Electronics Guide

Active filters combine passive components with active devices, typically operational amplifiers, to achieve frequency-selective behavior with significant advantages over purely passive implementations. By incorporating gain elements, active filters can provide signal amplification while filtering, eliminate the need for bulky and lossy inductors at low frequencies, achieve precise frequency response characteristics, and offer easy tunability. These attributes make active filters the preferred choice for most audio, instrumentation, and low-frequency signal processing applications.

The design of active filters involves selecting an appropriate topology, determining the filter order and response type, and calculating component values to achieve the desired cutoff frequency, quality factor (Q), and gain. This article explores the major active filter topologies, their characteristics, design equations, and practical implementation considerations that enable engineers to create precise frequency-selective circuits for diverse applications.

Fundamentals of Active Filter Design

Active filters use operational amplifiers or other active devices in conjunction with resistors and capacitors to implement filter transfer functions. The active device provides gain that compensates for losses in the RC network and enables the realization of complex pole locations that would require inductors in a passive implementation.

Key Parameters

Several parameters characterize active filter performance:

• Cutoff Frequency (f_c): The frequency at which the response falls 3 dB below the passband level, marking the transition between passband and stopband. Chebyshev designs are the exception, because their cutoff is defined at the edge of the ripple band rather than at the 3 dB point
• Quality Factor (Q): A measure of how sharply the pole pair resonates, with higher Q producing a steeper transition and a more pronounced peak near the cutoff
• Passband Gain: The amplification provided by the filter in the passband region
• Filter Order: The number of poles in the transfer function, determining the ultimate rolloff rate (20 dB/decade per pole)
• Sensitivity: How much the filter response changes with component value variations

Biquadratic Sections

Most active filter designs are based on second-order sections, also called biquadratic or biquad sections. The general second-order transfer function takes the form:

H(s) = (a₂s² + a₁s + a₀) / (s² + (ω₀/Q)s + ω₀²)

Here ω₀ is the pole frequency and Q is the quality factor. The numerator coefficients select the filter type: a₂ alone gives a high-pass, a₁ alone a band-pass, a₀ alone a low-pass, a₂ together with a₀ a notch, and the complete set with the sign of a₁ reversed an all-pass. Higher-order filters cascade second-order sections, each contributing one pole pair, and an odd order adds one real pole.

One distinction repays attention throughout the article. The parameter ω₀ is the pole frequency, which is not in general the 3 dB frequency. The two coincide for a Butterworth alignment. For Bessel and Chebyshev alignments they differ, which is why denormalization tables list the pole frequency and the cutoff frequency as separate columns.

Sallen-Key Configuration

The Sallen-Key topology, developed by R.P. Sallen and E.L. Key at MIT Lincoln Laboratory in 1955, is one of the most popular active filter configurations. It uses a single operational amplifier in a voltage follower or non-inverting amplifier configuration, with an RC network providing the frequency-selective behavior. The Sallen-Key structure is valued for its simplicity, low component count, and ease of design.

Low-Pass Sallen-Key Filter

The low-pass Sallen-Key filter consists of two resistors and two capacitors arranged around a non-inverting amplifier. The basic configuration places one resistor-capacitor pair in series from input to the non-inverting input, with a second capacitor from this node to ground, and another resistor providing positive feedback from the output. The transfer function provides a second-order low-pass response with:

• Cutoff frequency: f_c = 1/(2π√(R₁R₂C₁C₂))
• Quality factor: Q = √(R₁R₂C₁C₂)/(C₂(R₁ + R₂) + R₁C₁(1 − K))
• Passband gain: K = 1 + R_f/R_g, set by the non-inverting amplifier's feedback divider

For the unity-gain configuration (voltage follower), K = 1, and the quality factor depends only on the passive component ratios. This simplifies tuning but limits Q to relatively low values. For higher Q values, gain is increased by adding feedback resistors, though stability margins must be carefully managed.

Equal-Value Component Design

A practical design approach uses equal-value resistors (R₁ = R₂ = R) and equal-value capacitors (C₁ = C₂ = C), which simplifies procurement and matching. Under that constraint the cutoff frequency becomes f_c = 1/(2πRC), and the quality factor of the unity-gain version is fixed at Q = 0.5, a critically damped response with two coincident real poles. Reaching a higher Q therefore requires gain, and substituting K into the general expression collapses it to

Q = 1/(3 − K)

This relationship exposes a stability boundary. As the non-inverting gain K approaches 3 the Q rises without bound, and at K = 3 the equal-component stage oscillates. Practical designs keep K well below 3 and match the gain-setting divider closely, because Q grows steadily more sensitive to K as the limit approaches.

The table below gives the pole Q of the common second-order alignments and the equal-component gain each one demands. Because this design fixes the passband gain at K, the choice is not free: a Butterworth section built this way carries a gain of 1.586, about 4 dB, which the surrounding level plan must absorb.

High-Pass Sallen-Key Filter

The high-pass version exchanges the positions of the resistors and the capacitors, so capacitors now sit in series with the signal path and block DC. The pole frequency keeps exactly the same form, f_c = 1/(2π√(R₁R₂C₁C₂)), and the damping term is the dual of the low-pass one. For the unity-gain high-pass, Q = √(R₁/R₂)√(C₁C₂)/(C₁ + C₂), and with equal components the familiar Q = 1/(3 − K) applies unchanged.

One caution has no counterpart in the low-pass case. A Sallen-Key high-pass cannot hold its passband above the amplifier's own bandwidth, so the circuit is in truth a wide band-pass whose upper corner belongs to the op-amp rather than to the design. Where the passband must reach well into the megahertz region, the amplifier, not the RC network, decides the result.

Sensitivity Analysis

Because ω₀ varies as the inverse square root of a product of four components, the frequency sensitivity to any single one of them is one half: a 1 percent error in one resistor or capacitor moves the pole frequency by about 0.5 percent. That figure is benign, and it does not grow with Q.

The sensitivity of Q to the gain is the problem. Differentiating the equal-component result Q = 1/(3 − K) gives a normalized sensitivity of KQ, which is the same as 3Q − 1. At the Butterworth value Q = 0.707 the sensitivity is 1.1, so a 1 percent gain error costs 1 percent of Q and nobody notices. At Q = 5 the sensitivity is 14. At Q = 10 it is 29, so a 1 percent error in K becomes a 29 percent error in Q, which is the difference between a flat passband and a visible peak. Holding Q to a few percent at Q = 5 already calls for a gain-setting divider matched to roughly 0.1 percent.

This sensitivity, rather than component count, is the real reason designers leave the Sallen-Key form behind for high-Q stages. The multiple-feedback section has no positive-feedback gain path for an error to be multiplied through, and the state-variable section distributes the pole across a loop so that no single element dominates. As a working boundary, the Sallen-Key topology suits stages of Q up to about 5, remains usable to roughly 10 with precision components, and should be avoided above that.

Multiple Feedback Configuration

The multiple feedback (MFB) topology, also known as the infinite-gain multiple feedback or Rauch filter, uses an operational amplifier in an inverting configuration with multiple feedback paths. This arrangement provides both the desired filter response and signal inversion, making it useful in applications where phase inversion is acceptable or desired.

Low-Pass MFB Filter

The low-pass MFB filter uses three resistors and two capacitors around a single inverting amplifier. The input resistor R₁ drives an intermediate node; a second resistor R₂ links that node to the inverting input, which the feedback holds at virtual ground; a feedback resistor R₃ returns the output to the intermediate node; a feedback capacitor C₁ connects the output back to the inverting input; and a capacitor C₂ runs from the intermediate node to ground. The arrangement yields a second-order low-pass response.

Writing the node equations with the inverting input held at virtual ground and eliminating the intermediate node gives the transfer function

H(s) = −(1/(R₁R₂C₁C₂)) / (s² + s(1/R₁ + 1/R₂ + 1/R₃)/C₂ + 1/(R₂R₃C₁C₂))

from which the design parameters follow directly:

• Cutoff frequency: f_c = 1/(2π√(R₂R₃C₁C₂)). The input resistor R₁ does not appear, so gain can be set last without moving the corner
• Quality factor: Q = √(R₂R₃C₁C₂)/(C₁(R₂ + R₃ + R₂R₃/R₁))
• DC gain: K = −R₃/R₁. The feedback resistor R₃ and the input resistor R₁ together fix both the magnitude of the gain and its sign

The inversion is not incidental. Signal reaches the output only through the amplifier's inverting input, so every multiple-feedback section inverts, and a cascade of an even number of them restores the original polarity. Where polarity matters and the order is odd, a unity-gain inverter or a non-inverting first section fixes it.

The capacitor ratio is not free either. Substituting R₁ = R₃/|K| and maximizing Q over the ratio R₂/R₃ shows that Q peaks when R₂/R₃ = 1/(1 + |K|), and that its largest attainable value is √(C₂/C₁)/(2√(1 + |K|)). A design is therefore realizable only when

C₂ ≥ 4Q²(1 + |K|)C₁

The grounded capacitor must exceed the feedback capacitor by a factor that grows with the square of Q, and this, rather than any sensitivity argument, is what limits the multiple-feedback low-pass to moderate Q. A unity-gain section of Q = 5 already needs a ratio of 200 to 1, which strains the range of practical capacitor values.

A worked example makes the procedure concrete. For a unity-gain inverting Butterworth section, Q = 0.707 and |K| = 1, so C₂ must be at least four times C₁ and the optimum resistor ratio is R₂/R₃ = 0.5. Taking C₁ = 1.0 nF, C₂ = 4.0 nF, R₂ = 1.0 kΩ, and R₁ = R₃ = 2.0 kΩ yields Q = 0.707, a DC gain of −1, and f_c = 56.3 kHz. Scaling both capacitors by a common factor moves the cutoff frequency without disturbing Q or gain, which is the usual way to reach a target frequency once the ratios are settled.

Band-Pass MFB Filter

Replacing the input resistor with a capacitor, or placing one in series with it, turns the same network into a band-pass section. The added zero at the origin removes the constant term from the numerator and leaves a response that peaks at the center frequency and falls at 20 dB per decade on either side. A third resistor from the intermediate node to ground then allows the center-frequency gain and the Q to be chosen separately, which the simpler two-resistor arrangement does not permit.

The band-pass form is widely used in tone detection, audio equalizer sections, and channel filtering. Its practical ceiling is loop gain: the open-loop gain the amplifier must still possess at the center frequency rises roughly with the square of Q, so discrete designs rarely push a single multiple-feedback band-pass section past a Q of 10 to 20. Beyond that a state-variable section is the sounder choice, and it yields the low-pass and high-pass responses as a bonus.

Advantages and Limitations

The MFB topology offers several advantages:

• Lower sensitivity: Generally exhibits lower sensitivity to component variations than Sallen-Key, especially for Q values above 2
• Good high-frequency performance: The inverting configuration provides better bandwidth utilization of the op-amp
• Inherent gain: Provides adjustable gain as part of the filter function

The limitations are the inverting output, a comparatively low input impedance set by the input resistor and the network behind it, somewhat more involved design equations, and the capacitor ratio constraint described above. The virtual ground, moreover, holds only while the amplifier retains loop gain, so the section degrades as the signal frequency climbs toward the gain-bandwidth product, quietly rather than abruptly.

State-Variable Filters

State-variable filters use multiple operational amplifiers configured as integrators and summers to implement second-order filter functions. This topology derives its name from the state-space representation of differential equations, where each integrator output represents a state variable. State-variable filters provide simultaneous low-pass, band-pass, and high-pass outputs from a single circuit, offering exceptional flexibility and tunability.

Basic State-Variable Structure

The classic state-variable filter uses three operational amplifiers:

• First amplifier: Configured as a summing amplifier that combines the input signal with feedback from the filter outputs
• Second amplifier: Configured as an integrator, producing the band-pass output
• Third amplifier: Also an integrator, producing the low-pass output

The high-pass output is available at the summing amplifier output. Each output provides a specific frequency response characteristic, all sharing the same cutoff frequency and Q value.

Frequency and Q Control

The cutoff frequency comes from the integrator time constants, f_c = 1/(2πRC) for each integrator when both use the same R and C. Changing the two integrator resistors together tunes the frequency and leaves Q alone. The quality factor comes from a resistor ratio in the damping network around the summing amplifier, so Q can be trimmed without altering either time constant and therefore without moving the cutoff frequency. The input resistor sets the gain independently of both.

This separation of frequency, Q, and gain is the topology's defining advantage, and it makes the section tunable and trimmable rather than merely calculable. Laboratory and prototype builds often fit potentiometers to two of the three.

Three amplifiers cost board area, quiescent current, and noise, and the topology repays them in three ways. It presents the low-pass, band-pass, and high-pass responses at the same time, from three low-impedance amplifier outputs, so one circuit serves where three would otherwise be needed. Each parameter answers to its own element. And the pole pair is formed by a loop rather than by a single positive-feedback path, so the passive sensitivities stay near unity even at high Q instead of growing in proportion to it as they do in the Sallen-Key form. For a stage of Q = 20 or more, that last property alone settles the choice.

Notch and All-Pass Outputs

Summing the low-pass and high-pass outputs in equal measure, which takes a fourth amplifier, produces a band-stop response with its notch at the pole frequency. The notch width follows from Q, but the depth does not follow from the design at all: it is decided by how exactly the two summed signals cancel, so component matching and amplifier offset, rather than the transfer function, determine how deep the null goes. Weighting the two outputs unequally moves the zero off the pole frequency, which is how the low-pass-notch and high-pass-notch sections of an elliptic filter are built. Adding the band-pass output with reversed sign and unit weight instead produces the all-pass response, constant in magnitude and sweeping 360 degrees in phase.

Biquad Filters

Biquad filters (biquadratic filters) represent a general class of second-order active filters that implement the standard biquadratic transfer function. While the term "biquad" technically applies to any second-order filter, it commonly refers to specific topologies that provide independent control of all transfer function coefficients.

Standard Biquad Configuration

The standard biquad uses two integrators in a feedback loop, similar to the state-variable filter, but with modified input and feedback arrangements. The circuit typically provides access to adjust poles and zeros independently, enabling implementation of arbitrary second-order transfer functions including low-pass, high-pass, band-pass, notch, and all-pass responses.

Programming Biquad Parameters

Biquad filters are particularly useful when the filter response must be precisely controlled or changed during operation. The key parameters are:

• Center frequency (f₀): Controlled by the integrator resistors or capacitors
• Quality factor (Q): Set by feedback resistor ratios
• Zero locations: Determined by input summing network coefficients
• Gain: Adjusted independently of frequency response shape

This parameter independence makes biquads well suited to adaptive filters, tunable equalizers, and any application whose response must change while the circuit runs.

The name carries over into digital signal processing, where a cascade of second-order sections, each also called a biquad, is the standard realization of a high-order infinite-impulse-response filter. The reason is the same in both domains. A long polynomial evaluated directly is intolerably sensitive to its coefficients, whereas the same response split into second-order factors is not.

Tow-Thomas and KHN Structures

The Tow-Thomas and Kerwin-Huelsman-Newcomb (KHN) filters represent refined state-variable implementations that offer specific advantages in practical applications.

Tow-Thomas Biquad

The Tow-Thomas biquad, which developed from the work of J. Tow (1968) and L. C. Thomas (1971), uses two integrators and an inverting gain stage in a two-integrator loop (sometimes called the "ring of three") to implement a second-order transfer function. The circuit provides low-pass and band-pass outputs simultaneously, with the following characteristics:

• Independent Q adjustment: A single damping resistor R_Q across the lossy integrator's capacitor sets Q without touching either integrator time constant
• Independent gain: The input resistor fixes the passband gain without disturbing f₀ or Q
• Frequency tuning, with a caveat: For matched integrators, ω₀ = 1/RC and Q = R_Q/R, so retuning through the integrator resistors scales Q along with the frequency. Tuning through the integrator capacitors, or scaling R_Q with R, leaves Q where it was
• Low sensitivity: The passive sensitivities remain of order unity, and no positive-feedback path exists whose gain error would be multiplied by Q

The Tow-Thomas topology is widely used in voltage-controlled filters for synthesizers and in tunable electronic crossovers. The frequency-determining resistors can be replaced by voltage-controlled elements, most often operational transconductance amplifiers or junction field-effect transistors biased in their ohmic region, which gives electronic tuning across several decades. The usable range is bounded above by the amplifier's gain-bandwidth product and below by integrator leakage, input bias current, and offset, all of which the integrating capacitor amplifies as the time constant grows.

KHN Filter

The Kerwin-Huelsman-Newcomb (KHN) filter, introduced by W. Kerwin, L. Huelsman, and R. Newcomb in 1967, provides the complete set of second-order outputs (low-pass, band-pass, and high-pass) with excellent parameter control. The standard KHN configuration uses three operational amplifiers and offers:

• Simultaneous outputs: All three basic filter types available concurrently
• Orthogonal tuning: Frequency and Q adjustable without interaction
• Low output impedance: Each output comes directly from an op-amp
• Easy notch creation: Summing LP and HP outputs produces notch response

The KHN structure forms the basis for many commercial universal active filter integrated circuits, which include on-chip resistor networks for convenient frequency and Q programming.

All-Pass and Phase-Shift Networks

All-pass filters pass all frequencies with equal amplitude while introducing frequency-dependent phase shift. Unlike conventional filters that modify amplitude response, all-pass networks are used specifically to control signal phase without affecting magnitude. Applications include group delay equalization, phase compensation in feedback systems, and creating specific phase relationships in audio and communication systems.

First-Order All-Pass Filter

A first-order all-pass section sweeps 180 degrees in total, from 0 degrees at DC to −180 degrees well above the characteristic frequency, passing through −90 degrees exactly at that frequency. A single op-amp with one resistor-capacitor pair implements it:

• Transfer function: H(s) = (ω₀ − s)/(ω₀ + s)
• Magnitude: |H(jω)| = 1 at every frequency, because the numerator and denominator are complex conjugates of equal modulus
• Phase: −2 arctan(ω/ω₀), falling continuously from 0 to −180 degrees
• Characteristic frequency: f₀ = 1/(2πRC)
• Group delay: 2/ω₀, equal to 2RC, at low frequencies, falling toward zero well above f₀

The zero of this transfer function sits in the right half plane, the mirror image of the pole. That is the defining feature of an all-pass network and the reason the magnitude is flat: the zero contributes exactly the magnitude the pole removes, and twice the phase.

Second-Order All-Pass Filter

A second-order all-pass sweeps the full 360 degrees. Its numerator places a pair of zeros as the mirror image of the denominator's poles across the imaginary axis, so the two quadratics match in every coefficient except the sign of the linear term:

H(s) = (s² − (ω₀/Q)s + ω₀²) / (s² + (ω₀/Q)s + ω₀²)

The magnitude is again unity everywhere. The phase falls from 0 degrees at DC through −180 degrees at f₀ to −360 degrees well above it, and Q governs how abruptly that happens: a high-Q section concentrates its phase swing into a narrow band and produces a correspondingly tall, narrow peak in group delay near f₀. That peak is the designer's instrument. In practice the section is often taken from a state-variable block, whose outputs need only be summed with the band-pass term inverted.

Group Delay Equalization

Group delay is the negative derivative of phase with respect to frequency, and it measures the time a narrowband component centered at that frequency takes to traverse the network. Group delay that varies across the band smears a pulse, because its components arrive at different times, and the effect grows worse as the amplitude response grows sharper. A Chebyshev or elliptic filter buys its steep skirt with a pronounced group-delay peak near the cutoff, which is the trade a Bessel alignment refuses to make.

All-pass sections repair that damage without touching the amplitude response. A cascade of them, each placed at its own frequency and Q, adds delay where the main filter has too little, flattening the total. The technique is standard in digital communication links, high-fidelity audio crossovers, and radar pulse-compression chains, and it always costs absolute delay: a delay equalizer can only add, never subtract.

Phase-Shift Oscillators

Cascading multiple first-order all-pass sections creates a phase-shift network that can form the basis of an oscillator. When the total phase shift reaches 360 degrees at a specific frequency, and the loop gain equals unity, the system oscillates. This principle underlies simple phase-shift oscillators used in audio and low-frequency test applications.

Switched-Capacitor Filters

Switched-capacitor filters replace resistors with capacitors that are periodically switched by clock signals, using the principle that a capacitor switched at high frequency between two nodes behaves like a resistor. This technique enables the integration of precision analog filters on standard digital CMOS processes without requiring the accurate resistors that are difficult to fabricate in integrated circuits.

Operating Principle

When a capacitor C is switched at frequency f_clk between two voltage nodes, it transfers charge at an average rate equivalent to a resistor of value

R_eq = 1/(f_clkC)

The numbers explain why the technique took over integrated filtering. A 1 pF capacitor switched at 1 MHz behaves as a 1 MΩ resistor, a value that would consume an impractical area as a diffused or polysilicon resistor and would still hold a tolerance no better than roughly 20 percent. Capacitor ratios on the same die, by contrast, match to a fraction of a percent. The equivalence holds only while the clock runs much faster than the signal, since it is an average over many switching cycles rather than an instantaneous identity.

The filter's cutoff frequency becomes proportional to the clock frequency rather than absolute RC time constants. This ratio-metric behavior means that filter characteristics depend only on capacitor ratios (which can be controlled precisely in IC fabrication) and the clock frequency (which can be generated accurately). This principle enables:

• Accurate responses without trimming: Accuracy rests on capacitor ratios, which photolithography holds to a fraction of a percent, rather than on absolute values
• Clock tuning: The cutoff frequency tracks the clock, so one part covers a wide range and a crystal-derived clock gives it crystal stability
• Monolithic integration: A complete high-order filter fits on one die with no external frequency-determining components
• Low power: CMOS switches dissipate only switching losses, which fall with the clock rate

Anti-Aliasing Requirements

Because a switched-capacitor filter samples its input at the clock rate, it aliases like any other sampled system. Commercial parts standardize on clock-to-cutoff ratios of 50 to 1 and 100 to 1, which places the sampling rate far above the passband but not out of reach of out-of-band interference. A continuous-time filter ahead of the switched-capacitor section is therefore usual, to attenuate energy near the clock frequency and its multiples before it folds into the passband. A simple smoothing filter after the section is equally usual, because the output is a staircase held at the clock rate rather than a continuous waveform.

Clock Feedthrough

Switching transients can couple through to the output, creating artifacts at the clock frequency and its harmonics. Careful circuit design, balanced switch configurations, and post-filtering minimize this effect. Modern switched-capacitor filter ICs incorporate internal measures to reduce clock feedthrough to acceptable levels for most applications.

Common Applications

Switched-capacitor filters are extensively used in:

• Anti-aliasing for ADCs: Integrated filters preceding analog-to-digital converters
• Voice-band filtering: Telephone and audio codec applications
• Data acquisition systems: Programmable signal conditioning
• Communication receivers: Channel selection and IF filtering

Commercial switched-capacitor filter ICs are available with configurable response types, adjustable cutoff frequencies, and multiple filter orders, providing convenient solutions for many filtering requirements.

Tunable and Programmable Filters

Many applications require filters whose characteristics can be adjusted during operation, either manually for calibration and setup or automatically as part of adaptive signal processing. Several techniques enable real-time filter tuning while maintaining stable and predictable performance.

Voltage-Controlled Filters

Voltage-controlled filters (VCFs) use voltage-variable elements to adjust filter parameters in response to control signals. Common approaches include:

• OTA-based filters: Operational transconductance amplifiers whose transconductance varies with a control current, enabling voltage control of filter cutoff frequency
• JFET variable resistors: Junction field-effect transistors operating in their ohmic region as voltage-controlled resistors
• Multiplying DAC tuning: Digital-to-analog converters used to scale reference voltages and effectively vary component values
• Switched-capacitor clock tuning: Varying the clock frequency of switched-capacitor filters to shift cutoff frequency

Digitally Programmable Filters

Digitally programmable filters use digital control signals to select component values or filter configurations. Implementation methods include:

• Switched resistor/capacitor arrays: Banks of components selected by analog switches under digital control
• Digital potentiometers: Programmable resistive dividers that replace fixed resistors
• DAC-controlled VCFs: Digital-to-analog converters generating control voltages for voltage-controlled filters
• Filter coefficient registers: Dedicated filter ICs with digitally loaded coefficients

Automatic Tuning Systems

Some filter systems incorporate automatic tuning to compensate for component variations, temperature drift, or aging effects. A reference signal at a known frequency is applied to the filter, and feedback circuits adjust component values to maintain the desired response. This technique is commonly used in high-performance communication receivers and precision measurement systems.

Gyrator-Based Inductance Simulation

At low frequencies, inductors become impractically large, lossy, and expensive. Gyrators are active circuits that simulate inductive behavior using capacitors, enabling the implementation of inductor-based filter topologies without actual inductors. This technique is essential for realizing LC filter responses at audio and sub-audio frequencies.

Gyrator Fundamentals

A gyrator is a two-port network that transforms impedance by inversion. When a capacitor C is connected to one port of an ideal gyrator with gyration resistance R, the impedance seen at the other port is:

Z_sim = sR²C = sL_sim

The port therefore behaves as an inductor of value L_sim = R²C. The square on the gyration resistance gives the technique an enormous reach: a 10 kΩ gyration resistance with a 100 nF capacitor simulates 10 H, and 100 kΩ with 1 µF simulates 10,000 H, values no wound component could supply in any reasonable volume or with any usable self-resonant frequency.

Single Op-Amp Gyrator

The simplest gyrator implementation uses a single operational amplifier with two resistors and a capacitor. The circuit presents an inductive impedance at its input that can replace an inductor in many filter applications. Limitations include:

• Grounded inductance only: One terminal of the simulated inductor must connect to ground
• Finite Q: Finite open-loop gain and non-zero output impedance appear as a loss resistance in series with the simulated inductance, so the element never reaches the Q of a good air-cored coil
• Limited frequency range: The simulated reactance tracks sL only while the amplifier has loop gain to spare, and the element's Q collapses as the signal approaches the gain-bandwidth product
• Signal handling: The amplifier must swing the full voltage across the simulated inductor and supply its current, which caps the power the element can pass

Floating Inductance Simulation

Applications requiring inductors with neither terminal grounded (floating inductors) need more complex circuits. Two-op-amp and four-op-amp configurations can simulate floating inductors, enabling implementation of bridged-T, twin-T, and other topologies that require non-grounded reactive elements.

Applications

Gyrator-simulated inductors are commonly used in:

• Audio equalizers: Parametric and graphic equalizer sections
• Telephone hybrid circuits: Impedance matching networks
• Active crossover networks: Speaker system frequency dividers
• Low-frequency oscillators: LC oscillator circuits below audio frequencies

Generalized Impedance Converters

Generalized impedance converters (GICs) extend the gyrator concept to provide more flexible impedance transformation capabilities. A GIC can multiply or divide an impedance by a factor that may itself be frequency-dependent, enabling creation of frequency-dependent negative resistors (FDNRs) and other exotic elements useful in filter synthesis.

GIC Structure

The standard GIC uses two operational amplifiers and five impedances (resistors and capacitors) arranged in a specific topology. The input impedance looking into one terminal depends on the product and quotient of the five impedances:

Z_in = Z₁Z₃Z₅/(Z₂Z₄)

By selecting which elements are resistors and which are capacitors, various impedance behaviors can be created:

• Inductance simulation: Making one denominator impedance, Z₂ or Z₄, a capacitor and the rest resistors yields an input impedance proportional to s, equivalent to a grounded inductor
• FDNR, the D element: Making two numerator impedances capacitors, for example Z₁ and Z₃, creates an element whose impedance is proportional to 1/s²
• Frequency-dependent impedances: Other combinations of resistors and capacitors yield further specialized elements used in filter synthesis

FDNR Filter Design

The frequency-dependent negative resistor, usually called the D element, has impedance Z = D/s², a behavior no passive component exhibits. Its use rests on the Bruton transformation, which scales every impedance in a passive LC prototype by 1/(sk), where k is a free constant chosen for convenient element values. Scaling every branch by the same factor leaves the voltage transfer function untouched, but it converts each element into a different one:

• An inductor, sL, becomes L/k, a plain resistor
• A resistor, R, becomes R/(sk), a capacitor of value k/R
• A capacitor, 1/(sC), becomes 1/(s²kC), a D element of value kC

The direction matters, and it is easy to reverse by accident. The whole purpose is to remove the inductors, so inductors must end up as resistors; a transformed network still full of inductors means the scaling went the wrong way. The D elements the transformation creates are realized with GICs, and because the capacitors of a low-pass ladder prototype are its shunt branches, the resulting D elements are grounded, which is exactly the case the two-amplifier GIC handles directly.

One consequence needs attention in practice. The source and termination resistors of the prototype become capacitors, which leaves the amplifier inputs without a DC return path and lets offset and bias currents charge the network without limit. A large resistor placed in parallel with each transformed termination restores the bias path at the cost of a small, calculable error at the low-frequency end of the response.

What the method yields is a filter whose response matches the passive prototype exactly while using only resistors, capacitors, and op-amps. It is particularly effective for high-order Butterworth, Chebyshev, and elliptic low-pass filters, because the ladder prototype's celebrated insensitivity to component tolerance in the passband carries over intact to the active realization, which no cascade of biquads can claim.

Practical Considerations

GIC-based filters offer excellent performance but require careful attention to:

• Component matching: Accuracy of the impedance transformation depends on precise component ratios
• Op-amp selection: Both amplifiers must have adequate bandwidth and output drive capability
• DC bias: Proper biasing ensures op-amps operate in their linear range
• Noise: Multiple amplifier stages can accumulate noise, requiring low-noise designs for sensitive applications

Practical Implementation Considerations

Translating active filter designs into working circuits requires attention to several practical factors that affect real-world performance.

Op-Amp Selection

The operational amplifier choice significantly impacts filter performance:

• Gain-bandwidth product: The dominant constraint, and one that scales with Q as well as with cutoff frequency, as the next subsection sets out
• Slew rate: Must exceed the maximum rate of change the signal demands, or large-signal response departs from the small-signal design
• Input offset voltage: Multiplied by the stage gain and by every following stage, so it matters most in DC-coupled low-frequency chains
• Input bias current: Flows through the frequency-determining resistors and appears as an offset; FET-input devices are the usual answer for high-impedance sections
• Output impedance: Rises with frequency as loop gain falls, which limits the stopband depth of Sallen-Key sections and the Q of gyrators
• Noise: Voltage noise dominates at low source impedance and current noise at high, so the resistor level chosen for the filter also chooses the amplifier

Amplifier Bandwidth and Q Enhancement

Every design equation above treats the amplifier as ideal. A real amplifier's open-loop gain falls at 20 dB per decade above its dominant pole, so the gain block inside the filter carries a pole of its own at the gain-bandwidth product divided by the closed-loop gain. Carrying that extra pole through the second-order analysis produces two effects together: the pole frequency settles slightly below the designed value, and the Q rises above it. The second is the dangerous one, because it turns a flat design into a peaked one and, in the extreme, into an oscillator.

The size of the error grows with Q. For the unity-gain Sallen-Key low-pass, an approximate analysis gives a fractional Q error of about Qf₀/GBW, so the margin the amplifier needs is not a fixed multiple of the cutoff frequency but a multiple that scales with Q. A section of Q = 10 at f₀ = 100 kHz built with a 10 MHz amplifier, a hundred times the cutoff frequency and by the customary rule generous, still lands near Q = 11, which is roughly 0.8 dB of unwanted peaking. Holding that stage to 1 percent takes about 100 MHz, or 100Qf₀.

A workable rule follows. Allow 50 to 100 times the cutoff frequency for low-Q sections, and 100Q times the cutoff frequency wherever Q must be accurate. Two-integrator-loop topologies suffer the same mechanism by a different route, through the excess phase lag each integrator acquires as the signal approaches the amplifier's bandwidth, and their Q enhancement likewise scales with Q. At high Q the ideal design equations stop being a check at all; simulation against the manufacturer's amplifier macromodel is the only reliable one.

Stopband Limits and Feedthrough

Falling loop gain also puts a floor under the stopband. In a Sallen-Key low-pass the feedback capacitor returns the output to the RC network, and as the amplifier's closed-loop output impedance rises with frequency that path ceases to behave as a low-impedance source. The measured response falls at the expected 40 dB per decade, flattens, and then climbs again at 20 dB per decade, so the deep rejection the design promised never materializes. Antialiasing filters ahead of a converter are the usual casualty, since they are specified precisely on far-stopband rejection.

The remedies are straightforward once the cause is understood: a passive RC section after the stage, an amplifier with wider bandwidth and lower output impedance, or a topology whose output comes straight from an amplifier, as in the multiple-feedback and state-variable forms.

Component Selection

Passive component quality affects filter accuracy and stability:

• Resistor tolerance: 1 percent metal film suits most work; 0.1 percent is needed for precision filters and for the gain-setting divider of any Sallen-Key stage above a Q of about 3
• Capacitor type: NPO/C0G ceramic or polypropylene film for stability; avoid Z5U and Y5V ceramics in frequency-determining positions
• Temperature coefficients: Match coefficients between components to minimize drift over temperature
• Capacitor ESR: Significant at high frequencies; affects Q and response accuracy

Layout Considerations

PCB layout affects filter performance, particularly at higher frequencies:

• Ground planes: Provide low-impedance return paths and shielding
• Component placement: Keep frequency-determining components close together to minimize parasitic effects
• Power supply decoupling: Local bypass capacitors for each op-amp prevent supply-coupled oscillation
• Input/output separation: Prevent coupling between high-gain sections

Testing and Tuning

Verifying filter performance requires appropriate measurement techniques:

• Frequency response: Network analyzers or swept-frequency generators with level detectors
• Phase response: Vector network analyzers or oscilloscope phase measurement
• Transient response: Step and pulse testing reveals ringing and settling behavior
• Noise measurement: Spectrum analyzers or specialized noise measurement systems

Comparing Active Filter Topologies

Different topologies offer distinct advantages for specific applications:

The choice depends on required performance, available component count and power budget, output phase requirements, tunability needs, and integration constraints.

Conclusion

Active filter topologies provide powerful tools for implementing precise frequency-selective circuits across a broad range of applications. From the simple Sallen-Key configuration suitable for basic filtering tasks to sophisticated state-variable and GIC-based structures for demanding applications, engineers have numerous options for realizing desired frequency responses with the advantages of gain, low-frequency operation without inductors, and ease of tunability.

Successful design requires both halves of the subject. The theoretical half supplies transfer functions, pole-zero placement, and the sensitivity figures that decide which topology can carry a given Q. The practical half supplies the amplifier's finite gain-bandwidth product, its rising output impedance, component tolerance, and layout, each of which quietly alters the answer the equations gave. The single most common failure in active filter work is to design from ideal equations and then be surprised by a peaked response and a shallow stopband, both of which the amplifier, not the arithmetic, produced. Filter design software and circuit simulation close that gap cheaply, and a simulation run against the manufacturer's amplifier model costs far less than a board respin.

As electronic systems continue to demand more precise signal conditioning and filtering, active filter technology remains essential. Whether implementing anti-aliasing filters for data converters, audio equalizers for sound systems, or channel filters for communication receivers, the topologies presented here provide the foundation for meeting diverse filtering requirements with accuracy, flexibility, and cost-effectiveness.

Related Topics