Multipath Fading

The time difference between the arrival of the first and last significant multipath component. Delay spread causes frequency-selective fading when it exceeds the symbol period, resulting in inter-symbol interference (ISI) that distorts the received constellation diagram.

• Measured in microseconds for outdoor macro-cells
• Measured in nanoseconds for indoor environments
• Directly impacts the required equalizer complexity

The spectral broadening caused by relative motion between transmitter and receiver. Doppler spread introduces time-selective fading, making the channel response vary within a single transmission frame.

• Maximum Doppler shift: f_d = v/λ (velocity divided by wavelength)
• Causes fast fading when the channel changes within a symbol period
• Creates spectral regrowth that complicates modulation identification

The frequency range over which the channel response remains approximately constant. Coherence bandwidth is inversely proportional to delay spread and determines whether fading is flat or frequency-selective.

• Flat fading: Signal bandwidth < Coherence bandwidth (all frequencies fade together)
• Frequency-selective fading: Signal bandwidth > Coherence bandwidth (different frequencies fade independently)
• Critical for determining if a single-tap equalizer suffices

Two fundamental statistical models for multipath channels. Rayleigh fading models scenarios with no dominant line-of-sight (LOS) path, producing deep fades. Rician fading includes a dominant LOS component characterized by the K-factor.

• Rayleigh: Amplitude follows Rayleigh distribution (NLOS urban environments)
• Rician: Amplitude follows Rician distribution (suburban with partial LOS)
• K-factor = Power of LOS component / Power of scattered components

Classification based on the relationship between signal bandwidth and channel coherence bandwidth. Flat fading preserves the signal's spectral shape but varies amplitude, while frequency-selective fading creates a non-uniform frequency response.

• Flat fading: B_signal << B_coherence (simple amplitude scaling)
• Frequency-selective: B_signal > B_coherence (complex equalization required)
• Deep learning classifiers must be trained on both conditions for robust deployment

Multipath fading severely degrades automatic modulation classification accuracy by distorting the constellation diagram and introducing inter-symbol interference. Deep learning models must learn channel-invariant features.

• ISI smears constellation points, making QAM orders indistinguishable
• Time-varying channels require adaptive or recurrent neural architectures
• Data augmentation with simulated multipath profiles improves robustness
• Cyclostationary features offer inherent resilience to frequency-selective fading

The primary degradation mechanism caused by multipath fading. When delayed signal replicas overlap with subsequent symbols, the receiver cannot correctly distinguish discrete symbol states.

• Mechanism: A delayed path from symbol n arrives during the sampling instant of symbol n+1.
• Impact on AMC: ISI distorts the constellation diagram, smearing decision boundaries and making modulation schemes like 16-QAM and 64-QAM visually indistinguishable.
• Mitigation: Equalization filters and OFDM guard intervals are standard countermeasures.

A statistical model for multipath propagation where a dominant line-of-sight (LOS) component exists alongside scattered paths.

• K-factor: The ratio of LOS power to scattered power. A high K-factor (e.g., K=10) indicates a strong direct path and less severe fading.
• AMC Context: Classifiers trained purely on Rayleigh fading often fail in Rician environments. The non-zero mean of the amplitude distribution shifts feature statistics.
• Scenario: Suburban macro-cells and satellite links typically exhibit Rician fading.

A worst-case multipath model with no dominant LOS component. The received signal envelope follows a Rayleigh distribution, characterized by deep fades of 30-40 dB.

• Assumption: Dense scattering environment where all paths are independent and uniformly distributed in angle.
• AMC Challenge: Deep fades drop the instantaneous SNR to near-zero, causing catastrophic misclassification bursts. Temporal averaging or diversity combining is essential.
• Feature Impact: Higher-order cumulants become unreliable during deep fades.

The frequency range over which the channel response is considered flat (highly correlated). It is inversely proportional to the delay spread.

• Flat Fading: Signal bandwidth < Coherence Bandwidth. All frequencies fade simultaneously. Simpler to equalize.
• Frequency-Selective Fading: Signal bandwidth > Coherence Bandwidth. Different frequencies fade independently, causing severe ISI.
• AMC Design Rule: For wideband signals, a classifier must learn frequency-diverse features or rely on pre-equalized IQ samples.

The time difference between the arrival of the first significant multipath component and the last. Quantifies the temporal dispersion of the channel.

• Typical Values: Indoor (50-200 ns), Urban (1-10 µs), Hilly Terrain (up to 50 µs).
• ISI Severity: If delay spread exceeds ~10% of the symbol period, ISI becomes a dominant error source.
• AMC Training: Synthetic datasets must model realistic delay spread profiles (e.g., ITU Vehicular A) to ensure robust classification.

The spectral broadening caused by time-varying multipath due to relative motion between transmitter and receiver. Determines the channel's coherence time.

• Fast Fading: Coherence time < Symbol period. The channel changes within a single symbol, causing irreducible error floors.
• AMC Impact: Doppler shifts rotate the constellation and introduce inter-carrier interference in OFDM. Classifiers must learn Doppler-invariant features or use pilot-based channel estimation.
• Max Doppler: Calculated as f_d = v/λ, where v is velocity and λ is wavelength.