An ML Approach for Crosstalk-Aware Modulation Format Selection in SDM-EONs

Dual optimization for enhancing TRA in MCF-based SDM-EONs

“…First, machine learning (ML) can be used to replicate the strategy of selection of RCs by TRA and find RCs and optimal weights together without the need to re-run the network for each combination of weights and RCs. Second, a ML-aided standard template to optimize any RMCSA algorithm, similar to our previous work [10], can be used to obtain guaranteed improvements.…”

“…XT due to parallel transmissions on overlapping spectrum only affects neighboring cores, and cores that are not adjacent remain unaffected. For instance, in a three-core fiber, each core has two neighboring cores [10]. Similarly, in a seven-core fiber, the outer cores are surrounded by three adjacent cores, while the center core is encircled by six adjacent cores [10].…”

“…For instance, in a three-core fiber, each core has two neighboring cores [10]. Similarly, in a seven-core fiber, the outer cores are surrounded by three adjacent cores, while the center core is encircled by six adjacent cores [10]. Moreover, we ensure spatial continuity, meaning that each lightpath is assigned to the same core throughout its path.…”

“…However, each parameter in TC contributes differently in improving the performance of TRA. Thus, we weighted the three parameters in (9) with α, β and (1 − α − β) as shown in (10) such that 0 ≤ α, β, (α + β) ≤ 1. All the results shown in [13] are obtained when the weights are 1 in (9), which is similar to α=β= 1 3 in (10).…”

“…The weighted TC is given in (10) and α and β are used to weight the parameters. The optimal weights of α and β lead to better performance of TRA as compared to all the weights are equal.…”