank at the output. PR conditions impose strict constraints on the prototype filter pair:

• Nyquist(M) property: The M-fold decimated filter autocorrelation must be a perfect impulse.
• Power complementarity: The squared magnitude responses of the analysis and synthesis filters must sum to a constant. PR PFBs are critical in transmultiplexers, sub-band coding, and any application requiring lossless signal decomposition and recombination.

A robust sensing method that exploits the periodic statistical properties of modulated signals to distinguish them from stationary noise. The parallel narrowband outputs of a polyphase filter bank are ideal for computing the Spectral Correlation Function (SCF) across a wide bandwidth.

• Effective at very low signal-to-noise ratios (SNR)
• Discriminates signals based on symbol rate and carrier frequency
• Computationally efficient when paired with a channelizer

A technique enabling the reconstruction of a sparse wideband spectrum from sub-Nyquist rate samples. A polyphase filter bank can be integrated into a Modulated Wideband Converter (MWC) architecture to physically segment the spectrum before low-rate sampling.

• Drastically reduces the hardware burden for wideband sensing
• Relies on the inherent sparsity of the spectrum
• Enables real-time GHz-bandwidth monitoring with slower ADCs

A neural network architecture that directly processes complex-valued IQ data in its native domain. The sub-band outputs of a polyphase filter bank retain their complex baseband representation, making them a natural input for CVNNs that preserve phase information.

• Avoids information loss from I/Q to real-valued conversion
• Learns richer representations of phase-modulated signals
• Well-suited for automatic modulation classification tasks

A multi-dimensional spatial database integrating geolocated spectrum sensing data. A network of sensors using polyphase filter banks can simultaneously measure power across hundreds of channels to construct a high-resolution, real-time map of spectrum activity.

• Integrates propagation models and transmitter locations
• Provides a comprehensive view of spectrum occupancy across a region
• Enables proactive, map-based dynamic spectrum access

A high-resolution subspace-based algorithm for estimating the direction of arrival (DoA) of signal sources. When applied to the narrowband outputs of a polyphase filter bank, MUSIC can resolve closely spaced emitters in the angular domain for each frequency sub-band.

• Exploits the eigenstructure of the input covariance matrix
• Provides super-resolution beyond the Rayleigh limit
• Enables joint frequency-spatial spectrum cartography