We present Darts, a Python machine learning library for time series, with a focus on forecasting. Darts offers a variety of models, from classics such as ARIMA to state-of-the-art deep neural networks. The emphasis of the library is on offering modern machine learning functionalities, such as supporting multidimensional series, fitting models on multiple series, training on large datasets, incorporating external data, ensembling models, and providing a rich support for probabilistic forecasting. At the same time, great care goes into the API design to make it user-friendly and easy to use. For instance, all models can be used using fit()/predict(), similar to scikit-learn.

Power-line communications (PLC) are becoming a key component in home networking, because they provide easy and high-throughput connectivity. The dominant MAC protocol for high data-rate PLC, the IEEE 1901, employs a CSMA/CA mechanism similar to the backoff process of 802.11. Existing performance evaluation studies of this protocol assume that the backoff processes of the stations are independent (the so-called decoupling assumption). However, in contrast to 802.11, 1901 stations can change their state after sensing the medium busy, which is regulated by the so-called deferral counter. This mechanism introduces strong coupling between the stations and, as a result, makes existing analyses inaccurate. In this paper, we propose a performance model for 1901, which does not rely on the decoupling assumption. We prove that our model admits a unique solution for a wide range of configurations and confirm the accuracy of the model using simulations. Our results show that we outperform current models based on the decoupling assumption. In addition to evaluating the performance in steady state, we further study the transient dynamics of 1901, which is also affected by the deferral counter.

Several technologies, such as WiFi, Ethernet and power-line communications (PLC), can be used to build residential and enterprise networks. These technologies often co-exist; most networks use WiFi, and buildings are readily equipped with electrical wires that can offer a capacity up to 1 Gbps with PLC. Yet, current networks do not exploit this rich diversity and often operate far below the available capacity.

Power-line communications are employed in home networking to provide easy and high-throughput connectivity. The IEEE 1901, the MAC protocol for power-line networks, employs a CSMA/CA protocol similar to that of 802.11, but is substantially more complex, which probably explains why little is known about its performance. One of the key differences between the two protocols is that whereas 802.11 only reacts upon collisions, 1901 also reacts upon several consecutive transmissions and thus can potentially achieve better performance by avoiding unnecessary collisions. In this paper, we propose a model for the 1901 MAC. Our analysis reveals that the default configuration of 1901 does not fully exploit its potential and that its performance degrades with the number of stations. Based on analytical reasoning, we derive a configuration for the parameters of 1901 that drastically improves throughput and achieves optimal performance without requiring the knowledge of the number of stations in the network. In contrast, 802.11 requires knowing the number of contending stations to provide a similar performance, which is unfeasible for realistic traffic patterns. We confirm our results and enhancement with testbed measurements, by implementing the 1901 MAC protocol on WiFi hardware.

It has recently been shown that "flexible channelization", whereby wireless stations adapt their spectrum bands on a per-frame basis, is feasible in practice. In this paper, we propose TF-CSMA/CA, an algorithm for flexible channelization that schedules packets in time and frequency domains. TF-CSMA/CA is a simple extension of the CSMA/CA protocol used by IEEE 802.11. Contrary to existing channelization schemes, it is entirely distributed and it reacts only to packet collisions, successful transmissions and carrier sensing. With TF-CSMA/CA, when a station is involved in a collision, it performs backoff in both time and frequency domains. Backing off also in the frequency domain allows the transmitters to be much more efficient and aggressive in the time domain, which significantly reduces the severe overheads present with recent 802.11 PHY layers. The main challenge, however, is that the stations need some level of self-organization in order to find spectrum bands of variable widths that minimize interference, while still efficiently using the available spectrum. Using analysis and simulations, we show that such an extension of CSMA/CA to the frequency domain drastically improves both throughput and fairness. Notably, it enables the stations to find interference-free spectrum bands of appropriate size using no communication -- relying only on collisions and successes as implicit signals.

Accurate prediction of wireless network performance is important when performing link adaptation or resource allocation. However, the complexity of interference interactions at MAC and PHY layers, as well as the vast variety of possible wireless configurations make it notoriously hard to design explicit performance models. In this paper, we advocate an approach of “learning by observation” that can remove the need for designing explicit and complex performance models. We use machine learning techniques to learn implicit performance models, from a limited number of real-world measurements. These models do not require to know the internal mechanics of interfering Wi-Fi links. Yet, our results show that they improve accuracy by at least 49% compared to measurement-seeded models based on SINR. To demonstrate that learned models can be useful in practice, we build a new algorithm that uses such a model as an oracle to jointly allocate spectrum and transmit power. Our algorithm is utility-optimal, distributed, and it produces efficient allocations that significantly improve performance and fairness.

Power-line communications are employed in home networking to provide easy and high-throughput connectivity. IEEE 1901, the MAC protocol for power-line networks, employs a CSMA/CA protocol similar to that of 802.11, but is substantially more complex, which probably explains why little is known about its performance. One of the key differences between the two protocols is that whereas 802.11 only reacts upon collisions, 1901 also reacts upon several consecutive transmissions and thus can potentially achieve better performance by avoiding unnecessary collisions.

Power-line communications are becoming a key component in home networking. The dominant MAC protocol for high data-rate power-line communications, IEEE 1901, employs a CSMA/CA mechanism similar to the back off process of 802.11. Existing performance evaluation studies of this protocol assume that the back off processes of the stations are independent (the so-called decoupling assumption). However, in contrast to 802.11, 1901 stations can change their state after sensing the medium busy, which introduces strong coupling between the stations and, as a result, makes existing analyses inaccurate. In this paper, we propose a new performance model for 1901, which does not rely on the decoupling assumption. We prove that our model admits a unique solution. We confirm the accuracy of our model using both test bed experiments and simulations, and we show that it surpasses current models based on the decoupling assumption. Furthermore, we study the trade off between delay and throughput existing with 1901. We show that this protocol can be configured to accommodate different throughput and jitter requirements, and we give systematic guidelines for its configuration.

We investigate the IEEE 1901 MAC protocol, the dominant protocol for high data rate power-line communications. 1901 employs a CSMA/CA mechanism similar to - but much more complex than - the backoff mechanism of 802.11. Because of this extra complexity, and although this mechanism is the only widely used MAC layer for power-line networks, there are few analytical results on its performance. We propose a model for the 1901 MAC that comes in the form of a single fixed-point equation for the collision probability. We prove that this equation admits a unique solution, and we evaluate the accuracy of our model by using simulations.

We consider the problem of jointly allocating channel center frequencies and bandwidths for IEEE 802.11 wireless LANs (WLANs). The bandwidth used on a link affects significantly both the capacity experienced on this link and the interference produced on neighboring links. Therefore, when jointly assigning both center frequencies and channel widths, there is a trade-off between interference mitigation and the potential capacity offered on each link. We study this tradeoff and we present SAW (spectrum assignment for WLANs), a decentralized algorithm that finds efficient configurations. SAW is tailored for 802.11 home networks. It is distributed, online and transparent. It does not require a central coordinator and it constantly adapts the spectrum usage without disrupting network traffic. A key feature of SAW is that the access points (APs) need only a few out-of-band measurements in order to make spectrum allocation decisions. Despite being completely decentralized, the algorithm is self-organizing and provably converges towards efficient spectrum allocations. We evaluate SAW using both simulation and a deployment on an indoor testbed composed of off-the-shelf 802.11 hardware. We observe that it dramatically increases the overall network efficiency and fairness.

We present PIE, a scalable routing scheme that achieves 100% packet delivery and low path stretch. It is easy to implement in a distributed fashion and works well when costs are associated to links. Scalability is achieved by using virtual coordinates in a space of concise dimensionality, which enables greedy routing based only on local knowledge. PIE is a general routing scheme, meaning that it works on any graph. We focus however on the Internet, where routing scalability is an urgent concern. We show analytically and by using simulation that the scheme scales extremely well on Internet-like graphs. In addition, its geometric nature allows it to react efficiently to topological changes or failures by finding new paths in the network at no cost, yielding better delivery ratios than standard algorithms. The proposed routing scheme needs an amount of memory polylogarithmic in the size of the network and requires only local communication between the nodes. Although each node constructs its coordinates and routes packets locally, the path stretch remains extremely low, even lower than for centralized or less scalable state-of-the-art algorithms: PIE always finds short paths and often enough finds the shortest paths.

The goal of jointly providing efficiency and fairness in wireless networks can be seen as the problem of maximizing a given utility function. In contrast with wired networks, the capacity of wireless networks is typically time-varying and not known explicitly. Hence, as the capacity region is impossible to know or measure exactly, existing scheduling schemes either under-estimate it and are too conservative, or they over-estimate it and suffer from congestion collapse. We propose a new adaptive algorithm, called Enhance & Explore (E&E). It maximizes the utility of the network without requiring any explicit characterization of the capacity region. E&E works above the MAC layer and it does not demand any modification to the existing networking stack. We first evaluate our algorithm theoretically and we prove that it converges to a state of optimal utility. We then evaluate the performance of the algorithm in a WLAN setting, using both simulations and real measurements on a testbed composed of IEEE 802.11 wireless routers.