Abstract
Centralized Radio Access Networks (C-RAN) offer significant benefits for network management but introduce critical latency bottlenecks in the fronthaul link. Standard Hybrid Automatic Repeat Request (HARQ) protocols, which require rapid feedback, are often incompatible with the round-trip delays to a centralized Baseband Unit (BBU). This paper proposes a novel architecture that separates the "control" of retransmissions from the "data" decoding process. By enabling Remote Radio Heads (RRHs) or User Equipments (UEs) to make local, low-latency HARQ decisions based on channel estimates, the system significantly reduces latency with minimal impact on throughput. The analysis utilizes finite-blocklength capacity bounds to model real-world short packets in MIMO systems.
The Latency Challenge in C-RAN
In a C-RAN architecture, baseband processing is centralized in a cloud-based BBU, while Remote Radio Heads (RRHs) handle radio frequency functions at the cell site. These components are connected via a "fronthaul" link.
The primary obstacle is latency. Standard HARQ protocols require the receiver to quickly signal whether a packet was successfully decoded (ACK) or requires retransmission (NAK). In LTE/5G, this feedback loop must typically be completed within a few milliseconds. However, the propagation and processing delay over a long-distance fronthaul link to a cloud BBU often exceeds these limits, rendering standard HARQ ineffective. Waiting for the cloud to finish decoding before initiating a retransmission leads to significant network sluggishness and inefficiency.
Proposed Method: Separation of Control and Data Planes
The authors propose a functional split where the control of retransmissions is decoupled from the actual data decoding process.
- Local Control Decisions: Instead of waiting for the BBU's decoding result, the RRH estimates the Uplink Channel State Information (CSI) and predicts the probability of a decoding error ($P_e$) at the BBU.
- Decision Rule: If the predicted error probability is below a predefined threshold ($P_{th}$), the RRH immediately sends a local ACK to the UE, even though the cloud-based decoding is still in progress.
- Architectural Variants:
- BBU Hoteling: Each RRH is associated with a dedicated BBU, and the RRH makes the ACK/NAK decision.
- C-RAN with Joint Processing: Multiple RRHs serve a UE. The paper explores "Hard Feedback" (1-bit ACK/NAK per RRH) and "Soft Feedback" (quantized CSI sent to the UE for a final decision).
Analytical Framework
The study employs finite-blocklength capacity bounds to model the performance of short packets, which are prevalent in 5G and IoT applications. This approach provides a more realistic assessment than traditional Shannon capacity analysis, which assumes infinite packet lengths. The evaluation covers three HARQ modes: Type-I, Chase Combining (CC), and Incremental Redundancy (IR).
Key Insights
- CSI as a Success Proxy: A key finding is that the RRH does not need to perform complex decoding to predict success; channel conditions (SNR) are often sufficient for high-quality local decisions.
- The Mismatch Trade-off: Local decisions can lead to "false ACKs" (BBU fails to decode) or "unnecessary NAKs" (wasted retransmission). Optimizing the threshold $P_{th}$ is crucial to balancing these risks.
- Soft Feedback Superiority: In joint-processing C-RAN, "Soft Feedback" significantly outperforms "Hard Feedback" because it allows the UE to synthesize information from multiple RRHs.
Key Results
- Latency Reduction: The proposed scheme can reduce average latency by a factor of $1 + L_f$ (where $L_f$ is the fronthaul delay), potentially leading to a 4x or greater improvement in practical scenarios.
- Minimal Throughput Impact: In MIMO systems, the throughput loss compared to an ideal (zero-delay) feedback scenario is often less than 0.3 bit/symbol.
- Antenna Scaling: Increasing the number of antennas in MIMO systems enhances the reliability of local decisions, further narrowing the gap with ideal cloud-based feedback.
Conclusion and Future Work
This research demonstrates that separating control and data planes is a viable strategy for overcoming fronthaul latency in C-RAN. Future work will address interference modeling in dense massive MIMO environments and the integration of these techniques into random access scenarios.
Reference: Khalili, S., & Simeone, O. (2016). Uplink HARQ for Cloud RAN via Separation of Control and Data Planes. arXiv:1508.06570v5.
Summary updated