Correction to “Nonlinear Distortion Equalization in Multi‐Span Optical Links Via a Feed‐Forward Photonic Neural Network”

[E. Staffoli, E. Ferri, S. Gretter, L. Pavesi, Laser Photonics Rev. 2025;19:e00855.] This note clarifies a limited set of points in the published article. All experimental and simulated datasets, figures, performance trends, and conclusions are unchanged. The correction concerns (i) the interpretation of the parameter P i n $P_{in}$ in the experimental recirculating-loop description, (ii) two parameter values/signs reported in Table 2, (iii) the distinction between experimental and simulated power restoration across loops, and (iv) a few typos in the text. We apologize for any inconvenience that may have occurred. In the experimental sections, P i n $P_{in}$ must be read as the optical power entering the fiber span at the first loop only. In the measurements, the loop gain was adjusted for every transmission scenario to ensure that the power reaching the receiver matches the same reference value used for the 0 km case (including the effect of transients in the final amplification stages). As a consequence, the power entering the fiber is not strictly constant over all recirculations and can deviate from the nominal value set by P i n $P_{in}$ . The deviation is limited: in the worst-case configuration ( P i n = 9 $P_{in}=9$ dBm at the first loop), the input power in the fiber decreases by a total maximum of 4 dB across the loops, while still preserving the intended nonlinear propagation regime. Therefore, throughout the paper, any statement suggesting that P i n $P_{in}$ is identically restored at every loop in the experimental setup should be interpreted as “the loop gain is tuned to meet a fixed RX reference power”, not as “a constant per-span fiber input power”. Table 2 contains two reporting inaccuracies. The value of the “Losses in a single loop round-trip” must be l s = 26 $l_s=26$ dB (not 18 dB). In addition, the group-velocity-dispersion coefficient β 2 $\beta _2$ must carry a negative sign (consistent with the sign convention implied by a positive dispersion parameter D $D$ ). These are table/reporting corrections only; they do not alter any plotted results or conclusions. The simulations differ from the experimental loop with respect to power restoration. In the simulated scenarios, the input power in the fiber is restored exactly to the target values (e.g., P p r $P_{pr}$ and, when applicable, P p u $P_{pu}$ ) at every loop. Thus, statements in the paper about exact per-loop restoration apply to the simulated environment, whereas the experimental environment exhibits the limited loop-to-loop power variation described above. With this interpretation, the comparison between experiments and simulations remains valid, and the main conclusions of the work are unchanged. Finally, typo corrections are applied as follows. In the first paragraph of Section 3.2, the worst-case scenario for SPM equalization is obtained by fixing the probe input power in fiber to P p r = 10 $P_{pr} = 10$ dBm. In the penultimate sentence at the end of Section 4, the 13 dB BER reduction provided by the PNN in 100G SPM equalization (simulated) is obtained for P p r ≈ 8 $P_{pr} \approx 8$ dBm and L ≈ 350 $L \approx 350$ km. After Equation (4) in Section 5, the correct VOA to be included among the contributions to the terms l s $l_s$ is VOA 2, according to Figure 13.