Waveform Morphology Descriptors
Waveform Morphology Descriptors are analytical tools used to characterize and classify signal shapes by extracting key features from their temporal patterns.
Waveform Morphology Descriptors provide explicit characterizations of the local or cycle-level shape of a declared time-indexed waveform, pulse, transient, event profile, envelope, or other one-dimensional signal form over an identified support interval. Morphology concerns how amplitude organizes with respect to the waveform’s independent coordinate, including landmarks, baseline-relative amplitudes, prominence, widths, plateaus, rise/fall form, slopes, curvature-like shape, area, symmetry, internal component relations, cycle form, and template similarity. Importantly, waveform morphology is not spatial trajectory geometry, not event count or recurrence, not a value-distribution summary, not a spectrum, and not a behavioral label or physiological mechanism by definition.
Meaning and Morphological Object Identity
A Waveform Morphology Descriptor is a reproducible characterization of shape or internal profile structure computed from an identified waveform object under declared support, coordinate, preprocessing, polarity, baseline, landmark, normalization, and descriptor conventions. The waveform object may be a raw or prepared scalar trace, vector magnitude, envelope, pulse, beat, event-centered profile, speed or force profile, physiological waveform, or local descriptor contour, provided the resulting object has explicit one-dimensional amplitude-versus-coordinate semantics.
Morphological object identity requires reproducibility and explicit definition of the source signal identity/version, entity or channel, support interval, coordinate or time basis, units, preprocessing state, baseline rule, polarity convention, event or cycle identity, landmark set, and validity state. Two numerically identical waveform segments can remain scientifically distinct when they arise from different signals, channels, units, preprocessing pipelines, reference configurations, event definitions, or coordinate conventions.
Waveform morphology must be distinguished from waveform timing and spatial geometry. Morphology descriptors can use temporal distances between landmarks when those describe the internal shape of a single pulse or event profile—for example, rise time or width—without becoming general event-occurrence descriptors. Similarly, a velocity or speed profile is a waveform morphology object as a one-dimensional function of time, whereas the spatial trajectory from which it was derived possesses geometric properties that are scientifically different.
| Descriptor Category | Domain | Primary Object | Property Characterized | Critical Non-Equivalence |
|---|---|---|---|---|
| Waveform Morphology | Time-indexed signals | Local waveform profile | Internal waveform shape, amplitude organization | Not event count, not spectrum, not spatial geometry |
| Temporal/Event Descriptor | Event timing | Event timestamps | Occurrence timing, intervals | Not waveform shape, not amplitude profile |
| Statistical Distribution Descriptor | Statistical analysis | Amplitude value set | Amplitude distribution properties (mean, skew) | Not time-indexed shape, not temporal sequence |
| Spectral Descriptor | Frequency domain | Signal spectrum | Frequency content, power at frequencies | Not time-domain waveform shape |
| Spatial/Trajectory Geometry | Spatial coordinates | Trajectory path | Spatial geometry, curvature, path shape | Not time-amplitude waveform shape |
| Annotation/Reference | Metadata | External labels, events | Behavioral or physiological labels, references | Not waveform shape or internal profile |
| Model Output | Computational models | Predicted labels or features | Model-based inference or classification | Not raw waveform morphology |
Preprocessing dependence is fundamental for morphology. Filtering, smoothing, differentiation, integration, rectification, envelope extraction, detrending, baseline correction, normalization, resampling, interpolation, clipping repair, and device-side conditioning can alter peaks, troughs, widths, slopes, shoulders, notches, plateaus, and area. Therefore, a morphology descriptor characterizes the declared prepared waveform, not an abstract original signal whose shape may no longer be recoverable.
Landmarks, Baselines, Polarity, and Shape References
Waveform landmarks are explicitly defined shape anchors such as peaks, troughs, shoulders, plateaus, notches, inflection-like points, baseline intersections, threshold crossings, pulse feet, internal component peaks, and cycle boundaries. Landmarks can be directly detected by algorithms, imported from existing event sources, manually defined, or derived by other procedures. Morphology descriptors inherit these landmark definitions and their uncertainty.
Local extrema differ from scientifically admissible morphology landmarks. Noise can create many local maxima and minima; plateaus can contain multiple equal-valued candidates; broad peaks can have uncertain centers; shoulders may not fulfill ordinary extremum rules; and multiple nearby extrema can represent a single composite structure or multiple events. When landmark identity is not self-evident, threshold, prominence, scale, neighborhood, persistence, template, or other declared criteria must be specified.
Baseline and reference-level semantics vary. A descriptor can use signal zero, a global baseline, local pre-event baseline, post-event baseline, linearly varying baseline, fitted baseline, trough level, prominence contour, fraction of baseline-to-peak amplitude, or another reference. The baseline forms a part of the descriptor identity: raw peak value, peak above baseline, and peak-to-trough amplitude are different quantities even if numerically similar in one instance.
Waveform polarity affects analysis. Positive and negative deflections can be treated directly, sign-inverted under a declared convention, or analyzed through polarity-specific rules. Absolute value should not be used merely to simplify detection when sign carries scientific information. Peak/trough terminology must remain consistent with the analyzed polarity and waveform construction.
| Landmark | Operational Definition Needed | Morphology Use | Common Ambiguity |
|---|---|---|---|
| Peak | Yes | Primary amplitude maximum | Broad peaks with uncertain center |
| Trough | Yes | Primary amplitude minimum | Plateaus or flat troughs |
| Shoulder | Yes | Secondary slope change or plateau edge | May lack clear extremum; often subjective |
| Plateau | Yes | Flat-top or constant-amplitude region | Sensor saturation or clipping can mimic plateaus |
| Notch | Yes | Local dip or indentation | Distinguishing real notch vs. noise artifact |
| Inflection-Like Point | Yes | Curvature sign change point | Noise can produce spurious inflections |
| Baseline Intersection | Yes | Crossing of baseline reference level | Baseline definition ambiguity |
| Cycle/Event Boundary | Yes | Delineation of one waveform cycle or event | Boundary uncertainty, overlapping events |
Detector dependence and circularity must be considered. Morphology values can change when peak or event identity changes under different smoothing, threshold, prominence, minimum-distance, width, plateau, or template criteria. Detector or landmark parameters should not be tuned solely to obtain morphology values that correlate best with the same downstream labels or outcomes later used to claim descriptor quality.
Amplitude, Prominence, Width, and Plateau Morphology
The baseline-relative peak amplitude is defined as:
Here, x(t) is the declared waveform, t_p is the identified peak coordinate, b(t) is the declared baseline or reference-level function, and A_p is the peak amplitude relative to that baseline at the peak. Note that raw peak value x(t_p), baseline-relative amplitude A_p, peak-to-trough excursion, and normalized amplitude are distinct descriptor definitions.
Peak and trough amplitude descriptors include absolute extremum value, baseline-relative height or depth, peak-to-trough excursion, local amplitude range, and normalized amplitude. Each requires specifying the reference level, polarity convention, units, normalization denominator, and landmark identity. Multiplicative gain preserves normalized shape but alters absolute morphology values.
Peak prominence is the vertical extent by which a declared peak stands above a surrounding contour level determined by a declared prominence algorithm or neighborhood rule. Prominence differs from raw peak height, threshold relative to adjacent samples, baseline-relative amplitude, and peak-to-trough amplitude. Restricting the neighborhood used to find bases can intentionally produce a more local prominence definition.
Generic waveform width at a declared evaluation level is defined as:
Here, p is the identified waveform peak or pulse, h_p is the declared evaluation level used to determine the width, t_L and t_R are the left and right valid crossing or interpolated intersection coordinates satisfying the width rule at h_p with t_R > t_L, and W_p is the resulting width in the coordinate’s units. Full width at half maximum, width at half prominence, width at another relative height, baseline width, and sample-count width are not interchangeable unless their evaluation-level definitions coincide.
Width conventions vary in depth: fraction of absolute peak, fraction of baseline-to-peak amplitude, fraction of prominence, fixed physical threshold, and detector-defined contour level can create different widths. The interpolation method, crossing policy, plateau handling, multiple-crossing rule, units, and conversion from samples or frames to physical time must be declared. Sampling density limits width precision even when interpolation returns sub-sample coordinates.
Plateau and flat-top descriptors include plateau level, plateau width or duration, edge coordinates, plateau fraction, within-plateau variation, and left/right edge asymmetry. Genuine flat morphology must be distinguished from sensor saturation, clipping, quantization, digital ceiling/floor effects, or excessive smoothing. A flat top should not be given physiological or behavioral meaning before acquisition limitations are considered.
| Descriptor | Reference Needed | Units | Characteristic Misinterpretation |
|---|---|---|---|
| Raw Peak Value | No | Waveform units | Confused with baseline-relative amplitude |
| Baseline-Relative Amplitude | Yes (baseline) | Waveform units | Mistaken for raw peak or prominence |
| Peak-to-Trough Excursion | Yes (peak & trough) | Waveform units | Treated as separate amplitude rather than difference |
| Prominence | Yes (neighborhood) | Waveform units | Assumed equal to raw peak height |
| Width at Half Maximum | Yes (evaluation level) | Coordinate units | Confused with width at half prominence |
| Width at Relative Prominence | Yes (prominence level) | Coordinate units | Treated as general width without reference level |
| Baseline Width | Yes (baseline) | Coordinate units | Overlooking baseline definition |
| Plateau Width | Yes (plateau edges) | Coordinate units | Interpreted as genuine signal shape without clipping check |
Rise, Fall, Slope, Curvature, Area, and Symmetry
Rise and fall morphology describe waveform transitions between declared landmarks or amplitude levels. Relevant descriptors include rise time, fall time, time to peak within the waveform object, recovery time, rise/fall ratio, and level-specific transition times such as 10–90% or 20–80% when scientifically justified. Percentage levels are conventions chosen with specific purposes and should not be presented as universal standards for every behavioral signal.
A declared rise-slope descriptor is defined as:
Here, x(t) is the declared waveform, t_a and t_b are declared rising-side landmark or level-crossing coordinates with t_b > t_a, and S_r is the average rise slope over that interval. Local derivative, maximum slope, fitted slope, and average level-to-level slope are distinct descriptors. Slope units follow waveform units divided by coordinate units.
Fall slope, maximum and minimum local slope, tangent-like edge slope, and slope profiles serve as morphology descriptors when their estimation method, smoothing, differentiation, support, and units are explicit. Derivatives amplify high-frequency noise and can shift or create apparent shape landmarks after filtering. Do not infer physical velocity, flow, or force rate merely because a waveform slope has quantity-per-time units.
Curvature-like and inflection descriptors for a scalar waveform include second-derivative sign or magnitude, curvature of the graph under declared coordinate scaling, number and location of inflection-like points, convex/concave phase duration, and local sharpness. Graph curvature differs from spatial trajectory curvature, and rescaling time or amplitude can change geometric graph curvature unless normalization is built into the definition.
Baseline-relative signed waveform area is defined as:
Here, x(t) is the declared waveform, b(t) is the declared baseline function, t_a and t_b are ordered waveform-object boundaries with t_b > t_a, and A is the signed baseline-relative area over that support. Absolute area, positive-only area, normalized area, and phase-specific areas are different definitions; area has waveform-unit × time-unit dimensions and is not physical energy unless justified by the signal and physical model.
Waveform symmetry and asymmetry around a declared center such as the peak, midpoint, landmark, or optimized alignment center can be defined by comparing rise/fall durations, left/right areas, mirrored profiles, corresponding landmark offsets, or normalized residuals. Waveform asymmetry differs from statistical distribution skewness, which characterizes amplitude-value distribution and can remain unchanged after temporal reordering.
| Descriptor | Required Reference or Landmark | Units/Scale | Primary Sensitivity |
|---|---|---|---|
| Rise Time | Two rising-side landmarks | Coordinate units | Rise duration |
| Fall Time | Two falling-side landmarks | Coordinate units | Fall duration |
| Rise/Fall Ratio | Rise and fall times | Dimensionless (ratio) | Relative rise vs. fall speed |
| Average Slope | Two landmarks or levels | Waveform units / coordinate units | Mean slope over interval |
| Maximum Local Slope | Local slope extrema | Waveform units / coordinate units | Sharpest instantaneous slope |
| Inflection/Curvature-Like Descriptor | Inflection points or intervals | Dimensionless or waveform units / coordinate units² | Shape curvature and phase changes |
| Signed Area | Integration boundaries | Waveform units × coordinate units | Net area above/below baseline |
| Absolute Area | Integration boundaries | Waveform units × coordinate units | Total magnitude of waveform area |
| Mirror-Profile Asymmetry | Center reference | Dimensionless or waveform units | Left-right shape asymmetry |
Multi-Landmark, Pulse, and Cycle Morphology
Multi-landmark waveform morphology applies to profiles containing several ordered peaks, troughs, shoulders, notches, or secondary components. Relevant descriptors include amplitude ratios, landmark-to-landmark intervals internal to one waveform object, relative component areas, prominence relations, ordering, and normalized component positions. Component identity must be preserved; secondary peaks should not be named by physiological or behavioral role unless justified by signal modality and evidential model.
Pulse and beat morphology, after pulse or beat boundaries are established, can be characterized by foot, rise, primary peak, decline, notch, secondary component, end, amplitude, width, area, and internal ratios when those landmarks are observable. Morphology of one pulse must be distinguished from beat-to-beat interval, count, recurrence, or rate descriptors.
Cycle morphology for repeated waveforms, after cycle identity and boundaries are fixed, includes cycle amplitude, normalized cycle shape, internal landmark phase, peak/trough arrangement, intra-cycle asymmetry, area fractions, and deviation from a cycle template. Cycle-to-cycle variability can be summarized separately, but individual cycle morphology should remain identifiable when heterogeneity matters.
Time or phase normalization of cycles involves resampling each cycle to a common normalized coordinate to facilitate shape comparison and template construction. This removes or transforms absolute duration, local timing, cadence, and potentially genuine phase variation. The original cycle duration and warping/normalization rule must be preserved so normalized shape is not mistaken for native timing.
Morphology ratios such as rise/fall duration, amplitude/width, area/amplitude, prominence/amplitude, secondary/primary amplitude, or left/right area are valid only when numerator and denominator have declared semantics and denominator validity. Ratios can reduce selected scale effects but can become unstable near zero and discard scientifically meaningful absolute magnitude.
| Morphological Object | Required Landmark Structure | Representative Descriptor | Main Identity Risk |
|---|---|---|---|
| Single-Peak Profile | One primary peak | Peak amplitude, width | Peak misidentification due to noise or smoothing |
| Multi-Peak Pulse | Multiple ordered peaks and troughs | Amplitude ratios, internal intervals | Component merging or splitting |
| Notched Pulse | Primary peak with notch and shoulder | Notch depth, prominence relation | Mislabeling notch vs. noise |
| Repeated Cycle | Cycle boundaries, multiple landmarks | Cycle amplitude, intra-cycle asymmetry | Cycle boundary uncertainty |
| Normalized Cycle | Normalized coordinate, landmark phases | Normalized shape vector, template distance | Misinterpretation of normalized timing |
| Composite Event Profile | Multiple events or sub-events | Aggregate amplitude ratios, durations | Overlapping or conflated events |
Template, Structured, and Normalized Shape Descriptors
Template- and prototype-based morphology descriptors quantify distances, similarities, residuals, or landmark deviations between a waveform object and a declared reference shape after explicit alignment and normalization procedures. These include pointwise error, correlation-like shape similarity, amplitude-normalized distance, landmark deviation, derivative-profile similarity, and warped distance. Template identity and construction provenance form part of the descriptor definition.
Alignment choices for waveform-shape comparison include translation in time, baseline alignment, amplitude scaling, polarity correction, phase normalization, landmark alignment, and constrained temporal warping. Each alignment removes selected differences before shape comparison. Flexible warping can manufacture similarity by erasing genuine duration or phase differences; therefore, permitted deformation and scientific invariances must be explicit.
Structured morphology descriptors preserve more shape information than a single scalar, e.g., normalized sample vectors, landmark-coordinate vectors, piecewise-linear or polynomial coefficients, spline coefficients, basis coefficients, derivative profiles, and multi-level width profiles. The output remains a descriptor when the characterized shape property and construction are explicit. An arbitrary waveform vector or learned embedding is not automatically an interpretable morphology descriptor.
Invariance and equivariance choices affect morphology. Baseline translation, amplitude scaling, polarity reversal, time shift, time scaling, gain, and monotonic amplitude transformation can alter morphology descriptors differently. A normalized descriptor can intentionally remove one nuisance transformation while retaining another property, but greater invariance is not universally preferable when the removed magnitude or timing is behaviorally meaningful.
Morphology Uncertainty, Sensitivity, and Fidelity
Waveform-morphology uncertainty arises from noise, sampling interval, quantization, baseline uncertainty, detector or landmark ambiguity, interpolation, smoothing, filtering, clipping, saturation, incomplete support, low dynamic range, multi-component overlap, cycle-boundary uncertainty, template uncertainty, and alignment choices. Uncertain morphology is distinct from an undefined descriptor, absent landmark, censored waveform boundary, clipped peak, and structurally inapplicable descriptor.
Morphology fidelity refers to preservation of scientifically relevant waveform shape through acquisition and preprocessing. A constant gain or offset can preserve normalized shape while altering absolute amplitudes; nonlinear response, clipping, phase distortion, smoothing, limited bandwidth, integration, interpolation, or resampling can alter peaks, slopes, widths, shoulders, and notches. Similar-looking plots do not guarantee morphology preservation at descriptor-relevant resolution.
Sensitivity analysis for morphology descriptors should compare scientifically plausible baselines, landmark detectors, prominence neighborhoods, width levels, interpolation methods, smoothing strengths, derivative estimators, support boundaries, polarity rules, clipping exclusions, time/amplitude normalizations, template choices, and alignment constraints. Material changes in descriptor values or scientific conclusions under defensible alternatives should remain visible rather than being hidden by selecting the most favorable configuration.
Verification, Worked Interpretation, and Provenance
Consider a pulse-like waveform with a nonzero baseline, a primary peak, a shoulder or secondary feature, finite rise and fall phases, and one version with clipping or excessive smoothing.
- Raw peak value differs from baseline-relative amplitude and prominence because the baseline shifts the reference, and prominence considers surrounding contour levels.
- Width measured under two declared reference-level conventions (e.g., half maximum and half prominence) yields different values.
- Rise time and fall time descriptors quantify the finite transitions between landmarks.
- Signed and absolute area quantify net and total energy-like measures relative to baseline.
- Asymmetry descriptors compare rise and fall durations or left/right area.
- A structured descriptor such as a template similarity score or normalized shape vector captures overall shape beyond scalar summaries.
- Clipping or smoothing alters peak height, prominence, width, and slope, demonstrating preprocessing dependence.
- Changes in baseline definition or landmark ambiguity alter descriptor values and interpretation.
Verification of formulas on synthetic shapes with known landmarks confirms correctness. Morphology-definition/version, source signal and preprocessing version, support/object identity, coordinate/time units, polarity, baseline, landmark/event source and detector parameters, prominence and width conventions, interpolation, rise/fall levels, area support and integration rule, cycle normalization/alignment, template identity, clipping/saturation status, implementation/version, uncertainty, and sensitivity findings must be preserved with every descriptor extraction.
Numerical reproducibility verifies descriptor computation but does not verify behavioral or physiological interpretation of the shape.