Interference Mitigation via Rate-Splitting and Common Message Decoding in Cloud Radio Access Networks

TL;DR

This paper introduces a novel interference mitigation strategy in cloud radio access networks using rate-splitting and common message decoding, optimizing transmission schemes under practical constraints to significantly improve network performance.

Contribution

It proposes a new joint optimization framework for rate-splitting, common message decoding, clustering, and beamforming in C-RAN, addressing a complex non-convex problem with advanced relaxation techniques.

Findings

Significant performance gains over traditional interference mitigation methods.

Effective optimization approach for joint transmission scheme design.

Enhanced interference management in dense C-RAN deployments.

Abstract

Cloud-radio access networks (C-RAN) help overcoming the scarcity of radio resources by enabling dense deployment of base-stations (BSs), and connecting them to a central-processor (CP). This paper considers the downlink of a C-RAN, where the cloud is connected to the BSs via limited-capacity backhaul links. The paper proposes splitting the message of each user into two parts, a private part decodable at the intended user only, and a common part which can be decoded at a subset of users, as a means to enable large-scale interference management in CRAN. To this end, the paper optimizes a transmission scheme that combines rate splitting (RS), common message decoding (CMD), clustering and coordinated beamforming. The paper focuses on maximizing the weighted sum-rate subject to per-BS backhaul capacity and transmit power constraints, so as to jointly determine the RS-CMD mode of transmission, the cluster of BSs serving private and common messages of each user, and the associated beamforming vectors of each user private and common messages. The paper proposes solving such a complicated non-convex optimization problem using $l_0$-norm relaxation techniques, followed by inner-convex approximations (ICA), so as to achieve stationary solutions to the relaxed non-convex problem. Numerical results show that the proposed method provides significant performance gain as compared to conventional interference mitigation techniques in CRAN which treat interference as noise (TIN).

Full text

I Introduction

I-A Overview

Motivated by the scarcity of radio resources and the ever increasing need for higher data rates and reliable wireless services, C-RAN provides a practical network architecture capable of boosting the spectral and energy efficiency in next generation wireless systems (5G and beyond) [6824752, Le2015, 6736747]. By connecting many BSs to the CP, C-RANs enable spatial reuse through dense deployment of small cells, and exploit the emerging cloud-computing technologies for managing large networks [6923535, 7487951].

With ultra dense deployment of small cells, the distance between the base station (BS) and the end user decreases, which results in a better quality of the direct channel. This comes, however, at the cost of increasing inter-BS interference due to proximity of the BSs in neighbouring cells. Furthermore, in C-RAN, the performance of the system is also limited by the finite capacity of backhaul links, [6786060, 7096298, 7105959, 7488289, 7499119, 7593140, 7925639]. Intuitively, in the extreme case when the backhaul capacity goes to infinity, the C-RAN is equivalent to a broadcast channel (BC). In the other extreme in which the backhaul links have zero-capacity, the C-RAN becomes equivalent to an interference channel (IC), the capacity of which is still a well-known open problem, even for the simple two-user IC, where treating interference as noise (TIN) is known to be a suboptimal strategy, especially in high-interference regimes [Carleial78, 1056307, 4675741, 7422790]. With limited backhaul capacity, C-RAN bridges the two extremes. With this observation in mind, we investigate in this paper a transmission scheme which improves the performance of C-RAN in different regimes, i.e., in backhaul limited regimes and interference limited regimes.

In the rate splitting strategy, initially introduced by [Carleial78] for the IC, the message of each user is split into two parts: a private part decodable at the intended user only and a common part which can be decoded by other user. Such a strategy is shown to approach the capacity region of the IC in the seminal works of [1056307, 4675741]. Motivated by this fact, this paper studies rate-spitting in the realm of a C-RAN. It proposes splitting the message of each user into two parts, a private part decodable at the intended user only, and a common part which can be decoded at a subset of users.

Since the CP is connected to the BSs in cloud-enabled networks, C-RAN becomes a particularly suitable platform for the physical implementation of rate-splitting strategies. In the context of our paper, all rate splitting and common message decoding (RS-CMD) techniques are adopted for the sole purpose of reducing large-scale interference. As the CP is connected to the BSs via finite capacity backhaul links, it becomes equally important to determine the set of BSs (i.e., cluster) which serves each user, jointly with selecting the mode of transmission of each user (i.e., private, common, or both).

This work considers the RS-CMD problem in the downlink of a C-RAN, where the CP is connected to several BSs, each equipped with multiple antennas. The CP applies central encoding to user’s messages and establishes cooperation between a cluster of BSs by joint design of linear precoding in a user-centric clustering fashion, also known as data-sharing strategy [Simeone2009, 6588350, 5997324, 6920005], as it achieves a better performance compared to classical transmission schemes [7809154]. The paper then considers the problem of maximizing the weighted sum-rate (WSR) across the network, subject to per-BS backhaul capacity and transmission power constraints. The goal of this optimization is to jointly determine the RS-CMD mode of transmission, the cluster of BSs serving private and common messages of each user, and the associated beamforming vectors of each user private and common information. The paper provides an in-depth numerical investigation of the impact of RS-CMD strategy on the achievable rate in C-RANs, and compares it with the conventional strategies which treat interference as noise.

I-B Related Work

The contributions of this paper are related to works on rate splitting and common message decoding, clustering, and beamforming; topics which are studied in the literature of wireless systems, both individually and separately.

In rate-splitting schemes, the data of each user is divided into two parts: a private message which is decoded only at the intended user, and a common message which is decodable at the intended user and a subset of the unintended users. Reference [1056307] shows that such a RS-CMD technique leads to the largest known achievable rate-region in a 2-user IC. Such splitting strategy is further shown in [4675741] to achieve rates within one-bit from the capacity of the 2-user IC. Although being based on simple networks, those information-theoretical studies show the benefits of using RS-CMD techniques in high interference regimes. For instance, inspired by the theoretical works in [1056307, 4675741], the authors in [5910112] generalize this RS-CMD scheme to a practical multi-cell network showing significant achievable rate improvement by jointly designing the beamforming vectors for private and common information in RS-CMD as compared to beamforming design using TIN. In [6125368], the authors apply RS ideas to a practical setup of heterogeneous wireless networks. The results in [6125368] suggest that a significant performance gain can be reached by applying RS as compared to rank-1 coordinated beamforming schemes that adopt TIN strategy. The work in [7101259] uses common message decoding and successive interference cancellation techniques to maximize the sum rate in multi-cell multi-user MIMO system. The difference of convex optimization technique is used to efficiently solve the difficult underlying optimization problem. Recently, RS-CMD has also gained a noteworthy attention in the literature of medium access schemes. For instance, the authors in [journals/corr/abs-1710-11018] propose a novel RS multiple access (RSMA) scheme, which generalizes and outperforms conventional multiple access schemes such as Space-Division Multiple Access (SDMA) and Non-Orthogonal Multiple Access (NOMA). Based on these results, the authors in [8491100] show that RSMA is more energy efficient than SDMA and NOMA. Reference [2018arXiv180205567M], on the other hand, shows that linearly precoded RS is more efficient than the conventional Multi-User Linear Precoding (MU-LP) in terms of spectral and energy efficiency. Through numerical simulations, the authors in [2018arXiv180205567M] particularly show that, with no increase in receiver complexity, RS achieves better performance metrics as compared to both NOMA and MU-LP systems. The above works, i.e., [1056307, 4675741, 5910112, 6125368, 7101259, journals/corr/abs-1710-11018, 8491100, 2018arXiv180205567M], however, do not address cloud-enabled scenarios, as they ignore the physical-layer considerations induced by RS-CMS in C-RANs, and do not account for determining the set of common messages. This paper, therefore, focuses on the study of the joint resource allocation problem in C-RAN, together with evaluating the impact of RS-CMD techniques. The paper further develops a well-chosen heuristic procedure to determine the set of common messages that each user needs to decode.

In general, most of the existing works (e.g., [4443878, 5463229, 5594708, 5935083, 5962539, 6512541]) on multi-cell interference mitigation in practical networks focus on doing so through jointly allocating resources (e.g., beamforming vectors and transmit power) in order to maximize a network utility. References [4443878, 5463229, 5594708, 5935083, 5962539, 6512541], however, often adopt the strategy of TIN and assume an infinite backhaul capacity. Towards this end, the impact of finite backhaul links capacity is studied in the downlink of C-RAN in [6920005]. The problem studied in [6920005] turns out to be a mixed-integer non linear problem (MINLP), which is solved by relaxing the discrete non-convex per-BS backhaul constraints using re-weighted $l_{1}$-norm, and then by applying a generalized weighted minimum mean square algorithm (WMMSE). The authors in [6786060] consider the joint design of BSs’ clusters and beamforming vectors to minimize the network-wide transmit power cost. The trade-off between the backhaul traffic and transmit power is also investigated in references [6920005, 7105959, 7488289, 7499119, 7593140, 7925639], all of which adopt TIN to decode the received messages. At this point, it becomes essential to investigate how adopting RS-CMD can influence the design of clusters of BSs and the beamforming vectors associated with the private and common messages in a C-RAN setup. Towards this end, our current paper investigates the downlink C-RAN by utilizing a RS-CMD strategy, and focuses on evaluating its impact on jointly optimizing the beamforming vectors, the clustering and the transmission mode, so as to maximize the weighted-sum rate (WSR) across the network. To the best of authors’ knowledge, this is the first work on C-RAN which studies both the application of RS-CMD coupled with joint clustering and beamforming, and numerically illustrates the potential gain provided by RS-CMD over TIN.

I-C Contributions

In this paper, we propose using RS-CMD in downlink C-RAN to jointly design user centric clusters of BSs, so as to explore RS-CMD benefits in large-scale interference management. We formulate a WSR maximization problem subject to per-BS backhaul capacity and per-BS transmit power constraints, so as to determine the RS splitting mode, the cluster of BSs which serves each user, and the beamforming vectors associated with the private and common messages parts. Such a problem is generally NP-hard due to its mixed discrete and continuous optimization nature, in additional to the non-convexity of the constraints. Our paper proposes solving such a problem using a heuristic based on $l_{0}$-norm approximation to tackle the discrete part, followed by a polynomial time algorithm based on inner convex approximations, so as to find a stationary solution to the resulting non-convex continuous problem. The paper subsequently shows the numerical benefits of the proposed RS-CMD scheme in improving the achievable rates in C-RAN compared to the state-of-the art TIN strategy, both in the backhaul-limited and in the interference-limited regimes. Our main contributions can be summarized as follows:

• •

Common Message Decoding (CMD) Set: We propose a heuristic procedure for ordering the set of strongest interferers for each user, which consequently allows for determining the set of common messages to be decoded.

• •

Clustering: Based on $l_{0}$-norm relaxation and inner-convex approximation framework, we propose a dynamic clustering approach. In the context of RS-CMD, we determine the set of BSs serving the private message and the set of BSs which serves the common message for each scheduled user. As opposed to the static clustering scheme described in [6920005], dynamic clustering procedure forms the clusters by taking into account the CMD set of each user, which can significantly affect the network connectivity. To deal with the non-convex backhaul constraint, the paper particularly proposes a surrogate convex function to approximate the backhaul constraint. The paper then compensates for such approximations using proper outer-loop updates in an iterative manner.

• •

Beamforming: Even when the clusters are fixed, the WSR problem with RS-CMD is not convex. This is because the private and common rate functions are non-convex in the private and common beamforming vectors, respectively. The paper, therefore, proposes solving such issue using an algorithm that applies well-chosen inner-convex approximations. The proposed algorithm is proven to converge in polynomial time to a stationary solution.

• •

Numerical Simulations: We show through extensive numerical simulations that our proposed solution outperforms the classical TIN in C-RAN. In both the interference limited and the backhaul limited regimes, we illustrate that RS-CMD makes a better use of the network resources in order to achieve higher rates as compared with TIN for different network parameters.

The rest of the paper is organized as follows. Section II illustrates the system model. Section III introduces the transmission scheme adopted in this work and formulates the WSR problem accordingly. The proposed solution is introduced in section IV. Section V presents the numerical simulations, and section VI concludes the paper.