Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
IEC 62754-2017: Computation of Waveform Parameter Uncertainties
This international standard provides a comprehensive methodology for computing uncertainties of waveform parameters derived from digitized measurements, including state levels, transition durations, pulse parameters, and aberration measurements. It bridges the gap between raw waveform acquisition and metrologically valid parameter extraction.
1. Waveform Parameter Uncertainty Framework
IEC 62754-2017 establishes a systematic framework for propagating measurement uncertainties through the entire waveform parameter extraction process. The standard, prepared by IEC TC 85 (Measuring equipment for electrical and electromagnetic quantities), builds upon the measurement uncertainty principles of ISO/IEC Guide 98-3 (GUM) and applies them specifically to the waveform measurement domain defined in IEC 60469 (Pulse techniques and apparatus).
The standard distinguishes three fundamental waveforms in the measurement chain: the measured waveform (raw digitized data from the acquisition system), the corrected waveform (after applying calibration corrections including gain, offset, and linearity compensation), and the reconstructed waveform (an estimate of the original input signal after corrections for known systematic effects such as bandwidth limitations and impedance mismatch).
Key waveform parameters covered include amplitude parameters (state levels, waveform amplitude, impulse amplitude, overshoot, undershoot), temporal parameters (initial instant, reference level instants, transition duration, pulse duration, pulse separation, waveform delay), and combined parameters (transition settling error, settling duration).
| Parameter Category | Specific Parameters | Typical Uncertainty Sources |
|---|---|---|
| State levels | High state (s1), Low state (s2) | Noise, quantization, offset drift, gain error |
| Waveform amplitude | Vamp = |s1 − s2| | State level uncertainties (correlated) |
| Transition duration | tr (10%–90%), tf (90%–10%) | Bandwidth limitations, jitter, interpolation |
| Pulse duration | tp (50%–50%) | Reference level instant uncertainty |
| Overshoot / Undershoot | Vos, Vus | State level + peak amplitude uncertainty |
| Pulse separation | Δt | Combined reference level instant uncertainties |
A defining feature of this standard is its emphasis on correlated uncertainties. When calculating waveform amplitude Vamp = |s1 − s2|, the state levels s1 and s97 (defining the lower boundary of the high state) may share common uncertainty contributions from the reference voltage and the digitizer gain. Ignoring these correlations can lead to overestimating the amplitude uncertainty by a factor of 2 or more.
2. State Level Determination Methods
The standard describes three distinct methods for determining state levels from digitized waveforms, each with different uncertainty characteristics:
Histogram mode method: The most common approach, which constructs amplitude histograms for each state and determines the state level by identifying the histogram peak or calculating the mean of the samples within each state region. For a bimodal histogram (typical of pulse waveforms), the two modes correspond to s1 (high state) and s2 (low state). The uncertainty depends on the bin width, sample count, and noise distribution. The standard provides detailed uncertainty calculations for this method, including the standard deviation of the mean and the effect of histogram bin quantization.
Shorth method: A robust statistical approach that estimates state levels as the mean of the shortest half of the data (the shorth). This method is less sensitive to outliers and non-stationary noise within the state region. Annex B provides the computation of L and Y statistics for shorth-based uncertainty estimation.
Parametric fit method: The waveform within a state region is fit to a constant, linear, or quadratic function using least-squares regression. This method can account for drift within the state region (e.g., capacitor charging slope in a pulse) but requires careful selection of the fit model and validation of goodness-of-fit.
| Method | Best For | Uncertainty Type | Computational Complexity |
|---|---|---|---|
| Histogram mode | Stationary states with additive noise | Type A (statistical) | Low (O(n)) |
| Shorth | States with outliers or non-Gaussian noise | Type A (robust) | Medium (O(n log n)) |
| Parametric fit | States with known drift or trend | Type A + Type B | Medium (O(n)) |
Engineering Insight: For oscilloscope-based measurements, the histogram mode method is the recommended default because it naturally handles the amplitude quantization of the digitizer (typically 8-bit for real-time oscilloscopes). However, when the number of samples per state is small (fewer than 100), the shorth method often produces more reliable uncertainty estimates. A practical rule of thumb: use the histogram method with ≥ 500 samples per state region.
3. Monte Carlo Method and Expanded Uncertainty
The standard introduces the Monte Carlo method as a powerful approach for computing waveform parameter uncertainties, particularly useful when:
- The measurement function is non-linear or has no closed-form analytical solution
- Input probability distributions are non-Gaussian (e.g., uniform distribution from digitizer quantization, or t-distribution from small sample sets)
- The parameter involves complex correlation structures between multiple waveform features (e.g., overshoot depending on state level, transition duration, and peak detection simultaneously)
Expanded uncertainty is computed using a coverage factor k that depends on the effective degrees of freedom. The standard provides a table (Table 1, based on ISO/IEC Guide 98-3) relating the coverage factor to the degrees of freedom for a 95.45 % confidence level (k=2 for large degrees of freedom). When effective degrees of freedom are below 20, the standard recommends using the t-distribution to determine the appropriate coverage factor.
The waveform epoch (the total time window of analysis) uncertainty combines the sampling interval uncertainty (timebase calibration), trigger jitter, and interpolation uncertainty. For accurate transition duration measurements, the standard emphasizes that the combined temporal uncertainty must be significantly smaller than the specified transition duration tolerances.
A common pitfall in waveform parameter uncertainty: The reference level instants (e.g., 10 %, 50 %, 90 % points on a transition) depend on both the state level determination and the interpolation method used when no sampled point falls exactly at the reference level. Linear interpolation between adjacent samples underestimates the uncertainty compared to more sophisticated methods (sinc interpolation, cubic spline). When the sampling rate is less than 5x the signal bandwidth (i.e., approaching Nyquist-limited conditions), linear interpolation can introduce errors of 10–30 % in transition duration measurements. The standard recommends using the difference between linear and sinc interpolation as a Type B uncertainty contribution.
4. FAQs
Q: How does IEC 62754 relate to ISO/IEC Guide 98-3 (GUM)?
A: IEC 62754 is a domain-specific implementation of GUM principles applied to waveform measurement. While GUM provides the general framework for uncertainty evaluation (Type A, Type B, propagation, expanded uncertainty), IEC 62754 provides the specific measurement models, correlation structures, and computational methods for waveform parameters defined in IEC 60469.
Q: Does the standard cover digitizer bandwidth and probe effects on uncertainty?
A: Yes, the waveform correction process (Clause 5) explicitly addresses corrections for bandwidth limitations (via deconvolution or frequency-domain equalization), probe loading effects, and impedance mismatch corrections. Each correction introduces its own uncertainty contribution that must be propagated through to the final parameter uncertainty.
Q: What is the recommended number of waveform acquisitions for statistical uncertainty evaluation?
A: For Type A uncertainty evaluation of state levels, the standard recommends at least 10 repeated acquisitions, with 30 or more preferred when the underlying distribution is non-Gaussian. For transition duration measurement uncertainty, a minimum of 20 acquisitions is recommended due to the higher sensitivity to jitter and noise.
Q: How should uncertainty be reported for waveform parameters?
A: The standard follows GUM conventions: report the estimated value and expanded uncertainty (e.g., Vamp = 1.023 V ± 0.014 V). The coverage factor k and effective degrees of freedom should be stated. For waveform parameters with asymmetrical uncertainty distributions (common in overshoot measurements), the upper and lower expanded uncertainties should be reported separately.
