Source-linked AI summary
Cloud Radio Access Network: Virtualizing Wireless Access for Dense Heterogeneous Systems
Osvaldo Simeone, Andreas Maeder, Mugen Peng, Onur Sahin, Wei Yu
TL;DR
C-RAN addresses the cost and flexibility challenges of dense heterogeneous radio access by virtualizing base-station functions in centralized cloud processors. This paper surveys C-RAN research across fronthaul, processing, access control, resource allocation, architecture, and standardization, highlighting both centralization opportunities and fronthaul and protocol constraints. Its reviewed results include theoretical capacity characterizations, reported throughput gains, reduced fronthaul requirements, and candidate functional splits.
Problem
Dense C-RAN deployment and centralized processing must operate under constrained fronthaul capacity and latency, while resource-allocation problems can be non-convex and combinatorial.
Method
The paper provides a state-of-the-art overview covering fronthaul compression, baseband processing, medium access control, resource allocation, system-level considerations, and standardization.
Results
The review reports capacity characterizations, nearly doubled edge-cell throughput from distributed source coding in one scenario, fronthaul reductions up to factor 20 for some Layer 2 splits, and two leading split candidates.
Takeaways & Limitations
C-RAN centralization can support flexible resource sharing, coordinated transmission, and lower-cost dense deployments, but practical designs must match functional splits to fronthaul and protocol constraints.
Abstract
from arXiv · showhide
Cloud Radio Access Network (C-RAN) refers to the virtualization of base station functionalities by means of cloud computing. This results in a novel cellular architecture in which low-cost wireless access points, known as radio units (RUs) or remote radio heads (RRHs), are centrally managed by a reconfigurable centralized "cloud", or central, unit (CU). C-RAN allows operators to reduce the capital and operating expenses needed to deploy and maintain dense heterogeneous networks. This critical advantage, along with spectral efficiency, statistical multiplexing and load balancing gains, make C-RAN well positioned to be one of the key technologies in the development of 5G systems. In this paper, a succinct overview is presented regarding the state of the art on the research on C-RAN with emphasis on fronthaul compression, baseband processing, medium access control, resource allocation, system-level considerations and standardization efforts.
I. INTRODUCTION
C-RAN virtualizes base-station functions in centralized cloud processors while leaving low-cost RUs with radio functions. The architecture offers deployment, resource-management, coordination, and maintenance benefits, but fronthaul bandwidth and latency constrain realization of these gains.
- Architecture: C-RAN centralizes baseband and higher-layer operations on processors, while RUs retain radio functionalities and are managed by a reconfigurable CU.This architecture is an instance of network function virtualization applied to the RAN.
- Advantages: Replacing full-fledged base stations with RUs reduces deployment costs for dense heterogeneous networks through lower space and energy requirements.
- Advantages: A CU can flexibly allocate radio and computing resources across connected RUs, producing statistical multiplexing gains through load balancing.
- Advantages: Centralization facilitates coordinated and cooperative transmission across RUs, including eICIC and CoMP, thereby boosting spectral efficiency.
- Constraints: Fronthaul bandwidth and latency can limit cloud-processing benefits, including by preventing standard closed-loop error recovery under two-way communication latency.A more flexible RU-CU split may move FFT/IFFT, demapping, synchronization, or higher-layer functions to the RU.
- Scope: The paper surveys C-RAN research on fronthaul compression, baseband processing, medium access control, resource allocation, system-level issues, and standardization.
II. C-RAN SIGNAL MODELS
This section introduces the basic C-RAN signal models used to analyze the physical layer and referenced throughout the paper.
- Signal models: The paper begins with basic C-RAN signal models that are typically used for physical-layer analysis.
A. Uplink
The uplink model describes clustered RUs receiving multi-antenna UE transmissions over single-hop or multi-hop fronthaul, with noise, inter-cluster interference, channel variability, and finite link capacities represented explicitly.
- Topology: RUs are partitioned into CU-managed clusters, and uplink UEs transmit wirelessly to the RUs over single-hop or multi-hop fronthaul.
- Signal model: The uplink received signal is represented using a discrete-time complex baseband IQ model involving channel matrices, transmitted UE samples, thermal noise, and inter-cluster interference.
- Assumptions: The model can account for RU oversampling and different channel-matrix time variability determined by mobility and transmission parameters.
- Fronthaul: In a single-hop topology, RU i has a fronthaul capacity of C_i bits/s/Hz, allowing nC_i bits over an uplink coding block of n symbols.
B. Downlink
The downlink model describes multi-RU baseband transmission to each UE, with channel response, additive noise, inter-cluster interference, and fronthaul represented analogously to the uplink.
- Signal model: Each downlink UE receives a discrete-time baseband signal generated by the aggregate signal vector transmitted by all RUs in the cluster.
- Signal model: The downlink model includes the channel response from all cluster RUs to UE k, thermal noise, and interference from other clusters.
- Fronthaul: The downlink fronthaul network is modeled in the same fashion as the uplink fronthaul network.
III. FRONTHAUL COMPRESSION
This section reviews methods for transporting digitized IQ baseband signals over C-RAN fronthaul links, progressing from CPRI basics to point-to-point and network-information-theoretic compression.
- The review covers CPRI, point-to-point compression, and advanced network-information-theoretic fronthaul solutions.
A. Scalar Quantization: CPRI
CPRI standardizes CU–RU fronthaul communication using sampled, scalar-quantized baseband signals transmitted over a constant-bit-rate serial interface. Its rates scale with signal bandwidth, receive antennas, and quantization precision, creating capacity challenges for demanding LTE deployments.
- CPRI standardizes the CU–RU interface by sampling and scalar-quantizing baseband signals for constant-bit-rate serial transmission.
- LTE base-station CPRI rates can exceed 9.8 Gbps and standard fiber capacity with carrier aggregation, multiple sectors, and multiple antennas.The required rate depends on signal bandwidth, receive-antenna count, and bits per sample; LTE uses 8–20 bits per I or Q sample.
B. Point-to-Point Compression
Point-to-point compression addresses the high fronthaul rates produced by conventional CPRI by reducing bit rate while limiting distortion in the quantized signal.
- Point-to-point compression schemes reduce CPRI data-stream rates while limiting distortion in the quantized signal.
B.1 Compressed CPRI
Compressed CPRI combines signal-processing and quantization techniques to reduce fronthaul rates, while alternative functional splits move selected PHY processing to the RU. These approaches can substantially outperform conventional compressed CPRI under light loads.
- Compressed CPRI techniques include filtering and downsampling, per-block scaling, optimized non-uniform quantization, noise shaping, and lossless compression.The methods exploit oversampling, signal dynamic range, signal statistics, temporal correlation, and residual entropy.
- The alternative splits are organized as different CU–RU divisions of physical-layer functionality.
- Compressed CPRI techniques reduce fronthaul rates by around a factor of 3.
- Alternative PHY functional splits place operations such as frame synchronization, FFT/IFFT, or resource mapping and demapping at the RU.These splits are explored to obtain further fronthaul-rate reductions beyond point-to-point compression in the conventional architecture.
- Up to 30-fold compression ratios were reported for resource-aware functional splitting under small system loads.The RU can apply channel-specific quantization and omit unused resource blocks in lightly loaded frames.
C.1 Uplink
The uplink section examines how correlated RU observations can be compressed and jointly processed at the CU under finite fronthaul constraints. It presents distributed/Wyner-Ziv coding, signal models, and information-theoretic characterizations of achievable rates.
- Distributed source coding: Correlated observations across RUs, especially in dense networks, enable distributed source coding for uplink fronthaul compression.Once the CU recovers one RU’s signal, it can use it as side information to decompress another RU’s signal.
- Distributed source coding: Wyner-Ziv coding lets each RU exploit receiver-side correlation without knowing the side information available at the CU.The RU needs only the correlation between its received signal and the CU’s side information.
- Distributed source coding: Nearly double the edge-cell throughput was obtained numerically with distributed source coding for fixed average spectral efficiency and fronthaul capacities.The reported setting was a single macrocell overlaid with multiple smaller cells.
- Distributed source coding: Wyner-Ziv implementation requires the CU to communicate correlation-dependent quantizer or compressor choices to each RU, leaving codebook and selection-rule design open.The required correlation depends on the channel state information of the involved RUs.
- Uplink signal model and processing: Joint processing across RUs can mitigate inter-cell interference through cooperative reception, but it depends on instantaneous CSI and precise synchronization.Downlink synchronization is described as imperative for synchronous cooperative beamforming, while uplink timing differences can theoretically be corrected digitally.
- Uplink signal model and processing: The uplink model uses RU clusters jointly decoding UE transmissions over finite-capacity fronthaul, where compressed IQ samples introduce additional quantization noise.The analysis considers single-hop links and characterizes linear receive beamforming and successive interference cancellation with independent or Wyner-Ziv quantization.
B. Downlink
The downlink C-RAN review characterizes beamforming and quantization under finite fronthaul constraints, including multivariate compression and alternative functional splits. It also relates Layer 2 splitting to fronthaul reduction, centralized resource management, and timing constraints.
- Downlink processing: Downlink baseband processing uses beamforming or dirty paper coding to reduce interference, becoming a broadcast channel when fronthaul capacity is unlimited.With finite-capacity fronthaul, the analysis becomes more involved and must account for signal quantization.
- Downlink compression: Multivariate compression introduces correlated quantization noises that may cancel over the wireless channel and improve overall user rates.Unlike independent uplink encoding, downlink compression is performed centrally at the CU, which enables correlated quantization noises but requires extra fronthaul capacity.
- Rate characterization: Rate expressions characterize downlink capacity under fronthaul constraints for linear beamforming and dirty paper coding with per-link or multivariate quantization.These formulations support joint design of transmit covariances, quantization covariances, scheduling, power control, and beamformers, although the resulting optimization is non-convex.
- Alternative functional splits: Data-sharing can be effective when fronthaul capacity is limited and cooperation clusters are relatively small, despite inefficient replication when clusters are large.The standard split keeps RUs as remote antenna heads performing compression rather than UE-data encoding and decoding.
- Layer 2 functional splits: Synchronous Layer 2 protocols require delivery within 1 ms transmission intervals, whereas asynchronous sub-layers have latency requirements on the order of 20 ms.The distinction creates different latency and jitter requirements for possible RU-CU functional splits.
A. Constraints and Requirements
Layer 2 functional splits must balance fronthaul, timing, protocol-integration, and resource-management constraints against centralization gains. The review identifies splits A and C as the principal candidates, with different latency, cost, and coordination trade-offs.
- Fronthaul requirements: 150 Mbps is the theoretical maximum overall fronthaul rate for a 20 MHz FDD LTE system with two transmit antennas under worst-case traffic.The estimate includes approximately 10% Layer 2 control-plane overhead and assumes fully occupied radio resources with the highest MCS.
- Implementation constraints: Scheduling spans upper PHY, MAC, and RLC functions, requiring bidirectional information exchange within a fraction of the 1 ms TTI.This cross-layer dependence constrains functional splits unless the LTE protocol stack is reconsidered.
- Timing constraints: HARQ feedback imposes a below-3 ms round-trip budget and approximately 1 ms maximum fronthaul one-way latency when CU processing power is sufficient.The budget includes fronthaul transmission, CU processing, and frame building, excluding jitter buffering.
- Centralization gains: Centralization gains include multiplexing gains from aggregated traffic and processing demand, plus coordination gains from joint radio-resource management across RU clusters.Coordination can include scheduling, ICIC, DTX, admission control, load balancing, and hand-over functions at different split levels.
- Candidate splits: Split A supports centralized RRM but requires low backhaul latency, whereas split C is cost efficient and standardized as LTE dual connectivity.The review presents these two options as the main Layer 2 candidates among the considered splits.
- Resource-management constraints: Centralized RRM is difficult because weighted sum-rate optimization can be non-convex or combinatorial under interference and limited-fronthaul constraints.Relevant variables include downlink beamforming, uplink user association, RU clustering, and RU activation.
- Dynamic RRM: Dynamic RRM can be modeled as an average-cost MDP over CSI and queue states, but Bellman solutions suffer from a state-space curse of dimensionality.Approximate MDP, stochastic learning, and continuous-time MDP methods are identified as possible remedies; partial observability adds further complexity.
A. C-RAN Network Architecture
A C-RAN connects radio access sites to a centralized cloud unit through fronthaul, with backhaul linking the CU to the core network. Its architecture centralizes baseband pooling while transport choices constrain cost, latency, distance, and coordination scope.
- Architecture: The C-RAN architecture comprises access, fronthaul, backhaul, and packet-core segments, with cell sites connected to the CU through fronthaul links.RUs may implement different Layer 1 and Layer 2 functions; in the basic deployment they perform RF operations such as conversion, sampling, and amplification.
- Fronthaul transport: Fiber-optic and microwave links are leading fronthaul media, with transport selection shaped by cost, latency, and distance between radio sites and the cloud center.Dedicated fiber offers high data rate and low latency but has limited deployment because of cost.
- Cloud processing: Multiple baseband units can be collocated in the CU to coordinate virtualized access-network operations using DSP or general-purpose processor platforms.The CU also interfaces with the backbone network, including the LTE Evolved Packet Core, through the backhaul.
- Coordination scope: C-RAN clustering across multiple CUs can support coordinated transmission over wider areas, although inter-CU coordination is less efficient than intra-RU coordination.Latency and capacity limitations in the fronthaul constrain inter-CU coordination.
B. Next-Generation C-RAN Network Architecture
Next-generation C-RAN architecture addresses heterogeneous network-management constraints through flexible function placement and unified SDN control. The review frames this evolution as a balance between centralized virtualization and edge processing, supported by adaptable transport technologies and standardization efforts.
- Motivation: Heterogeneous technologies across fronthaul, backhaul, and core segments make flexible, dynamic resource management complex, costly, and inefficient.SDN advances motivate using software-defined management tools to address this limitation.
- SDN-based architecture: A unified SDN control plane dynamically assigns virtualized RU functions to network nodes using network- and link-level abstractions.These abstractions are populated through southbound and northbound interfaces and conveyed to the SDN controller.
- Industry development: C-RAN development has involved substantial industrial research and development alongside academic work over the last decade.The review situates standardization and deployment efforts within this broader development trajectory.
- Standardization: CPRI standardizes a serial, bidirectional, constant-rate digitized I/Q interface between the CU-side controller and radio equipment on the fronthaul.It specifies control-plane support with synchronization and low-latency transmission requirements.
- Standardization: Open Base Station Architecture Initiative and Open Radio Interface are competing standards and industry associations for base-station transceiver interfaces and functional descriptions.These efforts complement CPRI in the standardization landscape reviewed by the paper.
- Virtualization frameworks: ETSI NFV ISG targets telecom-network virtualization that supports multi-tenancy, dynamic functional allocation, lower deployment cost, and easier operations and maintenance.The framework is described as directly applicable to C-RAN architecture.
- Design trade-off: The review emphasizes a tension between centralized virtualization, which raises fronthaul capacity and latency requirements, and edge processing, which reduces delays and architectural constraints.Flexible RAN and fronthaul/backhaul technologies are presented as a way to adapt operation to traffic type and system conditions.