Abstract
This paper introduces two novel lattice-based coding schemes, Quantized-Compute-and-Forward (QCoF) and Jointly-Quantized-Compute-and-Forward (JQCoF), designed to optimize uplink transmission in Cloud Radio Access Networks (CRAN) with limited backhaul capacity. By integrating the denoising benefits of Compute-and-Forward (CoF) with lossy Wyner-Ziv compression, these schemes overcome the traditional limitations of CoF, particularly in dense networks where the number of users exceeds the number of base stations.
Introduction
In the evolving landscape of cellular networks, Cloud Radio Access Network (CRAN) architectures have emerged as a promising solution for centralized interference management. In a CRAN, multiple base stations (BSs) serve as simple relays, forwarding processed signals to a Central Processor (CP) for final decoding. However, the performance of such systems is strictly constrained by the finite capacity of the backhaul links connecting the BSs to the CP.
Existing strategies like Compress-and-Forward (CF) often waste backhaul bandwidth on noise, while standard Compute-and-Forward (CoF) requires a high number of base stations ($K \ge L$) and strict linear independence of equations to recover user messages. This paper addresses these challenges by proposing a hybrid approach that utilizes lossy compression on "denoised" linear combinations of signals.
Proposed Methodology
The authors propose two primary schemes based on nested lattice codes:
Quantized-Compute-and-Forward (QCoF)
In QCoF, each BS decodes a linear combination (an "equation") of the users' messages. Instead of attempting lossless transmission, which would require excessive backhaul capacity, the BS remaps the equation to the input symbol level and applies lossy compression. The CP then employs a multi-user decoder with Successive Interference Cancellation (SIC) to recover the original messages. This architecture relaxes the requirement for linear independence between equations at different BSs, making it suitable for scenarios where $L > K$.
Jointly-Quantized-Compute-and-Forward (JQCoF)
JQCoF is a generalization of QCoF where each BS jointly compresses both the decoded equation and the raw received signal. This allows the system to capture "innovation"—information present in the raw signal that was not captured by the decoded equation. By dynamically balancing these two components, JQCoF can adapt to varying backhaul capacities and channel conditions.
The selection of optimal integer coefficients for the equations is performed using the Lenstra-Lenstra-Lovász (LLL) algorithm, which provides a computationally feasible approximation for this NP-hard problem.
Performance Evaluation
The proposed schemes were evaluated using Monte Carlo simulations in a Gaussian CRAN model with $L=3$ users and $K=2$ base stations. Key findings include:
- Superiority of JQCoF: JQCoF consistently outperformed both Successive Wyner-Ziv (a standard CF approach) and QCoF across various backhaul capacities.
- Handling User Density: Unlike standard CoF, QCoF and JQCoF maintain high performance even when there are more users than base stations.
- Saturation and Bounds: While pure CoF-based methods tend to plateau as backhaul capacity increases, JQCoF continues to scale, approaching the theoretical cut-set upper bound of the network.
Conclusion
By combining the denoising properties of Compute-and-Forward with the efficiency of lossy compression, QCoF and JQCoF provide a robust framework for uplink processing in backhaul-constrained CRANs. These schemes offer significant gains in data rates and network flexibility, particularly in dense user environments where traditional relaying strategies fail. Future work will focus on optimizing coefficient selection beyond the LLL algorithm and evaluating performance under non-ideal channel conditions.
Summary updated