Analytic Signal

The analytic signal z(t) is constructed by combining the original real signal x(t) with its Hilbert transform H[x(t)] as the imaginary component:

• Formula: z(t) = x(t) + j·H[x(t)]
• The Hilbert transform applies a 90-degree phase shift to all positive frequency components
• The resulting complex signal contains only positive frequencies, eliminating the symmetric negative spectrum
• This construction preserves all information from the original real signal while enabling complex analysis techniques
• The real part represents the in-phase component, while the imaginary part represents the quadrature component

The instantaneous amplitude A(t) is computed as the magnitude of the analytic signal:

• Formula: A(t) = |z(t)| = √(x²(t) + H[x(t)]²)
• Represents the envelope of the signal at each time instant
• Provides a smooth curve that traces the signal's amplitude variations
• Critical for modulation analysis and transient detection in RF fingerprinting
• Unlike peak detection methods, the analytic envelope is mathematically rigorous and unambiguous
• Used to extract turn-on/turn-off transients for device identification

The instantaneous phase φ(t) is derived from the argument of the analytic signal:

• Formula: φ(t) = arg(z(t)) = arctan(H[x(t)] / x(t))
• Requires phase unwrapping to remove 2π discontinuities and obtain a continuous function
• Reveals subtle phase distortions caused by hardware impairments in transmitters
• Phase noise patterns serve as unique device fingerprints due to oscillator imperfections
• Enables detection of intentional and unintentional phase modulation characteristics
• Sensitive to I/Q imbalance and local oscillator leakage in transmitter chains

The instantaneous frequency f(t) is the time derivative of the unwrapped instantaneous phase:

• Formula: f(t) = (1/2π) · dφ(t)/dt
• Represents the dominant frequency at each moment in time
• Provides a single, unambiguous frequency value for non-stationary signals
• Essential for analyzing frequency-modulated signals and chirp characteristics
• Reveals frequency drift patterns caused by thermal effects in oscillators
• Used in radar signal analysis and emitter identification through frequency agility profiling

A defining property of the analytic signal is the complete elimination of negative frequency components:

• The Fourier transform of an analytic signal is zero for all negative frequencies
• This suppression removes the Hermitian symmetry inherent in real-valued signals
• Eliminates ambiguity in associating frequency content with physical phenomena
• Enables the use of complex baseband representations in communication systems
• Reduces bandwidth requirements for signal processing by half without information loss
• Forms the mathematical foundation for single-sideband modulation and efficient spectrum usage

Two critical theorems govern the validity of instantaneous parameter extraction from analytic signals:

• Bedrosian's Theorem: The Hilbert transform of a product of a low-frequency envelope and high-frequency carrier equals the envelope multiplied by the Hilbert transform of the carrier, provided their spectra do not overlap
• Nuttall's Theorem: The quadrature error between the actual signal and its quadrature approximation is minimized when the signal is narrowband
• These theorems define the conditions for accurate envelope and phase extraction
• Violations lead to physically meaningless negative frequencies or distorted instantaneous parameters
• Critical for understanding when analytic signal analysis is valid for wideband or multi-component signals

The analytic signal enables the calculation of instantaneous amplitude (envelope), instantaneous phase, and instantaneous frequency at every sample point. These features capture the subtle, time-varying distortions caused by hardware impairments.

• Envelope analysis reveals power amplifier non-linearities and transient overshoot
• Phase trajectory exposes oscillator phase noise and I/Q imbalance
• Frequency deviation highlights clock jitter and modulation errors

These instantaneous parameters serve as direct inputs to 1D convolutional neural networks (CNNs) and recurrent architectures for device classification.

The analytic signal is constructed by adding the original real signal to its Hilbert transform as the imaginary component: z(t) = x(t) + jH{x(t)}. The Hilbert transform applies a 90-degree phase shift to all positive frequency components.

• Suppresses negative frequency content entirely
• Creates a complex-valued signal with a one-sided spectrum
• Preserves all information from the original real waveform

This transformation is lossless and reversible, making it a foundational preprocessing step before time-frequency analysis or neural feature extraction.

The magnitude of the analytic signal, |z(t)|, provides the instantaneous envelope. Power amplifier (PA) non-linearities manifest as envelope-dependent amplitude and phase distortions unique to each transmitter.

• AM-AM distortion: Deviation from linear gain as a function of input amplitude
• AM-PM distortion: Unintended phase shift correlated with envelope magnitude
• Spectral regrowth: Out-of-band emissions caused by envelope clipping

Deep learning models trained on envelope statistics can identify devices by their characteristic PA compression curves and memory effects.

The instantaneous phase, φ(t) = arg[z(t)], reveals the phase noise profile of the transmitter's local oscillator. Each oscillator has a unique phase noise signature due to manufacturing variances in crystal resonators and phase-locked loops.

• Close-in phase noise reflects low-frequency flicker noise characteristics
• Phase discontinuities indicate PLL transient behavior
• Unwrapped phase drift captures long-term frequency stability

Recurrent neural networks (RNNs) and transformers process phase sequences to learn temporal dependencies that distinguish identical device models.

The time derivative of the unwrapped instantaneous phase yields the instantaneous frequency: f(t) = (1/2π) · dφ/dt. This parameter exposes unintentional frequency modulation caused by DAC clock jitter and power supply ripple.

• Micro-frequency deviations during symbol transitions reveal DAC slew rate limitations
• Periodic frequency ripple correlates with switching power supply artifacts
• Frequency settling time after channel changes indicates PLL loop bandwidth

These fine-grained frequency signatures are robust to channel effects and provide a stable biometric for long-term device tracking.

The analytic signal provides a natural complex baseband representation that separates the carrier from the modulation content. Deep learning architectures process this as dual-channel (I/Q) input tensors.

• Complex-valued neural networks preserve phase relationships between I and Q components
• I/Q constellation analysis reveals modulator imbalance and DC offset fingerprints
• Complex spectrograms computed from the analytic signal provide time-frequency features

Modern architectures like complex ResNets and attention-based models operate directly on analytic signal representations for end-to-end emitter identification.

The linear operator that creates the quadrature component of the analytic signal by applying a 90-degree phase shift to all positive frequency components and a -90° shift to negative ones. Its impulse response is h(t) = 1/(πt), making it a non-causal filter with a singularity at the origin.

• Implemented in discrete time via FIR filters using Parks-McClellan or Remez exchange algorithms
• The Fourier transform of the Hilbert kernel is H(ω) = -j·sgn(ω), a pure phase-shifter
• Essential for single-sideband modulation and envelope detection in communication receivers

Defined as the time derivative of the unwrapped instantaneous phase of the analytic signal: f_i(t) = (1/2π) · dφ(t)/dt. This scalar function reveals how the dominant spectral component evolves moment by moment, making it indispensable for analyzing non-stationary signals like frequency-modulated radar pulses.

• Requires careful phase unwrapping to avoid 2π discontinuities
• Physically meaningful only for monocomponent signals or narrowband components
• Forms the basis for time-frequency ridge extraction and synchrosqueezing transforms

The magnitude of the analytic signal: A(t) = |z(t)| = √(x²(t) + ẋ²(t)), where ẋ(t) is the Hilbert transform of the real signal x(t). This envelope traces the upper boundary of the waveform's energy, independent of the rapid carrier oscillations.

• Used in RF fingerprinting to extract transient turn-on/turn-off signatures
• Enables modulation recognition by analyzing envelope statistics and dynamics
• The squared envelope |z(t)|² gives instantaneous power for energy-based feature extraction

A fundamental constraint stating that the product of two signals has a Hilbert transform equal to one signal times the Hilbert transform of the other only if their spectra are disjoint. Specifically, if a(t) is a low-frequency envelope and cos(φ(t)) is a high-frequency carrier, then H{a(t)·cos(φ(t))} = a(t)·sin(φ(t)).

• Validates the physical interpretation of instantaneous amplitude and frequency
• Fails when envelope and carrier spectra overlap, producing nonsensical negative frequencies
• Critical for understanding when analytic signal decomposition is mathematically legitimate

The unique complex extension of a real signal that yields a non-negative instantaneous frequency for all time. Constructed by factorizing the signal into amplitude and frequency-modulated components using the Gabor-Ville decomposition rather than the standard Hilbert-based approach.

• Resolves the paradox of negative instantaneous frequencies that can arise from the standard analytic signal
• Uses the relationship z(t) = A(t)·exp(jφ(t)) where A(t) and φ'(t) are guaranteed non-negative
• Preferred for high-precision RF fingerprinting where instantaneous frequency sign carries physical meaning

A data-driven alternative to the analytic signal approach that decomposes a signal into Intrinsic Mode Functions (IMFs) before applying the Hilbert transform. Each IMF is a zero-mean oscillatory mode with symmetric envelopes, making it suitable for instantaneous frequency computation.

• The complete Hilbert-Huang Transform applies the Hilbert transform to each IMF individually
• Avoids Bedrosian's theorem violations by ensuring each IMF is narrowband
• Widely used in RF fingerprinting for separating transient events from steady-state carrier components