Uncertainty-Aware Dimensionality Reduction for Channel Charting with Geodesic Loss

Florian Euchner,Phillip Stephan,Stephan ten Brink

2024-12-03

Abstract:Channel Charting is a dimensionality reduction technique that learns to reconstruct a low-dimensional, physically interpretable map of the radio environment by taking advantage of similarity relationships found in high-dimensional channel state information. One particular family of Channel Charting methods relies on pseudo-distances between measured CSI datapoints, computed using dissimilarity metrics. We suggest several techniques to improve the performance of dissimilarity metric-based Channel Charting. For one, we address an issue related to a discrepancy between Euclidean distances and geodesic distances that occurs when applying dissimilarity metric-based Channel Charting to datasets with nonconvex low-dimensional structure. Furthermore, we incorporate the uncertainty of dissimilarities into the learning process by modeling dissimilarities not as deterministic quantities, but as probability distributions. Our framework facilitates the combination of multiple dissimilarity metrics in a consistent manner. Additionally, latent space dynamics like constrained acceleration due to physical inertia are easily taken into account thanks to changes in the training procedure. We demonstrate the achieved performance improvements for localization applications on a measured channel dataset

Information Theory,Signal Processing

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to solve several key problems in wireless Channel Charting technology, especially the challenges encountered when dealing with data sets with non - convex low - dimensional structures. Specifically:

1. **Mismatch between Euclidean distance and geodesic distance**:
- In the existing pseudo - distance - based Channel Charting methods, when data points are non - convexly arranged in physical space, there are significant differences between the Euclidean distance and the geodesic distance. This will lead to distortion in the learned channel chart.
- To solve this problem, the author proposes to use the Geodesic Loss function to ensure that the actual length of the path is considered when calculating the loss.

2. **Introducing uncertainty modeling**:
- Existing methods treat the pseudo - distance as a deterministic quantity and ignore its uncertainty. This practice may lead to inaccurate estimation of certain path lengths by the model.
- The author suggests modeling the pseudo - distance as a probability distribution and incorporating its uncertainty into the learning process, thereby improving the robustness and accuracy of the model.

3. **Introducing acceleration constraint**:
- Current methods do not consider the influence of physical inertia on the motion patterns of data points, resulting in unrealistic motion patterns that may occur in the channel chart.
- The author proposes an Acceleration Constraint, which improves the learning process of the channel chart by simulating physical inertia.

4. **Improving the training architecture**:
- The existing Siamese neural network architecture only compares two data points at a time, which limits the performance of the model.
- The author proposes a new Batch - wise Loss Training Architecture, which can consider multiple data points simultaneously, thereby better capturing the overall structure of the data.

### Summary

Through these improvements, the author hopes to significantly improve the performance of Channel Charting in positioning applications, especially when dealing with data sets with complex structures. In addition, these improvements may also have a positive impact on other applications based on Channel Charting.

### Formula summary

1. **Acceleration constraint loss function**:
\[
L_{\text{acc}} = -\frac{1}{L} \sum_{l = 1}^{L} \ln \left( e^{-\frac{(\|a^{(l)}\| - \mu_{\text{acc}})^2}{2\sigma^2_{\text{acc}}}} + e^{-\frac{(\|a^{(l)}\| + \mu_{\text{acc}})^2}{2\sigma^2_{\text{acc}}}} \right)
\]
where \( v^{(l)}=\frac{z^{(l)} - z^{(l - 1)}}{t^{(l)} - t^{(l - 1)}} \) and \( a^{(l)}=\frac{v^{(l)} - v^{(l - 1)}}{t^{(l)} - t^{(l - 1)}} \) represent the instantaneous velocity and acceleration in the latent space respectively.

2. **Geodesic loss function**:
\[
L_{\text{geo}}=\sum_{(i,j)\in B} \left( \frac{\|\rho^{(i,j)}_{\text{geo}} - \Delta^{(i,j)}_{\text{geo}}\|^2}{\Delta^{(i,j)}_{\text{geo}}+\beta} \right)
\]
where \(\rho^{(i,j)}_{\text{geo}}=\sum_{m = 1}^{M(i,j)} \|z(q^{(i,j)}_{m - 1}) - z(q^{(i,j)}_m)\|\) is from data point \(i\) to