Liquid State Machine Learning for Resource and Cache Management in LTE-U Unmanned Aerial Vehicle (UAV) Networks

Mingzhe Chen, Walid Saad, Changchuan Yin

TL;DR

This paper introduces a decentralized machine learning framework using Liquid State Machines (LSM) to manage resource allocation and content caching for UAV-based LTE-Unlicensed (LTE-U) networks. By predicting user demand and autonomously switching between licensed and unlicensed bands, the proposed approach significantly increases the number of users with stable data connections while converging much faster than traditional reinforcement learning methods.

What Problem Does This Paper Address?

Deploying Unmanned Aerial Vehicles (UAVs) as flying base stations is a promising way to provide on-demand coverage. However, two major bottlenecks exist:

  1. Fronthaul Constraints: The wireless link between the cloud/core network and the UAV often has limited capacity, slowing down content delivery.
  2. Spectrum Scarcity: Licensed bands are crowded. While LTE-U allows UAVs to tap into unlicensed 5GHz bands (shared with WiFi), managing this interference while ensuring "queue stability" (where data service rate meets or exceeds arrival rate) is incredibly complex.

Existing solutions often ignore the limited-capacity cloud-UAV link or fail to account for the highly dynamic nature of user content requests. This paper seeks to jointly optimize user association, spectrum allocation (licensed vs. unlicensed), and content caching to maximize the number of users with stable service.

Methods

The authors propose a distributed framework based on Liquid State Machines (LSM), a type of reservoir computing that uses spiking neural networks. The method is split into two functional parts:

  • Content Request Prediction: The cloud uses an LSM to predict the probability distribution of user requests based on "context information" (age, gender, device type, etc.). Because LSMs have an internal "liquid state" that evolves over time, they are exceptionally good at capturing temporal patterns in user behavior.
  • Self-Organizing Resource Allocation: Each UAV acts as an agent in an LSM-based reinforcement learning loop. The UAVs observe the network state and autonomously choose association and spectrum strategies.
  • Search-Based Optimization: Within the learning loop, the authors developed a specific search-based algorithm to find the optimal resource allocation for a given user association, which they mathematically proved to be optimal.

Unlike standard Q-learning, which can be slow to converge in high-dimensional spaces, the LSM's "memory" allows it to map complex network states to optimal actions more efficiently.

Datasets

The study utilizes a real-world dataset from Youku (the "YouTube of China") via the China Network Video Index. This dataset provides the content request patterns used to train and validate the LSM’s prediction accuracy. The simulation environment models a 200m radius area with 20 users and varying numbers of UAVs (1 to 9).

Key Insights and Ideas

  • The "Liquid" Advantage: Standard neural networks often struggle with the "snapshot" nature of wireless states. LSMs treat the network state as a continuous flow, where past inputs influence current decisions. This "fading memory" is perfect for predicting content trends that shift throughout the day.
  • Caching as a Fronthaul Bypass: The paper demonstrates that caching popular content directly on the UAV isn't just a luxury—it's a necessity to overcome the "bottleneck" of the cloud-to-UAV link.
  • LTE-U Duty Cycling: To play nice with WiFi users, the UAVs use a "duty cycle" method, only occupying the unlicensed band for a fraction of the time. The LSM learns exactly how much of that fraction is needed to keep user queues stable without crushing the local WiFi performance.

Key Results

  • Stability Gains: The LSM approach yielded a 33.3% gain in the number of users with stable queues compared to Q-learning with caching, and a 50.3% gain over Q-learning without caching.
  • Faster Convergence: One of the most critical metrics for mobile networks is how fast the AI learns. The LSM reached convergence 33.3% faster than conventional Q-learning.
  • Prediction Accuracy: The LSM predicted content request distributions with an accuracy within 10% of the real-world data, outperforming Echo State Networks (ESN) by reducing mean squared error by 13.4%.
  • Resource Efficiency: The results showed that users on the licensed band were 50% more likely to have stable queues than those on the unlicensed band, highlighting the inherent challenge of the interference-heavy unlicensed spectrum.

Limitations and Future Work

  • Static Positioning: The paper assumes UAVs are deployed at fixed locations. In reality, UAVs are mobile, and optimizing their flight trajectories alongside caching and spectrum remains a wide-open challenge.
  • Uniform Content Size: The model assumes all content files are of equal size ($L$). Variable file sizes (e.g., short clips vs. 4K movies) would add another layer of complexity to the caching logic.
  • Energy Constraints: UAVs have limited battery life. Future iterations of this work would need to balance the computational cost of the LSM and the transmission power against the UAV's remaining flight time.

Summary updated

Abstract

In this paper, the problem of joint caching and resource allocation is investigated for a network of cache-enabled unmanned aerial vehicles (UAVs) that service wireless ground users over the LTE licensed and unlicensed bands. The considered model focuses on users that can access both licensed and unlicensed bands while receiving contents from either the cache units at the UAVs directly or via content server-UAV-user links. This problem is formulated as an optimization problem, which jointly incorporates user association, spectrum allocation, and content caching. To solve this problem, a distributed algorithm based on the machine learning framework of liquid state machine (LSM) is proposed. Using the proposed LSM algorithm, the cloud can predict the users' content request distribution while having only limited information on the network's and users' states. The proposed algorithm also enables the UAVs to autonomously choose the optimal resource allocation strategies that maximize the number of users with stable queues depending on the network states. Based on the users' association and content request distributions, the optimal contents that need to be cached at UAVs and the optimal resource allocation are derived. Simulation results using real datasets show that the proposed approach yields up to 17.8% and 57.1% gains, respectively, in terms of the number of users that have stable queues compared with two baseline algorithms: Q-learning with cache and Q-learning without cache. The results also show that the LSM significantly improves the convergence time of up to 20% compared with conventional learning algorithms such as Q-learning.
BibTeX
@article{chen2019liquid,
  author = {Mingzhe Chen and Walid Saad and Changchuan Yin},
  title = {{Liquid State Machine Learning for Resource and Cache Management in LTE-U Unmanned Aerial Vehicle (UAV) Networks}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2019},
  url = {https://openreview.net/forum?id=PEmRiII6Da_N},
}

Loading related content...

Figure