When faced with growing satellite-frequency congestion, Entarian System Engineer Nahal Maleki set out to address the underlying cause: satellite and terrestrial LTE/5G services are increasingly operating in neighboring frequency ranges. As this overlap expands, conventional spectrum monitoring techniques—such as Automatic Modulation Classification (AMC)—struggle to remain accurate in these more complex, interference-heavy environments. Through her work on NOAA’s Radio Frequency Interference Monitoring System (RFIMS) project, Nahal and her team were tasked with overcoming these limitations and developing a more resilient solution.
To meet this challenge, Nahal and her team built on prior research in deep-learning AMC and developed a new framework tailored to the demands of satellite and terrestrial spectrum monitoring. Their approach uses bandwidth-specific Convolutional Neural Network (CNN) models to classify signals directly from raw I/Q samples received by a radio receiver and spanning a wide range of occupied bandwidths. A key addition is support for Orthogonal Frequency-Division Multiplexing (OFDM) LTE/5G signals, including cyclic-prefix OFDM (CP-OFDM) and DFT-spread OFDM (DFT-s-OFDM), across a broad range of occupied bandwidths. This extension enables the framework to more effectively handle the diversity of signal bandwidths encountered in real environments. The team further validates the approach using real RFIMS measurements.
A key aspect of the framework is its use of bandwidth‑specific CNN models. Because CNNs are designed to recognize spatial patterns, the system first determines the occupied bandwidth of the received signal and then selects the corresponding CNN model. Each model specializes in signals with similar bandwidth characteristics, resulting in notably higher classification accuracy than a single CNN trained across all bandwidths. The selected network produces a probability distribution over all possible modulation types, and a Softmax decision layer chooses the highest‑probability class. As Nahal explains:
“Rather than relying on manually designed features, the CNN automatically learns the characteristics that best distinguish different modulation types, improving its ability to generalize across diverse signal conditions.”
Beyond improving accuracy, the framework is more robust to varying operational environments because it learns features directly from the waveform rather than depending on manually engineered descriptors. This reduces the need for extensive feature design and tuning when incorporating new modulation types or communication standards. Nahal notes:
“From an operational perspective, the framework provides fast, automated signal identification that can be integrated into spectrum monitoring systems such as the RFIMS, enabling continuous monitoring with minimal human intervention.”
As wireless communication technologies continue to evolve, spectrum environments are becoming both more congested and more dynamic. Automated signal classification will play an increasingly essential role in helping spectrum operators understand who is transmitting, what technologies are being used, and whether interference or unauthorized activity is occurring.
CNN‑based AMC offers a scalable path forward, adapting more readily to emerging communication standards than conventional feature‑based methods. This capability supports a wide range of applications, including spectrum monitoring, interference detection, spectrum sharing, regulatory enforcement, satellite communications, defense, and cognitive radio systems.
As Vir Thanvi, Senior Vice President of Integrated Solutions, states:
“Entarian is leading development of a CNN‑driven AMC framework that classifies satellite and terrestrial signals directly from raw I/Q observations, supporting a wide range of applications including spectrum monitoring, interference detection and mitigation, spectrum sharing, regulatory enforcement, satellite communications, defense, and cognitive radio systems—especially in congested and contested spectrum domains—and plans to continue future research with additional datasets and mission‑intent A.I.”
In the long term, AI‑enabled modulation classification could serve as a foundational capability for autonomous spectrum management systems, enabling monitoring platforms to continuously identify, characterize, and respond to rapidly changing RF environments with greater speed and accuracy than traditional approaches.
Upcoming Presentation at SmallSat 2026
Nahal will be presenting this work at the 40th Annual Small Satellite Conference in Salt Lake City, Utah, taking place August 23–26, 2026. This year’s conference highlights the technologies and people driving the evolution of satellite constellations and the broader small satellite ecosystem.
Representing Entarian’s RFIMS project, Nahal will share her team’s advances in modulation classification as part of the conference’s ongoing focus on innovation, collaboration, and the future of small satellite systems.
