Abstract
The vast bandwidths of millimeter-wave (mmWave) FR2 bands (24-71 GHz) and emerging FR3 bands (7-24 GHz) are critical for high-speed 5G/6G connectivity. However, the high path loss at these frequencies necessitates narrow, directional beams, which are difficult to maintain for mobile users like connected vehicles. This paper explores "spectrum mobility"—the dynamic switching between frequency bands and beam directions—to optimize connectivity. By modeling the problem as a Partially Observable Markov Decision Process (POMDP) and applying Point-Based Value Iteration (PBVI), the authors develop a policy that balances the high-bandwidth potential of narrow mmWave beams with the robustness of wider cmWave beams. Numerical results demonstrate that this joint management approach significantly improves data rates and link reliability under mobility uncertainty.
Introduction: The Challenge of Mobile mmWave Connectivity
Next-generation wireless networks are pushing into higher frequency territories to meet the insatiable demand for data. While FR2 and FR3 bands offer massive bandwidth, they suffer from significant signal attenuation. To overcome this, base stations employ large antenna arrays to focus energy into narrow "pencil beams."
For mobile targets, such as vehicles cruising down a highway, these narrow beams present a significant challenge: alignment. Even slight movements or estimation errors can lead to a complete loss of signal. Traditional approaches focus on improving beam tracking within a single band. This paper proposes a more holistic strategy: leveraging the physical properties of different frequencies. Lower frequencies naturally produce wider beams that are easier to track but offer less bandwidth, while higher frequencies provide massive bandwidth through narrower, more fragile beams. The goal is to "cruise" across these frequencies to maximize throughput while minimizing the risk of disconnection.
Methodology: A POMDP Approach to Spectrum Mobility
The authors formulate the joint frequency and beam selection problem as a Partially Observable Markov Decision Process (POMDP). In this framework, the base station does not know the user's exact location (the hidden state) but maintains a "belief"—a probability distribution over possible locations—based on observed Signal-to-Noise Ratio (SNR) values.
The system's actions involve choosing both a frequency band and a beam direction. A critical hardware assumption is the use of a shared-aperture antenna array. Since the physical size of the array is fixed, higher frequencies (with shorter wavelengths) allow for more antenna elements to be packed into the same space, resulting in higher gain and narrower beams. This coupling means that choosing a frequency inherently dictates the beam's width and reach.
To find an optimal policy, the authors utilize the Point-Based Value Iteration (PBVI) algorithm. PBVI approximates the optimal solution by focusing on a set of representative belief points, allowing the base station to make decisions that balance immediate data rate rewards with the need to "explore" and refine its belief about the user's location.
Key Results and Insights
The proposed "Spectrum Mobility" policy was evaluated against single-frequency baselines using Monte Carlo simulations. The results highlight several key advantages of the dynamic approach.
- Superior Throughput: As shown in the performance graphs, the spectrum mobility policy consistently outperforms static single-frequency allocations. By intelligently switching bands, it achieves a higher average rate across varying levels of mobility uncertainty ($p$).
- Adaptive Spectrum Usage: The system adapts its channel utilization based on the environment. In high-uncertainty scenarios, the system frequently reverts to the "safer" 15 GHz band to maintain a stable link. As uncertainty decreases, it aggressively shifts to the 60 GHz band to capitalize on high bandwidth.
- Robustness to Model Mismatch: The policy proved resilient even when the user's actual movement (constant speed) differed from the stochastic model used for training. This suggests that the POMDP-based approach captures fundamental trade-offs that generalize well to real-world conditions.
Conclusion and Future Directions
This work demonstrates that treating frequency as a dynamic resource for uncertainty management—rather than a static assignment—is vital for robust mobile connectivity in the mmWave era. By jointly managing spectrum and antenna arrays, networks can provide high-speed links that remain stable despite user movement.
Future research will look into integrating Reinforcement Learning to adapt to unknown mobility patterns in real-time. Additionally, the authors plan to extend the framework to multi-user scenarios and develop more computationally efficient online algorithms to handle the complexities of larger-scale deployments.
Summary updated