Digital Twin

The digital twin replicates the complete impairment fingerprint of a specific device by mathematically modeling its analog front-end non-idealities. This includes:

• I/Q imbalance: Gain and phase mismatches in the quadrature modulator
• Local oscillator leakage: Unintended carrier feedthrough causing DC offset
• Power amplifier non-linearity: AM-AM and AM-PM distortion curves with memory effects
• Phase noise: Oscillator instability characterized by a phase noise mask
• DAC quantization error: Finite bit-depth conversion artifacts

The result is a synthetic signal that mirrors the real transmitter's Error Vector Magnitude (EVM) and spectral regrowth profile.

The digital twin generates the intrinsic transmitter signature in isolation, before any channel effects are applied. This clean, impaired waveform serves as the canonical reference for that device. Key properties:

• No multipath: The base model excludes fading, delay spread, and Doppler
• Controlled SNR: Noise is added parametrically during training, not baked into the twin
• Deterministic replay: The same impairment parameters always produce the same signature

This separation allows a single digital twin to be convolved with thousands of different Channel Impulse Responses (CIRs) to generate diverse, labeled training datasets.

The primary purpose of an RF digital twin is to solve the cold-start enrollment problem. A single twin can generate unlimited volumes of labeled I/Q data by:

• Domain randomization: Varying SNR, carrier frequency offset, and sampling clock offset across a defined range
• Channel convolution: Applying Tapped Delay Line (TDL) models for Rician and Rayleigh fading
• Augmentation: Injecting controlled levels of AWGN and narrowband interference

This enables training a deep learning fingerprinting model on millions of diverse examples before the real device is ever deployed in the field.

To ensure the digital twin is indistinguishable from its physical counterpart, a Hardware-in-the-Loop (HIL) validation loop is employed:

• The synthetic waveform is transmitted through a vector signal generator
• A high-fidelity spectrum analyzer captures the over-the-air signal
• A discriminator model attempts to distinguish synthetic from real captures
• Impairment parameters are iteratively refined until the discriminator performs at chance level (50%)

This closes the sim-to-real gap and certifies the twin for operational use.

A critical operational advantage: once a device's impairment profile is characterized and modeled, the digital twin can be enrolled into a fingerprinting system without requiring the physical device to be present. This supports:

• Supply chain authentication: Verifying components before they ship
• Fleet pre-enrollment: Onboarding thousands of IoT devices from a single golden-unit characterization
• Adversarial robustness testing: Generating spoofing attacks against the twin to harden the classifier

The twin becomes the root of trust for the device's physical-layer identity.

Real transmitters age. Their impairments drift due to temperature variation, voltage fluctuation, and component degradation. An advanced digital twin models this temporal evolution:

• Thermal drift curves: Parameter variation as a function of operating temperature
• Aging profiles: Long-term shifts in power amplifier gain and oscillator stability
• Voltage sensitivity: Impairment changes under battery drain conditions

This allows the fingerprinting model to track a device's identity continuously across its operational lifecycle, preventing false rejections due to natural drift.

The mathematical simulation of gain and phase mismatches between the in-phase and quadrature signal paths. This is a primary hardware impairment used to generate unique, synthetic transmitter fingerprints.

• Models amplitude error (ε) and phase error (φ) in the quadrature modulator
• Creates asymmetric constellation distortion that is device-specific
• Essential for replicating low-cost transmitter behavior in simulation

The emulation of amplitude and phase distortion in a transmitter's final stage, characterized by AM-AM and AM-PM conversion curves and memory effects.

• Replicates device-specific spectral regrowth and adjacent channel interference
• Modeled using Volterra series or memory polynomial frameworks
• Critical for capturing the most distinctive hardware signature in high-power transmitters

A neural network architecture where a generator creates synthetic impaired signals and a discriminator attempts to distinguish them from real ones. This adversarial process produces highly realistic training data.

• Generator learns to replicate subtle hardware impairment distributions
• Discriminator provides a learned loss function superior to simple MSE
• Enables generation of signals indistinguishable from real transmitter emissions

A training strategy that varies the parameters of a synthetic impairment simulator—such as noise levels, channel models, and SNR—to force a fingerprinting model to learn invariant, robust features.

• Prevents overfitting to specific simulation parameters
• Improves generalization from synthetic to real-world signals
• Randomizes CFO, phase noise, and multipath profiles per training batch

A time-domain representation of a multipath channel's effect on a transmitted signal, used as a filter kernel to synthetically impose delay spread and fading on a clean waveform.

• Implemented via Tapped Delay Line (TDL) structures
• Each tap represents a resolvable multipath component with specific delay and amplitude
• Convolved with the synthetic signal to emulate real-world propagation

A simulation methodology that integrates physical RF components—such as a vector signal generator—with a real-time software channel emulator to validate fingerprinting models against live hardware.

• Bridges the gap between pure simulation and field testing
• Validates digital twin fidelity by comparing synthetic and physical emissions
• Enables repeatable testing with controlled impairment injection