With the substantial increase in the number of antennas, polar-domain channel modeling for extremely large-scale antenna array (ELAA) systems has been introduced to capture both angular and distance information in near-field environments. The fine-grained polar-domain channel provides additional sources of randomness, making it well-suited for physical layer key generation (PLKG). To minimize the pilot overhead in multi-user key generation and leverage the randomness from the polar-domain channel paths, we herein design a zero-forcing (ZF)-based precoding scheme to mitigate inter-path and inter-user interference. Using ZF precoding, we derive an analytical expression for the sum secret key rate (SKR) as a function of power allocation variables, and then optimize these variables in the presence of eavesdroppers. Our simulations validate the proposed precoding design and power allocation methods in terms of sum SKR versus the transmit power, antenna configurations, and spatial correlation between legitimate and eavesdropping channels.

Group selection in virtual reality is an important means of multi-object selection, which allows users to quickly group multiple objects and can significantly improve the operation efficiency of multiple types of objects. In this paper, we propose a group selection method based on multiple rounds of probability permutation, in which the efficiency of group selection is substantially improved by making the object layout of the next round easier to be batch-selected through interactive selection, object grouping probability computation, and position rearrangement in each round of the selection process. We conducted ablation experiments to determine the algorithm coefficients and validate the effectiveness of the algorithm. In addition, an empirical user study was conducted to evaluate the ability of our method to significantly improve the efficiency of the group selection task in an immersive virtual reality environment. The reduced operations also indirectly reduce the user task load and improve usability.

In audiobook synthesis, it is important to have the ability to differentiate between dialog and narration or different characters. In this work, we propose dialog modeling methods for audiobook synthesis. The proposed approach consists of two stages. First, a text-based dialog style classifier is employed to predict narration vs. dialog from text, and further predict the corresponding characters into soprano and baritone. Then, a dialog style adaptor is added to the text-to-speech (TTS) model to allow synthesizing speech with the corresponding styles. With a speaker verification (SV) based style adaptor, we can even control the strength of a given style. We evaluated the proposed approach in audiobook synthesis with a mean opinion score (MOS) listening test using 9 carefully designed questions. The results show an improvement of 0.35 MOS on dialog distinction without degradation in other aspects. Also a comparative MOS (CMOS) test is conducted to verify the effectiveness of the proposed method.

Multiple description coding (MDC) is a type of error-resilient source coding that is advantageous for communications over channels with high loss and long delays. While neural network-based MDC promises higher compression efficiency and better semantic awareness, the aspect of semantic awareness is under-investigated. This paper is among the first efforts to study MDCs with competing performance goals known as the distortion-classification tradeoff. A new design concept called semantic polarization is proposed to contradict the traditional philosophy of balanced side encoder designs. We present a simple conceptual model to demonstrate the advantage of polarized MDC in having higher probabilities of satisfying at least one performance goal. We also propose a detailed implementation of polarized MDC based on SRGANs. Experiments on public datasets show that compared with state-of-the-art single description neural image codecs, the proposed MDC has multiple benefits in terms of enlarged rate range, superior semantic protection, and better perception quality, at the cost of slightly reduced but competitive distortion performance.

While sensing imagery in space missions has broad applications, the growing image resolution and data volume have caused a major challenge due to limited deep space channel capacities. To address this challenge, semantics-aware image compression becomes a promising direction. This paper is motivated to explore lossy compression at the ultra-low rate regime, which is a deviation from the high-fidelity- oriented tradition. Specifically, we propose an ultra-low rate deep image compression (DIC) codec by synthesizing multiple neural computing techniques such as style generative adversarial network (GAN), inverse GAN mapping, and contrastive disentangled representation learning. In addition, a residual-based progressive encoding framework is proposed to enable smooth transitions from the ultra-low rate regime to near- lossless regime. Experiments on the FFHQ and DOTA dataset demonstrate that compared with existing DICs, the proposed DIC can push the minimum rate boundary by about one order of magnitude while preserving the semantic attributes and maintaining a high perception quality. We further elaborate the design considerations for cross-rate-regime progressive DIC. Our study confirm that a semantically disentangled DIC holds the promise to bridge multiple rate regimes.

Predicting trajectories of pedestrians based on goal information in highly interactive scenes is a crucial step toward Intelligent Transportation Systems and Autonomous Driving. The challenges of this task come from two key sources: (1) complex social interactions in high pedestrian density scenarios and (2) limited utilization of goal information to effectively associate with past motion information. To address these difficulties, we integrate social forces into a Transformer-based stochastic generative model backbone and propose a new goal-based trajectory predictor called ForceFormer. Differentiating from most prior works that simply use the destination position as an input feature, we leverage the driving force from the destination to efficiently simulate the guidance of a target on a pedestrian. Additionally, repulsive forces are used as another input feature to describe the avoidance action among neighboring pedestrians. Extensive experiments show that our proposed method achieves on-par performance measured by distance errors with the state-of-the-art models but evidently decreases collisions, especially in dense pedestrian scenarios on widely used pedestrian datasets.

Named entity recognition (NER) is usually developed and tested on text from well-written sources. However, in intelligent voice assistants, where NER is an important component, input to NER may be noisy because of user or speech recognition error. In applications, entity labels may change frequently, and non-textual properties like topicality or popularity may be needed to choose among alternatives. We describe a NER system intended to address these problems. We test and train this system on a proprietary user-derived dataset. We compare with a baseline text-only NER system; the baseline enhanced with external gazetteers; and the baseline enhanced with the search and indirect labelling techniques we describe below. The final configuration gives around 6% reduction in NER error rate. We also show that this technique improves related tasks, such as semantic parsing, with an improvement of up to 5% in error rate.