Locally Orthogonal Training Design Based on Graph Coloring in Cloud-RANs
Abstract: In Cloud Radio Access Networks (C-RANs), the traditional approach of assigning globally orthogonal training sequences to users leads to a pilot overhead that scales linearly with the number of users. This "pilot contamination" problem poses a significant bottleneck for massive-scale 5G/6G deployments. This paper introduces a novel channel estimation strategy based on local orthogonality. By leveraging the spatial sparsity of interference and modeling the pilot assignment as a graph coloring problem, the authors demonstrate that the required training overhead scales only logarithmically ($O(\ln K)$) with the number of users. Numerical results validate this scaling and demonstrate significant throughput gains over conventional methods.
What Problem Does This Paper Address?
The architecture of a C-RAN involves a central Baseband Unit (BBU) pool coordinating hundreds or thousands of Remote Radio Heads (RRHs) to serve a massive user base. To perform effective beamforming and interference management, the BBU needs accurate channel state information (CSI), obtained via uplink training sequences (pilots) transmitted by users.
In traditional cellular systems, users are assigned unique, orthogonal sequences to avoid interference. However, the length of these sequences—and thus the training overhead—typically grows linearly with the number of users ($K$). In a dense C-RAN, this overhead can quickly consume the entire channel coherence time, leaving no resources for actual data transmission. The authors address this scalability crisis by rethinking the requirement for orthogonality.
Methods: From Global to Local Orthogonality
The core innovation lies in the observation that "interference is local." Due to path loss, an RRH is primarily affected by users in its immediate vicinity. Users located far away can reuse the same training sequences without causing significant pilot contamination.
The Graph Coloring Framework
The authors formalize this intuition through a multi-step process:
- Channel Sparsification: A neighborhood is defined for each RRH using a distance threshold $r$. Only users within this radius are considered potential interferers for that RRH.
- Conflict Graph Construction: A graph is built where each node represents a user. An edge is drawn between two users if they both fall within the neighborhood of at least one common RRH.
- Vertex Coloring: Assigning orthogonal sequences becomes a vertex coloring problem. Each "color" corresponds to a specific orthogonal sequence. The goal is to color the graph such that no two adjacent nodes share a color, using the minimum number of colors (the chromatic number).
- DSATUR Algorithm: Since optimal graph coloring is NP-complete, the authors utilize the DSATUR (Degree of Saturation) algorithm. This greedy heuristic provides near-optimal coloring with low computational complexity, making it suitable for real-time C-RAN coordination.
Key Insights: Logarithmic Scalability
The most profound theoretical contribution of this work is the proof that the minimum training length required for local orthogonality scales as $O(\ln K)$. This is a dramatic improvement over the $O(K)$ scaling of globally orthogonal designs. By exploiting the spatial distribution of users and the decaying nature of wireless signals, C-RANs can support orders of magnitude more users without a proportional increase in overhead.
Key Results
Numerical simulations were conducted to evaluate the proposed scheme in a random deployment of RRHs and users.
- Scaling Validation: As shown in the results, the training length required by the DSATUR algorithm follows the predicted logarithmic trend and remains strictly below the theoretical upper bounds derived from Random Geometric Graph (RGG) theory.
- Throughput Enhancement: The locally orthogonal scheme significantly outperforms globally orthogonal baselines. In high-SNR scenarios, it achieved over a 20% performance boost compared to random pilot assignment.
- Refined Sparsification: A "refined" version of the algorithm, which allows RRHs to estimate channels from additional nearby users if their sequences happen to be orthogonal, provided an additional 10% throughput gain.
Limitations and Future Work
While the results are promising, several practical challenges remain. The current model assumes an ideal, unlimited-capacity fronthaul, which may not be realistic for all deployments. Additionally, the distance threshold $r$ is currently a global constant; future research could explore adaptive, user-centric thresholds to further optimize performance. Finally, while the design is applicable to the downlink via channel reciprocity, practical implementation in Frequency Division Duplex (FDD) systems requires further investigation.
Summary updated