Hughes Network Systems

The Indian linguistic diversity is a major problem for scalable conversational AI systems, especially for low-resource languages and multimodal interaction. It is costly and infeasible to train and retain different models for each language and modality combination. The paper discusses a pivot-based multilingual AI architecture that can handle both speech and text interaction using the Indic languages and semantic reasoning using a high-resource pivot language. The proposed system proposes a modality-neutral normalization process that transforms speech and text input into a common textual form through which a common reasoning process can be followed regardless of the input or output modality. Semantic inferences are made in the pivot language space, and the responses are then produced in the target language and translated into the desired modality (text or speech). We give a formal mathematical description of the end-to-end system, semantic consistency conditions, and analysis of error propagation, and show how multilinguality and multimodality may be unrelated to reasoning. It is a low-resource architecture that simplifies the complexity of systems, allows them to be scaled to handle low-resource Indic languages, and serves as a center of thought to develop viable speech- and text-enabled multilingual assistants in AI.

Quick UDP Internet Connection (QUIC) is an emerging end-to-end encrypted, transport-layer protocol, which has been increasingly adopted by popular web services to improve communication security and quality of experience (QoE) towards end-users. However, this tendency makes the traffic analysis more challenging, given the limited information in the QUIC packet header and full encryption on the payload. To address this challenge, a novel rule-based approach is proposed to estimate the application-level traffic attributes without decrypting QUIC packets. Based on the size, timing, and direction information, our proposed algorithm analyzes the associated network traffic to infer the identity of each HTTP request and response pair, as well as the multiplexing feature in each QUIC connection. The inferred HTTP attributes can be used to evaluate the QoE of application-layer services and identify the service categories for traffic classification in the encrypted QUIC connections.

There is an increased interest in the deployment and use of hybrid architectures to benefit the end-user community. Several Low Earth Orbit (LEO) systems are being deployed that provide low latency to end-users. In the GEO world, Very High Throughput Satellite (VHTS) systems are being deployed that maximizes capacity density. In parallel, 3GPP has been actively involved in the development of Non-Terrestrial Transport networks that include satellites. No single technology can provide the best solution in terms of latency, throughput, efficiency, coverage, capacity density and availability for all applications, user terminal types and use cases. A hybrid architecture that uses a combination of LEO, GEO and Terrestrial is desired that optimizes in all dimensions is discussed above. To this end, a user terminal with multiple radio access technologies is desired and a smart logic in user terminals that determine the best path for a session is needed. Various aspects of this Smart Network Edge are described in this paper.

In this article, the theory, design, and validation of planar array antennas capable of producing dual-circularly-polarized (dual-CP) dual-beam radiation with a single feeding network are reported. Importantly, the properties of the formed CP beams with different handedness can be independently controlled. Such a capability is achieved by exploiting both the in-plane rotation-induced Berry phase and excitation phase for all the linearly polarized array elements, which can be regarded as a generalization of the well-known sequential rotation arrays. Microstrip patch arrays with such single-feed dual-CP dual-beam functionalities can be synthesized through a straightforward analytical approach that accurately predicts the radiation patterns for both right- and left-handed CP beams. The proposed technique and design methodology were verified by two proof-of-concept microstrip arrays consisting of 8×8 linearly polarized circular patch elements operating in the K-band, one with uniform excitation amplitudes and the other with tapered excitations for sidelobe suppression. Both arrays were fabricated and measured, experimentally achieving a joint S 11 <; -10 dB, axial ratio <; 3 dB, and 3 dB gain bandwidth of more than 22.5%/20.0% and 16.7%/15.8%, respectively, for the right-and left-handed CP beams. Within this broad operational band, the beam squinting is on average within ±3°. The proposed dual-CP dual-beam arrays have fully planarized structures, low profiles, wide operational bandwidths, and high polarization purity, making them useful candidates for applications including satellite communications, point-to-multipoint communications, and so on.