Motivation

Classical, model-based techniques for resource management and coordination in wireless networks often rely on a priori assumptions that are too idealized for real deployment scenarios. These techniques generally fail to capture the broad range of parameters and channel access controls that impact network performance, especially in situations involving coexistence of heterogeneous technologies on a shared spectrum. Machine learning (ML) techniques, on the other hand, enable the adaptation of a wide range of control parameters (features), with no or minimal assumptions on the underlying channel and network dynamics. ML techniques have recently been studied and applied to numerous wireless systems, from sensor networks to 5G millimeter-wave (mmWave) transmissions. The stringent quality of service (QoS) requirements associated with certain use cases of 5G+ wireless systems (e.g., ultra-reliable low-latency communications and eMBB) along with the integration of edge computing (MEC) platforms into these systems have triggered intensive research on the use of ML for wireless communications. 

The overarching goal of our research endeavor is to explore ML techniques for fast protocol adaptation, resource allocation, and spectrum sharing in both mid-band (sub-6 GHz) as well as high-band (mmWave) wireless systems. Our research strategy is to combine model-free ML tools (“black box”) with robust and highly generic modeling assumptions (hence, a “grey box” overall design) so as to achieve computationally efficient real-time adaptation of complex wireless networks under limited information. We leverage the flexibility of ML techniques to incorporate a large set of diverse but observable inputs into the learning engine. In contrast to existing state-of-the-art ML-based techniques that over-emphasize performance over complexity, our proposed designs target fast/real-time adaptation. Specific themes in our ML-based research agenda include, but are not limited to:

• User Tracking and Link/Beamwidth Adaptation in 5G mmWave Systems 
• Interference Mitigation in Contention-based Wireless Networks 
• Intelligent Scheduling of Time/Frequency Resources in OFDMA Wireless Systems
• Misbehavior Detection and Protocol Compliance for Unlicensed Wireless Networks
• Lightweight Machine Learning for Efficient Frequency-Offset-Aware Demodulation

Related Publications:

1. Irmak Aykin, Berk Akgun, Mingjie Feng, and Marwan Krunz, “MAMBA: A multi-armed bandit framework for beam tracking in millimeter-wave systems,” accepted to appear in the Proc. of the IEEE INFOCOM 2020, Beijing, China, April 2020.
2. Mohammed Hirzallah and Marwan Krunz, "Intelligent tracking of network dynamics for cross-technology coexistence over unlicensed bands," accepted to appear in the Proc. of the IEEE ICNC 2020, Big Island, Hawaii, Feb. 2020.
3. Amir Hossein Yazdani Abyaneh, Mohammed Hirzallah, and Marwan Krunz, "Intelligent-CW: AI-based framework for controlling contention window in WLANs," in the Proc. of the IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN’19), Newark, Nov 2019.
4. Peyman Siyari, Hanif Rahbari, and Marwan Krunz, ''Lightweight machine learning for efficient frequency-offset-aware demodulation,'' IEEE Journal on Selected Areas in Communications (JSAC) - Special Issue on Machine Learning in Wireless Communications, vol. 37, no. 11, pp. 2544-2558, Aug. 2019.
5. Hanif Rahbari, Peyman Siyari, Marwan Krunz, and Jerry Park, "Adaptive demodulation for wireless systems in the presence of frequency-offset estimation errors," Proc. of the IEEE INFOCOM 2018 Conference, Honolulu, HI, 2018, pp. 1592-1600 (acceptance rate 19.2%).