FRF

1 FRF Definition and Interpretation

A frequency response function (FRF) describes how a dynamic system converts an input excitation into an output response as a function of excitation frequency. In its most common use, the FRF is a complex-valued quantity whose magnitude indicates response level and whose phase indicates timing relationships between input and output. FRFs are widely used to characterize resonances, modal behavior, and frequency-dependent dynamics in vibration, structural dynamics, and related signal-processing tasks.

1.1 Complex-valued transfer perspective

In the frequency domain, the system is treated through linear transfer relationships between Fourier-transformed signals. The FRF represents this relationship as a complex ratio, combining both amplitude scaling and phase shift. Interpreting the FRF therefore requires viewing it not just as a “gain,” but as a frequency-dependent complex transfer that embeds system physics and measurement interactions.

1.2 Magnitude and phase representations

Because FRFs are complex-valued, they are typically presented through separate views of magnitude and phase. Magnitude plots help reveal where the system amplifies input energy (for instance near resonances), while phase trajectories help distinguish modal contributions and identify the effective direction of energy flow through the system. Together, these views support interpretation of stability, coupling effects, and the qualitative form of the response.

1.3 FRF vs. other transfer function forms

FRFs are closely related to more general transfer functions, but the term “FRF” is often used in experimental contexts where the transfer function is identified from measured frequency-domain data. Depending on conventions, FRFs may be defined with respect to specific input/output pairs, and they may emphasize steady-state response behavior under harmonic excitation. Other forms—such as impulse-response descriptions—address time-domain behavior and can be transformed into frequency-domain representations under appropriate conditions.

2 Mathematical Foundations

FRFs are typically defined by ratios of cross- and auto-spectral quantities derived from measured data, or by direct transfer relationships for an assumed linear model. The foundational mathematics depends on whether the experiment uses one input and one output (SISO) or multiple input and output channels (MIMO).

2.1 Single-input single-output (SISO) FRF

2.1.1 Frequency-domain ratio formulation

In a SISO setting, the FRF between an input \(u(t)\) and an output \(y(t)\) is commonly expressed in the frequency domain as \[ H_{yu}(\omega) = \frac{Y(\omega)}{U(\omega)}, \] where \(Y(\omega)\) and \(U(\omega)\) are Fourier transforms of the measured signals. Under linear time-invariant assumptions, this ratio corresponds to the system’s transfer behavior at each angular frequency \(\omega\).

2.2 Multiple-input multiple-output (MIMO) FRFs

2.2.1 Matrix form and cross-coupling concepts

For multiple channels, FRFs are arranged in matrix form. Each element corresponds to the transfer from a particular input channel to a particular output channel, capturing how signals on different channels influence each other. In practice, cross-coupling can be significant: energy injected through one actuator can produce responses at multiple sensors, and correlations among inputs can affect which spectral relationships are meaningful.

2.3 Relation to impedance/admittance (when applicable)

In some engineering domains, FRFs connect naturally to impedance and admittance concepts. For mechanical systems, relationships between force and velocity or displacement can be expressed through frequency-domain ratios, effectively yielding FRFs of different physical signal pairs. These representations are useful when the excitation and sensing channels match mechanical driving variables, such as forces from impact or controlled actuators.

2.4 Estimation in the frequency domain

Because real measurements are noisy and limited in duration, FRFs are usually estimated rather than known exactly. Estimators typically rely on spectral relationships, producing complex-valued estimates that converge toward the true transfer behavior as data length and signal quality improve. Regularization and averaging strategies may be applied to stabilize estimates when spectral quantities are small or dominated by noise.

3 Measurement and Experimental Setup

Experimental FRF measurement requires careful choices about excitation, sensor placement, signal conditioning, and post-processing. The goal is to obtain spectral estimates that faithfully represent the underlying linear response while minimizing artifacts from leakage, nonstationarity, or poor signal-to-noise ratio.

3.1 Excitation methods

Excitation can be applied using impulsive sources (such as hammer impacts), swept sine signals, or broadband random inputs. The excitation spectrum should cover the frequency range of interest with sufficient energy at each frequency. Selection depends on whether the system is expected to behave linearly over the amplitude range and whether the experiment prioritizes speed, energy distribution, or modal resolution.

3.2 Sensing and response measurement

Outputs are measured using sensors such as accelerometers, velocity transducers, displacement sensors, or other transducers appropriate to the variables of interest. Signal conditioning often includes amplification, filtering, and anti-aliasing. Sensor placement affects the observed response, so mapping between physical points and measured channels is essential for interpreting modes and for comparing experiments across configurations.

3.3 Sampling, windowing, and frequency resolution

Data acquisition uses discrete sampling at rates high enough to satisfy Nyquist constraints for the highest frequency component. Windowing functions reduce spectral leakage caused by finite time records. Frequency resolution is determined by record length: longer records sharpen spectral bins, while shorter records broaden them. Practical trade-offs balance resolution with the need to average multiple records for robustness.

3.4 Coherence and data quality checks

Coherence measures how consistently the output relates to the input at each frequency, based on spectral statistics. High coherence indicates that the FRF estimate is supported by repeatable input-output coupling, while low coherence suggests noise domination, missing excitation, or system nonlinearities. Coherence thresholds are often used as diagnostics to flag unreliable frequency regions.

4 FRF Estimation Techniques

FRF estimation methods translate raw time-series measurements into complex spectral estimates. Different approaches vary in bias, variance, and sensitivity to noise, especially when the input is contaminated or when cross-spectral quantities are poorly estimated.

4.1 Spectral averaging approaches

Spectral averaging improves stability by reducing random fluctuations in estimated spectra. Common strategies include segmenting time records into blocks and averaging their spectral estimates. Averaging can be performed in ways that balance statistical reliability with the desire to preserve time stationarity.

4.2 Cross-spectral density formulation

A widely used approach forms the FRF estimate from cross-spectral density (between output and input) and auto-spectral density (of the input). For SISO, a typical estimator has the form \[ \hat{H}(\omega) = \frac{G_{yu}(\omega)}{G_{uu}(\omega)}, \] where \(G_{yu}\) is the cross-spectrum and \(G_{uu}\) is the input auto-spectrum. This formulation naturally yields complex-valued FRFs and integrates data-quality information through spectral relationships.

4.3 Least-squares and alternative estimators

Least-squares estimators are also used, particularly when framing FRF estimation as a parameter-fitting problem in the frequency domain. Alternative methods include weighted estimators that downweight frequency bins where spectral estimates are less reliable, and regularized formulations when the input spectrum is small or noisy. The choice of estimator affects how uncertainties propagate into the final FRF.

4.3.1 Practical trade-offs (bias vs. variance)

Estimators face competing objectives: increasing averaging generally reduces variance but may increase bias if the system changes across records or if the analysis assumptions are violated. Bias can also arise from leakage, unmodeled delays, nonlinear behavior, or correlations introduced by measurement hardware. Practical workflows therefore pair estimation choices with diagnostics such as coherence and residual checks.

5 Modal Analysis Using FRFs

FRFs serve as a bridge between frequency-domain measurements and modal properties like natural frequencies and mode shapes. The underlying idea is that resonant behavior manifests as characteristic features in the FRF, which can be modeled or extracted using modal models.

5.1 Linking FRFs to resonances and modes

In many linear mechanical systems, resonances correspond to frequencies where the FRF magnitude increases and phase changes rapidly. Mode shapes determine how different sensor locations respond to the same modal excitation. By analyzing FRFs across multiple input-output pairs, one can infer which modes are active and how they contribute to measured responses.

5.2 Damping estimation concepts

Damping broadens resonance peaks and influences the phase behavior around them. While exact damping extraction depends on the adopted model (such as single-degree-of-freedom approximations or multi-modal parametric fits), damping parameters are typically inferred from the curvature and width of FRF features, rather than from magnitude alone.

5.3 Identification workflow from measured FRFs

A common workflow begins with measured FRFs, checks data quality via coherence, selects frequency bands containing dominant dynamic behavior, and fits modal models to the complex FRF data. The process may involve estimating residues, selecting model order, and verifying that the fitted parameters reproduce both magnitude and phase trends. For multi-mode systems, stable identification often requires careful selection of model structure and robust averaging.

5.4 Model updating with FRF targets

Model updating adjusts parameters of a computational model so that its predicted frequency-domain behavior matches measured FRFs. The calibration objective typically minimizes differences between predicted and measured FRFs over selected frequencies. This approach helps validate whether the model captures relevant stiffness, mass distribution, and damping characteristics.

6 Visualization and Diagnostics

Visualization supports interpretation by translating complex frequency-domain data into intuitive plots. Diagnostics help detect estimation problems, identify unreliable bins, and distinguish genuine system dynamics from artifacts.

6.1 Nyquist and polar plots

Nyquist plots trace the real and imaginary parts of an FRF as frequency varies, revealing loop-like structures around resonances. Polar representations show magnitude and phase in a consolidated view, helping identify regions where phase transitions align with resonance behavior. These plots are especially informative when analyzing complex-valued structure and modal contributions.

6.2 Bode-style FRF plots

Bode-style plots typically show magnitude (often in decibels or absolute units) and phase versus frequency, separated into two subplots. This format is useful for quickly locating resonant peaks, observing attenuation trends, and comparing multiple FRFs across configurations. When several FRFs are overlaid, differences in peak locations and phase offsets become immediately visible.

6.3 Detecting nonlinearity and artifacts

If the system deviates from linear behavior, FRFs may vary with excitation amplitude or with specific trial conditions, producing inconsistent spectral shapes. Artifacts such as leakage, sensor saturation, poor synchronization, or time-varying boundary conditions can also distort FRF estimates. Diagnostic plots and consistency checks across repetitions help separate physical effects from measurement artifacts.

6.3.1 Spurious peaks and leakage considerations

Spurious peaks can arise from narrowband noise, insufficient record length, or windowing mismatch. Spectral leakage spreads energy across frequency bins, potentially creating false features near resonance regions. Mitigation includes using appropriate window functions, ensuring adequate excitation bandwidth, and using averaging and coherence checks to confirm that peaks correspond to consistent input-output coupling.

7 Applications in Engineering and Science

FRFs are used wherever dynamic behavior in the frequency domain is relevant. Common applications include mechanical vibration characterization, dynamic system identification, and validation of models used for control design or predictive maintenance.

7.1 Vibration and structural dynamics

In structural applications, FRFs help identify natural frequencies, assess how components respond to external excitation, and evaluate changes due to added mass, altered stiffness, or damage-like conditions. By tracking FRFs across time or configurations, practitioners can quantify shifts in dynamic properties.

7.2 Mechanical system characterization

Mechanical systems such as beams, frames, rotating machinery components, and mounted assemblies are often characterized using FRFs between applied forces and measured displacements or accelerations. The complex transfer behavior provides a frequency-dependent map of how vibrations propagate through the structure.

7.3 Control systems frequency-domain analysis

Control-oriented analyses can use FRFs to understand how plant dynamics behave across frequencies, particularly when controllers are designed with frequency response in mind. Comparing experimental and modeled FRFs supports validation of assumptions used in linear control design approaches.

7.4 Diagnostics and condition monitoring concepts

FRFs contribute to condition monitoring by providing measurable signatures tied to the system’s dynamic response. Changes in resonance locations, damping-related peak widths, or phase characteristics can indicate alterations in structural properties. In many workflows, FRF trends are monitored alongside coherence and other quality measures to ensure detected changes reflect real dynamics rather than measurement drift.

8 Limitations and Best Practices

While FRFs are powerful tools, their reliability depends on assumptions about linearity, time invariance, and measurement quality. Best practices focus on maintaining conditions under which FRF interpretation remains meaningful.

8.1 Assumptions behind linear time-invariant (LTI) usage

FRF interpretation is most straightforward when the system can be approximated as linear and time invariant over the frequency range and excitation amplitude used. Significant nonlinearities, time-varying behavior, or changing boundary conditions can break the premise that a single complex transfer relation explains the data across trials.

8.2 Noise sensitivity and coherence thresholds

Noise affects spectral estimates and can inflate uncertainty where input energy is weak. Coherence helps identify frequencies where the FRF estimate is trustworthy. Applying coherence criteria prevents overinterpretation of frequency bins dominated by noise or unrelated disturbances.

8.3 Repeatability and calibration

Calibration ensures that sensor scaling, phase alignment, and measurement gains are accurate. Repeatability across trials supports confidence that the FRFs represent the same underlying dynamics. Changes in mounting, sensor coupling, or cable routing can introduce differences that must be controlled or accounted for.

8.4 Interpreting FRFs for model validation

When comparing measured and predicted FRFs, it is important to align definitions of inputs and outputs, confirm consistent reference points, and select comparable frequency ranges. A good match in magnitude alone is not sufficient; phase behavior and resonance shape are often needed to demonstrate that a model captures the system’s true dynamic structure.

9 Related Concepts and Terminology

FRFs sit within a broader framework of transfer function analysis and time-frequency relationships. Understanding related terms helps clarify notation, units, and how different representations connect.

9.1 Transfer function and Bode plots

A transfer function is the general frequency-domain mapping between an input and an output under a linear model. Bode plots are a common visualization that separates magnitude and phase versus frequency. In experimental practice, FRFs are often treated as measured counterparts to theoretical transfer functions and plotted similarly using Bode-style graphics.

9.2 Impulse response vs. frequency response

The impulse response characterizes a system’s output when given a brief input in the time domain. Frequency response describes the same dynamics in the frequency domain; under suitable conditions, the two are Fourier transform pairs. This connection supports cross-domain interpretation, even though many experiments directly estimate FRFs from steady-state spectral data.

9.3 FRF-based system identification overview

System identification using FRFs aims to fit models to measured frequency-domain behavior. The model can be physical (modal or state-space) or empirical. The identification process typically uses measured complex FRFs as targets and estimates parameters that best reproduce the observed dynamics across selected frequencies.

9.4 Common notation and unit conventions

Notation varies by field, but common conventions distinguish between input/output spectra and between FRFs defined as displacement/acceleration/velocity ratios or as force-to-response mappings. Units depend on the chosen signal pair; ensuring consistent units and understanding whether acceleration, velocity, or displacement is used are essential for correct interpretation and comparison across studies.

10 Quick Reference (Common FRF Use Cases)

This section summarizes frequent practical tasks accomplished with FRFs, focusing on what questions FRF analysis can answer.

10.1 Finding resonant frequencies

Resonant frequencies often appear as peaks in FRF magnitude and as characteristic phase transitions. By scanning the FRF across the frequency band of interest, peaks and phase behavior help locate where the system most strongly responds.

10.2 Comparing designs or configurations

FRFs allow side-by-side comparison of how design changes affect dynamic behavior. Shifts in peak locations, changes in amplitude levels, and differences in phase patterns provide evidence of how stiffness, mass distribution, damping, or boundary conditions have been modified.

10.3 Verifying model predictions with measurements

Measured FRFs serve as targets for validation. Comparing predicted FRFs to experimental ones—using complex-valued agreement criteria such as both magnitude and phase similarity—helps confirm whether the model captures resonance structure and frequency-dependent behavior.