Channel Charting in Real-World Coordinates with Distributed MIMO

TL;DR

The paper tackles embedding channel charts into real-world coordinates without ground-truth UE positions by introducing two anchor losses: a bilateration loss $\mathfrak{L}_{\text{bi}}$ and an LoS bounding-box loss $\mathfrak{L}_{\text{box}}$, integrated with a timestamp-based triplet loss $\mathfrak{L}_{\text{t}}$ for robust chart quality. It demonstrates a weakly-supervised learning framework in a D-MIMO setting, relying on known AP locations and approximate LoS areas rather than geometric propagation models or labeled UE positions. Across simulated outdoor/indoor and measurement-based indoor datasets, the proposed P2 method (combining $\mathfrak{L}_{\text{bi}}$, $\mathfrak{L}_{\text{box}}$, and $\mathfrak{L}_{\text{t}}$) achieves comparable latent-space and positioning performance to semi-supervised baselines while avoiding ground-truth labels, and it highlights the limitations of affine-transform mappings when channel charts are non-affine. The work advances practical channel charting for applications like handovers, beam management, and network planning by providing real-world coordinate interpretations without heavy labeling or synchronization assumptions.

Abstract

Channel charting is an emerging self-supervised method that maps channel-state information (CSI) to a low-dimensional latent space (the channel chart) that represents pseudo-positions of user equipments (UEs). While channel charts preserve local geometry, i.e., nearby UEs are nearby in the channel chart (and vice versa), the pseudo-positions are in arbitrary coordinates and global geometry is typically not preserved. In order to embed channel charts in real-world coordinates, we first propose a bilateration loss for distributed multiple-input multiple-output (D-MIMO) wireless systems in which only the access point (AP) positions are known. The idea behind this loss is to compare the received power at pairs of APs to determine whether a UE should be placed closer to one AP or the other in the channel chart. We then propose a line-of-sight (LoS) bounding-box loss that places the UE in a predefined LoS area of each AP that is estimated to have a LoS path to the UE. We demonstrate the efficacy of combining both of these loss functions with neural-network-based channel charting using ray-tracing-based and measurement-based channel vectors. Our proposed approach outperforms several baselines and maintains the self-supervised nature of channel charting as it neither relies on geometrical propagation models nor on any ground-truth UE position information.

Figures (7)

• Figure 1: Simulated D-MIMO scenarios: (a) outdoor scenario with $14\,642$ UE positions and $A = 6$ APs; (b) indoor scenario with $14\,606$ UE positions and $A = 8$ APs. Each AP (green boxes) is equipped with $M_R = 4$ antennas.
• Figure 2: Receive power (in decibels) at one AP for the simulated outdoor scenario with respect to the AP-to-UE distance in the xy-plane.
• Figure 3: Results for the simulated outdoor scenario: (a) ground-truth UE positions (green-to-red gradient colored area), AP positions (blue triangles), and the LoS bounding box of all APs; and (b-f) the channel charts or positioning estimates for the proposed (P) and baseline (B) methods. Since the output of baseline B2 is in arbitrary coordinates, the AP positions are not shown in (d). The proposed method P2 achieves comparable results in real-world coordinates as the semi-supervised baseline B3 but without requiring known UE positions during training.
• Figure 4: Receive power (in decibels) at one AP for the simulated indoor scenario with respect to (a) timestamps and (b) AP-to-UE distance in the xy-plane. In (a), the red line designates the power threshold chosen to identify LoS APs. In (b), the power values above the threshold (highlighted with green) indicate instances where the AP is estimated to be in LoS.
• Figure 5: Results for the simulated indoor scenario: (a) ground-truth UE positions (green-to-red gradient colored area), AP positions (blue triangles), and the LoS bounding box of three APs (blue solid and dashed lines); and (b-f) the channel charts or positioning estimates for the proposed (P) and baseline (B) methods. Since the output of baseline B2 is in arbitrary coordinates, the AP positions are not shown in (d). The proposed method P2 achieves comparable results in real-world coordinates as the semi-supervised baseline B3 but without requiring known UE positions during training.