Analysis of the Sallen-Key Architecture by James Karki Created 2026-10-20 Updated 2026-11-05
Primary application report deriving the active-filter transfer functions and discussing component ratios, gain, quality factor and nonideal operation. Sections 3 and 3.4 treat the low-pass and equal-component cases.
Base-excited oscillator Created 2026-10-20 Updated 2026-10-24
A base-excited oscillator has a moving support rather than a directly applied periodic force. If its damping depends on velocity relative to the support, obeys . A model with damping against absolute velocity instead obeys after subtracting its static equilibrium. These conventions give different transfer functions.
Bounded-input bounded-output stability means that every bounded plant input produces a bounded plant output from zero initial state. For a causal continuous-time linear time-invariant system with a proper rational transfer function, BIBO stability is equivalent to every pole of the reduced transfer function having strictly negative real part. Cancel common factors first: an apparent denominator root may be absent from the reduced transfer function. For with , the only pole is , so the first-order low-pass filter is BIBO stable. [ref]
Internal asymptotic stability also requires every unforced internal state to decay. A canceled factor can hide a growing internal state. For example, consider the linear system of ordinary differential equations
Its zero-state transfer function can be written and has BIBO stability. Yet from zero initial conditions and a unit-step plant input, grows without bound while the plant output remains bounded. The reduced transfer function conceals that second state. Even a transfer function with no visible cancellations can be implemented with additional hidden states; its poles alone do not establish internal asymptotic stability of an arbitrary implementation.
An original reduced model makes the calibration explicit. Let the torsion pendulum have restoring torque , with torsion constant , and viscous damping torque , with . Allow an unwanted axial torque :
Here is the undamped natural angular frequency and the quality factor. The forced harmonic oscillator response must be included before a twist can be interpreted as a torque.
Using the complex amplitude convention and similarly for and , differentiation multiplies by . Substitution gives
This is the torque-to-twist transfer function. With and real positive , the twist amplitude is
For , the twist phase relative to is . At it is zero, although the reaction torque leads by . A phase measured at the mirror is not automatically the phase of the applied torque.
A fixed-amplitude reference torque measures directly. The Einstein-de Haas effect instead contains an extra factor when stays fixed. Consequently its twist amplitude peaks exactly at in this model, whereas the fixed-torque peak is at when . To verify the first statement, divide the squared denominator of by : the quantity to minimize is , minimized at . Both response curves tend to zero at high frequency; only the magnetomechanical curve also tends to zero at zero frequency.
Figure 1.
Calibrated forced torsion: distinguish the mechanical response from the magnetic drive
. Original calculated curves, not historical measurements or fitted data. A constant reference torque and a constant-amplitude magnetic dipole moment share the same mechanical transfer function. The lower plot uses each input's own cosine as its phase reference. Markers at show equal twist amplitudes but different phases. Details.
For a fully stipulated example, choose , , , , and . These teaching inputs give , , reaction torque amplitude at , and
An optical lever at distance gives small-displacement spot amplitude ; its full peak-to-peak width is . For a fast reversal from to , the same inputs give and initial , provided the external angular impulse is negligible.
One independent mechanical calibration adds known moment of inertia , keeping unchanged. If are the undamped oscillation periods before and after adding it, then
For in the stipulated example, and , recovering the chosen and . Real damped periods require a damping correction: a free decay has angular envelope and oscillation angular frequency . Measuring that decay determines . Calibrating the optical scale, mechanical response and change of magnetic dipole moment separately is what makes the inferred gyromagnetic ratio meaningful.
The capacitive bridging complex-error bandwidth is the largest frequency over which a receiving port preserves the source phasor within a specified complex relative gain error. The criterion includes phase shift as well as amplitude. A large DC input impedance alone cannot establish this bandwidth when input capacitance matters. The manufacturer's input models explicitly include resistive and capacitive paths.[ref]
For the scalar impedance bridging with capacitive input model, assume a positive real source electrical resistance , receiver resistance , and input capacitance in parallel. Let be the unloaded source phasor. A voltage divider gives
For tolerance , the complex relative gain error is
Its square is strictly increasing in , since its derivative is . Thus satisfying defines a continuous frequency interval starting at DC, whenever it is feasible. Rearranging gives
The DC error floor is . If , no frequency meets the tolerance; if , only DC does. The boxed positive bandwidth applies when .
For an original near-limit example, choose , and . Then and the DC error is about 0.9901 percent. A 1 percent full-complex tolerance leaves only
An amplitude-only 1 percent criterion, , instead allows about . The half-power (approximately ) signal bandwidth, relative to DC, is . These are three distinct specifications of this same ideal model.
Figure 1.
Complex-error, amplitude-error and half-power bandwidths can differ greatly
. Original 10 kilohm source, 1 megohm receiving resistance and 100 picofarad input-capacitance model. A one-percent complex error permits about 225.65 hertz; one-percent amplitude error permits about 2273.47 hertz. The half-power point relative to DC is about 160.75 kilohertz. Details.
The square-root collapse of capacitive bridging bandwidth near the DC error floor follows by factoring the numerator:
For fixed and ,
This asymptotic equivalence explains the sharp loss of usable bandwidth when the resistive divider has already consumed nearly all the error allowance. Increasing or reducing changes that floor; reducing only rescales a feasible bandwidth but cannot remove infeasibility at DC.
The criterion also controls a complete waveform. Suppose a source waveform is both a square-integrable function and a bandlimited function, with spectrum supported where . For the corresponding zero-state response , the Fourier transform of is . The Plancherel theorem, or the norm bound for a Fourier multiplier on L2, gives
Here the Fourier transform uses the frequency-in-hertz convention, so the displayed Plancherel theorem has no extra normalization factor. This bound includes waveform changes caused by phase. It applies to the zero-state response of the stated frequency response. Any additional homogeneous solution from initially stored energy contributes its own error.
A bench check therefore needs both amplitude ratio and phase shift, with probe and cable capacitance included in the port model. Magnitude-only observations determine without determining . The formula describes this chosen passive input network; additional amplifier dynamics require their own transfer function.
For a nonzero reference transfer function , the complex relative gain error is the relative error
It measures the distance from the complex ratio to one, so it responds to both amplitude error and phase error. A vector vector network analyzer measures both quantities; replacing its complex result by a magnitude discards part of the measured information.[ref]
Write , where and is the phase difference. Multiplying by the complex conjugate gives
Thus a pure five-degree phase error has , about , even though the magnitude ratio is exactly one. For small and small measured in radians, . Adding the two unsigned errors is generally not the same operation.
The metric also controls signals. Suppose two linear time-invariant systems have frequency responses and act on the same input in L2 space. On its frequency support assume almost everywhere and almost everywhere. Applying the norm bound for a Fourier multiplier on L2 to the reference output yields
This is an output-relative bound, so input frequencies suppressed by a zero of require separate treatment. Near a reference zero, a small absolute error can produce an enormous complex relative error; an absolute or explicitly fixed-scale tolerance is then more informative.
The continuous-time frequency response is the frequency response of a continuous-time linear time-invariant system. From zero initial state, a BIBO stable system driven by a sustained sinusoid approaches the steady response
Any decaying initial or switching transient is separate. The modulus is the amplitude gain; the complex argument is the shift in oscillation phase. Thus evaluating a transfer function at measures both how much of an oscillation passes through and how far it lags.
For the first-order low-pass filter, with , direct complex number arithmetic gives
The phase uses the inverse tangent branch between and . Low angular frequencies have gain close to and little lag. High angular frequencies have gain approximately and lag approaching . This is why the model is a first-order low-pass filter: it suppresses fast variation compared with slow variation. Rowell relates frequency response gain and phase to poles and then introduces Bode plots in MIT 2.004, lecture 32, sections 1 and 2.
At the corner angular frequency,
For and , , the gain is , and the lag is degrees. The angular frequency is measured in radians per second; ordinary frequency is measured in hertz (cycles per second), with . Consequently the same corner is , not .
Figure 1.
Bode plot of the first-order low-pass filter with and
. Gain and oscillation phase are plotted against angular frequency on a logarithmic horizontal axis. The vertical line marks , where the gain is and the phase is degrees. The gain panel shows ; its corner is decibels below the low-frequency gain. Details.
Dylan F. Williams 2026-11-05
Dylan F. Williams is a metrology researcher identified with NIST in a 2001 paper coauthored with Paul D. Hale and Tracy S. Clement. Their work connects measurement calibration of an optical receiver with corrections for electrical loss and electrical impedance mismatch.[ref]
Why does the electrical connection matter when the original signal is light? A photodiode produces a photocurrent, but the measured signal also passes through cables, connectors and an electrical load. A simple original model writes the measured transfer function as
For this scalar model, if the connection has amplitude gain and the measured amplitude gain is at one frequency, the inferred receiver amplitude gain is . Treating as the receiver result would incorrectly include the connection's attenuation. A real mismatched connection requires a fuller model than this product.
The distinction between amplitude and power also matters. At the same real electrical impedance, an amplitude ratio corresponds to a power ratio . Both give the same change in decibels when the appropriate definitions are used:
Using on the amplitude ratio would instead give about and describe a different quantity. These are illustrative calculations, rather than measurements attributed to Williams; the documented contribution is his coauthorship of the shared receiver-metrology work.
An electroacoustic transducer converts between an electrical signal and acoustic motion. A loudspeaker or a pair of headphones converts electrical input into sound; a microphone converts acoustic input into an electrical output. The name describes the conversion role, not one universal physical mechanism.
In an idealized moving-coil device, a current-carrying conductor in a magnetic field experiences a force. If the active conductor length is and the relevant field strength is , the perpendicular-component model gives . With and current , the force is . A diaphragm and its mechanical load turn that force into motion and sound pressure; force alone does not determine the complete acoustic output.
Conversion therefore has several linked stages: electrical drive, mechanical motion and acoustic coupling. A pressure transfer function of the complete device can depend on frequency and its surroundings. Headphone measurements on an acoustic mannequin, for example, varied when the cushions were repositioned, despite using the same nominal transducer. This demonstrates a change in the coupled system rather than requiring a change in the electrical recording itself.[ref]
The distinction is useful in calibration. A quoted electrical sensitivity, mechanical displacement or acoustic pressure is meaningful only with its input, measurement point and loading conditions specified. An electroacoustic transducer transfers energy and information through a physical interface; its useful behavior belongs to the device together with that interface.
First-order low-pass filter Created 2026-10-14 Updated 2026-10-20
A first-order low-pass filter with positive constant gain and time constant obeys the ordinary differential equation
The derivative term limits how quickly the plant output follows a changing plant input. A larger time constant means slower adjustment. This standard form, with , is developed in Derek Rowell, MIT 2.004, lecture 20, section 1.1: first-order systems; the following calculation keeps the gain explicit.
Write . The Laplace transform of the derivative is , so
Setting the initial condition gives the transfer function . Its single simple pole is , which has negative real part. The unforced response decays, and the input/output map has BIBO stability. The eventual gain for a constant plant input is .
For a unit step, represented by the Heaviside step function , the Laplace transform is . From zero initial condition,
Take and , with dimensionless plant input and plant output. Then rises toward . After one time constant, at , it reaches , or of its final value. At , it reaches , or . These percentages describe the rise from zero, rather than an instantaneous jump to the target.
With a nonzero initial condition, the same step instead gives
For and , compare with . Their plant outputs differ by despite having the same transfer function and the same plant input. The extra exponential decay term must be supplied from the initial condition; and alone cannot determine it.
Figure 1.
One transfer function, one unit step, two initial conditions
. The first-order low-pass filter has and time constant . The blue curve starts at and the orange curve at . The marked and levels refer to the blue curve's final value ; the two curves' difference is the decaying initial-state contribution. Details.
Substitution into an algebraic transfer function can produce finite amplitude and phase even when the causal system has unstable poles. It then describes a sinusoidal particular solution whose growing homogeneous response is not generically absent. A physically settled frequency response requires transient asymptotic stability; for an unstable causal transfer function, its Laplace region of convergence need not include the imaginary axis.
The group delay introduced by a capacitive amplifier load is the negative angular frequency derivative of the extra wave phase produced by that load. For the resistive output model in reactive-load phase shift in a finite-output-impedance amplifier, divide the loaded transfer function by the unloaded one:
The added group delay is therefore
This subtraction of unloaded wave phase prevents the amplifier's intrinsic group delay from being attributed to the capacitor. The terminal model uses a Thévenin equivalent; the group delay definition differentiates an unwrapped wave phase, as in the SciPy signal-processing documentation. Its digital-filter function reports samples, whereas the analog derivative above uses radians per second and gives seconds.[ref][ref]
For an original example, choose and , hence . At , corresponding to , the added group delay is . The same operating point has : a single sinusoid's phase displacement and the delay of a narrow frequency packet need not agree.
A practical test applies a weakly modulated carrier, varies its central frequency, and measures the output modulation displacement while comparing loaded and unloaded operation. The group delay describes that narrow-band envelope only when the frequency response varies slowly across its bandwidth. The corresponding causal impulse response begins at zero time, for , and vanishes for ; it is not a rigidly translated impulse. Thus group delay versus pulse-front arrival for an RC load matters when a measured pulse is broad enough to change shape.
A headphone transfer function describes how an input electrical signal becomes acoustic pressure at a specified point near or within the listener's ear. For a linear small-signal description,
where is complex sound pressure amplitude and the input electric potential difference. This transfer function includes both magnitude and phase. With these variables its units are pascals per volt, rather than a dimensionless gain.[ref]
For example, at one frequency let and apply a sinusoidal input. The pressure amplitude is under the model. A compensating filter for a desired pressure would ideally supply , but division by a small or uncertain response can demand a large, fragile correction.
The coupling to the ear is part of the system. Abhijit Kulkarni and H. Steven Colburn repeatedly positioned headphone cushions on a KEMAR acoustic mannequin and measured changes in response. Their measurements showed that repositioning could complicate compensation using a fixed inverse filter. A transfer function measured for one fitting need not remain identical after the device is moved.[ref]
This differs from a head-related transfer function, which characterizes transformation from an external sound source to an ear location. Reproducing a desired spatial sound over headphones can require accounting for both transformations, rather than confusing the headphone's coloration with the directional effect being simulated.
High-pass filter Created 2026-10-20 Updated 2026-10-20
A high-pass filter attenuates low frequencies relative to a specified higher-frequency pass-band. A simple first-order RC high-pass has the displayed transfer function, with a zero at zero frequency and a pole at . Combining a high-pass with a low-pass makes a band-pass, but its attainable selectivity depends on pole structure rather than gain alone.
Integrating amplifier 2026-11-05
An integrating amplifier produces an output proportional to the time integral of its input over an intended operating range. A common circuit uses an operational amplifier, an input resistor and a feedback capacitor . With the non-inverting input held at zero, ideal negative feedback maintains a virtual ground at the inverting input. Negligible amplifier input current then gives
The initial output records the capacitor's initial charge. Texas Instruments' circuit note derives this inverting relation and explains why real integration is limited by the circuit time constant and amplifier bandwidth.[ref]
For an original example, choose and , so . A constant0.20volt input lasting2milliseconds changes the output by . Repeating ten such pulses without intervening input or integrator reset changes it by in the ideal model. The output retains information about accumulated pulse area, rather than only the present input value.
A small offset also accumulates. With the same components, a residual input-equivalent1millivolt gives a slope of . Starting from zero, an idealized negative output limit at5volts would be reached in5seconds. The finite supply range causes amplifier clipping; an ideal integral extending indefinitely is not physically available.
Putting a resistor in parallel with the feedback capacitor supplies a DC path. Under ideal amplifier assumptions its zero-state transfer function is
At frequencies where , this approaches ; at low frequency it instead approaches the finite gain . The resistor prevents unrestricted DC accumulation while changing the lowest usable integration frequency.[ref]
An integrating amplifier can therefore turn a short detector pulse into a measurable accumulated output, but its initial condition, reset, baseline, frequency range and output limits belong to the measurement model. A charge amplifier is a related current-input arrangement in which collected charge is converted through feedback capacitance.
Inverting two-pole band-pass filter Created 2026-10-20 Updated 2026-11-05
An ideal inverting operational amplifier with series input and parallel feedback has the displayed transfer function, where and . Its peak is at with voltage gain . A wide input-high-pass/feedback-low-pass separation produces a broad plateau, not a narrow resonance. Validity requires stable negative feedback, sufficient loop gain and linear output operation.
Linear-light RGB 2026-11-05
In Linear-light RGB, each component is proportional to the amount of its corresponding primary light. Adding or averaging components therefore corresponds to adding or averaging those light contributions, provided the same primaries and reference conditions are used. Nonlinear display encodings such as sRGB must first be decoded for this arithmetic.[ref]
For a nonnegative sRGB channel in , the linear value is when , and otherwise. Thus encoded mid-gray represents approximately0.214 linear light. Averaging encoded black and white gives0.5, while averaging their light gives0.5 linear, encoded as approximately0.735. This explains dark fringes produced by averaging nonlinear channel values directly. The primaries, white point and transfer rule together determine a color space; linearity alone does not specify which colors an RGB triple represents.
Mechanical driving-point impedance describes the relation between a driving force and motion at the same mechanical port. Here the convention is
where are Laplace transforms with zero initial conditions, and is the signed relative velocity of the two terminals. Choose signs so is entering power. The units of this mechanical impedance are . Its inverse is the mechanical driving-point mobility. Conventions must be checked explicitly: in The Inerter: A Retrospective by Malcolm C. Smith, the force–current analogy calls the mechanical impedance and the admittance.[ref]
For a spring, dashpot and inerter sharing the same terminals, relative velocities agree and forces add. Hence
At a sinusoidal angular frequency , using complex amplitudes proportional to ,
The real part determines mean absorbed power, while the imaginary part records the out-of-phase reactive response. With peak velocity amplitude , the mean entering power is .
For an original example, choose , and . At , the inertial and elastic terms cancel, giving . A relative velocity amplitude requires force amplitude , and mean absorbed power is . The individual spring and inerter still store and exchange energy.
A mechanical impedance measurement drives the chosen port and records synchronized force and motion there. A load cell measures force; displacement or an accelerometer can supply motion. For an acceleration complex amplitude , , so the phase as well as magnitude matters. Fixture inertia, sensor measurement calibration and the second terminal's motion belong in the measurement model. Using a response at a different point instead defines a transfer relation, not this driving-point relation. Initial stored motion also prevents interpreting an arbitrary transient quotient as a zero-initial-state transfer function.
Multiplicative uncertainty represents a model error relative to a nominal controlled plant. For a scalar transfer function, one useful form is
where is nominal, shapes the permitted relative error across frequency, and belongs to a specified bounded uncertainty class. Unlike an additive error, the correction is multiplied by the nominal plant. MathWorks's ucover reference distinguishes input multiplicative, output multiplicative and additive forms.[ref]
For an original static-error example, take , and a real constant . The allowed family is
The uncertainty changes gain while preserving the pole in this family. A dynamic can instead introduce frequency-dependent magnitude and phase changes; its admissible dynamics must be stated separately.
An omitted actuator lag illustrates a dynamic relative error. If , then
At , its magnitude is about0.0995; at , it is about0.995. A constant10percent relative-error bound would cover the first frequency but fail badly at the second. A shaping function should reflect that discrepancy rather than extrapolate a low-frequency fit.
For a matrix-valued plant, multiplication order matters. Input uncertainty is placed on the right of , while output uncertainty is placed on the left; the identity matrices and uncertainty dimensions must match the corresponding signal spaces.[ref] These forms generally cannot be interchanged when matrix products do not commute.
The scalar ratio also needs care near zeros of . A tiny absolute discrepancy can be a large relative discrepancy there. Choosing multiplicative uncertainty is therefore a modeling decision about which relative errors and frequency ranges a robust control argument is intended to cover.
Nyquist stability criterion Created 2026-09-24 Updated 2026-11-05
The Nyquist stability criterion applies the argument principle to a specified negative-feedback return ratio . Assume a proper rational loop with a well-posed interconnection, no poles or closed-loop characteristic zeros on the imaginary axis, and no singularity at the closing part of the contour. Let count poles of in the open right half-plane and let count zeros of there, both with multiplicity. For a positively oriented contour enclosing that half-plane, let be the signed counterclockwise winding of its image under around . Then
A clockwise contour convention reverses the sign of , so the orientation must accompany the formula. Closed-loop input-output stability requires and exclusion of imaginary-axis zeros; simply passing through or encircling is not a conclusion without the pole count and convention.
Poles on the imaginary axis require an indented contour and a corresponding boundary analysis. Also, cancellations in a reduced transfer function can conceal unstable internal states, so a Nyquist conclusion about alone does not prove internal stability of an arbitrary nonminimal realization. The bounded-input bounded-output stability example exhibits that distinction.