Signal Analysis

Signal analysis forms the foundation of modern electronic measurement, enabling engineers to extract meaningful characteristics from captured waveforms. By applying mathematical transforms, statistical methods, and automated algorithms to digitized signals, analysis systems reveal information about signal behavior that would be difficult or impossible to observe directly. From determining frequency content through spectral analysis to understanding signal variability through statistical measures, signal analysis transforms raw data into actionable engineering insights.

The power of digital signal analysis lies in its ability to process signals in ways that were impractical with analog techniques. A digital oscilloscope can simultaneously display a time-domain waveform, its frequency spectrum, and statistical parameters derived from thousands of acquisitions. Modern instruments perform these analyses in real time, allowing engineers to observe how signal characteristics change dynamically with operating conditions. This comprehensive view accelerates debugging, validates designs against specifications, and provides the quantitative data needed for rigorous engineering decisions.

This article explores the key techniques used in digital signal analysis: Fast Fourier Transform (FFT) processing for frequency-domain analysis, time-frequency methods that reveal how spectral content evolves over time, statistical analysis for characterizing signal variability, histogram generation for amplitude distribution visualization, parameter extraction for quantitative measurement, and automated measurement systems that apply these techniques consistently and efficiently.

FFT Processing

The Fast Fourier Transform has revolutionized frequency-domain analysis by making spectral computation practical for real-time instruments. The FFT efficiently decomposes time-domain signals into their constituent frequency components, revealing harmonic content, noise characteristics, and spectral signatures that are invisible in waveform displays.

FFT Fundamentals

The FFT computes the Discrete Fourier Transform exactly, but with far less arithmetic. Direct evaluation of the DFT costs O(N²) operations, because each of the N output bins is a sum over all N samples. A radix-2 FFT splits the work into log₂ N stages of N/2 butterflies each, reducing the cost to O(N log₂ N). For a 1,024-point record, direct evaluation needs 1,048,576 complex multiplications and a radix-2 FFT at most 5,120, a reduction of about 200 times that makes real-time spectral display practical.

Power-of-Two Length: Classic radix-2 algorithms require records of 2ⁿ samples; mixed-radix algorithms handle other lengths that factor into small primes
Complex Output: FFT produces complex values encoding both magnitude and phase at each frequency bin
Frequency Range: For a real input the magnitude spectrum is symmetric about half the sample rate, so instruments display only the bins from DC to the Nyquist frequency, f_s/2

The bin spacing follows from the record length. An N-point record sampled at f_s spans N/f_s seconds, and its bins fall at multiples of the reciprocal of that span:

Δf = f_s/N

A 1,024-point record sampled at 1 MS/s therefore has bins about 977 Hz apart. Bin spacing is often called frequency resolution, but it is not the same as the ability to separate two tones: the window's main lobe spans several bins, so two components of similar amplitude must usually be a few bins apart before they appear as distinct peaks.

Windowing Functions

Window functions taper the ends of a finite record toward zero to reduce spectral leakage. Each trades the width of its main lobe, which limits how closely two tones can be separated, against the height of its sidelobes, which limits how small a component can be seen beside a large one:

Rectangular Window: No modification, narrowest main lobe but highest sidelobes (−13 dB)
Hanning Window: The raised-cosine window, called the Hann window in the signal-processing literature; a good general-purpose choice, with a −32 dB peak sidelobe and sidelobes that fall at 18 dB per octave
Hamming Window: Lower peak sidelobe (about −43 dB), but its sidelobes fall at only 6 dB per octave, so distant leakage is higher than with Hanning
Blackman Window: Lower sidelobes (−58 dB) at cost of wider main lobe
Flat-Top Window: Nearly flat main-lobe peak, so a tone reads within a small fraction of a decibel of its true amplitude wherever it falls between bins; its very wide main lobe suits amplitude calibration rather than resolving close tones
Kaiser Window: Adjustable parameter trades main lobe width against sidelobe level

Spectral Leakage

Understanding leakage is crucial for accurate spectral analysis:

Cause: Discontinuity at record boundaries when signal period does not match record length
Effect: Energy spreads from true frequency into adjacent bins
Scalloping Loss: A tone midway between two bins reads low, by up to 3.92 dB with the rectangular window and 1.42 dB with the Hanning window
Coherent Sampling: When the record holds an exact integer number of signal cycles, the fundamental and its harmonics fall on bin centers and do not leak, even with the rectangular window
Window Selection: Choose window based on whether frequency resolution or amplitude accuracy is more important

FFT Averaging

Averaging multiple FFT records improves measurement reliability:

RMS Averaging: Averages power (magnitude squared) across records, reducing the variance of the noise floor so the trace appears smoother while stable signal peaks remain prominent; it does not lower the average noise floor, and a hundred averages reduce the fluctuation of the noise trace about tenfold
Vector Averaging: Averages complex values, requires phase-coherent triggering, and rejects non-coherent noise, lowering the displayed noise floor by 10 dB for each tenfold increase in the number of averages
Peak Hold: Retains maximum value at each bin, useful for capturing intermittent signals
Exponential Averaging: Weighted average emphasizing recent acquisitions for tracking changing spectra

FFT Display Formats

Various display formats serve different analysis needs:

Linear Magnitude: Direct voltage or power representation
Logarithmic (dB): Compresses dynamic range for viewing small signals near large ones
dBm: Power referenced to 1 milliwatt, common in RF applications
dBV: Voltage referenced to 1 volt RMS
Phase Spectrum: Phase angle versus frequency
Power Spectral Density: Power per unit bandwidth for noise characterization

Practical FFT Considerations

Achieving accurate FFT results requires attention to several factors:

Record Length Selection: Longer records improve frequency resolution but increase acquisition and processing time
Zero Padding: Appending zeros interpolates the spectrum between frequency bins without improving true resolution
Overlap Processing: Overlapping sequential records for continuous spectral monitoring
Dynamic Range: ADC resolution and noise floor limit ability to see small spectral components
Aliasing Prevention: Anti-aliasing filter must attenuate frequencies above Nyquist before sampling

Time-Frequency Analysis

While FFT analysis reveals frequency content, it provides no information about when those frequencies occur. Time-frequency analysis techniques simultaneously show both spectral content and its temporal evolution, essential for analyzing signals with changing frequency characteristics such as chirps, transients, and modulated waveforms.

Short-Time Fourier Transform

The STFT applies FFT to successive windowed segments of the signal:

Sliding Window: Window moves through signal, computing FFT at each position
Spectrogram Display: Two-dimensional plot with time, frequency, and color-coded magnitude
Window Length Trade-off: Short windows give good time resolution but poor frequency resolution, and vice versa
Overlap: Overlapping windows provide smoother temporal display
Time-Frequency Uncertainty: The Gabor limit, the signal-processing counterpart of the Heisenberg uncertainty principle, bounds the product of a window's RMS time spread σ_t and RMS frequency spread σ_f: σ_t × σ_f ≥ 1/(4π), with equality only for a Gaussian window

Spectrogram Applications

Spectrograms reveal time-varying spectral behavior:

Frequency Hopping: Visualize signal transitions between frequencies
Chirp Analysis: Track frequency sweeps through time
Transient Characterization: See spectral content of short-duration events
Vibration Analysis: Identify changing mechanical resonances
EMI Investigation: Correlate interference with system events

Wavelet Analysis

Wavelets provide adaptive time-frequency resolution:

Multi-Resolution: Good time resolution at high frequencies, good frequency resolution at low frequencies
Wavelet Functions: Various mother wavelets (Morlet, Daubechies, Mexican hat) suit different signal types
Continuous Wavelet Transform: Scalogram shows wavelet coefficients across time and scale
Discrete Wavelet Transform: Efficient decomposition into frequency bands
Transient Detection: Wavelets excel at localizing short-duration events

Wigner-Ville Distribution

This quadratic time-frequency representation offers high resolution:

High Resolution: Uses no analysis window, so it escapes the STFT's window trade-off and concentrates a linear chirp exactly along its instantaneous-frequency line
Cross-Term Interference: Every pair of components produces an oscillating cross-term midway between them in the time-frequency plane, however far apart they are
Smoothed Versions: The pseudo-Wigner-Ville, Choi-Williams, and other kernels reduce cross-terms at resolution cost
Single-Component Signals: The unsmoothed distribution is best suited to mono-component signals

Instantaneous Frequency

Tracking frequency as a continuous function of time:

Hilbert Transform: Derives analytic signal for instantaneous frequency calculation
Phase Derivative: Instantaneous frequency is rate of change of instantaneous phase
FM Demodulation: Extract modulating signal from frequency-modulated carriers
Phase Noise Analysis: Characterize oscillator frequency stability

Persistence Display

Digital persistence accumulates spectral information over time:

Intensity Grading: Brightness indicates frequency of occurrence at each point
Intermittent Signal Detection: Rare events visible even among dominant signals
Jitter Visualization: Signal variation appears as thickness in persistence display
Variable Decay: Adjustable decay rate emphasizes recent or historical behavior

Statistical Analysis

Statistical analysis characterizes signal variability and distribution, providing insights beyond single-acquisition measurements. By processing many acquisitions, statistical methods reveal patterns in signal behavior, quantify measurement uncertainty, and detect anomalies that might not appear in individual waveforms.

Basic Statistical Measures

Fundamental statistics describe signal amplitude distribution:

Mean: Average value, indicates DC component or center of distribution
Median: Middle value, robust to outliers
Standard Deviation: Spread around the mean; for a signal it equals the RMS of the AC component once the DC mean is removed
Variance: Square of standard deviation, useful for noise power calculations
RMS: Root-mean-square value; its square is proportional to average signal power, capturing both DC and AC content
Peak-to-Peak: Maximum minus minimum value

Higher-Order Statistics

Higher moments reveal additional distribution characteristics:

Skewness: Asymmetry of distribution, zero for symmetric distributions
Kurtosis: Tailedness of the distribution; high values indicate heavy tails and impulsive outliers, making it a useful metric for impact and bearing-fault detection
Crest Factor: Peak to RMS ratio, indicates impulsiveness; √2, about 1.414, for a sine wave
Form Factor: RMS value divided by the rectified average, indicates waveform shape; about 1.11 for a sine wave

Acquisition Statistics

Statistics computed across multiple acquisitions reveal timing variations:

Time Interval Error: Deviation of each edge from its ideal position, the record from which the other timing statistics derive
Period Jitter: Deviation of each period from the ideal or mean period; the standard deviation of the period measurements is the RMS period jitter
Cycle-to-Cycle Jitter: Difference in duration between adjacent periods
Rise Time Statistics: Distribution of transition times across acquisitions

Statistical Process Control

SPC techniques monitor measurement stability:

Control Charts: Track parameter values over time against control limits
Capability Indices: Cp and Cpk indicate process capability relative to specifications
Trend Analysis: Detect drift before parameters exceed limits
Out-of-Control Detection: Automatic alerts when statistics indicate abnormal behavior

Correlation Analysis

Correlation reveals relationships between signals:

Auto-Correlation: Signal correlated with time-shifted version of itself
Cross-Correlation: Correlation between two different signals
Periodicity Detection: Auto-correlation peaks indicate repeating patterns
Time Delay Measurement: Cross-correlation peak location indicates delay between signals
Coherence: Frequency-domain measure of linear relationship between signals

Probability Distributions

Fitting distributions to measurement data enables prediction:

Gaussian (Normal): Bell curve, common for noise and random variations
Uniform: Equal probability across range, typical for quantization error when the signal spans many code levels
Bimodal: Two peaks, indicates switching between states
Rayleigh: Magnitude of noise vector, common in RF envelope detection
Log-Normal: Multiplicative random processes

Histogram Generation

Histograms visualize amplitude distributions by counting how often signal values fall into discrete bins. This graphical representation reveals distribution shape, identifies multiple modes, and provides intuitive understanding of signal variability that complements numerical statistics.

Histogram Fundamentals

Basic histogram concepts and parameters:

Bins: Amplitude range divided into discrete intervals
Bin Width: Width of each amplitude interval
Counts: Number of samples falling within each bin
Normalization: Converting counts to probability density
Resolution Trade-off: More bins show detail but require more samples for statistical significance

Vertical Histogram

Amplitude histogram along the vertical axis of waveform display:

Noise Distribution: Noise appears as Gaussian spread around signal level
Logic Level Analysis: Digital signals show peaks at high and low states
Duty Cycle: Relative heights of logic peaks indicate duty cycle
Glitch Detection: Histogram counts at unexpected levels reveal glitches
Eye Diagram Integration: Histograms at specific sampling points show eye opening distribution

Horizontal Histogram

Time-based histograms show timing distributions:

Edge Histogram: Distribution of edge timing relative to trigger
Jitter Measurement: Width of edge histogram indicates timing uncertainty
Deterministic Components: Multiple peaks indicate deterministic jitter sources
Period Histogram: Distribution of clock period measurements

Histogram Analysis Techniques

Extracting information from histogram shape:

Peak Detection: Identify modes in multi-modal distributions
Gaussian Fitting: Fit normal distribution to extract mean and sigma
Dual-Dirac Model: Separate deterministic and random jitter components
Tail Analysis: Extreme values indicate bit error rate in communications
Area Integration: Compute probability of values within specified range

Eye Diagram Histograms

Histograms applied to eye diagrams characterize serial data quality:

Vertical Eye Opening: Histogram at sampling point shows voltage margin
Horizontal Eye Opening: Edge histogram shows timing margin
Bathtub Curve: BER versus sampling point derived from histogram tails
Contour Plots: Two-dimensional histograms show probability density across eye

Histogram Acquisition Modes

Different modes optimize histogram collection:

Full Waveform: Include all sample points in histogram
Gated Histogram: Histogram only samples within specified time window
Measurement Histogram: Histogram of computed measurements rather than raw samples
Infinite Persistence: Accumulate histogram indefinitely
Windowed: Rolling window retains recent samples only

Parameter Extraction

Parameter extraction algorithms automatically measure signal characteristics, providing quantitative values for design verification, specification compliance testing, and performance monitoring. These algorithms convert raw waveform data into meaningful engineering parameters.

Time-Domain Parameters

Measurements characterizing waveform timing and shape:

Rise Time: Time for signal to transition from low to high reference levels (typically 10% to 90%)
Fall Time: High to low transition time
Pulse Width: Time between the rising and falling crossings of the middle reference level (typically 50%)
Period: Time between corresponding edges of successive cycles
Frequency: Reciprocal of period
Duty Cycle: Ratio of positive pulse width to period
Slew Rate: Rate of voltage change across a transition, usually computed between the lower and upper reference levels

Amplitude Parameters

Measurements characterizing voltage levels:

Peak-to-Peak: Maximum minus minimum voltage
Amplitude: Difference between top and base levels
High Level: Steady-state voltage of logic high (top)
Low Level: Steady-state voltage of logic low (base)
Overshoot: Excursion beyond the high level after a rising edge, expressed as a percentage of amplitude
Undershoot: Excursion below low level after falling edge
Preshoot: Excursion before edge in opposite direction
Ringing: Damped oscillation following transition

Reference Levels

Defining measurement reference points:

Absolute Thresholds: Fixed voltage levels specified by user
Percentage Thresholds: Levels defined as percentage of amplitude (10%, 50%, 90%)
Hysteresis: Different thresholds for rising and falling edges prevent multiple triggering
Top and Base: The two state levels, usually computed from the modes of the amplitude histogram, which keeps overshoot and ringing from biasing them
Standard Definitions: IEEE Std 181 defines the terms for transitions, pulses, and related waveforms, together with algorithms for computing state levels, transition durations, and aberrations

Timing Measurements

Relationships between edges and events:

Delay: Time between edges on different signals
Setup Time: Data stable before clock edge
Hold Time: Data stable after clock edge
Propagation Delay: Time from input to output transition
Phase: Timing difference expressed as degrees or radians
Skew: Timing difference between nominally simultaneous signals

Jitter Parameters

Every edge-timing statistic starts from the time interval error (TIE), the displacement of each edge from its ideal position as set by a recovered or fitted reference clock. Differencing the TIE sequence once gives period jitter, the deviation of each period from the ideal or mean period. Differencing it again gives cycle-to-cycle jitter, the change in duration between adjacent periods. Each difference acts as a high-pass filter, so period jitter barely registers the slow wander that can dominate TIE, while cycle-to-cycle jitter emphasizes abrupt timing steps. Any of the three can be reported as an RMS value or as a peak-to-peak value.

RMS Jitter: Standard deviation of the timing variations, the usual way to state random jitter
Peak-to-Peak Jitter: Range between the earliest and latest timing deviations; with any random component it grows with the number of cycles observed, so it is meaningful only alongside the sample count
Random Jitter: Unbounded component, modeled as Gaussian
Deterministic Jitter: Bounded component, divided into data-dependent jitter (intersymbol interference and duty-cycle distortion), periodic jitter, and bounded uncorrelated jitter
Total Jitter: Peak-to-peak jitter at a stated bit error rate, combining the random and deterministic components, commonly through the dual-Dirac model

Frequency-Domain Parameters

Measurements derived from spectral analysis:

Fundamental Frequency: Dominant spectral component
Harmonic Distortion: Level of each individual harmonic relative to the fundamental
Total Harmonic Distortion: Combined power of the harmonics, usually a stated number of them, relative to the fundamental
Signal-to-Noise Ratio: Signal power relative to noise power, usually with the harmonics excluded
SINAD: Signal power relative to the combined power of noise and distortion
Spurious-Free Dynamic Range: Ratio of fundamental to largest spurious component
Effective Number of Bits: ADC resolution implied by measured SINAD

The effective number of bits (ENOB) inverts the signal-to-noise ratio of an ideal N-bit converter driven by a full-scale sine wave, 6.02N + 1.76 dB, so that a measured SINAD can be read as a resolution:

ENOB = (SINAD − 1.76) / 6.02

SINAD is in decibels. A measured SINAD of 74 dB gives (74 − 1.76) / 6.02, or about 12.0 effective bits. The formula assumes a full-scale test tone; a smaller tone yields a lower SINAD and therefore a lower ENOB unless the result is corrected for the shortfall.

Automated Measurements

Automated measurement systems apply analysis algorithms consistently across acquisitions, enabling high-throughput testing, continuous monitoring, and repeatable characterization. Automation eliminates manual measurement variability and enables statistical analysis of large datasets.

Measurement Automation Fundamentals

Core concepts in automated signal analysis:

Measurement Definition: Specify parameters, thresholds, and conditions
Source Selection: Identify signal source (channel, math function, memory)
Gating: Define time or cycle region for measurement
Statistics Mode: Accumulate statistics across multiple acquisitions
Result Format: Numeric display, histogram, trend, or export

Measurement Gating

Controlling measurement scope within acquired data:

Full Record: Measure entire acquired waveform
Time Gates: Measure only within specified time window
Cursor Gates: Interactive gate placement
Cycle Gating: Measure specific cycles within multi-cycle acquisition
Logic Qualification: Measure only when qualifying signal meets condition

Measurement Sequencing

Organizing multiple measurements for efficient execution:

Measurement Tables: Configure multiple measurements displayed together
Measurement Groups: Related measurements computed simultaneously
Conditional Measurements: Perform measurements based on previous results
Measurement Prioritization: Order measurements by importance or dependency

Pass/Fail Testing

Automated comparison against specifications:

Limit Definition: Specify acceptable ranges for each parameter
Mask Testing: Define allowed waveform regions graphically
Statistical Limits: Pass/fail based on statistical parameters
Action on Fail: Stop acquisition, sound alarm, save data, or continue
Margin Analysis: Report how close measurements are to limits

Test Sequences

Scripted measurement procedures:

Instrument Setup: Configure acquisition parameters automatically
Stimulus Control: Coordinate with signal sources and power supplies
Measurement Execution: Perform defined measurements in sequence
Data Logging: Record results to file or database
Report Generation: Create formatted test reports automatically

Remote Control and Programming

External control of measurement systems:

SCPI Commands: Standard Commands for Programmable Instruments
IVI Drivers: Interchangeable Virtual Instruments software interface
LabVIEW Integration: Graphical programming environment connectivity
Python Scripting: Popular language for test automation
MATLAB Interface: Integration with analysis and simulation tools

Data Management

Handling measurement results and waveform data:

Waveform Storage: Save raw or processed waveform data
Measurement Logging: Record measurements with timestamps
Database Integration: Store results in SQL or specialized databases
File Formats: CSV, XML, proprietary formats for different applications
Network Transfer: Move data to servers for analysis and archiving

Advanced Analysis Techniques

Sophisticated analysis methods address specialized measurement challenges and extract additional information from acquired signals.

Deconvolution and De-embedding

Removing effects of measurement system from results:

Fixture De-embedding: Remove effects of test fixtures and cables
Channel Calibration: Compensate for probe and oscilloscope frequency response
S-Parameter De-embedding: Use network parameters to remove interconnect effects
Time-Domain Reflectometry: Locate and characterize the impedance discontinuities in a fixture or interconnect before its effect is removed

Signal Separation

Isolating components within composite signals:

Filtering: Frequency-selective isolation of signal components
Synchronous Detection: Extract signals coherent with reference
Independent Component Analysis: Separate statistically independent sources
Adaptive Filtering: Dynamic filter adjustment for changing interference

Pattern Recognition

Identifying specific features in waveforms:

Template Matching: Find occurrences of reference patterns
Anomaly Detection: Identify waveforms differing from normal behavior
Classification: Categorize waveforms into predefined groups
Machine Learning: Train systems to recognize complex patterns

Signal Reconstruction

Recovering signals from incomplete or corrupted data:

Interpolation: Estimate values between sample points
Extrapolation: Predict signal behavior beyond measured range
Noise Reduction: Enhance signal-to-noise ratio through processing
Compressed Sensing: Reconstruct signals that are sparse in some basis from fewer samples than the Nyquist rate would require

Application Examples

Signal analysis techniques serve diverse measurement applications.

Clock Signal Analysis

Characterizing clock quality and jitter:

Period Measurement: Statistical analysis of clock period stability
Jitter Decomposition: Separate random and deterministic jitter components
Phase Noise: Spectral analysis of timing variations
Spread Spectrum: Verify intentional frequency modulation for EMI reduction

Power Integrity Analysis

Characterizing power distribution quality:

Ripple Measurement: AC content on DC power rails
Transient Response: Voltage deviation during load changes
Impedance Characterization: Frequency-dependent power distribution impedance
Noise Correlation: Relationship between power noise and signal integrity

Serial Data Analysis

Characterizing high-speed serial communications:

Eye Diagram Analysis: Timing and voltage margins for data recovery
Jitter Analysis: Total jitter budget allocation
Bathtub Curve: Bit error rate versus sampling point
Compliance Testing: Verification against interface specifications

Vibration and Acoustic Analysis

Mechanical signal characterization:

Order Analysis: Frequency components locked to rotation speed
Resonance Identification: Detect and characterize mechanical resonances
Bearing Analysis: Spectral signatures of bearing wear
Modal Analysis: Structure vibration modes and natural frequencies

Summary

Signal analysis transforms raw waveform data into meaningful engineering information through mathematical processing and statistical methods. From FFT-based spectral analysis that reveals frequency content to statistical methods that characterize signal variability, these techniques provide the quantitative foundation for electronic measurement and characterization.

FFT processing efficiently decomposes signals into frequency components, with windowing functions and averaging methods optimizing accuracy for different analysis goals. Time-frequency techniques like spectrograms and wavelet analysis reveal how spectral content evolves over time, essential for non-stationary signals. Statistical analysis quantifies signal variability across multiple acquisitions, while histograms visualize amplitude and timing distributions graphically.

Parameter extraction algorithms automatically measure time-domain, amplitude, and frequency-domain characteristics, providing consistent quantitative results for design verification and compliance testing. Automated measurement systems apply these techniques efficiently across large datasets, enabling high-throughput testing and continuous monitoring with pass/fail assessment against specifications.

Advanced techniques including de-embedding, signal separation, and pattern recognition address specialized challenges, with machine learning among the methods applied to anomaly detection and waveform classification. Together, these signal analysis capabilities form the analytical foundation of digital instrumentation, enabling engineers to extract maximum insight from measured data.