Source-linked AI summary
Joint Energy Minimization and Resource Allocation in C-RAN with Mobile Cloud
Kezhi Wang, Kun Yang, Chathura Sarathchandra Magurawalage
TL;DR
The paper addresses the limited integration of MCC and C-RAN by proposing a joint approach to reduce total energy under task-time constraints. It reports that joint energy minimization outperforms separate solutions, improving performance and saving energy.
Problem
Research integrating MCC and C-RAN is rarely addressed, motivating joint reduction of total energy under given task time constraints.
Method
The paper proposes a novel C-RAN structure and models the whole mobile-cloud and mobile-network energy cost as an optimization problem considering time constraints.
Results
The proposed joint energy minimization outperforms separate solutions and improves system performance while saving energy.
Takeaways & Limitations
A C-RAN architecture involving mobile clones jointly minimizes energy across mobile cloud and mobile network operations.
Abstract
from arXiv · showhide
Cloud radio access network (C-RAN) has emerged as a potential candidate of the next generation access network technology to address the increasing mobile traffic, while mobile cloud computing (MCC) offers a prospective solution to the resource-limited mobile user in executing computation intensive tasks. Taking full advantages of above two cloud-based techniques, C-RAN with MCC are presented in this paper to enhance both performance and energy efficiencies. In particular, this paper studies the joint energy minimization and resource allocation in C-RAN with MCC under the time constraints of the given tasks. We first review the energy and time model of the computation and communication. Then, we formulate the joint energy minimization into a non-convex optimization with the constraints of task executing time, transmitting power, computation capacity and fronthaul data rates. This non-convex optimization is then reformulated into an equivalent convex problem based on weighted minimum mean square error (WMMSE). The iterative algorithm is finally given to deal with the joint resource allocation in C-RAN with mobile cloud. Simulation results confirm that the proposed energy minimization and resource allocation solution can improve the system performance and save energy.
I. INTRODUCTION
The paper motivates integrating C-RAN and mobile cloud computing to address rising traffic, computation demands, and system energy costs. It proposes joint energy minimization and resource allocation under task-time constraints using a WMMSE-based iterative approach.
- Motivation: Rising mobile traffic and computation-intensive applications increase energy and resource pressures across mobile devices and networks.The introduction links traffic growth, demanding applications, and operator energy expenditure.
- C-RAN and MCC: C-RAN centralizes intensive network computation in a BBU pool while distributed RRHs provide simplified radio access and coordinated transmission.The architecture separates RRHs, the BBU pool, and fronthaul links, enabling centralized processing and cooperation.
- C-RAN and MCC: Mobile cloud computing allows resource-limited users to offload computation-intensive tasks to powerful cloud platforms.Prior work considered energy savings, stochastic-channel optimization, computation offloading, and resource management for MCC.
- Paper objective: The paper addresses the limited prior integration of MCC and C-RAN by co-locating mobile-cloud virtual machines with BBUs.The mobile cloud executes tasks, while the BBU returns results through RRHs.
- Paper objective: The proposed method jointly minimizes total energy under task-time constraints and allocates mobile-cloud and C-RAN resources.The formulation includes computation and transmission energy and time, while accounting for beamforming-vector design in RRHs.
- Solution approach: The non-convex, NP-hard problem is transformed using WMMSE, and an iterative algorithm addresses joint resource allocation.The formulation separates power minimization and throughput maximization before converting throughput maximization to weighted MSE minimization.
II. SYSTEM MODEL
This section introduces the mobile-cloud and C-RAN system model, including computation, communication, energy, and time consumption. Quality of service is represented through a task-execution time constraint.
- System model: The system model covers mobile-cloud computation and C-RAN network behavior.It presents the overall design before specifying computation and communication consumption models.
- System model: Energy and time consumption are modeled for both cloud execution and network operation.These models support the paper’s subsequent resource-allocation formulation.
- QoS requirement: The quality-of-service requirement is expressed as a time constraint on completing the given task.The task deadline is the stated QoS constraint in the system model.
A. Mobile Clone and System Architecture
The architecture assigns each user a cloud mobile clone that executes computation-intensive tasks and returns results through C-RAN. The model captures clone computation, task data, capacity, and selected system assumptions.
- Mobile clone operation: Mobile clones execute users’ computation-intensive tasks in the cloud, reducing the mobile user’s execution overhead.A clone receives task instructions and configuration information, performs execution, and returns the result through C-RAN.
- Mobile clone architecture: Co-locating mobile clones with the BBU keeps task information and data in the cloud before execution.The paper proposes cloud-based virtual machines beside the BBU with the same software stack as their corresponding users.
- Mobile clone architecture: Clone-to-clone communication can replace some wireless communication and save wireless resources, energy, and time overhead.The proposed architecture allows mobile clones to communicate directly in the cloud.
- System assumptions: The system contains N single-antenna UEs and L RRHs with K ≥1 antennas connected to the BBU pool by high-speed fronthaul.Each UE is assumed to have one established mobile clone with the same software stack.
- Task model: Each task is characterized by required CPU cycles F_i and returning output-data size D_i.The output includes task results and associated output information transmitted to the UE through C-RAN.
- System assumptions: The model neglects uplink instruction overhead, assumes available CSI at the BBU pool, and excludes fronthaul energy and time while enforcing fronthaul constraints.Virtual-machine computation capacity is also bounded by a maximum capacity for each clone.
C. Network Model
The network model describes result transmission from mobile clones through RRHs, including wireless rates, beamforming, transmit-power limits, and fronthaul-capacity constraints.
- Result transmission: Execution results are transmitted from the mobile clone to each UE through the C-RAN.The model defines the returning data size D_i and its transmission time.
- Wireless model: The received signal model uses serving-RRH beamforming vectors, channel vectors, unit-power data symbols, and Gaussian noise.These quantities determine the SINR and achievable rate for each UE.
- Wireless model: UE i’s wireless rate is modeled as r_i = B_i log(1 + SINR_i), where B_i is its assigned channel bandwidth.The rate depends on the wireless bandwidth and signal-to-interference-plus-noise ratio.
- Power constraints: The serving RRHs’ transmit power and energy are linked to the beamforming vectors and are subject to per-RRH power constraints.The model specifies the power used to send each task and each RRH’s power limitation.
- Fronthaul constraints: Fronthaul constraints limit the aggregate data rates carried from the BBU to each RRH by C_j ≤ C_j,max.The paper also discusses l0-norm modeling, where nonzero beamforming entries represent transmitted data streams.
E. QoS Requirement
The paper defines QoS through a total completion-time constraint covering mobile-cloud execution and transmission of results to the UE, then minimizes their combined energy cost. It separately formulates mobile-clone and C-RAN minimization problems, with feasibility dependent on computation capacity.
- QoS formulation: QoS constrains the total time for task execution and returning results to the mobile user.The total time combines execution time and transmission time.
- Joint energy objective: The objective minimizes energy consumed by mobile-cloud execution and C-RAN transmission of task results.A weight η_i trades off mobile-cloud and C-RAN energy and represents power-amplifier inefficiency.
- Problem decomposition: The design first formulates energy minimization for the mobile clone and then for C-RAN with fronthaul constraints.Separate solutions are provided for the mobile clone and C-RAN.
- Mobile-clone allocation: The mobile-clone execution time is constrained by T_i,max^C, and its energy minimization is formulated subject to that bound.The supplied formulation identifies the mobile clone's time constraint as a condition for solving its energy problem.
- Feasibility boundary: If the required computation cannot be completed within the mobile-clone time limit, the problem has no solution under the assumed capacity.The paper states that increasing maximum computation capacity is the way to guarantee QoS in this case.
B. Energy Minimization for C-RAN
The C-RAN energy-minimization problem is non-convex and difficult to solve, so the paper uses approximations, SOC reformulation, reweighted sparsity, and an iterative algorithm to handle fronthaul constraints.
- Problem transformation: Problem P2 is non-convex and NP-hard, making direct solution difficult.The paper therefore introduces approximations for the energy-minimization formulation.
- Problem transformation: Ignoring interference and applying Cauchy-Schwarz yields an approximation that transforms the C-RAN problem toward power minimization.The transformation is described as conditional on the approximation steps.
- SOC reformulation: Arbitrary beamforming-vector phase rotation allows a P3 constraint to be rewritten as a second-order cone constraint.This supports an SOC-based formulation of the C-RAN optimization.
- RRH clustering: The non-convex l0-norm is approximated by a convex reweighted expression, increasing weights on weaker beamforming links and encouraging them toward zero.This process can form RRH clusters serving corresponding UEs.
- Iterative solution: P4 without the fronthaul constraint is an SOC problem, whereas the fronthaul-constrained version is addressed iteratively.Algorithm 1 uses interior-point SOCP optimization and updates transmission-related variables until convergence or the iteration limit.
- Complexity: O(M · (KNL)^3.5) is the approximate computational complexity of Algorithm 1 for M iterations.Most complexity comes from the SOCP optimization in Step 1.
IV. JOINT OPTIMIZATION SOLUTION
The joint optimization minimizes total mobile-cloud and C-RAN energy while satisfying task time, transmission, computation, and fronthaul constraints. Because the resulting problem is non-convex, the paper develops WMMSE-based iterative solutions.
- Joint objective: The paper solves energy minimization and resource allocation jointly between the mobile cloud and mobile network.The target is total energy consumed by task execution and transmission of processing results.
- QoS constraints: Task completion must satisfy a total QoS time constraint that includes execution time plus transmission time.The constraint is imposed on completing the task and returning its results to the UE.
- Energy model: The joint objective combines mobile-cloud execution energy and weighted C-RAN transmission energy as E_i = EC_i + η_iETr_i.The rate and remaining constraints are inherited from the preceding formulations.
- Solution approach: P5 is non-convex and difficult to solve, motivating WMMSE-based iterative solution methods.The paper states that subsequent reformulation produces a problem that can be handled with WMMSE iterations.
A. Problem Transformation
The paper relaxes the joint problem's time constraint and reformulates it using auxiliary variables and vector notation, producing P6 for WMMSE-based iterative solution.
- Time relaxation: The time constraint is relaxed because equality holds for the relaxed version of P5.The relaxation uses the computation and transmission time expressions from the energy and rate models.
- Rate feasibility: The reformulation derives minimum achievable rates under the task-time and computation-capacity conditions.The passages state positivity assumptions for T_i,max and cloud computation capacities.
- Notation simplification: The reformulated problem uses vector notation for beamforming and channel variables to simplify P5.The vectors collect RRH-to-UE beamforming coefficients and corresponding channel coefficients.
- WMMSE reformulation: After reformulation, the variable i no longer appears in P6, which can be solved using a WMMSE-based iterative method.The supplied passages identify P6 as the intermediate problem for the next solution stage.
B. WMMSE-based Solution
The paper reformulates the optimization as an equivalent WMMSE problem and solves it iteratively using block coordinate descent. The procedure alternates receiver, MSE-weight, and transmit-beamforming updates, with the final beamforming step handled as an SOCP.
- WMMSE reformulation: The non-convex optimization is reformulated as an equivalent WMMSE problem and addressed with block coordinate descent.The objective is decreasing in the mobile user’s data rate, and the resulting formulation is convex in each individual variable φi, vij, and ui.
- Block updates: With transmit beamforming fixed, the optimal receive beamforming vector is obtained using the MMSE receiver.The receiving beamforming vector ui is defined for a single-antenna user equipment, and the corresponding mean square error is formulated at user i.
- Block updates: With the transmit beamforming vector and MMSE receiver fixed, the MSE weight is updated through the relation between MSE covariance and data rate.The weighted sum MSE minimization follows the corresponding rate relation, including the terms BiTi,max log(ei)+Di log(2).
- Block updates: With the MSE weight and MMSE receiver fixed, transmit beamforming is solved as a QCQP that can be transformed into an SOCP.The overall WMMSE-based iterative method uses a small constant ε to guarantee convergence and follows the proposed Algorithm 2.
- Algorithm and complexity: The algorithm returns the RRH cluster, beamforming vectors, data rates, and computational capacities after the iteration limit is reached.Its computational complexity mainly comes from the Step 3 SOCP optimization and is approximately O(M · (KNL)3.5).
V. SIMULATION RESULTS
Simulations evaluate joint energy minimization for C-RAN with mobile cloud under varying task demands, time constraints, QoS requirements, and system sizes. The joint solution consistently outperforms separate energy-minimization approaches while energy trends reflect computation, transmission, and deadline tradeoffs.
- Simulation setup: The simulations model a C-RAN with L = 4 RRHs and N = 5 mobile users, with five mobile clones co-located with BBUs.Each clone shares its corresponding user's software stack and executes that user's task.
- Energy trends: Higher QoS levels require more energy, whereas data-size gaps remain small because of the selected tradeoff factor.This comparison is reported for total energy across mobile cloud execution and C-RAN transmission.
- Joint versus separate optimization: The joint energy-minimization solution achieves the best performance, outperforming separate solutions across the comparisons in Figs. 7 and 8.The separate-solution settings use Ti,max = 0.1s, with Di = 1000 in Fig. 7 and Fi = 1500 in Fig. 8.
- Scalability checks: With one additional user, the proposed optimization retains nearly the same performance gain, although all solutions use more power than in Fig. 7.Similar performance gains are also reported for different antenna numbers, but those figures are omitted because of limited space.
VI. CONCLUSION
The paper proposes a C-RAN architecture with mobile clones and jointly minimizes mobile-cloud and mobile-network energy under task time, QoS, and fronthaul constraints. The proposed resource-allocation solution improves system performance and saves energy, while future work extends the transmission and fronthaul models.
- Architecture: The proposed architecture incorporates mobile clones into C-RAN to combine mobile cloud computing with the cloud radio access network.Each UE has one task executed in a mobile clone, modeled by required CPU cycles and result data size.
- Joint optimization: The optimization jointly minimizes whole-system energy across mobile-cloud computation and mobile-network operation under task time and QoS constraints.The formulation also includes fronthaul constraints for obtaining RRH clusters.
- Results: The proposed energy-minimization and resource-allocation solution improves system performance and saves energy.The evaluation compares whole-system energy consumption and reports the energy-minimization solution as the best-performing solution.
- Future work: Future work will model the complete uplink and downlink transmission process between UEs, RRHs, and mobile clones.The planned extensions include fronthaul transmission time and energy-consumption models.