Session
Ground Systems
Location
Salt Palace Convention Center, Salt Lake City, UT
Abstract
Automatic modulation classification (AMC) is a critical capability for spectrum monitoring, interference detection, and signal identification in increasingly congested satellite and terrestrial communication environments. This paper presents a convolutional neural network (CNN)-based AMC framework that classifies analog, digital, and multicarrier communication signals directly from raw in-phase and quadrature (I/Q) samples. The proposed framework supports twelve modulation classes, including binary phase shift keying (BPSK), quadrature PSK (QPSK), 8PSK, 16 quadrature amplitude modulation (16QAM), 64QAM, pulse amplitude modulation (PAM)4, Gaussian frequency shift keying (GFSK), continuous phase FSK (CPFSK), broadcast FM (B-FM), single-sideband amplitude modulation (SSB-AM), double-sideband AM (DSB-AM), and orthogonal frequency-division multiplexing (OFDM).
To accommodate signals occupying a wide range of bandwidths, multiple bandwidth-specific CNN models are trained using synthetically generated waveforms and recorded analog signals. The training dataset consists of 5,000 waveform frames per modulation class and includes additive white Gaussian noise. LTE and 5G uplink signals are generated using both cyclic-prefix OFDM (CP-OFDM) and discrete Fourier transform spread OFDM (DFT-s-OFDM) waveform models. Each classifier operates on 40,000-sample I/Q frames and employs a 28-layer CNN architecture trained using stochastic gradient descent with momentum.
The proposed approach is evaluated using both simulated datasets and real measurements collected by the Radio Frequency Monitoring System (RFIMS) at the Table Mountain Test Range (TMTR) in Boulder, Colorado. The range is located at the National Telecommunications and Information Administration’s (NTIA) Institute for Telecommunication Sciences (ITS) Table Mountain Field Site and Radio Quiet Zone. The measurement dataset includes weather-satellite downlink signals from METOP-B, METOP-C, NOAA-18, NOAA-19, and SARAL satellites, as well as LTE and 5G uplink transmissions. RFIMS employs an eight-element antenna array with analog beamforming capability, enabling high signal to noise ratio (SNR) reception of satellite signals across the upper hemisphere.
Experimental results demonstrate that the proposed framework can reliably distinguish satellite and terrestrial communication signals over a wide range of operating conditions. OFDM waveforms exhibit the highest classification performance due to their distinctive multicarrier structure, while robust classification accuracy is also achieved for BPSK- and QPSK-based satellite downlink signals. Validation using real RFIMS measurements confirms the ability of the proposed AMC framework to correctly identify operational weather-satellite and cellular communication signals. The results demonstrate the suitability of the proposed approach for practical spectrum-monitoring, interference-detection, and coexistence-analysis applications in shared satellite-terrestrial communication environments.
Document Type
Event
Automatic Modulation Classification for Spectrum Sensing in Shared Satellite–LTE/5G Bands
Salt Palace Convention Center, Salt Lake City, UT
Automatic modulation classification (AMC) is a critical capability for spectrum monitoring, interference detection, and signal identification in increasingly congested satellite and terrestrial communication environments. This paper presents a convolutional neural network (CNN)-based AMC framework that classifies analog, digital, and multicarrier communication signals directly from raw in-phase and quadrature (I/Q) samples. The proposed framework supports twelve modulation classes, including binary phase shift keying (BPSK), quadrature PSK (QPSK), 8PSK, 16 quadrature amplitude modulation (16QAM), 64QAM, pulse amplitude modulation (PAM)4, Gaussian frequency shift keying (GFSK), continuous phase FSK (CPFSK), broadcast FM (B-FM), single-sideband amplitude modulation (SSB-AM), double-sideband AM (DSB-AM), and orthogonal frequency-division multiplexing (OFDM).
To accommodate signals occupying a wide range of bandwidths, multiple bandwidth-specific CNN models are trained using synthetically generated waveforms and recorded analog signals. The training dataset consists of 5,000 waveform frames per modulation class and includes additive white Gaussian noise. LTE and 5G uplink signals are generated using both cyclic-prefix OFDM (CP-OFDM) and discrete Fourier transform spread OFDM (DFT-s-OFDM) waveform models. Each classifier operates on 40,000-sample I/Q frames and employs a 28-layer CNN architecture trained using stochastic gradient descent with momentum.
The proposed approach is evaluated using both simulated datasets and real measurements collected by the Radio Frequency Monitoring System (RFIMS) at the Table Mountain Test Range (TMTR) in Boulder, Colorado. The range is located at the National Telecommunications and Information Administration’s (NTIA) Institute for Telecommunication Sciences (ITS) Table Mountain Field Site and Radio Quiet Zone. The measurement dataset includes weather-satellite downlink signals from METOP-B, METOP-C, NOAA-18, NOAA-19, and SARAL satellites, as well as LTE and 5G uplink transmissions. RFIMS employs an eight-element antenna array with analog beamforming capability, enabling high signal to noise ratio (SNR) reception of satellite signals across the upper hemisphere.
Experimental results demonstrate that the proposed framework can reliably distinguish satellite and terrestrial communication signals over a wide range of operating conditions. OFDM waveforms exhibit the highest classification performance due to their distinctive multicarrier structure, while robust classification accuracy is also achieved for BPSK- and QPSK-based satellite downlink signals. Validation using real RFIMS measurements confirms the ability of the proposed AMC framework to correctly identify operational weather-satellite and cellular communication signals. The results demonstrate the suitability of the proposed approach for practical spectrum-monitoring, interference-detection, and coexistence-analysis applications in shared satellite-terrestrial communication environments.
