Tapped Delay Line (TDL)

A TDL model decomposes the wireless channel into a finite set of resolvable multipath components. Each tap corresponds to a distinct propagation path with a specific delay, complex amplitude, and Doppler spectrum. The tap spacing is typically set to the inverse of the signal bandwidth, ensuring that each tap represents an independently fading path. This discrete approximation of the continuous Channel Impulse Response (CIR) enables computationally efficient simulation while preserving the essential time-dispersive nature of real-world channels.

Each tap's complex coefficient is generated as a wide-sense stationary uncorrelated scattering (WSSUS) random process. The amplitude follows a specified fading distribution:

• Rayleigh fading: No dominant path, envelope follows Rayleigh distribution
• Rician fading: Dominant line-of-sight component defined by a K-factor
• Nakagami-m fading: Generalized model for various severity levels The phase is uniformly distributed over [0, 2π]. Temporal correlation is imposed by shaping the Doppler spectrum, typically using the Jakes model or a custom Power Spectral Density (PSD).

The Power Delay Profile (PDP) defines the relative power and excess delay for each tap, parameterizing the TDL for specific environmental scenarios:

• ITU Pedestrian A: Low delay spread, 4 taps, flat fading
• ITU Vehicular A: Moderate delay spread, 6 taps
• ITU Typical Urban: High delay spread, 20 taps, severe frequency selectivity
• TDL-A/B/C/D/E: 3GPP 5G NR models with 23 taps, each with specific Ricean K-factors and Doppler spectra The PDP directly determines the coherence bandwidth of the emulated channel.

Each tap's temporal variation is governed by a Doppler spectrum that models the relative motion between transmitter, receiver, and scatterers. Common implementations include:

• Classical Jakes spectrum: U-shaped PSD for isotropic scattering, characterized by maximum Doppler shift f_d = v·f_c/c
• Flat spectrum: Uniform PSD for simplified testing
• Gaussian spectrum: For scenarios with dominant scatterer clusters
• Bell-shaped spectrum: For aeronautical channels The Doppler spread determines the coherence time, the interval over which the channel remains approximately constant.

To achieve continuous-time channel emulation, the discrete tap weights are processed through interpolation and anti-aliasing filters. Key techniques include:

• Polyphase interpolation: Efficient sample rate conversion for fractional delays
• Sinc interpolation: Ideal bandlimited reconstruction for arbitrary tap delays
• Linear interpolation: Low-complexity approximation for narrowband signals
• Farrow structure: Variable fractional delay filter enabling dynamic tap delay adjustment These filters ensure that the emulated channel accurately represents the continuous-time physics of multipath propagation without spectral artifacts.

The TDL model's computational cost scales with the number of taps and the signal bandwidth. Optimization strategies include:

• Tap truncation: Removing taps below a power threshold (e.g., -20 dB relative to strongest tap)
• Frequency-domain convolution: Using FFT-based overlap-add for long impulse responses
• Reduced-state models: Merging closely spaced taps with similar Doppler characteristics
• Fixed-point implementation: Quantizing tap weights for FPGA or ASIC deployment For a 20 MHz LTE signal with a 20-tap TDL, the real-time convolution requires approximately 400 million multiply-accumulate operations per second.

The time-domain signature of a multipath channel, representing the received signal when an ideal impulse is transmitted. A TDL directly implements a discrete-time approximation of the CIR, where each tap coefficient corresponds to the complex gain of a resolvable propagation path. The CIR is the fundamental kernel used to convolve a clean synthetic waveform with realistic environmental distortions.

A parameter set that defines the intensity and relative delay of multipath components. The PDP specifies the tap spacing and average power weights used to configure a TDL for specific environments:

• ITU Indoor Office: Short delays, low dispersion
• ITU Vehicular-A: Moderate delays, classic cellular testing
• ITU Pedestrian-B: Multiple closely-spaced taps
• TDL-A/B/C models: 3GPP standardized profiles for 5G NR testing

The process of convolving a synthetic signal with a time-varying TDL to replicate the destructive and constructive interference of real-world propagation. Each tap's coefficient is modulated by a Doppler spectrum (e.g., Jakes, flat, Gaussian) to simulate relative motion. This creates realistic fast fading (rapid amplitude fluctuations) and frequency-selective fading (deep nulls at specific frequencies) critical for training robust fingerprinting models.

Defines the frequency-domain characteristics of the time variation applied to each TDL tap. The spectrum models the distribution of Doppler shifts caused by scatterers around a moving receiver:

• Jakes (Classic): Isotropic scattering, U-shaped spectrum for mobile terminals
• Flat: Uniform distribution, used for simplified testing
• Gaussian: Models specific directional scattering clusters The maximum Doppler frequency f_d = v·f_c/c determines the fading rate.

Two statistical models implemented by configuring TDL tap statistics:

Rayleigh Fading: All taps have zero-mean complex Gaussian coefficients. Models dense scattering with no line-of-sight (NLOS) path. The envelope follows a Rayleigh distribution.

Rician Fading: One dominant tap represents the line-of-sight (LOS) component with a constant amplitude plus scattered components. Defined by the K-factor (ratio of LOS power to scattered power). K=0 reduces to Rayleigh; K→∞ approaches no fading.

The fundamental thermal noise model injected after the TDL filtering stage to emulate the receiver's noise floor. AWGN is statistically independent from the multipath fading and has a flat power spectral density across all frequencies. The Signal-to-Noise Ratio (SNR) parameter controls the noise power relative to the signal, enabling training of fingerprinting models across diverse operating conditions from clean (30+ dB) to harsh (< 0 dB) environments.