S-Transform

The core innovation of the S-Transform is its scalable Gaussian window whose width is inversely proportional to frequency. This means:

• Low Frequencies: The window widens, providing high frequency resolution to distinguish closely spaced spectral components.
• High Frequencies: The window narrows, providing high time resolution to precisely localize sharp transients and fast-changing events. This adaptive behavior directly links the analysis scale to the Fourier frequency, a property not found in the fixed-window STFT.

Unlike the Continuous Wavelet Transform (CWT), the S-Transform uniquely maintains an absolute reference to the phase of the signal. The phase information in the S-Transform domain is directly tied to the zero-time reference of the original time series. This is mathematically achieved by retaining the Fourier modulation kernel, allowing for:

• Direct cross-spectral analysis between channels.
• Meaningful phase extraction for signal synchronization.
• Construction of a time-frequency filter with a known phase response.

A fundamental property of the S-Transform is that averaging its output over the entire time axis perfectly reconstructs the Fourier spectrum of the signal. This provides a direct, mathematically lossless link between the time-frequency domain and the pure frequency domain. This property ensures that:

• No information is lost in the transformation.
• The time-frequency representation is a true decomposition of the Fourier components.
• It can be used as a rigorous tool for spectral analysis of non-stationary signals.

The S-Transform provides a multi-resolution analysis similar to wavelets but without the cross-term interference that plagues quadratic distributions like the Wigner-Ville Distribution (WVD). Because it is a linear transform, it does not generate spurious artifacts when analyzing multi-component signals. This results in a clean, interpretable time-frequency map where:

• Each signal component is represented independently.
• The energy distribution is non-negative.
• The resolution trade-off is governed by the Heisenberg-Gabor uncertainty principle, optimized by the Gaussian window.

The S-Transform can be efficiently computed using the Fast Fourier Transform (FFT) by leveraging its relationship to the Short-Time Fourier Transform. The algorithm involves:

• Computing the Fourier transform of the signal.
• Shifting the spectrum and multiplying by a frequency-dependent Gaussian window in the frequency domain.
• Applying an inverse FFT for each frequency sample. This frequency-domain implementation makes the S-Transform practical for processing long signal records, though it remains more computationally intensive than a standard STFT.

The S-Transform is a losslessly invertible transform. The original time-domain signal can be perfectly reconstructed from its S-Transform representation by integrating over all frequencies. This is a direct consequence of its connection to the Fourier spectrum. This property is critical for:

• Time-frequency filtering: modifying the S-Transform matrix and inverting it to obtain a filtered signal.
• Signal denoising: zeroing out noise-dominated regions in the time-frequency plane.
• Feature extraction: isolating specific time-frequency components for analysis.

The S-Transform excels at analyzing turn-on transients where frequency content changes rapidly. Its frequency-dependent Gaussian window automatically narrows at higher frequencies, capturing the sharp spectral changes during amplifier ramp-up. This multi-resolution property isolates the unique transient 'fingerprint' caused by DAC slewing and PLL locking dynamics without requiring manual window size tuning. The absolutely referenced phase information preserves the temporal structure of the transient, unlike the CWT which only provides relative phase.

Frequency-dependent I/Q imbalance manifests as mirror-frequency interference that varies across the bandwidth. The S-Transform's direct relationship to the Fourier spectrum allows engineers to track the magnitude of gain imbalance and quadrature skew as continuous functions of both time and frequency. This reveals how modulator impairments evolve during a transmission burst, providing a richer feature set for fingerprinting than static impairment estimates derived from a single constellation diagram.

The S-Transform preserves absolute phase information, making it directly applicable to extracting phase noise signatures from local oscillators. By analyzing the time-frequency representation around the carrier, the spreading effect of phase noise can be separated from amplitude noise. The multi-resolution windowing ensures that close-in phase noise (slow fluctuations) and far-out phase noise (fast fluctuations) are resolved with appropriate time-frequency trade-offs, revealing the unique spectral regrowth pattern of each transmitter's PLL loop filter.

Multipath fading causes frequency-selective notches that corrupt fingerprint features. The S-Transform's frequency-dependent resolution provides a framework for isolating signal components in time-frequency regions less affected by destructive interference. By applying time-frequency masking based on local SNR estimates, features can be extracted only from high-confidence regions. This selective analysis, combined with the transform's invertibility, enables reconstruction of a channel-compensated signal representation for more robust device identification.

Clock jitter in digital-to-analog converters introduces non-stationary timing errors that spread signal energy in the time-frequency plane. The S-Transform's ability to maintain a direct relationship with the Fourier spectrum allows for the calculation of instantaneous frequency deviation caused by jitter. By tracking the time-varying spectral broadening around symbol transitions, the statistical properties of the jitter process—unique to each device's clock generation circuit—can be quantified and used as a robust identifying feature.

AM-AM and AM-PM distortion in power amplifiers creates spectral regrowth that varies with the signal envelope. The S-Transform captures how these non-linear products evolve dynamically, revealing the amplifier's memory effects caused by thermal and bias circuit time constants. By analyzing the time-frequency distribution around spectral regrowth shoulders, the specific non-linear transfer characteristic of each amplifier can be profiled, providing a hardware-specific signature that persists even when transmitting standardized waveforms.

The foundational windowed Fourier method that computes the frequency content of local signal segments. Unlike the S-Transform, the STFT uses a fixed-width window for all frequencies, resulting in a constant, non-adaptive time-frequency resolution trade-off. This makes it computationally efficient but suboptimal for analyzing signals with both slow and fast transient components simultaneously.

A multi-resolution analysis technique that correlates a signal with scaled and shifted versions of a mother wavelet. The S-Transform is directly derived from the CWT but with a crucial distinction: it uses a frequency-dependent Gaussian window and multiplies by a phase correction factor, retaining absolute phase information that the CWT discards. This makes the S-Transform a hybrid bridge between the CWT and STFT.

The squared magnitude of the STFT, visualized as a two-dimensional heatmap of spectral power over time. While an S-Transform produces a similar visual output, the spectrogram's fixed window causes smearing of high-frequency transients or poor low-frequency localization. The S-Transform's adaptive window yields a sharper, more readable spectrogram-like representation without manual window size tuning.

A quadratic time-frequency distribution offering the highest possible joint resolution by computing the Fourier transform of the signal's instantaneous autocorrelation. However, it suffers severely from cross-term interference for multi-component signals. The S-Transform, being a linear transform, is completely free of these spurious artifacts, making it more reliable for analyzing complex real-world waveforms.

A post-processing reassignment technique that sharpens a spectrogram or scalogram by concentrating diffused energy onto the true instantaneous frequency ridges. The S-Transform can be combined with synchrosqueezing to produce a highly concentrated, denoised time-frequency representation that retains the S-Transform's absolute phase and multi-resolution properties, making it a powerful tool for mode extraction.

A transform with geometrically spaced frequency bins and a window length that varies inversely with frequency, mirroring the human auditory system. The S-Transform achieves a similar variable resolution effect but with a linear frequency scale and a Gaussian window that provides optimal Heisenberg uncertainty minimization. The CQT is favored in music analysis, while the S-Transform excels in general signal processing.