Photonic Network Communications manuscript No. (will be inserted by the editor) Local and Global Approaches for the Adaptive Control of a Cascade of Ampliﬁers Erick de A. Barboza, · Carmelo J. A. Bastos-Filho · Joaquim F. Martins Filho · Marcionilo J. da Silva · Leonardo D. Coelho · Uiara C. de Moura · Juliano R. F. de Oliveira Received: date / Accepted: date Abstract In this work we present and analyze three approaches for the adaptive control of the operating point of a cascade of Erbium-doped ﬁber ampliﬁers (EDFA), aiming optical networks performance enhance- ment. The ﬁrst approach is called Annealing Search Heuristic with Backpropagation and ﬂexible output (AsHB ﬂex ) and uses machine learning concepts to up- date the ampliﬁers gains through an iterative process. The second one (Exhaustive Method ) uses an exhaus- tive search to evaluate all possible solutions for the problem and obtain the optimum solution. The last one (MaxGain ) is a heuristic method that uses pre- vious knowledge about the problem to obtain the solu- tions. The ampliﬁer characteristics and speciﬁcations were obtained experimentally through measurements of gain, noise ﬁgure, gain ripple and power consump- tion on commercially available EDFAs. We performed comparisons among these approaches and others found in the literature, and the results show that the three proposals outperformed the previous ones in terms of noise ﬁgure, gain ripple and BER. For example, in a link with 4 ampliﬁers the Exhaustive Method achieved a reduction of the cascade noise ﬁgure from 10.05 dB to 5.18 dB, a reduction of the gain ripple from 24.08 dB to 18.56 dB and a reduction of the BER in almost two orders of magnitude, when compared with the tradi- Erick de A. Barboza, Joaquim F. Martins Filho, Mar- cionilo J. da Silva and Leonardo D. Coelho Department of Electronics and Systems, Federal University of Pernambuco, Recife, PE, 50740-550 Brazil E-mail: jfmf@ufpe.br C. A. J. Bastos-Filho Polytechnic School of Pernambuco, University of Pernam- buco, Recife, PE, 50720-001 Brazil U. C. Moura and J. R. F. Oliveira CPqD, Campinas, SP, Brazil tional approach, which deﬁnes the gain to compensate the loss of the previous link. However, the computation time of Exhaustive Method becomes prohibitive as the number of ampliﬁers in the link increases. Both Max- Gain and AsHB ﬂex obtained similar solutions, close to the optimum operation point in a reasonable time. Keywords Optical ampliﬁer, Dynamic Optical Network, Adaptive Control of Operating Point, Machine Learning, Heuristics, Optical communication. 1 Introduction The high levels of heterogeneity, from transmission tech- nologies to type of services, are creating new challenges for the optical network technologies. One of the key issues for this new kind of network (heterogeneous dy- namic network) is the creation of eﬃcient mechanisms to control and manage the network devices and con- ﬁgurations. In order to face this issue, while fulﬁlling user demands and guarantying the quality of service (QoS), the network needs to present some features, such as self-conﬁguration, self-healing, self-optimization and self-protection functionalities, which implies in an au- tonomic and reconﬁgurable network that can take de- cisions without human intervention [13]. The role of developing autonomic networks is having the attention of the researchers, and recent works have shown some proposals as autonomic signal modulation format recognition [9], autonomic QoT estimator [8] and autonomic virtual topology design [5]. Especially in these works, it is used the concept of a cognitive net- work, which can be described as an autonomic network that enhances its performance along the time due to learning processes.