Hilbert Transform Envelope

The foundation of envelope extraction. The Hilbert transform creates a quadrature-phase version of the real signal by applying a -90° phase shift to all positive frequency components and +90° to negative ones. The analytic signal is formed as: z(t) = x(t) + j * H{x(t)}, where x(t) is the real transient and H{x(t)} is its Hilbert transform. This complex representation eliminates negative frequencies, enabling unambiguous instantaneous amplitude and phase calculation.

The instantaneous envelope is the absolute value of the analytic signal: A(t) = |z(t)| = sqrt(x(t)² + H{x(t)}²). This produces a smooth, positive-definite curve that traces the peak power contour of the transient, completely removing the carrier oscillations. Key characteristics extracted include:

• Rise time (10%-90%): Reflects power amplifier slew rate
• Overshoot percentage: Indicates damping factor of bias network
• Settling time: Reveals PLL and AGC loop bandwidths
• Decay constant: Exposes power supply discharge path impedance

Beyond amplitude, the analytic signal yields the instantaneous phase: φ(t) = arctan(Im{z(t)} / Re{z(t)}). The derivative of the unwrapped phase gives the instantaneous frequency: f(t) = (1/2π) * dφ/dt. During transients, this reveals:

• Phase discontinuities at turn-on caused by oscillator startup
• Frequency settling profiles reflecting PLL loop filter dynamics
• VCO pulling effects as the power amplifier load impedance changes These phase-derived features are highly discriminative for emitter identification.

The Hilbert envelope enables precise temporal segmentation of a transient into its Attack, Decay, Sustain, and Release (ADSR) phases, borrowed from audio synthesis but directly applicable to RF transients:

• Attack: The initial energy rise, characterized by slope and inflection points
• Decay: The brief settling from overshoot to steady-state
• Sustain: The stable operating amplitude (if the burst is long enough)
• Release: The turn-off ramp-down profile Each segment's duration, curvature, and statistical moments form a unique hardware fingerprint.

The Hilbert envelope is robust to additive Gaussian noise, but its accuracy depends on Bedrosian's theorem: the envelope and carrier must have non-overlapping spectra for clean separation. In practice, wideband transient splatter can violate this condition. Mitigation strategies include:

• Bandpass filtering around the carrier before Hilbert transformation
• Wavelet-based denoising to isolate the transient from background noise
• Empirical Mode Decomposition (EMD) to separate the envelope from residual carrier leakage Proper pre-processing ensures the extracted envelope reflects true hardware behavior, not measurement artifacts.

In digital signal processing, the Hilbert transform is implemented via a finite impulse response (FIR) filter with anti-symmetric coefficients. Critical considerations:

• Filter order: Higher orders (e.g., 128+ taps) improve phase accuracy at low frequencies but increase latency
• Group delay: The filter introduces a delay of (N-1)/2 samples that must be compensated
• Edge transients: The FIR filter's own impulse response creates artifacts at the burst boundaries; zero-padding or mirror extension of the signal mitigates this
• Real-time constraints: For FPGA deployment, optimized polyphase filter structures minimize computational load

The complex-valued signal formed by combining the original real signal with its Hilbert transform as the imaginary part: z(t) = x(t) + jH{x(t)}. This representation eliminates negative frequency components, enabling unambiguous extraction of instantaneous attributes. The magnitude of the analytic signal yields the envelope, while the phase derivative provides instantaneous frequency. This mathematical construct is fundamental because it transforms a real oscillatory waveform into a rotating phasor in the complex plane, allowing the slow-varying amplitude modulation to be separated from the rapid carrier cycles without the distortion caused by simple rectification or peak detection.

The process of computing the time-varying magnitude of the analytic signal: A(t) = sqrt(x(t)² + H{x(t)}²). Unlike envelope detectors that rely on diode rectification and low-pass filtering, this method provides a mathematically exact envelope with no ripple or phase lag. For transient analysis, this precision is critical because it preserves the fine structure of the ramp-up and ramp-down profiles, including subtle inflection points and overshoot characteristics that serve as unique hardware identifiers. The technique correctly tracks the envelope even when the carrier frequency is higher than the modulation bandwidth, a condition where traditional filtering methods introduce unacceptable distortion.

Derived from the analytic signal's angle: φ(t) = arctan(H{x(t)} / x(t)). The instantaneous frequency is the time derivative: f(t) = (1/2π) dφ/dt. During transmitter turn-on transients, the phase trajectory reveals critical hardware dynamics including PLL settling behavior, VCO pulling effects, and phase discontinuities caused by non-ideal switching. These phase-domain features are often more discriminative than amplitude features alone, as they directly reflect the reactive component tolerances and oscillator physics unique to each device. Phase unwrapping algorithms are required to eliminate 2π discontinuities before differentiation.

Using the Hilbert envelope to precisely demarcate transient regions for feature extraction. The envelope's rising edge defines the attack phase (10% to 90% amplitude points), while the falling edge defines the decay phase. Key segmentation points include:

• Burst onset: where the envelope crosses a noise-floor threshold
• Overshoot peak: the maximum envelope excursion above steady-state
• Settling point: where the envelope stabilizes within ±1% of final value
• Ringing interval: the damped oscillation period following the main transition Accurate segmentation ensures that fingerprint features are extracted from consistent temporal regions across captures, eliminating alignment errors that degrade classification accuracy.

While the standard Hilbert transform requires narrowband signals for meaningful instantaneous frequency, the Hilbert-Huang Transform (HHT) extends the concept to broadband transients. HHT first decomposes the signal using Empirical Mode Decomposition (EMD) into Intrinsic Mode Functions (IMFs), then applies the Hilbert transform to each IMF individually. This adaptive approach is particularly valuable for transient signals containing multiple time-scale components, such as the simultaneous fast ringing artifact and slow thermal settling observed in power amplifier turn-on signatures. The resulting Hilbert spectrum provides a high-resolution time-frequency representation without the cross-term interference of Wigner-Ville distributions.

In modern software-defined radio systems, the analytic signal is often generated directly through quadrature downconversion rather than explicit Hilbert transform computation. The in-phase (I) and quadrature (Q) samples from the ADC naturally form the real and imaginary parts of the analytic signal at baseband. The envelope is then computed as sqrt(I² + Q²). Digital implementations must consider:

• FIR filter design for Hilbert transformers (typically odd-length, anti-symmetric coefficients)
• Group delay compensation to align I and Q channels
• CORDIC algorithms for efficient magnitude computation on FPGAs
• Fixed-point precision requirements to preserve transient detail without overflow