Signal Fidelity

Signal Fidelity is the degree to which recorded evidence preserves the scientifically relevant properties of the signal, event structure, state sequence, spatial pattern, or other observable manifestation that the acquisition process is intended to retain. Fidelity concerns the correspondence between the evidence available at an appropriate reference point and the evidence ultimately represented in the record. It is important to understand that fidelity is property-specific and purpose-dependent. It is not equivalent to overall signal quality, accuracy, precision, signal-to-noise ratio, completeness, integrity, calibration, or behavioral validity.

Meaning of Signal Fidelity

Signal fidelity is the preservation of the information-bearing relationships that matter for a stated scientific use. Depending on the evidence, relevant properties can include amplitude, relative amplitude, waveform morphology, timing, duration, frequency content, phase relationships, spatial configuration, event identity, event order, state transitions, or categorical structure. A signal can preserve some properties faithfully while distorting others.

The target phenomenon is the latent behavioral or physical process generating the observable manifestation. The observable manifestation is the measurable signal or event structure that reflects that phenomenon. The reference representation is a defined standard or idealized form of the evidence against which fidelity is judged. The measuring-system input is the physical or electronic signal entering the acquisition device. The measuring-system output is the digital or analog signal produced by the measurement system after processing. The retained record is the stored or transmitted representation finally available for analysis.

Fidelity can only be assessed relative to a declared reference relationship; the latent behavioral construct itself is not automatically a directly observable reference waveform. It is incorrect to treat an inaccessible or theoretically inferred phenomenon as though its exact signal realization were known.

Fidelity as Preservation of Relevant Information

Property-specific fidelity means that a transformation preserving event timing may alter amplitude; a system preserving average level can distort morphology; a representation preserving categorical state may discard within-state waveform detail. Every fidelity claim must identify which properties are expected to remain invariant, approximately preserved, or recoverably transformed.

Invariance to known reversible transformations is allowed. A constant unit conversion, declared gain factor, coordinate transformation, or invertible encoding can change numerical representation without necessarily reducing fidelity if the scientifically relevant information and transformation are preserved. This is a representational difference, not information loss or distortion.

Irrecoverable alterations include clipping, aliasing, unrecorded intervals, irreversible averaging, source mixing without sufficient separation information, or lossy encoding that permanently remove distinctions present before recording. A reconstructed estimate is not identical to directly preserved evidence merely because it appears plausible.

Fidelity is reference-dependent. It can be evaluated against generated test signals, higher-quality reference instruments, known physical inputs, simultaneous reference observations, source models under justified assumptions, or other appropriate standards. When no defensible reference exists, one can assess consistency or plausibility but should not claim exact fidelity to an unknown source.

Amplitude, Scale, and Linearity Fidelity

Amplitude fidelity is the preservation of relevant magnitude relationships. This includes absolute amplitude fidelity (correctness of absolute values), relative amplitude fidelity (preservation of ratios or differences between amplitudes), gain error (systematic multiplicative deviation), offset error (systematic additive deviation), and scale distortion (nonlinear changes in amplitude relationships). A constant gain or offset can preserve waveform shape while impairing absolute amplitude interpretation, so shape fidelity and amplitude accuracy are not identical.

Linearity refers to a measurement relation preserving proportional structure within its valid operating domain. Nonlinear distortion can change relative amplitudes, waveform morphology, harmonic structure, thresholds, or event measurements. Linearity contributes to fidelity but is not a universal requirement for every sensing principle.

Dynamic range, saturation, and clipping limit amplitude fidelity. Values near or exceeding the valid range can compress, flatten, truncate, or distort magnitude and morphology. A clipped waveform may retain coarse event timing but permanently lose peak amplitude or shape information.

Quantization and finite amplitude resolution impose representational limits. Quantization maps ranges of values to discrete levels, obscuring distinctions smaller than the step size. Higher bit depth does not guarantee higher effective fidelity when sensor noise, front-end limitations, scaling, saturation, or device processing dominate.

Temporal and Waveform Fidelity

Temporal fidelity preserves scientifically relevant event times, intervals, durations, ordering, transition timing, and within-signal temporal relationships. A constant known latency shifts a signal in time while preserving internal temporal structure, whereas variable latency, clock error, timestamp uncertainty, temporal averaging, or jitter can distort temporal relationships.

Waveform morphology fidelity preserves transient shape, slopes, peaks, troughs, widths, rise and fall behavior, curvature, and relative timing of waveform components when these properties carry scientific information. Visual similarity alone is insufficient if meaningful clinical, physiological, behavioral, or physical details are attenuated or displaced.

Temporal support and integration effects arise from exposure intervals, integration windows, rolling averages, device-side aggregation, or finite sensor response, representing activity over time rather than instantaneous values. These effects can be scientifically appropriate but limit fidelity for phenomena varying faster than the measurement integrates.

Transient response phenomena—overshoot, undershoot, ringing, and settling—are ways an acquisition system can alter rapid signal changes. A system may have adequate nominal bandwidth yet reproduce transients poorly. Fidelity should be evaluated from response properties relevant to the target signal rather than from one specification alone.

Frequency and Phase Fidelity

Frequency-response fidelity preserves relative amplitudes of frequency components needed to represent the target signal. Finite bandwidth, roll-off, resonances, filtering, sensor dynamics, and embedded conditioning can attenuate or emphasize components unequally. Bandwidth is necessary to interpret some fidelity questions but is not a complete measure of fidelity.

Phase and delay fidelity concern frequency-dependent phase shifts or group-delay variation that can change waveform morphology and relative timing among components even when amplitude response appears acceptable. A uniform delay of the entire signal differs from frequency-dependent temporal distortion.

Sampling and aliasing impose fidelity constraints. Sampling must preserve temporal and spectral distinctions needed for the intended evidence. Insufficient sampling can cause higher-frequency behaviors to produce indistinguishable patterns (aliasing). Increasing stored sample rate after aliasing does not restore lost distinctions.

Frequency-domain fidelity requirements are signal- and purpose-dependent. Preserving a slowly varying rate may require narrower bandwidth than preserving sharp morphology, rapid transients, phase relations, or derived timing landmarks. There is no universal bandwidth requirement for behavioral, physiological, or neurophysiological acquisition.

Spatial, Geometric, and Source Fidelity

Spatial fidelity preserves scientifically relevant spatial relationships, shapes, positions, trajectories, orientations, distributions, or source patterns. Finite spatial resolution, projection, lens or geometric distortion, spatial averaging, point-spread effects, registration error, placement changes, and occlusion can alter recorded spatial evidence.

Source fidelity preserves which physical, physiological, behavioral, or digital source contributed to the recorded evidence when source identity matters. Crosstalk, spatial mixing, volume conduction, acoustic overlap, shared channels, source leakage, or incorrect attribution can preserve substantial signal energy while reducing fidelity to the intended source.

Spatial fidelity differs from acquisition geometry and behavioral spatial meaning. Geometry determines how evidence becomes observable; spatial fidelity concerns how well relevant spatial relationships survive measurement and representation; behavioral interpretation assigns scientific meaning to those relationships. Accurate coordinates alone do not establish behavioral meaning of position, orientation, proximity, or gaze.

Fidelity of Events, States, and Digital Records

Event-based fidelity involves preservation of event presence and properties. Failures include missed events, spurious events, merged or split events, incorrect event identity, altered duration, wrong ordering, duplicate events, or inaccurate timestamps. A record can preserve event count while distorting timing or preserve timing for detected events while missing others.

State-based and categorical record fidelity involves preservation of state identity, transition order, transition time, dwell duration, allowed state distinctions, and persistence of uncertainty. Collapsing several distinct states into one category may be acceptable for one purpose but destructive for another.

Native digital traces and system-generated records have fidelity influenced by event logging policies, batching, caching, asynchronous writes, deduplication, aggregation, clock semantics, software updates, platform rules, and logging failures. Digitally generated records should not be assumed lossless or temporally exact merely because no analog sensor is involved.

Encoding, Compression, and Representation Fidelity

Lossless transformations alter storage representation while preserving recoverable information. Lossy transformations intentionally discard distinctions. Compression can be lossless or lossy. Encoding changes the form of representation. Derived representations preserve selected properties while discarding others.

Representation fidelity is the preservation of properties required after digitization, formatting, serialization, codec processing, unit conversion, coordinate transformation, aggregation, or other transformations performed before the record being evaluated. A compact or standardized representation is not necessarily faithful if it removes distinctions required by the scientific use.

Visually or numerically cleaner representations can have lower fidelity. Smoothing, compression, interpolation, denoising, normalization, or averaging can make signals appear more regular while attenuating real transients, changing extrema, altering phase relationships, or hiding uncertainty. Aesthetic cleanliness should never substitute for fidelity evidence.

Fidelity Assessment and Scientific Adequacy

Reference-based fidelity assessment uses known inputs, simultaneous higher-quality references, calibrated sources, replayable test signals, physical standards, or controlled reference events when appropriate. Assessment should compare particular scientifically relevant properties rather than collapse all differences into one generic error value.

Reference-free and indirect fidelity evidence should be used cautiously. Internal consistency, expected physical constraints, redundancy, cross-channel agreement, plausibility, known sensor characteristics, or stable reference behaviors can reveal possible fidelity problems when a direct reference is unavailable. Such evidence supports a fidelity judgment but should not be presented as an exact measurement of unknown ground-truth signals.

Single metrics are incomplete. Correlation can remain high despite gain or offset error; signal-to-noise ratio can be high despite nonlinear distortion; bandwidth can be wide despite poor transient or phase response; low average error can hide rare but critical distortions. Fidelity assessment should match metrics to the properties and support relevant to the scientific use.

Fidelity can differ across time intervals, channels, participants, spatial regions, event types, operating ranges, frequencies, amplitudes, or contexts. A global fidelity statement should not erase localized distortion, saturation, dropout, timing failure, or source ambiguity that affects scientifically important evidence.

Scientific adequacy refers to whether preserved fidelity suffices for the intended behavioral or physiological claim. The same record can be adequate for coarse event counting but inadequate for precise onset timing, adequate for average level but inadequate for morphology, or adequate for one spatial scale but inadequate for another. Universal fidelity thresholds detached from the target property and intended use should be avoided.