Why does the bilinear transform cause frequency warping?

The bilinear transform substitution s = (2/T)·(z-1)/(z+1) creates a nonlinear frequency mapping because it maps the infinite jω axis onto the finite unit circle. Evaluating on the unit circle (z = e^(jω_d·T)): s = (2/T)·(e^(jω_d·T) - 1)/(e^(jω_d·T) + 1) = (2/T)·(e^(jω_d·T/2) - e^(-jω_d·T/2))/(e^(jω_d·T/2) + e^(-jω_d·T/2))·(1/j)·j = j·(2/T)·tan(ω_d·T/2). So s = jω_a maps to: ω_a = (2/T)·tan(ω_d·T/2). For ω_d·T << 1 (frequency much below Nyquist): tan(x) ≈ x, so ω_a ≈ ω_d (no warping). Near Nyquist (ω_d → π/T): tan(ω_d·T/2) → ∞, so ω_a → ∞. This means the entire analog frequency range (0 to ∞) maps to the digital range (0 to f_s/2), with increasing compression at higher frequencies. Practical impact: a Butterworth lowpass designed for f_c = 10 kHz, sampled at f_s = 44.1 kHz. Without pre-warping: the -3 dB point shifts to f_d = (2/T)·arctan(ω_c·T/2)/(2π) = (44100/π)·arctan(π·10000/44100) = 9,260 Hz (7.4% error). With pre-warping: design the analog prototype at ω_a = (2/T)·tan(π·10000/44100) = 2π·10,800 Hz, which maps back to exactly 10,000 Hz digitally. The error increases as f_c approaches f_s/2. At f_c = 0.4·f_s: error without pre-warping exceeds 15%.

How does the bilinear transform compare to other analog-to-digital methods?

Three primary methods exist for converting analog prototypes to digital filters: 1) Bilinear transform: s = (2/T)·(z-1)/(z+1). Maps left half-plane → unit circle interior (stable → stable). No aliasing. Frequency warping requires pre-warping at critical frequencies. Preserves filter order. Best for: IIR filters where magnitude response shape matters (lowpass, bandpass, notch). Used in 95%+ of practical IIR filter designs. 2) Impulse invariance: h[n] = T·h_a(nT). Samples the analog impulse response directly. No frequency warping (exact frequency correspondence). Aliasing: analog frequency components above f_s/2 fold back, corrupting the response. Only usable for lowpass and bandpass filters (not highpass or bandstop, which have significant energy near f_s/2). Preserves filter order. Best for: filters where exact frequency mapping is critical and the analog prototype is bandlimited. 3) Matched z-transform: maps analog poles and zeros directly via z = e^(sT). Each analog pole s_k maps to z_k = e^(s_k·T). Preserves pole/zero locations exactly. Aliasing possible (same as impulse invariance). Simple to implement. Best for: control systems where pole locations are more important than frequency response shape. Comparison for a 5th-order Butterworth LPF at f_c = 0.1·f_s: Bilinear (pre-warped): exact -3 dB at f_c, maximum passband error <0.01 dB. Impulse invariance: -3 dB exactly at f_c, but stopband floor rises due to aliased images (typically -50 to -60 dB vs. -∞ for analog). Matched-z: -3 dB within 0.5% of f_c, similar aliasing issues.

Where is the bilinear transform used in RF systems?

The bilinear transform is embedded in virtually every DSP-based RF subsystem: 1) Software-defined radio (SDR) channel filters: the digital downconverter in an SDR receiver requires sharp channel-select filters. A 7th-order Chebyshev Type I with 0.5 dB ripple and 200 kHz bandwidth, designed as an analog prototype and converted via bilinear transform, provides >60 dB adjacent channel rejection. Pre-warping ensures the 200 kHz bandwidth is exact at the output sample rate. 2) Digital predistortion (DPD): linearizing RF power amplifiers requires modeling the PA's nonlinear memory effects with IIR filters. The memory polynomial model uses bilinear-transformed analog basis functions to capture the PA's frequency-dependent AM/AM and AM/PM characteristics. Typical DPD systems use 3rd-5th order IIR sections designed via bilinear transform. 3) Sigma-delta ADC loop filters: oversampled ADC architectures use high-order noise-shaping loop filters. These are designed as analog integrator chains and converted to digital using the bilinear transform (or its multi-rate equivalent). A 4th-order sigma-delta modulator at 10× oversampling achieves 80+ dB SNR. 4) Radar pulse compression: matched filters for chirp (LFM) waveforms can be designed as analog all-pass networks and digitized via bilinear transform. The frequency warping is pre-compensated to maintain the matched filter's time-bandwidth product. 5) Audio codec filters in RF transceivers: the analog anti-alias and reconstruction filters in baseband I/Q processing chains are often implemented digitally using bilinear-transformed Butterworth or Bessel prototypes for linear phase response.

DSP & Filter Design

Bilinear Transform

Q: How does the bilinear transform compare to other analog-to-digital methods?

Three primary methods exist for converting analog prototypes to digital filters: 1) Bilinear transform: s = (2/T)·(z-1)/(z+1). Maps left half-plane → unit circle interior (stable → stable). No aliasing. Frequency warping requires pre-warping at critical frequencies. Preserves filter order. Best for: IIR filters where magnitude response shape matters (lowpass, bandpass, notch). Used in 95%+ of practical IIR filter designs. 2) Impulse invariance: h[n] = T·h_a(nT). Samples the analog impulse response directly. No frequency warping (exact frequency correspondence). Aliasing: analog frequency components above f_s/2 fold back, corrupting the response. Only usable for lowpass and bandpass filters (not highpass or bandstop, which have significant energy near f_s/2). Preserves filter order. Best for: filters where exact frequency mapping is critical and the analog prototype is bandlimited. 3) Matched z-transform: maps analog poles and zeros directly via z = e^(sT). Each analog pole s_k maps to z_k = e^(s_k·T). Preserves pole/zero locations exactly. Aliasing possible (same as impulse invariance). Simple to implement. Best for: control systems where pole locations are more important than frequency response shape. Comparison for a 5th-order Butterworth LPF at f_c = 0.1·f_s: Bilinear (pre-warped): exact -3 dB at f_c, maximum passband error <0.01 dB. Impulse invariance: -3 dB exactly at f_c, but stopband floor rises due to aliased images (typically -50 to -60 dB vs. -∞ for analog). Matched-z: -3 dB within 0.5% of f_c, similar aliasing issues.

Q: Where is the bilinear transform used in RF systems?

The bilinear transform is embedded in virtually every DSP-based RF subsystem: 1) Software-defined radio (SDR) channel filters: the digital downconverter in an SDR receiver requires sharp channel-select filters. A 7th-order Chebyshev Type I with 0.5 dB ripple and 200 kHz bandwidth, designed as an analog prototype and converted via bilinear transform, provides >60 dB adjacent channel rejection. Pre-warping ensures the 200 kHz bandwidth is exact at the output sample rate. 2) Digital predistortion (DPD): linearizing RF power amplifiers requires modeling the PA's nonlinear memory effects with IIR filters. The memory polynomial model uses bilinear-transformed analog basis functions to capture the PA's frequency-dependent AM/AM and AM/PM characteristics. Typical DPD systems use 3rd-5th order IIR sections designed via bilinear transform. 3) Sigma-delta ADC loop filters: oversampled ADC architectures use high-order noise-shaping loop filters. These are designed as analog integrator chains and converted to digital using the bilinear transform (or its multi-rate equivalent). A 4th-order sigma-delta modulator at 10× oversampling achieves 80+ dB SNR. 4) Radar pulse compression: matched filters for chirp (LFM) waveforms can be designed as analog all-pass networks and digitized via bilinear transform. The frequency warping is pre-compensated to maintain the matched filter's time-bandwidth product. 5) Audio codec filters in RF transceivers: the analog anti-alias and reconstruction filters in baseband I/Q processing chains are often implemented digitally using bilinear-transformed Butterworth or Bessel prototypes for linear phase response.

/by-LIN-ee-ur/ — Tustin's Method

Maps analog s-domain transfer functions to digital z-domain via s = (2/T)·(z−1)/(z+1). Guarantees stability preservation (LHP → unit circle interior). Frequency warping: ω_a = (2/T)·tan(ω_d·T/2). Pre-warping corrects critical frequencies. Standard method for IIR digital filter design from Butterworth, Chebyshev, and elliptic prototypes. Used in SDR channel filters, DPD, and sigma-delta modulators.

Map: s → z

Warp: tan(ωT/2)

Stable: Guaranteed

Understanding the Bilinear Transform

Designing digital filters from scratch in the z-domain is mathematically challenging. The bilinear transform provides an elegant shortcut: start with a well-understood analog filter prototype (Butterworth, Chebyshev, elliptic) with decades of design tables and closed-form solutions, then convert it to a digital filter through a simple algebraic substitution. The resulting digital filter inherits the analog prototype's frequency-response shape, stability, and filter order.

The trade-off is frequency warping. The substitution compresses the infinite analog frequency axis onto the finite digital frequency range (0 to f_s/2), with increasing distortion at higher frequencies. Pre-warping compensates for this by adjusting the analog prototype's critical frequencies before transformation, ensuring the digital filter's cutoff frequencies land exactly where intended. For RF DSP applications where filters operate well below the Nyquist frequency, the warping is negligible.

Core Equations

Bilinear Transform Substitution:
s = (2/T)·(z − 1)/(z + 1)

Inverse:
z = (1 + sT/2)/(1 − sT/2)

Frequency Warping Relation:
ω_a = (2/T)·tan(ω_d·T/2)

Pre-Warping (at critical freq ω_c):
ω_a,pre = (2/T)·tan(π·f_c/f_s)

Example (f_c = 10 kHz, f_s = 44.1 kHz):
No pre-warp: f_d = 9,260 Hz (7.4% error)
Pre-warped: f_d = 10,000 Hz (exact)

Analog-to-Digital Filter Methods

Method	Mapping	Aliasing	Warping	Best For
Bilinear	s = (2/T)(z−1)/(z+1)	None	Yes	IIR LP/BP/HP/BS
Impulse invariance	h[n] = T·h_a(nT)	Yes	None	Bandlimited LP/BP
Matched z	z = e^sT	Yes	None	Control systems
Step invariance	Step response match	Yes	None	DAC reconstruction
FOH transform	First-order hold	Partial	Partial	Smoothed sampling

Common Questions

Frequently Asked Questions

Why frequency warping?

Maps infinite jω axis to finite unit circle. ω_a = (2/T)tan(ω_dT/2). Near DC: ω_a ≈ ω_d. Near Nyquist: tan → ∞. Pre-warp at f_c eliminates cutoff error. At f_c = 0.1·f_s: <1% error without pre-warping.

Comparison to other methods?

Bilinear: no aliasing, warping, 95%+ of IIR designs. Impulse invariance: no warping but aliasing (LP/BP only). Matched-z: exact pole mapping, control-focused. Bilinear preserves filter order and stability.

RF applications?

SDR channel filters (Chebyshev, >60 dB rejection). DPD memory modeling (IIR basis functions). Sigma-delta loop filters (80+ dB SNR). Radar matched filters (chirp compression). Baseband I/Q anti-alias (Bessel for linear phase).

Related Terms

← Biconical Antenna Binary Phase-Shift Keying →

DSP Solutions

Precision RF Components

RF Essentials provides precision terminations and custom waveguide assemblies for SDR development, digital filter characterization, and DSP-based RF subsystem integration.

Request a Quote