Normalized Delivery Time in Fog Radio Access Networks

18 concepts11 papers1 proposal

In modern wireless networks, minimizing content delivery latency is a primary engineering goal. However, analyzing latency from an information-theoretic perspective introduces a fundamental mathematical challenge: as the Signal-to-Noise Ratio (SNR) approaches infinity, the transmission rate of a wireless link also approaches infinity, causing absolute delivery time to drop to zero. This makes it difficult to capture the structural bottlenecks of a network—such as limited edge caching capacity or constrained fronthaul links—in a clean, parameter-independent way.

To solve this problem, researchers developed the concept of Normalized Delivery Time (NDT). By scaling actual delivery latency against an ideal baseline, NDT isolates the structural overhead of network architectures in the high-SNR regime, providing a clear framework for evaluating storage-latency trade-offs.

The Intuition Behind NDT

The core idea of NDT, introduced by Sengupta et al., is to evaluate system performance by comparing the worst-case time required to deliver content in a cache-aided network against an ideal, interference-free system with infinite storage.

Mathematically, in a high-SNR regime, the capacity of a standard wireless link scales proportionally with $\log(\text{SNR})$. If we measure the absolute time $T(\text{SNR})$ to deliver a file, it scales inversely with $\log(\text{SNR})$. To eliminate this dependency on the operating SNR, NDT normalizes the actual delivery time by the time it would take to send the same file over a single baseline interference-free link with no caching constraints. Because both the numerator and denominator scale with $1/\log(\text{SNR})$, the SNR terms cancel out.

What remains is a constant metric that quantifies the worst-case latency penalty incurred by the network's physical architecture. As formalized by Sengupta et al., NDT provides a standardized, parameter-abstracted way to evaluate how edge caching and cloud processing interact. This framework can also be extended to map out an "NDT region"—a multi-dimensional space characterizing the latency trade-offs when different content types have varying timeliness requirements, as demonstrated by Goseling.

Balancing Fronthaul and Edge Caching

In a Fog Radio Access Network (F-RAN), content can either be fetched from a central cloud server via a capacity-constrained fronthaul link or served directly from local caches at wireless edge nodes (ENs). The minimum achievable NDT depends heavily on how these two pathways are coordinated.

Transmission strategies generally fall into two categories:

  • Serial Transmission: The ENs first fetch the missing content from the cloud over the fronthaul, and only after receiving the full data do they transmit it to the end users over the wireless channel.
  • Pipelined Transmission: ENs simultaneously transmit cached content to users while fetching the remaining uncached data from the cloud. He et al. demonstrate that this simultaneous execution effectively hides a portion of the fetching latency, significantly outperforming serial alternatives.

To minimize NDT within these modes, the network must optimize "multiplicity"—the number of edge nodes that cooperatively transmit a piece of content. Zhang and Simeone show that increasing cooperation allows multiple antennas or nodes to perform joint beamforming, which neutralizes wireless interference and speeds up the edge delivery phase. However, cooperative transmission requires multiple ENs to hold the same file, which either drains local cache capacity or forces more data to be fetched over the fronthaul.

By precisely balancing this trade-off, Zhang and Simeone proved that optimized caching and delivery schemes can achieve performance within a tight multiplicative factor of the absolute theoretical optimum (within $3/2$ for serial and $2$ for pipelined systems).

Serial Mode:
|--- Fetch from Cloud ---|--- Transmit to User ---| = Higher NDT

Pipelined Mode:
|--- Fetch from Cloud ---|
        |--- Transmit Cached Content ---|         = Lower NDT

Topology and Cooperative Caching

The theoretical limits of NDT shift when moving from fully connected, idealized networks to more complex, practical topologies.

In scenarios where users are only partially connected to a subset of edge nodes, interference management becomes more challenging. Roushdy et al. addressed this by designing a coded caching scheme that utilizes real interference alignment. Their work demonstrates that even when connectivity is limited, strategically increasing user-to-node connectivity drastically reduces NDT, provided that the resulting interference is aligned and canceled out.

Furthermore, caching does not have to be restricted to edge nodes. Shariatpanahi et al. analyzed systems where both the edge nodes and the end-user devices possess internal caches. By combining cooperative transmission from the ENs with side-information multicasting directly between users, the authors proved that a joint cloud-edge-user caching architecture can achieve near-optimal content delivery latency, maintaining a maximum $3/2$ performance gap from the information-theoretic optimum.

Accelerating Delivery via Device-to-Device Cooperation

An alternative way to bypass fronthaul bottlenecks is to allow end-user devices to assist one another via Device-to-Device (D2D) communication. If User A has a file cached that User B needs, they can share it locally without involving the cloud or the edge nodes.

In a series of analyses, Shamai mapped out the fundamental latency limits of D2D-aided F-RANs. A key insight from this line of work is that D2D communication does not automatically guarantee a latency reduction. As shown by Shamai, D2D cooperation only lowers the minimum NDT if the local D2D data rate exceeds a specific mathematical threshold determined by the network's fronthaul capacity and its inherent structural latency.

When the D2D links are fast enough to clear this threshold, they can be integrated into a pipelined architecture. By leveraging a compress-and-forward cooperation strategy, Shamai proved that a D2D-enhanced F-RAN can achieve an NDT within a factor of two of the absolute theoretical minimum, regardless of how large the network grows. This relationship can be explicitly characterized via closed-form expressions that map out exactly how fractional cache sizes, fronthaul rates, and D2D capacities interact to dictate the absolute floor of wireless content delivery latency (Shamai).

Key Foundational Work

Fog-Aided Wireless Networks for Content Delivery: Fundamental Latency Tradeoffs (2017) — contributed 2 concept elements to this direction.
Cache Aided Wireless Networks: Tradeoffs between Storage and Latency (2015) — contributed 1 concept elements to this direction.
Delivery Latency Trade-Offs of Heterogeneous Contents in Fog Radio Access Networks (2017) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_normalized_delivery_time_minimization_b82a0759,
  title = {normalized delivery time minimization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/normalized-delivery-time-minimization/rdr_b82a0759c2a540d18b06ac4153deb328},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_normalized_delivery_time_minimization_b82a0759_v1,
  title = {normalized delivery time minimization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/normalized-delivery-time-minimization/rdr_b82a0759c2a540d18b06ac4153deb328/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake