Deep learning cancels 6G interference with 60% less computational complexity
Researchers from Aristotle University and IMEC have developed a deep learning framework, NBI-CNet, designed to improve OFDM signal recovery in congested 6G FR3 spectrum environments. The physics-informed architecture reduces computational complexity by up to 60% compared to traditional compressive sensing methods while mitigating narrowband interference.
Key Takeaways
- NBI-CNet architecture achieves up to 60% reduction in computational complexity versus the state-of-the-art EOMP-IDS algorithm.
- Framework delivers a 33 dB coding gain under high spectral overlap by preventing traditional signal-peak confusion.
- Unified pipeline performs joint interference cancellation and soft demodulation without requiring prior knowledge of the interferer count.
- Scale-invariant design allows the neural network to generalize across different FFT sizes without additional retraining.
Why It Matters
Congestion in the 6G FR3 upper mid-band (7-24 GHz) creates severe interference between terrestrial networks and incumbent satellite services. This framework provides a path for high-capacity 6G data delivery without the heavy processing latency that historically plagued OFDM signal recovery. By shifting from rigid matrix-based calculations to adaptive deep learning, operators can maintain sub-millisecond reliability even in dense IoT and satellite-heavy environments. This tech enables the deep integration of non-terrestrial and cellular networks required for ubiquitous global streaming. Watch for pilot tests in combined satellite-cellular 6G testbeds scheduled for late 2026.
Additional Context
The development of NBI-CNet aligns with broader industry efforts to secure the 6G FR3 band, which is increasingly viewed as the backbone for high-performance wireless data. Per Nokia in January 2026, the 6-8 GHz range is critical for achieving the bandwidth required for 6G services, though the company warned that cellular and Wi-Fi coexistence in these frequencies risks substantial performance degradation. Similarly, the NTIA reported in May 2026 that significant headway has been made in studying the 7 GHz band for mobile use, following a 2025 executive memorandum aimed at identifying 600 MHz of licensed spectrum to support 6G leadership.
Technological innovation in this area is also moving toward AI-native network designs. According to Omdia in April 2026, more than 20% of European mobile operators are already trialing direct-to-device (D2D) satellite partnerships, emphasizing the need for robust interference mitigation as terrestrial and non-terrestrial networks converge. Industry projections from Juniper Research in June 2026 suggest that 6G connections will reach 4.1 million by 2029, scaling to 2.9 billion by 2035. As these networks scale, AI-driven solutions like NBI-CNet are expected to transition from research frameworks to standard components within the Radio Access Network (RAN).
IMEC's role in this research reflects its strategic pivot toward becoming a primary AI and semiconductor brand for the 6G era. In early 2026, IMEC CEO Luc Van den hove announced the establishment of imec.AI-labs to focus on agentic and physical AI that can adapt to dynamic environmental situations. This focus on adaptive learning matches the requirements of NBI-CNet, which is designed to handle non-stationary RF environments where interference parameters fluctuate symbol-to-symbol due to changing channel conditions and uncoordinated device behavior.
Read full article at arxiv.org
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source