More Capacity from Less Spectrum: Tapping into Optical-layer Intelligence in Optical Computing-Communication Integrated Network

Driven by massive investments and consequently significant progresses in optical computing and all-optical signal processing technologies lately, this paper presents a new architectural paradigm for next-generation optical transport network, entitled \textit{optical computing-communication integrated network}, which is capable of providing dual services at the optical layer, namely, computing and communication. This approach seeks to exploit the potential for performing optical computing operations among lightpaths that traverse the same intermediate node. \textit{Optical-layer intelligence concept} is thus introduced as the capability to perform computing / processing at the lightpath scale to achieve greater spectral and/or computing efficiency. A case study focusing on optical aggregation operation is introduced, highlighting the key differences between optical computing-communication integrated network and its current counterpart, optical-bypass ones. A mathematical formulation for optimal designs of optical-aggregation-enabled network is then provided and performance comparison with traditional optical-bypass model is drawn on the realistic NSFNET topology.

💡 Research Summary

The paper introduces a novel architectural paradigm for future optical transport networks called the Optical Computing‑Communication Integrated Network (OCCIN). While the widely deployed optical‑bypass architecture eliminates costly optical‑electrical‑optical (O‑E‑O) conversions and improves cost and energy efficiency, it still requires that lightpaths traversing a common intermediate node remain completely isolated in time, frequency, or space. This isolation prevents any form of in‑network optical signal processing or computing, limiting the ability to exploit the growing capabilities of photonic computing and all‑optical signal processing technologies.

To overcome this limitation, the authors propose the concept of “optical‑layer intelligence,” which endows optical switching nodes with the ability to perform computing operations directly on lightpaths as they pass through. The paper focuses on a concrete example: optical aggregation. Two 400 Gbps QPSK lightpaths originating from different sources and destined for a common node are combined at an intermediate node into a single 800 Gbps 16‑QAM lightpath on the same wavelength. The aggregated signal can later be de‑aggregated at the final destination to recover the original traffic. This operation reduces the number of required wavelengths from two to one and cuts the total wavelength‑link cost dramatically (e.g., from 12 wavelength‑link units in a conventional bypass scenario to 4 units after aggregation).

A rigorous integer linear programming (ILP) model is developed to capture the joint routing, wavelength assignment, and aggregation decisions. The model takes as input the physical graph G(V,E) and a set D of uniform‑capacity traffic demands. Decision variables include binary routing indicators xᵈₑ, aggregation indicators θᵈᵥ, aggregated‑path routing variables zᵈ,ᵥ,ₑ, and pairing variables fᵈ₂ᵈ₁ that denote whether two demands are aggregated. The objective minimizes total wavelength‑link usage, with a term that subtracts the saved wavelength‑link resources when aggregation occurs. Constraints enforce flow conservation for each demand, limit each demand to at most one aggregation, require paired demands to share the same destination, synchronize aggregation nodes, and ensure proper routing of the resulting aggregated lightpaths. The formulation extends standard routing ILPs by adding a substantial number of variables and constraints, making the problem NP‑hard and an order of magnitude more complex than conventional bypass planning.

The authors evaluate the proposed architecture on the realistic NSFNET topology. Traffic is generated using a two‑to‑many pattern: two random source nodes and a varying number of destination nodes (4, 8, and 12) to increase the likelihood of demands sharing a common destination—a prerequisite for aggregation. For each load level, ten random traffic instances are simulated, with each demand occupying a dedicated 50 GHz‑spaced wavelength in the C‑band. The ILP is solved optimally for both the OCCIN (aggregation‑aware) and the traditional bypass designs, and the total wavelength‑link cost is compared. Results show that the aggregation‑aware design consistently outperforms the bypass baseline across all load scenarios, achieving up to a 35 % reduction in wavelength‑link usage. The gains are most pronounced at moderate loads, where the probability of suitable aggregation pairs is highest.

The study demonstrates that embedding optical‑layer intelligence into network nodes can substantially improve spectral efficiency without sacrificing traffic demands, and it opens new avenues for in‑network photonic computing. The authors suggest several future research directions: extending the model to support multi‑way aggregation, heterogeneous line rates and modulation formats, dynamic real‑time aggregation scheduling under traffic fluctuations, and exploring other optical computing primitives (e.g., matrix multiplication, convolution) that could be performed directly in the transport layer. By blurring the traditional boundary between communication and computation, the OCCIN paradigm promises to reduce capital and operational expenditures, lower power consumption, and enable ultra‑low‑latency, high‑throughput services essential for emerging AI‑driven applications and next‑generation data‑center interconnects.