Machine Learning Enabled Spectrum Sharing in Dense LTE-U/Wi-Fi Coexistence Scenarios

Adam Dziedzic, Vanlin Sathya, Muhammad Iqbal Rochman, Monisha Ghosh, Sanjay Krishnan

This paper, authored by researchers at the University of Chicago, presents a machine-learning approach to a critical problem in wireless networking: how to make LTE and Wi-Fi play nice in the same frequency bands.

TL;DR

The researchers developed a deep-learning framework that allows LTE-Unlicensed (LTE-U) base stations to accurately count the number of nearby Wi-Fi networks by analyzing raw energy patterns. This enables LTE-U to adjust its transmission "duty cycle" for fair coexistence without the high cost and complexity of a full Wi-Fi receiver.

What Problem Does This Paper Address?

As mobile data demand skyrockets, cellular providers are moving into the unlicensed 5 GHz spectrum—traditionally the home of Wi-Fi. LTE-Unlicensed (LTE-U) coexists with Wi-Fi by using a "duty cycle" approach: it transmits for a period (ON) and then stays silent (OFF) to let Wi-Fi users through.

To be a "fair neighbor," an LTE-U base station must adjust its OFF time based on how many Wi-Fi networks (Basic Service Sets or BSSs) are active nearby. However, identifying the number of active Wi-Fi networks in real-time is difficult. Decoding Wi-Fi packets requires expensive hardware (a full Wi-Fi receiver inside an LTE base station), while simpler methods like Energy Detection (ED) are often too inaccurate in dense, noisy environments.

Methods

The authors treat the problem as a time-series classification task. Instead of decoding packets, they observe the energy values on the channel during the LTE-U "OFF" duration.

  • Baseline Comparison: They compared their approach against three standard methods: Header-Based (HD) decoding, Energy Detection (ED), and Autocorrelation (AC).
  • ML Architectures: They evaluated several models, including:
  • Fully Connected (FC) networks: Standard neural networks.
  • VGG: A modified version of the famous image-recognition CNN adapted for 1D time-series.
  • Fully Convolutional Networks (FCN): This proved to be the most robust, as it requires no specific feature engineering and excels at finding patterns in raw time-series data.
  • Optimization: To improve efficiency, they implemented FFT-based convolution with compression, reducing the computational load of the neural network while maintaining high accuracy.

Datasets

The researchers did not use a pre-existing dataset; they built a custom experimental testbed to generate real-world data.

  • Hardware: National Instruments (NI) USRP RIO boards were used to emulate the LTE-U base station and Wi-Fi access points.
  • Scenarios: They collected 3.4 GB of data covering scenarios with 0 to 5 active Wi-Fi APs.
  • Variables: The data includes variations in distance (6, 10, and 15 feet) and channel conditions, specifically Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) environments.

Key Insights and Ideas

  • Energy Fingerprinting: The core "aha moment" is that the distribution of energy over time contains a distinct "fingerprint" for the number of active transmitters. Even without reading the data inside the packets, the timing and overlap of energy bursts allow an ML model to distinguish between one, two, or five active networks.
  • Robustness in Density: Traditional methods like Energy Detection (ED) and Autocorrelation (AC) work reasonably well when only one or two Wi-Fi networks are present. However, their performance craters in "dense" scenarios (3+ APs). The ML models, particularly the FCN, remained highly accurate regardless of density.
  • Latency vs. Accuracy: The authors found that a "chunk size" of 512 samples provided the best trade-off, allowing for near-perfect accuracy with minimal inference delay.

Key Results

  • High Accuracy: The proposed ML approach achieved ~99% accuracy in identifying the number of Wi-Fi APs across almost all tested scenarios.
  • Superiority over Baselines: In dense scenarios (Case H to N), the accuracy of Energy Detection dropped as low as 56%, and Autocorrelation dropped to 63%, while the ML model maintained its 99% performance.
  • Reduced Delay: The ML-based system could detect and adapt the duty cycle in 3.1 seconds, significantly faster than Energy Detection (5.9s) or Autocorrelation (4.8s).
  • Compression: Using FFT compression, they could compress the data by 60% while maintaining over 99% accuracy for 2-3 classes.

Limitations and Future Work

While the results are promising, the authors acknowledge a few areas for further development:

  • Traffic Type Sensitivity: The current model counts the number of APs but doesn't necessarily distinguish between a "heavy" user (streaming 4K video) and a "light" user (sending a text). Future work aims to predict the type of Wi-Fi traffic (voice, video, etc.).
  • LTE-LAA Integration: This paper focuses on LTE-U (duty-cycling). The newer standard, LTE-LAA, uses "Listen-Before-Talk" (LBT), which presents a different set of coexistence challenges.
  • 6 GHz Band: With the recent opening of the 6 GHz band for unlicensed use, the authors suggest their ML framework will be essential for managing even more complex heterogeneous environments.

Summary updated

Abstract

The application of Machine Learning (ML) techniques to complex engineering problems has proved to be an attractive and efficient solution. ML has been successfully applied to several practical tasks like image recognition, automating industrial operations, etc. The promise of ML techniques in solving non-linear problems influenced this work which aims to apply known ML techniques and develop new ones for wireless spectrum sharing between Wi-Fi and LTE in the unlicensed spectrum. In this work, we focus on the LTE-Unlicensed (LTE-U) specification developed by the LTE-U Forum, which uses the duty-cycle approach for fair coexistence. The specification suggests reducing the duty cycle at the LTE-U base-station (BS) when the number of co-channel Wi-Fi basic service sets (BSSs) increases from one to two or more. However, without decoding the Wi-Fi packets, detecting the number of Wi-Fi BSSs operating on the channel in real-time is a challenging problem. In this work, we demonstrate a novel ML-based approach which solves this problem by using energy values observed during the LTE-U OFF duration. It is relatively straightforward to observe only the energy values during the LTE-U BS OFF time compared to decoding the entire Wi-Fi packet, which would require a full Wi-Fi receiver at the LTE-U base-station. We implement and validate the proposed ML-based approach by real-time experiments and demonstrate that there exist distinct patterns between the energy distributions between one and many Wi-Fi AP transmissions. The proposed ML-based approach results in a higher accuracy (close to 99% in all cases) as compared to the existing auto-correlation (AC) and energy detection (ED) approaches.
BibTeX
@article{dziedzic2020machine,
  author = {Adam Dziedzic and Vanlin Sathya and Muhammad Iqbal Rochman and Monisha Ghosh and Sanjay Krishnan},
  title = {{Machine Learning Enabled Spectrum Sharing in Dense LTE-U/Wi-Fi Coexistence Scenarios}},
  journal = {IEEE Open Journal of Vehicular Technology},
  year = {2020},
  doi = {10.1109/ojvt.2020.2981519},
}

Loading related content...

Figure