Discussion and Limitation¶
To have a realistic representation of the current cellular networks, we used publicly available datasets. Hence, we are limited with the limitations of the data. While national bodies are more reliable source of information on the MNOs’ infrastructures compared to crowdsourced datasets, the datasets might be incomplete, e.g., mobile coverage points to offer rural area coverage not being recorded in the dataset. Another caveat is that we had to make some assumptions and simplifications about the operation of these networks. For instance, we have not considered thoroughly the interference management and coordination approaches which are typically applied by the network operators. In real operation, there are many knobs (e.g., from interference cancellation to power adaptation) that would change the SINR and hence the channel capacity and consequently the achieved satisfaction of the user. Hence, our results should be interpreted with these shortcomings in mind.
We used the 3GPP path loss models to model the signal loss in different areas. However, our models only use an independent path loss probability for every link. Path loss by buildings or blockers is not independently distributed for every link and this affects the signal propagation. Nevertheless, we believe that our analysis maintains a good balance between realism and tractability.
To mitigate the time complexity while assessing the quality and resilience in a certain area such as a province or municipality, we considered each area as an isolated network. More particularly, the users, the BSs that are within the borders of that particular region and the BSs within 2000 meters of this border are considered. Consequently, users at the areas close to the borders of these regions can connect to the BSs in the neighboring regions. Since we have not considered the users of that neighboring region that would be connecting to those BSs, our performance results might be overestimating the reality.
Due to the limited availability of data about each MNO, we have assumed an equal user distribution among MNOs, which may be unrealistic. We chose not to use market reports that indicate the users shares per MNO, as it is still not public how mobile virtual operators use the physical operators’ network. Moreover, we have assumed that the number of users in a certain region is proportional to the population of that region. However, reality could differ from this distribution as the number of cellular users in an area depends also on business or social activities of that region, e.g., a hotspot business area attracting many people from other regions. Moreover, our analysis considered only a throughput and coverage perspective. For 5G networks, there are various new applications whose performance is assessed by other metrics such as packet loss or latency. To have a more rigorous understanding of the national cellular network performance and to identify the regions that should be prioritized in service enhancement, national agencies (e.g., Rijksinspectie Digitale Infrastructuur in the Netherlands or Ofcom in the UK) can maintain maps of cellular network availability and speed performance similar to the maps for broadband connectivity [5].
Finally, more knowledge on the temporal dynamics of a cellular network, e.g., number of served users during peak and off-peak hours, and application requirements can provide a more realistic performance assessment of the networks. For example, nation-wide crowd-sensing campaigns such as [2] could help with collecting this data by reflecting the users’ experience on different geographies including rural areas and with all cellular operators in the proximity of the user. These on-site measurements can be an input to both the national coverage maps and to MNOs for their network planning and assessing the potential benefits of infrastructure sharing. To show the benefits of network sharing, one could also investigate how much the number of resources in a shared network can be decreased compared to the situation nowadays. As MNOs typically over-provision their network to ensure user satisfaction even in case of a failure, network sharing can help in less over-provisioning and therefore lower energy usage.
To measure the full resilience of a cellular network under correlated failures caused by a disaster such as an earthquake or a flood, one cannot forget the backbone. We did not model the backbone and therefore our results only show a bestcase scenario: failures in the backbone network only degrades the performance, e.g., increasing FDP and decreasing FSP. Investigating and quantifying the effect of such correlated failures is therefore an interesting topic for further research.
为了尽可能真实地反映当前蜂窝网络的运行情况,我们采用了公开可用的数据集。因此,我们的研究结果也受到这些数据集本身局限性的约束。尽管相较于众包数据,国家级数据源在描述运营商基础设施方面更加可靠,但这些数据集依然可能存在不完整的情况。例如,某些用于覆盖农村区域的移动通信站点可能未被完整记录在内。
此外,我们在网络运行机制方面也做出了一些简化与假设。例如,本文未深入考虑运营商在实际部署中普遍采用的干扰管理与协同机制。现实中的网络运营涉及多种调节手段,如干扰消除、功率自适应等,这些都会对信噪比(SINR)产生影响,从而影响信道容量及最终的用户满意度。因此,本文的研究结论需在意识到上述简化前提的基础上加以解读。
在建模信号衰减方面,我们采用了 3GPP 的路径损耗模型用于不同区域的传播建模。但我们的模型仅考虑每个链路独立的路径损耗概率,未能涵盖建筑物或遮挡物带来的空间相关性衰减,而这些在实际中是非独立的。这一点会影响信号传播的准确性。尽管如此,我们认为本文在真实性与模型可计算性之间实现了较为合理的平衡。
为了降低在区域层面(如省或市)评估通信质量与韧性时的计算复杂度,我们将每个区域视为一个独立网络,具体而言,仅考虑该区域边界内以及边界外 2000 米范围内的用户与基站。因此,处于边界附近的用户仍可连接邻近区域的基站。但由于我们未考虑来自邻区的用户对这些基站的连接需求,因此可能导致性能评估结果存在一定程度的高估。
由于缺乏关于各运营商精确的用户分布数据,我们假设用户在不同运营商之间均匀分布,然而这一假设在现实中未必成立。我们也未采用市场报告中提供的用户市场占比数据,原因在于尚不明确虚拟运营商如何使用实体运营商的网络资源。此外,我们假设区域内用户数量与该区域人口数量成正比,但在现实中用户密度还受到商务活动或社交活动影响。例如,一个商业热点区域可能会吸引大量外来用户。本文的分析仅从吞吐量与覆盖率的角度出发,而对于 5G 网络而言,许多新兴应用场景还需关注其他性能指标,例如丢包率或时延等。
为了更全面地了解国家蜂窝网络的性能表现,并识别应优先提升服务质量的区域,建议由国家机构(如荷兰的 Rijksinspectie Digitale Infrastructuur 或英国的 Ofcom)负责维护类似于宽带连接覆盖图的蜂窝网络可用性与速度地图 [5]。
此外,若能进一步掌握蜂窝网络的时序动态信息,如高峰与非高峰时段的服务用户数量,以及各类应用的性能需求,将有助于提升网络评估的现实性。例如,参考文献 [2] 中提出的全国范围的用户众包测量计划,可以帮助收集反映用户在不同地理区域(尤其是农村地区)与不同运营商网络下的真实体验数据。这些实地测量数据不仅可用于完善国家覆盖地图,也可为运营商的网络规划提供依据,并辅助评估基础设施共享所带来的潜在益处。
为了展示网络共享的优势,还可进一步研究在共享网络架构下,可减少多少资源投入而仍维持当前的服务质量。由于运营商通常会进行过度部署以保证在故障时的用户体验,网络共享有望缓解这一过度部署的需求,从而降低能源消耗。
最后,要全面评估蜂窝网络在重大自然灾害(如地震或洪水)引发的相关区域性故障下的韧性,还必须考虑骨干网的作用。由于本文未对骨干网进行建模,故当前分析结果可视为理想最优情形。在实际中,骨干网络的故障将进一步加剧性能退化,例如 FDP 增加、FSP 降低。因此,系统性地研究和量化此类相关故障对整体网络的影响,是未来研究的重要方向。