Recent research shows that integrating artificial intelligence (AI) into wireless communication systems can significantly improve spectral efficiency. However, most AI-based receiver studies rely on simulated radio channel data for both training and validation, raising concerns about real-world generalization, which is vital for ensuring reliable field performance. In this study, we train DeepRx, a convolutional neural network (CNN)-based OFDM receiver, under various simulated channel scenarios and validate its performance over- the-air (OTA) using software-defined radio (SDR) technology in a small cell-type setup. To enhance receiver training, we investigate a randomized 3GPP TS38.901 channel model to diversify the training data, thereby improving performance over conventional receivers and matching or exceeding the performance of receivers trained on narrowly targeted channel models. These results demonstrate DeepRx's robust generalization capability and suggest that narrowly scoped, individual TS38.901 models can compromise both training and validation, underscoring the need for tailored channel models, careful training strategies, and OTA testing in learned receiver development.

Machine learning algorithms have recently been considered for many tasks in the field of wireless communications. Previously, we have proposed the use of a deep fully convolutional neural network (CNN) for receiver processing and shown it to provide considerable performance gains. In this study, we focus on machine learning algorithms for the transmitter. In particular, we consider beamforming and propose a CNN which, for a given uplink channel estimate as input, outputs downlink channel information to be used for beamforming. The CNN is trained in a supervised manner considering both uplink and downlink transmissions with a loss function that is based on UE receiver performance. The main task of the neural network is to predict the channel evolution between uplink and downlink slots, but it can also learn to handle inefficiencies and errors in the whole chain, including the actual beamforming phase. The provided numerical experiments demonstrate the improved beamforming performance.

In this article, we propose multiple machine learning (ML) based physical-layer receiver solutions for demodulating orthogonal frequency-division multiplexing (OFDM) signals that are subject to high level of nonlinear distortion. Specifically, three novel deep learning based convolutional neural network receivers are devised, containing layers in time- and/or frequency-domains, allowing to demodulate and decode the transmitted bits reliably despite the high error vector magnitude (EVM) in the transmit signal. Applicable training procedures are also described, such that the learned layers in the receiver processing properly generalize over different nonlinear distortion and multipath channel characteristics. Extensive set of numerical results is provided, in the context of 5G NR uplink (UL) incorporating also measured terminal power amplifier (PA) characteristics. The obtained results show that the proposed receiver systems are able to clearly outperform the classical linear minimum mean-squared error (LMMSE) receiver as well as the existing ML receiver approaches, especially when the EVM is high compared to modulation order. This is particularly so when the devised ML receiver is of hybrid nature with layers both in time and frequency. The proposed ML receivers can thus facilitate pushing the terminal PA systems deeper into saturation, and thereon improve the terminal power-efficiency, radiated power and network coverage. Through combining the obtained radio link performance results with link budget calculations, all carried out at the 28 GHz mmWave band, it is shown that the proposed ML receivers can enhance the network coverage in terms of maximum UL link distances by close to 100%, when compared to classical LMMSE receiver based networks.

Recently, deep learning has been proposed as a potential technique for improving the physical layer performance of radio receivers. Despite the large amount of encouraging results, most works have not considered spatial multiplexing in the context of multiple-input and multiple-output (MIMO) receivers. In this paper, we present a deep learning-based MIMO receiver architecture that consists of a ResNet-based convolutional neural network, also known as DeepRx, combined with a so-called transformation layer, all trained together. We propose two novel alternatives for the transformation layer: a maximal ratio combining-based transformation, or a fully learned transformation. The former relies more on expert knowledge, while the latter utilizes learned multiplicative layers. Both proposed transformation layers are shown to clearly outperform the conventional baseline receiver, especially with sparse pilot configurations. To the best of our knowledge, these are some of the first results showing such high performance for a fully learned MIMO receiver.

In this paper, we propose a machine learning (ML) based physical layer receiver solution for demodulating OFDM signals that are subject to a high level of nonlinear distortion. Specifically, a novel deep learning based convolutional neural network receiver is devised, containing layers in both time- and frequency domains, allowing to demodulate and decode the transmitted bits reliably despite the high error vector magnitude (EVM) in the transmit signal. Extensive set of numerical results is provided, in the context of 5G NR uplink incorporating also measured terminal power amplifier characteristics. The obtained results show that the proposed receiver system is able to clearly outperform classical linear receivers as well as existing ML receiver approaches, especially when the EVM is high in comparison with modulation order. The proposed ML receiver can thus facilitate pushing the terminal power amplifier (PA) systems deeper into saturation, and thereon improve the terminal power-efficiency, radiated power and network coverage.

Deep learning has solved many problems that are out of reach of heuristic algorithms. It has also been successfully applied in wireless communications, even though the current radio systems are well-understood and optimal algorithms exist for many tasks. While some gains have been obtained by learning individual parts of a receiver, a better approach is to jointly learn the whole receiver. This, however, often results in a challenging nonlinear problem, for which the optimal solution is infeasible to implement. To this end, we propose a deep fully convolutional neural network, DeepRx, which executes the whole receiver pipeline from frequency domain signal stream to uncoded bits in a 5G-compliant fashion. We facilitate accurate channel estimation by constructing the input of the convolutional neural network in a very specific manner using both the data and pilot symbols. Also, DeepRx outputs soft bits that are compatible with the channel coding used in 5G systems. Using 3GPP-defined channel models, we demonstrate that DeepRx outperforms traditional methods. We also show that the high performance can likely be attributed to DeepRx learning to utilize the known constellation points of the unknown data symbols, together with the local symbol distribution, for improved detection accuracy.

The purpose of this study was to analyze the factors affecting motivation and whether these factors support students' autonomy. Students were interviewed in order to study the impact of collaborative learning study arrangements on the their motivation. The data was then analyzed using self-determination theory. The study showed that it is possible to build structured courses utilizing collaborative learning and interactive engagement methods such that autonomy is supported. The interviews showed that if there is room for autonomously motivated students to choose their approach to work in the groups, the group work engages students through social relatedness.

Bidirectional recurrent neural networks (RNN) are trained to predict both in the positive and negative time directions simultaneously. They have not been used commonly in unsupervised tasks, because a probabilistic interpretation of the model has been difficult. Recently, two different frameworks, GSN and NADE, provide a connection between reconstruction and probabilistic modeling, which makes the interpretation possible. As far as we know, neither GSN or NADE have been studied in the context of time series before.As an example of an unsupervised task, we study the problem of filling in gaps in high-dimensional time series with complex dynamics. Although unidirectional RNNs have recently been trained successfully to model such time series, inference in the negative time direction is non-trivial. We propose two probabilistic interpretations of bidirectional RNNs that can be used to reconstruct missing gaps efficiently. Our experiments on text data show that both proposed methods are much more accurate than unidirectional reconstructions, although a bit less accurate than a computationally complex bidirectional Bayesian inference on the unidirectional RNN. We also provide results on music data for which the Bayesian inference is computationally infeasible, demonstrating the scalability of the proposed methods.

We combine supervised learning with unsupervised learning in deep neural networks. The proposed model is trained to simultaneously minimize the sum of supervised and unsupervised cost functions by backpropagation, avoiding the need for layer-wise pre-training. Our work builds on top of the Ladder network proposed by Valpola (2015) which we extend by combining the model with supervision. We show that the resulting model reaches state-of-the-art performance in semi-supervised MNIST and CIFAR-10 classification in addition to permutation-invariant MNIST classification with all labels.

The experiences from the first year of studies play a major role in determining the progress of later studies. Engaging study culture, requiring active participation in classes and active support from the teaching staff, can facilitate overcoming obstacles in the studies. In practice, the teacher has a key role in affecting the mindsets of the students. Implementing assessment as a continuous process permits collecting real-time information on students' performance during the course, which enables to monitor, and more importantly, to influence the students' behaviour and decrease the number of students dropping out from their studies. The students' progress was monitored in two big first-year courses (170-300 seats) on physics and circuit analysis starting from the first weeks of the studies. The goal was to recognise and to affect the potential drop-out candidates and support them to continue working towards completing their courses with the teacher's own actions and pedagogical decisions. Experiences gained during this project provide insight on appropriate actions in planning of teaching to facilitate students' success in studies.

Active learning, project-based teaching, and student collaboration are current trends in engineering education. Incorporating these have also been the goal of the basic studies development project EPOP started at the Aalto University School of Electrical Engineering in 2011. In the project, two obligatory basic courses in circuit analysis and electromagnetic field theory have been taught using interactive engagement during the spring of 2012. This paper presents the implementation of the teaching, including methods and evaluation with several concrete examples. As a result of the novel teaching, motivation and the engagement of students were at a high level during the whole course and learning results were better than those of the students participating the traditional lecture course.

Parasitic elements play a major role in advanced circuit design and pose considerable run-time and memory problems for the post-layout verification, especially in the case of full-chip extraction. This brief presents a realizable R(L)C(M)-netlist-in-R(L)C(M)-netlist-out method to sparsify and reduce the capacitive coupling parasitics in circuits with interconnect lines. The method is applicable in conjunction with partitioning-based model-order reduction algorithms to reduce the complete extracted netlists, or as a stand-alone tool to process only the capacitive coupling. It is shown that, by using the method, circuits with even dense capacitive coupling can be partitioned and reduced efficiently.

Some mobile devices automatically fetch content from social networking services and integrate it into the device user interface. In this paper, we explore how people consume integrated social networking services in a field study with an innovative aggregator named Linked Internet UI Concept, or LinkedUI. Twenty users completed the field study. We logged their activities and performed user interviews. The study reveals two main use cases: habitual checking where users frequently glanced at the integrated services at short intervals; and serendipitous content discovery where they come across some content when performing other tasks. The study also reveals that the users only attended to a small proportion of the full content set, such as content recently published and content from selected contacts. This indicates the user need of quick access to relevant content.

This paper presents a robust partitioning-based model-order reduction (MOR) method, PartMOR, suitable for reduction of very large RLC circuits or RLC-circuit parts of a non-RLC circuit. The MOR is carried out on a partitioned circuit, which enables the use of low-order moments and macromodels of few elements, while still preserving good accuracy for the reduction. As the method produces a positive-valued, passive, and stable reduced-order RLC circuit (netlist-in-netlist-out), it can be used in conjunction with any standard analysis tool or circuit simulator without modification. It is shown that PartMOR achieves excellent reduction results in terms of accuracy and reduced CPU time for RLC, RC, and RL circuits.

This paper presents the Linked Internet UI Concept, or LinkedUI for short, as a holistic user interface concept to facilitate social interaction on mobile devices. It aggregates social events from social networking services and communication channels and uses hypertext navigation for presentation and interaction. We describe the concept design principles, highlights of the design, and the prototype implementation. We conducted a user study to compare LinkedUI with the benchmark of web applications running in a web browser to follow their friends' activities in Twitter and Flickr and two other optional services on mobile devices. The study results reveal that the users performed tasks faster in LinkedUI and also liked it more than the benchmark. These findings support the design principles of LinkedUI in facilitating social interaction via mobile devices.

The increase in connected mobile computing devices has createdthe need for ubiquitous Web access. In many usagescenarios, it would be beneficial to interact multimodally.Current Web user interface description languages, such asHTML and VoiceXML, concentrate only on one modality.Some languages, such as SALT and X+V, allow combiningaural and visual modalities, but they lack ease-of-authoring,since both modalities have to be authored separately. Thus,for ease-of-authoring and maintainability, it is necessary toprovide a cross-modal user interface language, whose semanticlevel is higher. We propose a novel model, calledXFormsMM, which includes XForms 1.0 combined withmodality-dependent stylesheets and a multimodal interactionmanager. The model separates modality-independentparts from the modality-dependent parts, thus automaticallyproviding most of the user interface to all modalities.The model allows flexible modality changes, so that the usercan decide, which modalities to use and when.

World Wide Web Consortium is currently specifying XForms form language, which is intended to be the next generation language for Web forms. XForms is not currently supported by the major browsers, which prevents the early utilization of XForms. In this paper, we describe our implementation of XForms. The implementation is part of our X-Smiles browser, which is an open source XML browser.