Bessel Filter

A control system's anti-aliasing filter must not distort the timing of feedback signals. The Bessel filter is chosen specifically because it preserves pulse shape and signal timing better than any other standard filter approximation.

Core Concept

The Bessel filter is optimized for maximally flat group delay, also called linear phase response. Group delay τg(ω) = -dφ/dω represents the time delay experienced by different frequency components. If τg is constant across frequency, all components are delayed by the same amount and the signal shape is preserved.

Butterworth and Chebyshev filters optimize magnitude response. Their group delay varies significantly with frequency, meaning different frequency components arrive at the output at different times. For a square wave or pulse input, this causes overshoot, ringing, and pulse distortion at the output. Bessel filters avoid this by accepting a poorer magnitude roll-off in exchange for constant group delay.

Bessel filter poles are derived from Bessel polynomials Bn(s). The pole positions cluster near the negative real axis in the s-plane, which is why the magnitude roll-off is gentle. For a 4th-order Bessel low-pass filter with 1 kHz cut-off, implemented using two cascaded Sallen-Key stages with TL071 op-amps, the step response shows no overshoot at all, making it ideal for digital-to-analog output smoothing in servo drives.

Key Equations

Group delay definition: τg(ω) = -dφ(ω)/dω where φ(ω) is the phase of H(jω). For ideal Bessel, τg is constant across the pass-band.

Normalized Bessel polynomials (low-pass, ω = 1 rad/s reference): B1(s) = s+1, B2(s) = s^2+3s+3, B3(s) = s^3+6s^2+15s+15

General Bessel polynomial recursion: Bn(s) = (2n-1)B(n-1)(s) + s^2 × B(n-2)(s)

Transfer function of nth-order Bessel LPF: H(s) = B0 / Bn(s/ωc) where B0 = Bn(0) is the DC normalization constant and ωc = 2πfc.

Group delay at DC for nth-order Bessel (normalized to ωc = 1): τg(0) = (2n-1)!! / (2^n × n!) which sets the baseline flat group delay.

Given:
  2nd-order Bessel low-pass filter
  Normalized polynomial: B2(s) = s^2 + 3s + 3
  Desired fc = 1 kHz (3 dB frequency)
  Denormalization factor for 2nd-order Bessel to -3dB:
    scale factor ωscale = 1.3617 × 2πfc

Why this formula:
  Bessel poles are normalized to unit group delay, not unit -3dB.
  To set -3dB at fc, multiply all s-values by ωscale = 1.3617 × ωc.

Formula:
  H(s) = 3 / (s^2 + 3s + 3) after normalization to unit group delay
  Denormalized: replace s with s / (1.3617 × 2π × 1000)

Substitution:
  ωscale = 1.3617 × 2π × 1000
         = 1.3617 × 6283.2
         = 8555 rad/s

Denormalized H(s):
  H(s) = 3ωscale^2 / (s^2 + 3ωscale × s + 3ωscale^2)
  denominator natural frequency: ω0 = ωscale × sqrt(3) = 8555 × 1.732 = 14,817 rad/s
  Q = sqrt(3)/3 = 0.5774

Final Answer:
  2nd-order Bessel LPF sections: ω0 = 14,817 rad/s, Q = 0.577
  Implement as Sallen-Key with R = 10 kΩ, C = 6.75 nF (from ω0 = 1/RC).

Exam Tip: GATE questions on Bessel filters focus on comparing it to Butterworth and Chebyshev. Remember the hierarchy: Butterworth = maximally flat magnitude, Chebyshev = equiripple magnitude (sharpest roll-off for given order), Bessel = maximally flat group delay (linear phase, best pulse response). Bessel has the worst magnitude roll-off of the three. A Bessel filter's -3 dB frequency is not the same as its group-delay normalization frequency. The 2nd-order Bessel needs a scale factor of 1.3617 to move the -3 dB point to ωc.

Key Properties

• Maximally flat group delay across the pass-band. All frequencies are delayed by nearly the same time, preserving signal shape.
• Phase response is approximately linear with frequency, equivalent to a pure time delay in the pass-band.
• Step response has zero (or near-zero) overshoot, making Bessel ideal for pulse and digital signal filtering.
• Magnitude roll-off is the poorest of Butterworth, Chebyshev, and Bessel for the same filter order.
• Bessel poles cluster near the negative real axis. 2nd-order Bessel poles are at s = -1.5 ± j0.866 (normalized).
• Normalization: Bessel polynomials are normalized to unit group delay at DC. To set the -3 dB at fc, a frequency scaling factor must be applied: 1.3617 for n=2, 1.7557 for n=3, 2.1139 for n=4.
• Used in servo drive output filters, biomedical signal acquisition, and digital communication pulse shaping where timing accuracy matters more than sharp cut-off.

Quick Revision

• Bessel filter: maximally flat group delay. Linear phase in pass-band.
• τg(ω) = -dφ/dω is constant across pass-band for ideal Bessel.
• Step response: no overshoot. Best of all standard filters for pulse fidelity.
• Worst magnitude roll-off compared to Butterworth and Chebyshev at same order.
• 2nd-order Bessel polynomial: B2(s) = s^2 + 3s + 3. Q = 1/sqrt(3) = 0.577.
• Frequency scaling factors to set -3 dB: n=2: 1.3617, n=3: 1.7557, n=4: 2.1139.
• Applications: servo control, biomedical, pulse shaping, CCD clocking circuits.
• Exam trap: Students assume Bessel -3 dB frequency equals its group-delay normalization frequency. They are not the same. The Bessel polynomial is normalized to unit group delay at DC, not to unit -3 dB. Using the wrong normalization gives a filter with the -3 dB point shifted away from the intended fc.