ABSTRACT Depression, a prevalent mental disorder in modern society, significantly impacts people's daily lives. Recently, there have been advancements in developing automated diagnosis models for detecting depression. However, data scarcity, primarily due to privacy concerns, has posed a challenge. Traditional speech features have limitations in representing knowledge for depression diagnosis, and the complexity of deep learning algorithms necessitates substantial data support. Furthermore, existing multimodal methods based on neural networks overlook the heterogeneity gap between different modalities, potentially resulting in redundant information. To address these issues, we propose a multimodal depression detection model based on the Enhanced Cross‐Attention (ECA) Mechanism. This model effectively explores text‐speech interactions while considering modality heterogeneity. Data scarcity has been mitigated by fine‐tuning pre‐trained models. Additionally, we design a modal fusion module based on ECA, which emphasizes similarity responses and updates the weight of each modal feature based on the similarity information between modal features. Furthermore, for speech feature extraction, we have reduced the computational complexity of the model by integrating a multi‐window self‐attention mechanism with the Fourier transform. The proposed model is evaluated on the public dataset, DAIC‐WOZ, achieving an accuracy of 80.0% and an average F 1 value improvement of 4.3% compared with relevant methods.

Concurrent neural signal instrumentation withstanding neural stimulation artifacts is essential for bi-directional neural interfaces to guarantee signal integrity. In this work, different front-end structures and stimulation artifact mitigation techniques are firstly reviewed to benchmark their step response speed. Then, a mixed domain level-crossing scheme is proposed to achieve fast dynamic response with minimized hardware overhead. The benefit of extending the phase detection range of the phase detectors in VCO-based continuous time$\rm \Delta \Sigma $modulators is investigated with stability and noise consideration. Then a shift-register-based phase counter is proposed to extend the phase detectors’s detection range, thereby increase quantization resolution and stability margin for in-band noise optimization. The proposed VCO-based neural front-end was fabricated in a 180 nm CMOS process. The prototype achieves$6.38~\mu $Vrms input-referred noise over 0.5 Hz-10 kHz bandwidth. With a linear input range of 120 mVpp, it exhibits a SNDR of 71.6 dB and a DR of 77.0 dB, which could be further extended up to 100 dB in the artifact adaption mode. Measurements verify that the proposed neural front-end can recover from rail-to-rail differential mode or common mode artifacts within 10$\mu \text{s}$(minimum$6.25~\mu \text{s}$) while the superposed small signal can be recorded uninterruptedly.

High-density large-scale implantable electrode arrays enable neural recording with cellular level resolution. Active electrodes with integrated instrumentation and processing are necessary for single neuron activity recording to minimize interference and reduce bandwidth for the signal conversion and transmission. This introduces design challenges for circuits miniaturization to match the electrode density and meet the dynamic range requirement. This paper proposes a clockless, area efficient spike detection neural front-end enabled by a VCO-based level-crossing analog to digital converter (LC-ADC) and spike detection algorithm based on continuous time nonlinear energy operator (CT-NEO). Level-crossing events and corresponding timing information can be directly generated by the proposed LC-ADC. A continuous time NEO algorithm is proposed and implemented, featuring low complexity and high detection accuracy, MATLAB simulation shows that the algorithm can still achieve 70% accuracy at an estimated signal-to-noise ratio (SNR) of -4dB. Implemented in a 65 nm CMOS process, the proposed front-end consumes 3.15 $\mu$W and 0.00385 mm2active area.

The ever increasing mass of information leads to new challenges on analyzing or navigating the large data sets. Combining visual perception and interaction capabilities with the enormous storage and computational power of today's computer systems, especially with the rise of 3D virtual worlds, has great potential in providing deeper immersion and intuitive interactions with large data sets. In this demo, we exploit the potential of navigating large data-sets in a 3D virtual world, by transforming raw data sets into semantically rich, high level interactions and presenting data through rich, real-time visualization. We also explore the use of various digital devices that most users have available to build “distributed interfaces” and provide capabilities that make interactions within the 3D space, and with the data sets presented in the 3D space, more natural and expressive.

Dynamic load balancing holds the potential to scale virtual worlds flexibly by dynamic allocation of hardware to match load. In this paper, we study the benefits and overheads of space based load partitioning, in particular, distributed binary space partitioning (BSP). Our evaluation is based on Open Simulator, a virtual world system compatible with Second Life® viewers. Our work reveals that although simple and effective, distributed BSP has several limitations and suffers from high overhead. We then analyze the fundamental reasons of these limitations. To overcome the limitations, we argue that it is necessary to break away from the simulator-centric architecture used in today's virtual worlds, and present potential new directions.

Network selection is the decision process for a mobile terminal to handover between homogeneous or heterogeneous networks. This handover decision can be either mobile or network initiated. Today's decision is mainly based on the received signal strength (RSS). In the future as the number of available networks increase the selection process must evaluate additional factors such as monetary cost, offered services, network conditions, required energy to operate in a network, system conditions, user and operator's preferences. Upcoming IEEE or 3GPP standards provide facilities such as network maps, handover negotiation messages, network and client event reports and message exchange like handover triggers and handover negotiation. In this paper we investigate the use of a cost function to perform an optimal network selection using information provided by these standards, such as network coverage map or network properties. The cost function provides flexibility to balance different factors in the decision making, and our research is focused on improving both seamlessness and energy efficiency of the device and handovers. We evaluated our approach based on usage scenarios over WiFi, WiMax, and 3G networks using captured signal strength traces. The results of our simulations and early implementation show that our schemes select the optimal networks and handovers were triggered at appropriate times to increase overall network connectivity as compared to traditional triggering schemes, while at the same time optimizing energy consumption of multi-radio devices.

Multi-radio devices provide end-users the ability to achieve ubiquitous and seamless connectivity anytime, anywhere across heterogeneous networks. Minimal service disruption while a device roams across networks is key for a successful deployment. Todaypsilas roaming decisions are reactive and traditionally based on thresholds for signal strength or other radio properties, leading to undesirable handover delays. This paper introduces an Intel/BT client architecture to support seamless mobility for multi-radio devices and proposes a novel approach to predict when a device needs to take proactive actions to perform handovers. Significant improvement in reducing service discontinuity time is demonstrated by applying our approach to WiFi and WiMax networks. We also validate the architecture and algorithm benefits using our WiFi/WiMax multi-radio prototype in heterogeneous wireless network environments.

Dynamic switching between multiple radio interfaces has been shown to reduce power consumption and increase battery lifetime in mobile devices. In this paper, we first present empirical data measured on Sony UMPC and iPhone in a personal area network (PAN) environment to emphasize the need for dynamic multi-radio switching. We then present a collaborative power management architecture and design protocols that balance between performance and energy efficiency by exploiting energy characteristics of different radios. A prototype that implements the protocols is developed on a laptop and evaluated on a PAN testbed. Empirical results using typical MP3 workloads show a 32% reduction in the total platform energy consumption on the laptop and 72% reduction in energy consumed due to radio activity. Moreover, analysis shows that for UMPC and iPhone, potentially 2 to 4- fold reductions on platform energy consumed by radio activities could be achieved.

Roaming across heterogeneous wireless networks poses a challenging issue in mobility management such as seamless and efficient handover to reduce end-user's service interruption. Efficient techniques for seamless handover between access points (APs) when user roams from one area to another can be categorized into three stages in conventional way, i) initiation, ii) preparation, and iii) execution. However, frequent handover may overload the network with signalling traffic and causes call dropping to increase as well. This paper takes a deeper look at predictive triggers at the link layer. With an appropriate implementation of predictive triggering algorithms, early handover initiation and preparation could be performed. This paper addresses the investigation on predictive and accurate mechanisms to generate a handover trigger.

Hardware/software co-verification is becoming an indispensable tool for building highly trustworthy embedded systems. A stumbling block to effective co-verification using model checking is the lack of support to unified property specification for hardware, software, and entire embedded systems. In this paper, we develop xPSL, a unified property specification language for co-verification. xPSL extends the IEEE Property Specification Language (PSL) to support specification of temporal assertions over both hardware and software events. The semantics of hardware and software events and their temporal correlations are formalized based on translation of both hardware and software semantics to a common formal semantic basis. xPSL has been applied in co-verification to specify properties of hardware and software components, and furthermore entire embedded systems. Case studies have shown that xPSL is very effective in enabling co-verification of system-level properties and facilitating compositional reasoning.

In secure group communications, there are both rekey and data traffic. We propose to use application-layer multicast to support concurrent rekey and data transport. Rekey traffic is bursty and requires fast delivery. It is desired to reduce rekey bandwidth overhead as much as possible since it competes for bandwidth with data traffic. Towards this goal, we propose a multicast scheme that exploits proximity in the underlying network. We further propose a rekey message splitting scheme to significantly reduce rekey bandwidth overhead at each user access link and network link. We formulate and prove correctness properties for the multicast scheme and rekey message splitting scheme. We have conducted extensive simulations to evaluate our approach. Our simulation results show that our approach can reduce rekey bandwidth overhead from several thousand encrypted new keys (encryptions, in short) to less than ten encryptions for more than 90% of users in a group of 1024 users.

Measurement studies indicate a high rate of node dynamics in p2p systems. In this paper, we address the question of how high a rate of node dynamics can be supported by structured p2p networks. We confine our study to the hypercube routing scheme used by several structured p2p systems. To improve system robustness and facilitate failure recovery, we introduce the property of K-consistency, K ≥ 1, which generalizes consistency defined previously. (Consistency guarantees connectivity from any node to any other node.) We design and evaluate a failure recovery protocol based upon local information for K-consistent networks. The failure recovery protocol is then integrated with a join protocol that has been proved to construct K-consistent neighbor tables for concurrent joins. The integrated protocols were evaluated by a set of simulation experiments in which nodes joined a 2000-node network and nodes (both old and new) were randomly selected to fail concurrently over 10,000 seconds of simulated time. In each such "churn" experiment, we took a "snapshot" of neighbor tables in the network once every 50 seconds and evaluated connectivity and consistency measures over time as a function of the churn rate, timeout value in failure recovery, and K. Storage and communication overheads were also evaluated. We found our protocols to be effective, efficient, and stable for an average node lifetime as low as 8.3 minutes (the median lifetime measured for Napster and Gnutella was 60 minutes [10]).

In a system proposed by Plaxton, Rajaraman and Richa (PRR), the expected cost of accessing a replicated object was proved to be asymptotically optimal for a static set of nodes and pre-existence of consistent and optimal neighbor tables in nodes [9]. To implement PRR's hypercube routing scheme in a dynamic, distributed environment, such as the Internet, various protocols are needed (for node joining, leaving, table optimization, and failure recovery). In this paper we first present a conceptual foundation, called C-set trees, for protocol design and reasoning about consistency. We then present the detailed specification of a join protocol. In our protocol, only nodes that are joining need to keep extra state information about the join process. We present a rigorous proof that the join protocol generates consistent neighbor tables for an arbitrary number of concurrent joins. The crux of our proof is based upon induction on a C-set tree. Our join protocol can also be used for building consistent neighbor tables for a set of nodes at network initialization time. Lastly, we present both analytic and simulation results on the communication cost of a join in our protocol.