The future high-speed cellular networks require efficient and high-speed handover mechanisms. However, the traditional cellular handovers are based upon measurements of target cell radio strength which requires frequent measurement gaps. During these measurement windows, data transmission ceases each time, while target bearings are measured causing serious performance degradation. Therefore, prediction-based handover techniques are preferred in order to eliminate frequent measurement windows. Thus, this work proposes an ultrafast and efficient XGBoost-based predictive handover technique for next generation mobile communications. The ML algorithm in general prefers 70–30% of training and test data, respectively. However, always obtaining 70% of training samples in mobile communications is challenging because the channel remains correlated within coherence time only. Therefore, collecting training samples beyond coherence time limits performance and adds delay; thus, the proposed work trains the model within coherence time where this fixed data split of 70–30% makes the model exceed coherence time. Despite the fact that the proposed model gets starved of required training samples, still there is no loss in predication accuracy. The test results show a maximum delay improvement of up to 596 ms with enhanced performance efficiency of 68.70% depending upon the scenario. The proposed model reduces delay and improves efficiency by having an appropriate training period; thus, the intelligent technique activates faster with improved accuracy and eliminates delay in the algorithm to predict mmWaves’ signal strength in contrast to actually measuring them.

中文翻译：

---

使用基于动态学习周期的高效技术改善基于 AI 的切换延迟

未来的高速蜂窝网络需要高效、高速的切换机制。然而，传统的蜂窝切换基于目标小区无线电强度的测量，这需要频繁的测量间隙。在这些测量窗口期间，每次数据传输都会停止，而测量目标轴承会导致性能严重下降。因此，首选基于预测的切换技术以消除频繁的测量窗口。因此，这项工作为下一代移动通信提出了一种超快速且高效的基于 XGBoost 的预测切换技术。ML 算法通常分别喜欢 70-30% 的训练和测试数据。然而，在移动通信中始终获得 70% 的训练样本具有挑战性，因为信道仅在相干时间内保持相关。因此，收集超出一致性时间的训练样本会限制性能并增加延迟；因此，建议的工作在相干时间内训练模型，其中 70-30% 的固定数据拆分使模型超过了相干时间。尽管所提出的模型缺乏所需的训练样本，但预测准确性仍然没有损失。测试结果显示，最大延迟改进高达 596 ms，性能效率提高 68.70%，具体取决于场景。所提出的模型通过具有适当的训练周期来减少延迟并提高效率；因此，