[论文解读] Data Augmentation and Attention for massive MIMO-based Indoor Localization in Changing Environments
该论文提出两种数据增强技术以模拟被屏蔽的天线,并在基于CSI的室内定位的CNN/DNN中加入注意力模块,在静态数据经过随机衰减增强后训练,在环境变化时实现66 mm的平均误差
The demand for high-precision indoor localization has grown significantly with the rise of smart environments, industrial automation, and location-aware applications. While massive Multiple-Input and Multiple-Output (MIMO) systems enable millimeter-level accuracy by leveraging rich Channel State Information (CSI), most existing solutions are optimized for static environments, where users or devices remain fixed during data collection and inference. Real-world applications, however, often require real-time localization in changing environments, where rapid movement, unpredictable blockages, and dynamic channel conditions pose significant challenges. To address these challenges, we introduce two data augmentation techniques designed to resemble blocked antennas, enhancing the generalizability of localization models to dynamic scenarios. Additionally, we enhance an existing Deep Learning (DL) model by incorporating attention modules, improving its ability to focus on relevant channel features and antennas. We train our model on data from a static scenario, augmented with the proposed techniques, and evaluate it on a dataset collected in changing scenarios. We investigate the performance enhancements achieved by the data augmentation techniques and the Attention modules, and observe a localization accuracy improvement from a mean error of 286 mm, when trained without Attention and without data augmentations, to 66 mm, when trained with Attention and data augmentation. This shows that high localization accuracy can be maintained in changing environments, even without training data from those scenarios.
研究动机与目标
- 在动态环境中推动高精度室内定位,并解决从静态训练数据的泛化问题。
- 开发模拟被屏蔽天线和非视距(NLoS)效应的数据增强方法,以应对变化环境。
- 在Massive MIMO CSI中融入注意力机制,聚焦于有信息量的天线和子载波。
- 评估经增强的静态训练模型在变化环境数据上的鲁棒性。
- 提供可复现实验的开源代码。
提出的方法
- 将CSI表示为来自64天线线阵列的64×100复杂矩阵和100个子载波。
- 引入两种数据增强方案:(i) Vanilla: 在一个子集天线上随机置零CSI;(ii) Random Attenuation: 在一个子集天线上随机衰减CSI。
- 通过在现有深度学习定位模型中插入两个在子载波和天线层面运行的注意力模块来增强模型(先在子载波上注意,再在天线上注意)。
- 在静态环境数据上使用所提出的技术进行增强并训练模型,然后在变化环境数据上进行评估。
- 在不同增强策略下比较基线DenseNet(DN)与AttentionDenseNet(ADN)。
- 提供一个开源实现链接。
![Figure 1 : Our proposed enhancement of the DNN in [ 1 ] . We insert two Attention modules, shown in pink, to enable the model to focus on specific antennas and subcarriers. The rest follows the original model architecture.](https://ar5iv.labs.arxiv.org/html/2602.12954/assets/x1.png)
实验结果
研究问题
- RQ1数据增强是否能通过模拟被屏蔽天线来提升CSI基于室内定位在变化环境中的泛化能力?
- RQ2在子载波和天线上的注意力机制是否进一步提升动态信道条件下的定位性能?
- RQ3Vanilla与Random Attenuation两种增强对静态与变化环境性能的比较影响?
- RQ4在未见的动态场景中,哪种架构和增强的组合效果最好?
主要发现
| Model | Data augmentation | Test error (mm) |
|---|---|---|
| DN | None | 6 |
| ADN | None | 8 |
| DN | Vanilla | 15 |
| ADN | Vanilla | 18 |
| DN | RA | 4 |
| ADN | RA | 8 |
- 在静态场景中,DN的平均误差为6–8 mm,ADN为4–8 mm,具体取决于增强方式。
- 使用随机衰减增强且结合ADN时静态泛化最好(DN:RA得到4 mm;ADN:RA得到8 mm)。
- 在变化环境中,ADN在所有增强下的表现均优于DN;随机衰减配合ADN在变化数据上实现66 mm的平均误差。
- 上限参考(直接在变化数据上训练)对DN和ADN的平均误差为12–13 mm,表明在缺乏目标域数据的情况下仍有提升空间。
- 总体而言,注意力在未见变化中有助于降低误差;在增强方案中,随机衰减增强提供最强的一般化能力。
- 最佳模型(ADN+随机衰减)在变化场景中实现66 mm的平均误差,而在没有注意力或增强时为286 mm。
![Figure 2 : The setting of the nomadic dataset [ 3 ] . The black stars show the user positions, in blue the ULA , and in orange six trajectories of a human walking, each defining a new scenario in which data is collected. For the static scenario, no human is moving.](https://ar5iv.labs.arxiv.org/html/2602.12954/assets/fig/data.png)
更好的研究,从现在开始
从论文设计到论文写作,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。