{"id":"f70423cd-d0f7-4de4-9d78-bfbbc100ab48","arxiv_id":"2505.18753","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An optical computing-communication integrated network that aggregates two QPSK lightpaths into one 16-QAM lightpath at intermediate nodes uses fewer wavelengths than optical-bypass routing for computing requests.","lead":"This paper proposes a new optical network architecture where lightpaths can be combined and computed at intermediate nodes instead of just being passed through. Simulations on a realistic European topology suggest this integrated design reduces the number of wavelengths needed to serve distributed machine-learning aggregation compared to traditional optical-bypass routing.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim assumes 16-QAM aggregation at any node with no reach penalty; if the aggregated lightpath exceeds 16-QAM reach, the reported wavelength savings in Fig. 3 may be physically infeasible.","rationale":"The central claim is that integrated optical computing-communication networking uses fewer wavelengths than optical-bypass for computing requests. This claim is established only through the assumption that two QPSK lightpaths can be aggregated into a single 16-QAM lightpath at any intermediate node with no extra spectrum and no capacity loss. The reader's weakest assumption identifies this aggregation as the fragile point, noting that physical-layer constraints could erase the gains. I agree and sharpen the point: 16-QAM has a materially shorter reach than QPSK, so the aggregated lightpath's feasibility depends on the computing node's distance to the destination. The paper's MILP description includes no reach or OSNR constraints, and the illustrative results select computing nodes that appear close to the destination, which may not generalize. Because the manuscript already receives a CONDITIONAL verdict conditioned on such physical-layer issues, my stress-test does not move the verdict; it reinforces the specific condition that should be tested before acceptance. The proposed concrete test directly checks whether the reported savings survive a physically plausible reach constraint, which is the single most decisive check for the central claim.","tokens_in":13476,"tokens_out":9507,"duration_ms":90659,"concrete_test":"Re-run the COST239 experiments with an explicit reach constraint: for each lightpath, assign OSNR-limited reach for QPSK and 16-QAM (for example, using a GN-model with the assumed per-wavelength line rate and standard fiber parameters), and forbid any lightpath whose shortest-path length exceeds its modulation's reach. Compare the minimized wavelength counts against Fig. 3 for all 11 destination nodes. If any integrated-network wavelength count increases or matches the bypass count, the claimed consistent spectral gain depends on ignoring reach limits; if the counts remain unchanged, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's quantitative claim (Section 5, Fig. 3) compares two designs by the minimum number of wavelengths needed to serve computing requests on COST239. In the integrated design, a computing request is served by two QPSK lightpaths converging at a computing node and one 16-QAM lightpath from that node to the destination. The described model (Section 4) includes wavelength uniqueness, wavelength continuity, and one-computing-node selection, but no physical-layer constraint on the aggregated lightpath. However, 16-QAM requires roughly 6.7 dB higher OSNR than QPSK at the same symbol rate, and therefore has a shorter reach for a fixed transceiver and bit rate. In a realistic mesh, the optimal computing node minimizing wavelength count could place the 16-QAM lightpath on a route longer than its reach, especially for destinations with few incident fibers. In the illustrative tables, computing nodes are chosen at or near a source close to the destination, which may mask this issue. If the 16-QAM lightpath is infeasible, the network would need either a lower-order modulation (using additional wavelengths or a second hop) or regeneration, and the claimed 'improvement is consistently realized across all destination nodes' would not hold under physical constraints. The load-bearing assumption is therefore that the aggregation operation is universally available at any node at no spectral, power, or reach cost; this assumption is not justified in the manuscript and is not captured in the MILP as presented.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an 'optical computing-communication integrated network' in which two QPSK lightpaths arriving at an intermediate node are optically aggregated into a single 16-QAM lightpath, thereby providing in-network computation at the optical layer. It introduces a routing, wavelength, and computing assignment (RWCA) problem for serving computing requests in such networks, claims to formulate it as a mixed-integer linear program, and reports CPLEX-based results on the COST239 topology showing that the integrated design uses fewer wavelengths than the conventional optical-bypass design for all destination nodes. The motivating application is geo-distributed machine learning, where partial results from two source nodes are aggregated to a destination.","tokens_in":13684,"tokens_out":5359,"duration_ms":42801,"significance":"If the underlying optical aggregation operation is physically realizable at intermediate nodes with no additional spectral cost and no reach penalty, the paper would identify a new network design problem (RWCA) and a potentially useful architectural direction for in-network computing. The paper also gives a concrete traffic scenario and a comparative simulation setup, which are useful for framing the problem. However, the manuscript does not provide the claimed MILP formulation, does not justify the physical feasibility of the aggregation operation, and omits any physical-layer constraints (notably 16-QAM reach) that are essential for the reported wavelength savings to be credible. The paper's main quantitative claim therefore rests on unverified modeling assumptions rather than on a validated design methodology.","major_comments":[{"comment":"The paper states that 'the RWCA is formulated in the form of the mixed integer linear programming model' and that the model was solved by CPLEX, but no MILP is actually presented. Section 4 contains only a problem statement and Algorithm 1, which is a while-loop with no termination condition, no search procedure, and no substantive algorithmic content. Without the decision variables, objective function, and constraints (wavelength uniqueness, wavelength continuity, and one-computing-node selection are mentioned but never written mathematically), the CPLEX results in Section 5 cannot be independently checked or reproduced. This is a load-bearing omission because the paper's central comparison claims that 'all results for both designs are optimally collected.'","section":"Section 4"},{"comment":"The entire spectral-efficiency gain rests on the assumption that two QPSK lightpaths can be optically aggregated at any intermediate node into one 16-QAM lightpath with output capacity equal to the sum of the input capacities, at no additional spectral cost, no power penalty, and no reach penalty. The manuscript cites references [55,56] for coherent spectral superposition, but those works demonstrate modulation-format aggregation in a transmitter/electrical-signal context and do not establish that the operation can be performed on two arbitrary in-flight lightpaths at an intermediate network node. Moreover, the model includes no physical-layer constraint on the resulting 16-QAM lightpath: 16-QAM requires roughly 6.7 dB higher OSNR than QPSK at the same symbol rate and therefore has a shorter reach. In the COST239 topology, links can be several hundred kilometers long, and the 16-QAM lightpaths in Table 2 (e.g., 10-2-1) may exceed a realistic 16-QAM reach at 400 Gbps. If such lightpaths are physically infeasible, the claimed 'improvement is consistently realized across all destination nodes' could disappear or reverse.","section":"Section 3, Figure 1, and Section 5"},{"comment":"The traffic generation is underspecified. The text says that for each destination, 'from 10 remaining nodes forming 5 distinct computing requests (i.e., five pairs of sources (s1, s2))', but it does not state how the five pairs are chosen, whether the same five pairs are used for every destination node, or how many random instances were averaged. Since Figure 3 reports a single optimal wavelength count per destination, the claim that the improvement is 'consistently realized across all destination nodes' is only valid for the particular (unspecified) pair set. The authors should specify the exact traffic instances or report statistics over many randomly generated instances.","section":"Section 5, Figure 3"},{"comment":"The complexity statement that 'RWCA is one order of magnitude more complex in term of number of variables and constraints' than RWA is not backed by any variable/constraint counts or a formal complexity argument. While RWCA obviously extends RWA, the comparative complexity claim needs at least approximate counts from the actual formulation, which is currently absent.","section":"Section 4 and Section 5"}],"minor_comments":[{"comment":"Typo: 'ligthpath' should be 'lightpath'.","section":"Abstract"},{"comment":"The paper refers to 'the Section 5' twice: once for Simulation Results and once for the summary/conclusion; the latter should be Section 6.","section":"Introduction, last paragraph"},{"comment":"Algorithm 1 is not an algorithm in any standard sense; it is a loop over constraints with no search method, no initialization, and no termination condition. It should either be replaced with a description of the actual solution method (e.g., the CPLEX implementation details) or removed in favor of the explicit mathematical formulation.","section":"Algorithm 1"},{"comment":"The notation '(2,3,1)' for a computing request is not explicitly defined in the text; it would be helpful to state that the triple is (s1, s2, d).","section":"Section 3, Figure 1 and Table 2"},{"comment":"The paper claims to be 'the first contribution tapping into the opportunity of providing computing services at the optical layer' but immediately lists several preceding works by the same authors ([50,51,52]) and another arXiv preprint [47] on integrated computation and communication. The novelty claim should be qualified to reflect the specific new contributions of this manuscript.","section":"Section 2, contributions"},{"comment":"The mechanism for optical aggregation at an intermediate node is described only in vague terms ('coherent spectral superposition', 'electro-optic modulators'). Since this is the key enabling technology, a more detailed physical explanation or a reference to a demonstration of in-network aggregation (as opposed to transmitter-side aggregation) is needed.","section":"Section 3"}],"recommendation":"major_revision","confidential_remarks":"The paper relies heavily on the authors' own prior work ([50,51,52]) and makes a 'first' claim that may need careful positioning. More importantly, the absence of the actual MILP formulation and the lack of any physical-layer validation of the aggregation operation mean that the reported quantitative results cannot currently be trusted. If the authors can provide the full formulation, add a reach/OSNR constraint for the aggregated lightpaths, and rerun the experiments, the manuscript could become a useful contribution. If the optical aggregation at intermediate nodes proves physically infeasible, the central claim would need to be substantially rewritten."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does two things. It extends the authors' optical-computing-enabled network concept to a formulation of routing, wavelength and computing assignment (RWCA), and it reports a CPLEX-based comparison on COST239 showing the integrated network uses fewer wavelengths than optical bypass. The illustrative example is clear, and the idea of serving computing requests by optically aggregating two QPSK lightpaths into one 16-QAM lightpath is genuinely interesting. That aggregation is the new operational ingredient, and the wavelength savings in Figure 3 look plausible under the model's assumptions.\n\nWhat is missing: the MILP formulation. Section 4 gives a problem statement and a while-loop, not a mathematical program. So the CPLEX results are not independently checkable. Traffic generation is also underspecified: how the five source-pairs are selected per destination is not stated, which matters because the comparison has only 11 data points.\n\nThe soft spot that worries me most is the physical assumption. The model treats optical aggregation as free, lossless, and available at any node, with no reach penalty for the resulting 16-QAM lightpath. That is a strong assumption. 16-QAM needs about 6.7 dB more OSNR than QPSK, so at given symbol rate the reach is shorter. In the COST239 tables the computing node is often chosen at or near a source, which yields short aggregated lightpaths. But the optimization is free to choose any node, and for some destination/source combinations the optimal node might place the 16-QAM lightpath on a longer route. If that lightpath exceeds reach, the network would need regeneration or a lower-order format, and the reported savings could shrink or vanish. The paper does not discuss this, and the reader's concern is valid.\n\nOn the positive side, the paper is clearly written and the comparison is framed fairly: both designs are optimised by CPLEX, and the RWCA problem is a legitimate extension of RWA. The novelty is incremental relative to the authors' previous work, but it is real. I would consider this a conditional acceptance: the idea deserves a venue, but the revision must include the actual MILP, the traffic model, and a physical-layer constraint or at least an honest discussion of the reach issue.\n\nAs a colleague, I'd say send it back for major revision rather than desk reject. It is not a finished result, but it is a serious attempt at a new network capability, and the gaps are closable.","headline":"Interesting extension with a plausible quantitative result, but the missing MILP and the cost-free 16-QAM aggregation assumption keep the central claim conditional.","tokens_in":14265,"tokens_out":2642,"would_cite":false,"duration_ms":21793,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"By aggregating two lower-order optical signals into one higher-order signal at intermediate nodes, an optical network can serve computing requests with fewer wavelengths than optical-bypass routing on the COST239 topology.","keywords":["Optical communication-computing integrated network","Routing wavelength and computing assignment problem","Optical-computing-enabled network","In-network optical computing","Optical-bypass network","Optical-layer intelligence","Optical aggregation","Optical network design and planning 2.0"],"falsifier":"Measure whether the 16-QAM lightpath produced by coherent spectral superposition of two QPSK lightpaths at an intermediate node requires extra spectral resources, such as an additional guard band, higher launch power, or a different wavelength, compared with the two separate QPSK lightpaths; if it does, re-solving the COST239 instances with that physical constraint will show whether the wavelength savings survive.","tokens_in":1637,"feed_emoji":"🔆","tokens_out":2269,"duration_ms":104031,"temperature":0.7,"pith_summary":"This paper proposes an optical computing-communication integrated network in which intermediate optical nodes do more than route lightpaths past each other; they can also perform computing operations on lightpath entities, specifically optically aggregating two QPSK lightpaths into a single 16-QAM lightpath. The aim is to serve computing requests, such as distributed machine-learning aggregation of partial results from two source data centers to a destination, directly at the optical layer. To quantify the benefit, the paper formulates the routing, wavelength and computing assignment (RWCA) problem, an extension of the traditional routing and wavelength assignment (RWA) problem, and solves it optimally on the realistic COST239 topology. The central claim is that the integrated approach achieves greater spectral efficiency, measured as fewer wavelengths, than the optical-bypass model, and the paper reports that the improvement holds for every destination node in the studied cases.","feed_headline":"Optical in-network computing cuts wavelengths for computing requests","feed_subtitle":"Optical aggregation beats optical-bypass on every destination node in COST239 tests.","key_machinery":"The central object is the optical aggregation operation: an all-optical function, realized by coherent spectral superposition with electro-optic modulators, that combines two QPSK lightpaths (2 bits per symbol) into a single 16-QAM lightpath (4 bits per symbol) at an intermediate node. This operation converts what would be two wavelengths on shared links into one wavelength carrying the same combined information, which is the source of the spectral saving. It also introduces a new design dimension: choosing which node performs the computation. That choice, together with routing and wavelength assignment for the resulting lightpath segments, defines the routing, wavelength and computing assignment (RWCA) problem, which the paper models as a mixed integer linear program and contrasts with the traditional RWA problem.","core_discovery":"The central claim is that integrating optical computing into the optical layer can reduce the minimum number of wavelengths needed to serve computing requests. In the studied scenario, each computing request has two source nodes and one destination: in the optical-bypass network the request is treated as two communication demands carried by two separate lightpaths to the destination, where aggregation happens electronically; in the integrated network, the two sources send lightpaths to a common intermediate node, which performs an optical aggregation operation that turns two QPSK lightpaths (2 bits per symbol each) into one 16-QAM lightpath (4 bits per symbol) carrying both partial results, and that single lightpath continues to the destination. The wavelength saving is demonstrated by optimal solutions on COST239: across all destination nodes the integrated network uses fewer wavelengths than optical-bypass, and in a detailed traffic instance it uses 2 wavelengths instead of 3. The paper also presents this as the first formulation of serving computing services directly at the optical layer in fiber-optic networks.","pith_inferences":["Beyond the paper's two-to-one aggregation case, the same argument suggests that combining more than two lower-order lightpaths into one higher-order lightpath could give larger wavelength savings, but that scaling is not simulated.","The result is demonstrated for a many-to-one computing pattern (five source pairs, one destination); how the gain behaves under mixed communication and computing traffic or multiple simultaneous destinations remains an open question the paper does not address.","If physical-layer constraints force the aggregation to specific nodes or require extra guard bands, the wavelength saving may shrink; a sensitivity study varying the set of allowed computing nodes would test this."],"forward_implications":["On the COST239 instances studied, the integrated network uses fewer wavelengths than optical-bypass for every destination node, with a representative case needing 2 wavelengths instead of 3.","A computing request with two sources and one destination is served by choosing a computing node and establishing three lightpaths, or two if the computing node is one of the sources, which is why the new problem is called routing, wavelength and computing assignment.","The spectral saving comes from replacing two QPSK lightpaths on shared links with one 16-QAM lightpath, so the number of wavelength-link units drops while the carried information is preserved.","Higher destination nodal degree reduces the minimum number of wavelengths in both designs because more incoming fibers allow more wavelength re-use.","Because RWA is NP-hard and RWCA has an order of magnitude more variables and constraints, RWCA inherits that complexity, so the spectral gain comes with a higher optimization cost."],"supporting_citations":[{"why":"Supplies the optical-bypass architecture that the integrated network is compared against.","marker":"[11]"},{"why":"Defines the routing and wavelength assignment problem that RWCA extends and that serves as the baseline.","marker":"[17]"},{"why":"Shows integrated computation and communication in fiber-optic transmissions, the line of work this paper carries to serving computing requests at the optical layer.","marker":"[47]"},{"why":"Earlier work proposing the optical-computing-enabled network concept on which this architecture builds.","marker":"[50]"},{"why":"Earlier study of optical-computing-enabled networks from network-design perspectives that this paper extends with the RWCA formulation.","marker":"[51]"},{"why":"Describes optical channel aggregation by modulation-format conversion via coherent spectral superposition, the physical mechanism behind the QPSK-to-16-QAM operation.","marker":"[55]"},{"why":"Reports modulation-format aggregation of Nyquist channels by spectral superposition with electro-optic modulators, supporting the feasibility of the aggregation operation.","marker":"[56]"}],"fun_headline_variants":["Optical computing slashes wavelengths in integrated networks","Integrated optical computing needs fewer wavelengths","Optical-layer computing trims wavelength count","In-network optical computing saves wavelengths","Optical-layer computing beats bypass with fewer wavelengths"],"cache_read_input_tokens":16384,"weakest_assumption_plain":"The load-bearing premise is that any chosen intermediate node can aggregate two QPSK lightpaths into one 16-QAM lightpath at no extra spectral cost, with output capacity exactly equal to the combined input capacity, and with no physical-layer penalty.","fun_headline_variants_meta":{"raw":{"variants":["Optical computing slashes wavelengths in integrated networks","Integrated optical computing needs fewer wavelengths","Optical-layer computing trims wavelength count","In-network optical computing saves wavelengths","Optical-layer computing beats bypass with fewer wavelengths"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000756,"raw_usage":{"total_tokens":3403,"prompt_tokens":1030,"completion_tokens":2373,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":646,"completion_tokens_details":{"reasoning_tokens":2309}},"tokens_in":646,"tokens_out":2373,"duration_ms":15463,"temperature":1.0,"reasoning_tokens":2309,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:25:42.180672+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure whether the 16-QAM lightpath produced by coherent spectral superposition of two QPSK lightpaths at an intermediate node requires extra spectral resources, such as an additional guard band, higher launch power, or a different wavelength, compared with the two separate QPSK lightpaths; if it does, re-solving the COST239 instances with that physical constraint will show whether the wavelength savings survive.","supporting_citations":[{"cited_title":"Saleh, J","cited_arxiv_id":null,"evidence_quote":"Supplies the optical-bypass architecture that the integrated network is compared against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier work proposing the optical-computing-enabled network concept on which this architecture builds."},{"cited_title":"Misra, S","cited_arxiv_id":null,"evidence_quote":"Describes optical channel aggregation by modulation-format conversion via coherent spectral superposition, the physical mechanism behind the QPSK-to-16-QAM operation."},{"cited_title":"Misra, S","cited_arxiv_id":null,"evidence_quote":"Reports modulation-format aggregation of Nyquist channels by spectral superposition with electro-optic modulators, supporting the feasibility of the aggregation operation."}],"review_version":1}