Domain generalization aims to reduce the vulnerability of deep neural networks in the out-of-domain distribution scenario. With the recent and increasing data privacy concerns, federated domain generalization, where multiple domains are distributed on different local clients, has become an important research problem and brings new challenges for learning domain-invariant information from separated domains. In this paper, we address the problem of federated domain generalization from the perspective of domain hallucination. We propose a novel federated domain hallucination learning framework, with no additional data exchange between clients other than model weights, based on the idea that a domain hallucination with enlarged prediction uncertainty for the global model is more likely to transform the samples into an unseen domain. These types of desired domain hallucinations are achieved by generating samples that maximize the entropy of the global model and minimize the cross-entropy of the local model, where the latter loss is further introduced to maintain the sample semantics. By training the local models with the learned domain hallucinations, the final model is expected to be more robust to unseen domain shifts. We perform extensive experiments on three object classification benchmarks and one medical image segmentation benchmark. The proposed method outperforms state-of-the-art methods on all the benchmarks, demonstrating its effectiveness.

Wireless backhaul has recently gained a significant amount of interest as a cost-effective solution in comparison with conventional backhaul technologies with dedicated microwave links or fiber optics. Self-interference cancellation (SIC) is an enabling technology that allows wireless backhaul to operate in the more spectrum-efficient in-band full-duplex (IBFD) operation mode instead of the out-of-band mode. Compared to Wi-Fi IBFD transceivers, wireless in-band backhaul systems face some unique challenges, such as significantly higher transmission power and much larger propagation delay spread for the self-interference signal, especially in the low-frequency bands under 1 GHz, which often prevent accurate SIC performance. The SIC is often implemented with an interference-cancelling filter, where the filter weights are essentially the channel estimates of the self-interference signals. In this paper, a frequency-domain Radio Frequency (RF) SIC (RF-SIC) framework with a novel filter weight optimization algorithm is proposed to tackle the challenges of wireless in-band backhaul. The proposed RF-SIC does not require a dedicated training phase which needs to stop the transmission of the backhaul signal. Moreover, it has the capability of tracking the self-interference channel variation since the filter weights are updated in a block-by-block fashion.

The development of 5G system provides a unique opportunity to implement a unified wireless transmission platform that can simultaneous deliver unicast, multicast and broadcast services using the same spectrum and the same infrastructure. This can be achieved by developing a more capable point-to-multipoint transmission (PTM) system component, also called 5G Multimedia Broadcast Multicast Services (5G-MBMS). A properly designed 5G-MBMS could achieve the convergence of terrestrial mobile broadband and TV broadcasting systems.By inherently allowing dynamic spectrum allocation among different types of services, this converged system could offer significant higher spectrum utilization. In this paper, the power-based non-orthogonal multiplexing (P-NOM) technology is proposed in addition to the existing orthogonal time-division-multiplexing (TDM) scheme. Significant capacity gains could be achieved by using P-NOM in a 5G-MBMS system for delivering different types of broadcast services, and for delivering mixed unicast and broadcast services. The capacity benefits of a 5G-MBMS with P-NOM is demonstrated by both theoretical analysis and computer simulations. A key finding is that using a two-layer P-NOM can essentially deliver a broadcast service on top of a unicast network, each achieving nearly full capacity. Next, a complexity analysis will reveal that implementing P-NOM in a 5G-MBMS system does not require significant additional complexity at consumer devices. Finally, it will be shown that, different from the non-orthogonal-multiple-access (NOMA) technologies, using P-NOM in 5G-MBMS system requires little or no change to the existing radio resource allocation mechanisms in LTE and 5G, when one power layer is used for broadcast services delivery.

This paper describes systems, devices, and methods to implement a bi-directional integrated inter-tower wireless communications network (IITWCN). The described technology can be implemented in combination with the Broadcast Core Network (BCN) in next generation broadcast eco-system and, therefore, support new business cases for broadcast operators such as the delivery of flexible datacasting services or support broadcast or point-to-point Internet services. The introduced bi-directional inter-tower communications network (ITCN) extends the previous unidirectional in-band distribution links (IDL) and adopts the on-channel repeater (OCR) as a simplified backhaul solution in single frequency networks (SFN). The concept of the coordinated ITCN is also presented, aiming at future broadcast Internet services. The ITCN provides a scalable and configurable network solution embedded in a broadcast system, which becomes independent from any non-broadcasting telecommunication infrastructure. The described technology partially relies on the infrastructure of the underlying broadcast/multicast network, using the allocated service channels without requiring additional frequency bands or a separate frequency band. The bi-directional inter-transmitter communication links are therefore referred to as integrated transmission links and the corresponding network as an integrated network.

To meet the ever increasing demand for better service quality and availability, the next generation digital TV (DTV) broadcast, a.k.a., the ATSC 3.0, is developed with new capabilities to deliver 4K ultra-high-definition (4K-UHD) services to fixed receivers, robust HD-quality services to mobile receivers, as well as non-broadcast services (e.g., broadband), including Internet of Things (IoT), Connected Vehicles, etc. To realize these capabilities, it becomes necessary to evolve the existing single-transmitter high-power-high-tower (HPHT) network to multi-transmitter low-power-low-tower single-frequency-networks (SFN), which requires installing new transmitters. Each new transmitter needs a feederlink to receive the service data from the broadcast gateway (BG) for transmission. This feederlink contributes to a significant portion of installation and operational costs of the new transmitters. This paper proposes a wireless in-band distribution link (WIDL) technology, where the distribution data is delivered to the newly deployed transmitters from the existing DTV towers wirelessly sharing the same TV band with the broadcast services. The distribution signal is multiplexed with service signal in the same ATSC 3.0 waveform. Layered division multiplexing (LDM) technology is proposed to achieve more efficient transmission of the distribution signal. The WIDL offers better performance and more robust operation than the on-channel repeater (OCR) technologies. More importantly, it offers the possibility of delivering backhaul data for future applications over the DTV infrastructure, such as IoT and connected vehicles. Therefore, the proposed WIDL is one enabling technology to achieve convergence of broadcast services with broadband and other wireless services on the DTV spectrum.

This paper analyzes reliability and quality gains that can be achieved by a broadcast-unicast convergent platform operating in non-orthogonal multiple access (NOMA)-based cellular networks. The analysis is addressed in network scope, dealing with randomly distributed users and base stations (BSs) within a stochastic geometry framework. Additionally a well-known Zipf-distribution popularity model on user demands is also taken into account, which effectively reflects a broadcasting gain from reducing redundant resource expense toward the users demanding same contents. Based on the stochastic modeling, closed-form delivery success probabilities and the consequent transmission rates are analytically derived, where the derivations are obtained separately for the networks with different access control strategies. Furthermore, the optimal broadcast content selection strategy maximizing the transmission rates is proposed. By means of numerical calculations, NOMA is shown to achieve substantial gain at broadcast/unicast joint transmission compared to the conventional orthogonal multiple access (OMA) network, particularly for the service configuration with popular contents having lower target service rates than less popular contents as long as the number of broadcast programs is restricted to be equal to the number of unicast services in a single frame.