Time-Frequency Analysis

The foundational technique that segments a signal into short, quasi-stationary windows and applies the Fourier Transform to each. The result is a spectrogram: a visual representation of spectral power density over time.

• Resolution Trade-off: Governed by the Heisenberg-Gabor limit; a narrow window yields good time resolution but poor frequency resolution, and vice versa.
• Windowing Functions: Hann, Hamming, and Blackman windows mitigate spectral leakage caused by abrupt truncation.
• AI Application: Spectrograms serve as standard image inputs for Convolutional Neural Networks (CNNs) in interference classification, treating frequency patterns as visual textures.

A quadratic time-frequency representation that provides the highest possible joint resolution by correlating the signal with a time-reversed copy of itself. Unlike the STFT, the WVD is not dependent on a windowing function.

• Cross-Term Interference: The primary drawback; when analyzing multi-component signals, the WVD generates oscillatory cross-terms midway between true signal components, creating artifacts that can confuse classifiers.
• Smoothed Pseudo WVD: A variant applying independent smoothing in time and frequency to suppress cross-terms at the cost of reduced resolution.
• Use Case: Effective for classifying linear frequency modulated (LFM) chirp jammers where auto-terms are distinct.

A multi-resolution analysis technique that decomposes a signal using scaled and shifted versions of a mother wavelet—a finite, oscillatory waveform. Unlike the fixed-resolution STFT, wavelets provide adaptive time-frequency tiling.

• Scale vs. Frequency: Low scales capture high-frequency, short-duration transients; high scales capture low-frequency, long-duration trends.
• Discrete Wavelet Transform (DWT): Implements a dyadic filter bank for efficient computation, producing approximation and detail coefficients.
• Transient Detection: Excels at isolating and classifying impulsive interference like pulse jamming or bursty electromagnetic noise due to its variable windowing.

A member of the Cohen's class of quadratic distributions designed to reduce cross-term interference while preserving high time-frequency resolution. It applies an exponential kernel function in the ambiguity domain.

• Kernel Design: The exponential kernel acts as a low-pass filter in the ambiguity domain, suppressing cross-terms that typically appear far from the origin while retaining auto-terms.
• Parameter Tuning: A controllable smoothing parameter governs the trade-off between cross-term suppression and auto-term broadening.
• Classification Feature: CWD images provide cleaner inputs for deep learning models when classifying multiple simultaneous jamming signals, reducing false features.

A perceptually motivated time-frequency representation that maps the linear frequency axis to the Mel scale, which approximates logarithmic human auditory perception. Widely used in audio processing, it is adapted for RF classification.

• Mel Filter Bank: A series of overlapping triangular filters spaced according to the Mel scale, integrating spectral energy to reduce dimensionality.
• Mel-Frequency Cepstral Coefficients (MFCCs): A compact feature set derived by applying a Discrete Cosine Transform (DCT) to the log Mel-spectrogram, decorrelating features.
• RF Adaptation: Applied to baseband or demodulated audio signatures of interference to generate compact, robust feature vectors for lightweight classifiers.

A data-driven, adaptive decomposition method that breaks a non-stationary signal into a finite set of Intrinsic Mode Functions (IMFs) without a predefined basis function. The Hilbert-Huang Transform applies the Hilbert Transform to these IMFs.

• Sifting Process: An iterative algorithm that extracts zero-mean oscillatory components from the signal envelope, starting with the highest frequency.
• Hilbert-Huang Spectrum: A sharp, high-resolution time-frequency representation derived from the instantaneous frequency of each IMF.
• Non-Linear Interference: Highly effective for analyzing non-linear and chaotic jamming patterns where linear methods like STFT fail to capture the signal structure.

The foundational time-frequency representation that applies a sliding window function to segment a non-stationary signal before computing the Fourier transform of each segment. The result is a spectrogram—a 2D image of spectral power over time.

• Resolution Trade-off: A narrow window yields good time resolution but poor frequency resolution; a wide window does the opposite, governed by the Heisenberg-Gabor limit.
• Feature Engineering: Spectrograms serve as direct inputs to Convolutional Neural Networks (CNNs) for visual pattern recognition of jamming pulses, frequency hops, and chirps.
• Windowing Functions: Choices like Hamming, Hann, or Kaiser windows control spectral leakage and sidelobe levels, directly impacting classification accuracy.

A quadratic time-frequency distribution offering the highest possible joint resolution without the windowing trade-off inherent in the STFT. The WVD computes the Fourier transform of the signal's instantaneous autocorrelation function.

• Cross-Term Interference: The quadratic nature generates spurious cross-terms between multi-component signals, which can obscure true signal features and mislead classifiers.
• Mitigation: Smoothed variants like the Choi-Williams or Zhao-Atlas-Marks distributions apply kernel functions to suppress cross-terms at the cost of reduced resolution.
• Use Case: Excels at resolving closely spaced frequency components in linear frequency-modulated (LFM) jammers where STFT resolution is insufficient.

A multi-resolution analysis technique that decomposes a signal using scaled and shifted versions of a mother wavelet—a localized, oscillatory function. Unlike the fixed-resolution STFT, wavelets provide high time resolution at high frequencies and high frequency resolution at low frequencies.

• Continuous Wavelet Transform (CWT): Produces a scalogram, mapping signal energy across time and scale (inverse frequency). Ideal for detecting transient interference pulses and singularities.
• Discrete Wavelet Transform (DWT): Implements a dyadic filter bank for efficient, non-redundant decomposition, commonly used for denoising IQ samples before classification.
• Wavelet Families: Daubechies, Morlet, and Symlet wavelets offer different trade-offs between compactness and smoothness for matching specific interference signatures.

A technique exploiting the periodic statistical properties of modulated signals, which manifest as spectral correlation in the frequency domain. Cyclostationary analysis reveals features invisible to conventional power spectral density estimation.

• Spectral Correlation Function (SCF): A 2D function mapping spectral frequency against cycle frequency, uniquely characterizing each modulation scheme's symbol rate, carrier offset, and pulse shape.
• Robustness: Cyclostationary features remain detectable at very low signal-to-noise ratios (SNR) where spectrograms fail, making them critical for classifying weak or distant interference.
• Computational Cost: The FAM (FFT Accumulation Method) algorithm reduces SCF computation to a practical level for real-time systems by exploiting FFT efficiency.

A data-driven, adaptive decomposition that breaks a non-stationary signal into a finite set of Intrinsic Mode Functions (IMFs) without predefined basis functions. Each IMF represents a simple oscillatory mode embedded in the signal.

• Hilbert-Huang Transform (HHT): Applying the Hilbert transform to each IMF yields instantaneous frequency and amplitude, producing a high-resolution time-frequency-energy distribution.
• Interference Isolation: EMD excels at separating multi-component jamming signals into physically meaningful sub-signals, enabling per-component classification.
• Limitations: Susceptible to mode mixing—where a single IMF contains disparate frequency scales—requiring ensemble variants (EEMD) with noise-assisted stabilization.

A member of the Cohen's class of reduced-interference distributions that applies an exponential kernel to suppress WVD cross-terms while preserving auto-term resolution. The kernel is parameterized by a smoothing factor controlling the cross-term suppression level.

• Kernel Design: The Choi-Williams kernel is a 2D Gaussian in the ambiguity domain, satisfying key mathematical properties including time and frequency marginal preservation.
• Trade-off: Larger smoothing factors eliminate cross-terms but broaden auto-terms, reducing the ability to resolve closely spaced interference components.
• Application: Frequently used for analyzing frequency-hopping spread spectrum (FHSS) signals where multiple hop transitions create dense cross-term artifacts in the WVD.