Filter Design

This page covers the design methods for digital and analog filters.

Digital Filters

Digital filters process discrete-time signals. FIR (finite impulse response) filters are always stable and can achieve linear-phase characteristics.

Classification of FIR Design Methods

Method Optimization Criterion Characteristics
Parks-McClellan $L^\infty$ (minimax) Equiripple, worst-case guarantee
Least Squares $L^2$ (minimum mean-square error) Analytical, fast computation
Window Method None (empirical) Simple, intuitive
Frequency Sampling None (interpolation) FFT-based, fast
Chebyshev Interpolation Interpolation at nodes Theoretically grounded

Minimax (Maximum Error Minimization)

Minimizes the worst-case error. Ideal when a guaranteed lower bound on stopband attenuation is required.

Parks-McClellan Method

Optimal equiripple FIR filter design using the Remez exchange algorithm. Achieves the minimum maximum error for a given filter order.

Least Squares

Minimizes the average error. Can be solved analytically as a system of linear equations.

Least Squares Method

FIR filter design that minimizes the sum of squared errors. Extensions with weighting and constraints are also discussed.

Classical Methods

Computationally simple and easy to understand intuitively. Useful when requirements are relaxed or for educational purposes.

Window Method

Truncates the impulse response of an ideal filter with a window function. Simple and easy to implement.

Frequency Sampling Method

Specifies the desired frequency response at discrete points and obtains filter coefficients via the inverse DFT. Enables fast computation using the FFT.

Interpolation Methods

Designs that pass exactly through specified frequency points.

Chebyshev Interpolation

Filter design using interpolation at Chebyshev nodes. Avoids the Runge phenomenon and yields a stable approximation.

Comparison of Methods

Ripple Characteristics

Error characteristics comparison between Parks-McClellan and least squares methods
Figure 1: Error characteristics of the Parks-McClellan (equiripple) and least squares methods (21-tap FIR filter)

Window Method Comparison

The window method applies a window function when truncating the impulse response of an ideal filter to finite length. The choice of window trades off passband ripple, stopband attenuation, and transition width.

Comparison of FIR filter designs using different window functions
Figure 2: Window method comparison - Hamming, Blackman, and Kaiser windows (31 taps, fc = 0.4)

Frequency Sampling Method

The frequency sampling method samples the desired frequency response at discrete points and obtains the impulse response via the inverse DFT. Ripple can be reduced by introducing interpolation samples in the transition band. Fast computation via the FFT is possible.

FIR filter design using the frequency sampling method
Figure 3: Frequency sampling method (31 taps) - comparison with the Hamming window

Dolph-Chebyshev Window

The Dolph-Chebyshev window is the optimal window that produces equal-level sidelobes for a given main-lobe width. It yields equiripple sidelobe characteristics at a specified attenuation level.

FIR filter design using the Dolph-Chebyshev window
Figure 4: Dolph-Chebyshev window method (31 taps, 50 dB specified attenuation)

Comparison of All Methods

Comparison of FIR design methods
Figure 5: Magnitude response comparison of FIR design methods (31 taps, fc = 0.4)

Selection Guidelines

  • Guaranteed stopband attenuation required → Parks-McClellan method
  • Computational speed is critical → Window method or frequency sampling method
  • Average performance matters most → Least squares method
  • Precise control at specific frequencies → Frequency sampling or Chebyshev interpolation
  • Real-time updates (adaptive filtering) → Least squares (LMS, etc.)

Analog Filters

Analog filters process continuous-time signals. IIR (infinite impulse response) digital filters are often obtained by digitizing an analog filter design via the bilinear transform or similar techniques.

Classical Analog Filters

Filter Passband Stopband Characteristics
Butterworth Monotonic Monotonic Maximally flat response
Chebyshev Type I Equiripple Monotonic Sharp cutoff
Chebyshev Type II Monotonic Equiripple No passband ripple
Elliptic (Cauer) Equiripple Equiripple Sharpest cutoff

Selection Guidelines

  • Passband flatness is critical → Butterworth
  • Narrow transition band required → Chebyshev Type I or elliptic
  • Flat passband with tolerable stopband ripple → Chebyshev Type II
  • Meet the specification with minimum order → Elliptic filter
  • Group delay characteristics matter → Bessel

References