Intent-based networking (IBN) is a natural trend in communications networks with increasing automation levels. IBN provides an intermediate layer that abstracts complex engineering rules from the operator, interpreting high-level business intents and translating them into network configurations. Although IBN has received considerable attention for packet and wireless networks, the research in the field of optical IBN is still in its early stages of development. Indeed, the examination of optical-layer survivability as an intent criterion has not been explored yet. Recently, we showcased an intent-based survivability system built around a failure management module (FMM) connected to a network digital twin (NDT). In this paper, we expand upon this initial work by detailing the FMM and quantitatively assessing the metrics used in the demonstration. In accordance with the current practice, IBN systems should offer the user the ability to request traditional survivability metrics, such as the existence of protection or availability. However, we suggest that an IBN system should also warn the user of the limitations of these approaches. In particular, we show that using traditional asymptotic (steady-state) availability as an intent metric incurs high non-compliance risks. Therefore, we recommend the implementation of risk calculations in intent translators as an auxiliary resource for intent-based survivability.

In optical networks, assessing the quality of transmission (QoT) of future demands is critical for optimized network operation. Given the recent advances in network telemetry and Big Data processing, there has been a growing trend to leverage the telemetry of existing connections to enhance the QoT estimation for future demands. Currently, QoT estimation techniques are categorized into three classes: physics-informed analytical models, machine learning (ML) models, and hybrid models, combining the first two classes. Hybrid models have the potential to achieve accurate QoT estimation with reasonable explainability. In this work, we propose a hybrid QoT estimation technique based on gradient transfer between wavelengths using a simplified network digital twin. The proposed approach relies on the GN model to create a starting point NDT (greenfield), which is then updated with signal-to-noise ratio (SNR) telemetry from deployed transponders (brownfield) using gradient descent techniques. By assuming that neighboring channels have similar performance, the method enables learning from adjacent lightpaths. We use well-established ML optimizers to enhance the convergence rate and reduce steady-state estimation error. Compared with classical analytical models, simulation results indicate improved accuracy with reasonable explainability in network scenarios with imperfect amplifier parameters.

In the evolving landscape of 5G and 6G networks, the demands extend beyond high data rates, ultra-low latency, and extensive coverage, increasingly emphasizing the need for reliability. This paper proposes an ultra-reliable multiple-input multiple-output (MIMO) scheme utilizing quasi-orthogonal space-time block coding (QOSTBC) combined with singular value decomposition (SVD) for channel state information (CSI) correction, significantly improving performance over QOSTBC and traditional orthogonal STBC (OSTBC) when analyzing spectral efficiency. Although QOSTBC enhances spectral efficiency, it also increases computational complexity at the maximum likelihood (ML) decoder. To address this, a neural network-based decoding scheme using phase-transmittance radial basis function (PT-RBF) architecture is also introduced to manage QOSTBC's complexity. Simulation results demonstrate improved system robustness and performance, making this approach a potential candidate for ultra-reliable communication in next-generation networks.

We propose an unsupervised machine-learning (ML) method to localize faulty optical amplifiers within large optical networks using power time-series data. An amplifier is considered faulty if an anomaly is detected at its output power but not at its input power, indicating that the anomaly is generated in the monitored device. Time-series anomalies are detected using an ML-based prediction engine that has been adapted to replace a proprietary core with widely adopted algorithms. The results indicate effective anomaly detection performance and demonstrate the successful localization of a faulty amplifier within a real network.

Artificial neural networks (ANNs) have become a popular tool in digital signal processing (DSP). Among the widespread ANN architectures, complex-valued neural networks (CVNNs) have been extensively studied in image processing and telecommunications. Unlike their real-valued counterparts, CVNNs can handle signals directly in the complex domain. Due to this capability, CVNNs usually exhibit higher accuracy and improved convergence compared to real-valued neural networks (RVNNs). Despite their improved performance in several applications, CVNNs still lag behind RVNNs in terms of learning techniques and heuristics. In this context, we propose adaptive learning rate approaches for CVNNs, extending the well-known adaptive gradient (AdaGrad), root-mean-square propagation (RMSProp), AdaMax, AMSGrad, softplus AMSGrad (SAMSGrad), Nesterov-accelerated adaptive moment estimation (Nadam), and DiffGrad to the complex domain. Computational complexities of the proposed optimizers are analyzed for CVNN architectures. Results are compared in terms of mean-squared-error convergence.

ML-based QoT estimation has been recently investigated as a relevant method for the planning and operation of optical networks. While analytical Gaussian noise models offer suitable QoT estimates, their accuracy is considerably impaired if the model parameters differ substantially from those encountered in real physical networks. To alleviate this problem, ML-based QoT estimation offers better QoT estimates if the model is trained with performance parameters obtained from established field connections. The class of QoT estimation algorithms considered in this paper receives structural parameters as inputs and produces an estimated GSNR as output. We consider the case of greenfield network deployment, where a new network is implemented from scratch based on the QoT data collected from other networks. The impact of a heterogeneous training dataset with the accumulation of amplifier imperfections on the performance of the QoT estimator is investigated. We use Shapley additive explanations to quantify the contribution of particular structural parameters to ML-based GSNR estimation and reduce QoT estimation complexity. Our findings reveal that while QoT estimation for large networks remains challenging due to unlearnable random perturbations, even with training data from similarly large networks, estimation for small networks exhibits robust performance, even when trained with diverse datasets including larger networks.

With the advent of sixth-generation (6 G) communication networks, artificial intelligence (AI) is recognized as a promising technological enabler due to the paradigm shift it brings to various processing functions. Consequently, AIbased algorithms, including both real- and complex-valued neural networks (NNs), have been proposed in the literature, although they are typically evaluated using simulated data. In this work, an over-the-air (OTA) experimental setup is employed to gather real-world datasets to assess NNs applied to joint channel estimation and equalization in $M$-ary quadrature amplitude modulation (QAM) systems. A comparative analysis between real- and complex-valued NN architectures is also provided, carefully matched in terms of learnable parameters to ensure an unbiased evaluation. The channel environments consider both line-of-sight (LoS) and non-line-of-sight (nLoS) propagation scenarios, aiming to provide insights into the impact of realistic channel conditions on the performance of NN-driven physical layer functionalities.

In the ever-evolving field of digital communication systems, complex-valued neural networks (CVNNs) have become a cornerstone, delivering exceptional performance in tasks like equalization, channel estimation, beamforming, and decoding. Among the myriad of CVNN architectures, the phase-transmittance radial basis function neural network (PT-RBF) stands out, especially when operating in noisy environments such as 5G MIMO systems. Despite its capabilities, achieving convergence in multi-layered, multi-input, and multi-output PT-RBFs remains a daunting challenge. Addressing this gap, this paper presents a novel Deep PT-RBF parameter initialization technique. Through rigorous simulations conforming to 3GPP TS 38 standards, our method not only outperforms conventional initialization strategies like random, K-means, and constellation-based methods but is also the only approach to achieve successful convergence in deep PT-RBF architectures. These findings pave the way to more robust and efficient neural network deployments in complex digital communication systems.

In the ever-evolving field of artificial neural networks and learning systems, complex-valued neural networks (CVNNs) have become a cornerstone, achieving exceptional performance in image processing and telecommunications. More precisely, in digital communication systems, CVNNs have been delivering significant results in tasks like equalization, channel estimation, beamforming, and decoding. Among the CVNN architectures, the complex-valued radial basis function neural network (C-RBF) stands out, especially when operating in noisy environments such as 5G multiple-input multiple-output (MIMO) systems. In such a context, this paper extends the classical shallow C-RBF to deep architectures, increasing its flexibility for a wider range of applications. Also, based on the parameter selection of the phase transmittance radial basis function (PT-RBF) neural network, we propose an initialization scheme for the deep C-RBF. Via rigorous simulations conforming to 3GPP TS 38 standards for digital communications, our method not only outperforms conventional initialization strategies like random, K-means, and constellation-based methods but it also seems to be the only approach to achieve successful convergence for deep C-RBF architectures. These findings pave the way to more robust and efficient neural network deployments in complex-valued digital communication systems.

This paper presents an online method for joint channel estimation and decoding in massive MIMO-OFDM systems using complex-valued neural networks (CVNNs). The study evaluates the performance of various CVNNs, such as the complex-valued feedforward neural network (CVFNN), split-complex feedforward neural network (SCFNN), complex radial basis function (C-RBF), fully-complex radial basis function (FC-RBF) and phase-transmittance radial basis function (PT-RBF), in realistic 5G communication scenarios. Results demonstrate improvements in mean squared error (MSE), convergence, and bit error rate (BER) accuracy. The C-RBF and PT-RBF architectures show the most promising outcomes, suggesting that RBF-based CVNNs provide a reliable and efficient solution for complex and noisy communication environments. These findings have potential implications for applying advanced neural network techniques in next-generation wireless systems.

Deep learning is an essential artificial intelligence tool broadly used in engineering, physics, data science, biology, healthcare, agribusiness, finance, and many other areas. Current Python frameworks for deep learning, such as TensorFlow, Keras, PyTorch, and scikit-learn, only solve real-domain problems, representing a considerable part of real-world applications but not all. For instance, complex-valued signals are essential for current and future technologies in telecommunications. Thus far, numerous works employing real-valued neural networks adapted to complex-domain processing, end up generating sub-optimal results. To fulfill this demand, this article presents RosenPy, an open-source framework in Python for complex-valued neural networks.

We propose an SGD-based QoT estimation technique that operates on a network-wide scale by transferring gradients among neighboring wavelengths. Simulation results indicate effective and low-complexity QoT estimation using only transponder SNR telemetry.

The increasing demands of modern telecommunications require improvements in spectral efficiency and system throughput. In this context, our study introduces a novel decoding method for MIMO-OFDM systems employing parallel neural networks, which markedly enhances decoding speed and accuracy over previous models. Unlike serial decoding, which fails to address the unique characteristics of individual subcarriers, our method employs distinct phase-transmittance radial basis function (PT-RBF) neural networks for each subcarrier. This parallel processing approach significantly reduces decoding time and increases system adaptability by effectively managing nonlinear impairments and intersymbol interference. Simulation results show that our method outperforms conventional decoding techniques in reducing bit error rate (BER) across both linear and nonlinear scenarios.

We propose a DNN-based QoT estimation technique that operates on a network-wide scale. The DNN training data is composed of connection paths, frequency slots, and OSNR transponder telemetry, collected from both synthetic and physical connections. Simulation results indicate effective and low-complexity QoT estimation.

We demonstrate a packet-optical differentiated survivability mechanism implemented in P4 switches using gNMI telemetry. The controller detects the premium slice interruption and switches it to an alternative path reducing the throughput of the regular slice.

This work presents Chat-IBN-RASA, a conversational AI chatbot based on the open-source RASA framework, acting as an Intent Translator component for Intent-Based Networking (IBN) architectures. The IBN-driven RASA chatbot allows users to interact with the network management system using high-level language communication without the need for in-depth technical knowledge of the network. The chatbot is trained using Natural Language Understanding (NLU) models, a defined domain, stories, and custom actions for API communication and database queries. We present the prototype implementation in a use case of survivability intents in packet-optical networks. The custom actions featured include communication with the network database, path computation, and the recommended path for intent deployment, considering availability, protection type, data rate, and link distances.

In modern communication systems operating with Orthogonal Frequency-Division Multiplexing (OFDM), channel estimation requires minimal complexity with one-tap equalizers. However, this depends on cyclic prefixes, which must be sufficiently large to cover the channel impulse response. Conversely, the use of cyclic prefix (CP) decreases the useful information that can be conveyed in an OFDM frame, thereby degrading the spectral efficiency of the system. In this context, we study the impact of CPs on channel estimation with complex-valued neural networks (CVNNs). We show that the phase-transmittance radial basis function neural network offers superior results, in terms of required energy per bit, compared to classical minimum mean-squared error and least squares algorithms in scenarios without CP.

This letter proposes the use of quasi-orthogonal space-time block codes (QOSTBC) to enhance link quality and reliability in massive multiple-input multiple-output (m-MIMO) systems subject to independent fading in dynamic channels. It has been shown, however, that the computational complexity of classical decoding algorithms, such as maximum likelihood, can hinder the adoption of QOSTBC codes in systems with many antennas and high-order modulation schemes. Complex-valued neural networks (CVNNs) offer a promising alternative for joint decoding and channel estimation with competitive computational complexity. This letter presents an extension of our previously proposed CVNN with supervised training, which incorporates two semi-supervised learning techniques: hard inference learning (HIL) and Gaussian inference learning (GIL). By leveraging non-pilot-aided data, HIL and GIL enable the CVNNs to self-learn from useful information, increasing their tracking ability and robustness in dynamic channels.

Complex-valued neural networks (CVNNs) are nonlinear filters used in the digital signal processing of complex-domain data. Compared with real-valued neural networks~(RVNNs), CVNNs can directly handle complex-valued input and output signals due to their complex domain parameters and activation functions. With the trend toward low-power systems, computational complexity analysis has become essential for measuring an algorithm's power consumption. Therefore, this paper presents both the quantitative and asymptotic computational complexities of CVNNs. This is a crucial tool in deciding which algorithm to implement. The mathematical operations are described in terms of the number of real-valued multiplications, as these are the most demanding operations. To determine which CVNN can be implemented in a low-power system, quantitative computational complexities can be used to accurately estimate the number of floating-point operations. We have also investigated the computational complexities of CVNNs discussed in some studies presented in the literature.