34254 research outputs found
Sort by
Design and Implementation of Machine-Learning-Aided Auto-Mode-Selection Polar Decoder Chip Featuring Adaptive Multiple Modes
No full textWith the rise of AI, stable communication systems and chips remain vital. AI needs robust hardware and networks to connect to computing centers. Thus, advancing communication chips continues to be crucial in the AI era.
Channel coding improves transmission reliability in communication systems. Polar codes achieve the Shannon limit, the theoretical maximum for channel capacity. This thesis develops a multi-mode polar decoder chip, enabling broader application to complex communication environments.
This thesis proposes three technologies to improve the decoding efficiency and throughput of multi-mode polar code decoder chips. With the aid of machine learning, the channel environment is detected in real time and automatically switched to the applicable decoding mode, so that the chip always operates in the most efficient mode to maximize the benefits of multi-mode. Adaptive decoding mode and segmented packet design are introduced to improve the decoding throughput of the chip while maintaining decoding reliability. The sorting module is improved from the algorithm level. Unnecessary operations are removed and a two-dimensional matrix-based sorting method is added to further reduce the complexity of sorting and reduce the delay of sorting operations to improve the throughput of the chip. The combination of the above three technologies achieves a polar code decoder chip suitable for multi-scenario, easy to use and high throughput.
The above design is implemented with TSMC 40nm CMOS process to realize this chip, and FPGA is used to verify critical modules. Finally, the implementation results of the chip are compared and analyzed with other literature
DRL-Based Transmission Design for Distributed STAR-RIS-Aided Communication Under Hardware Impairments and Imperfect CSI
No full textWithin this study, we explore the functionality of distributed simultaneous transmitting and reflecting (STAR) reconfigurable intelligent surface (RIS)-assisted downlink multi-user
communication system consisting of a base station (BS) assisting multiple users (at both front and back ends).
In order to provide an energy-efficient transmission design for the considered distributed-STAR-RIS ON/OFF
system, we create a energy efficiency (EE) maximization problem
that tackles the issue of imperfect channel state information (CSI) and hardware impairments while ensuring a minimum rate quality of service (QoS) at each user under the total transmit power constraint at the BS. Considering the non-convex nature of the presented problem, we introduce a framework based on deep reinforcement learning (DRL) that provides optimum solution by incorporating the deep deterministic policy gradient (DDPG) and twin delayed DDPG (TD3) algorithms. The efficacy and complexity of the proposed algorithms are validated graphically obtained through the extensive simulations.
We provide a demonstration of how the performance of the system under consideration is impacted by variation in key system parameters, such as the maximum transmit power available at the BS, number of RIS, reflecting elements in each RIS and antenna at the BS. Thero simulation results showcase the efficiency of the suggested algorithms with the perfect CSI approach outperforming the imperfect CSI and hardware impairments.
The results further reveal the non-negligible impact of CSI imperfections, underscoring the critical significance of a intelligent design to uphold users QoS
The Relationship between Brand Community Experience, Brand Community Commitment, and Brand Attitude: The Moderating Effects of Social Media Platforms and Brand Personality
No full textPrevious research found managing social media communities alone does not automatically capture consumer attention. Studies on brand community commitments have focused on the direct impact of experiences on commitments and brand attitudes, overlooking the differentiation of brand personalities and social media platforms. This study aims to validate the relationships among brand community experiences, attitudes, commitments, and overall brand perception. It seeks to gain a comprehensive understanding of how diverse brand personalities and social media platforms used in marketing can influence user experiences, community commitments, and brand attitudes.
Using a quantitative analysis approach and adapt the questionnaire based on relevant research concepts, considering different brand personalities and social media platforms. The independent variables include brand personality (Excitement and Competence) and social media platform (Facebook, Instagram, LINE), resulting in a 2 x 3 factorial design. Participants was randomly assigned one of the six questionnaire variations. Collected data was analyzed using statistical software of JASP and SPSS Amos to explore the relationships among brand community experiences, commitments, and attitudes, while examining the influence of brand personalities and social media platforms on these factors.
Based on the analysis results, homophily experience in the overall social community environment show the most significant relationship with brand community commitments and brand attitudes. Regarding competence brands, the research indicates that information and homophily experience have the most substantial relationship with behavioral commitments and brand attitudes. On the other hand, excitement brands can enhance consumer commitment to brand communities and further improve brand attitudes through homophily experiences, information experiences, and entertainment experiences
Pulse Radar Using SIL Technology
No full textIn order to enhance national defense, the pulse waves used as transmitted signals in the radar system are combined with self-injection locking technology. Leveraging the advantages of self-injection locking technology to eliminate the impact of stationary clutter, the system sensitivity is improved. This paper conducts preliminary experiments on the feasibility of the radar system in this manner.
The proposed pulse self-injection locking radar demonstrates the ability to reflect different delay times through the self-injection locking mechanism in closed-loop experiments. The IQ frequency demodulator lowers the echo signal to baseband, significantly reducing the sampling rate requirements. The processed echo signal after demodulation and the reference signal from the switch undergo mutual correlation processing. The time difference between the two sets of delay lines, as measured by a network analyzer, corresponds to the signal processing time, indicating that the proposed pulse self-injection locking radar has a certain effect in distinguishing different delay times. This paper also integrates pulse compression radar with phase self-injection locking technology to achieve stretch processing, and the experimental results are as expected. From the two experiments mentioned above, it is evident that applying self-injection locking technology to radar systems has developmental feasibility
Students\ue2 learning outcomes under different teacher\ue2s classroom management
No full textCram schools are indispensable for most people in Taiwan. Students travel from school to cram school day in and out. There is a wide array of cram schools, with English cram schools being the most often sought after. Given the growing awareness among parents about the significance of English as a means of global communication, they are confident that the sooner their children are familiar with English, the more advantageous it will be. The purpose of this study is to understand whether teachers' classroom management will impact students' learning outcomes (English proficiency). Eighteen workers and eighty-one students in two cram schools were observed in the study. The study is based on two English cram schools in Kaohsiung, daily conversations with my colleagues and students, students' reactions in class, and learning outcomes. Two leadership styles have been observed being applied in different cram schools: servant leadership and directive leadership
A financial network investment strategy for Labor Pension Fund
No full textAccording to estimates by the National Development Council, the population of Taiwan aged 65 and over will reach 20% by 2025, entering a super-aged society, and may reach 40% by 2057, making the economic livelihood security of the elderly population an important issue for the nation. As of October 2023, the number of contributors to the "Labor Pension Fund" is 7.54 million, accounting for 71.48% of the employment of Taiwanese workers, and the operational surplus or deficit of the fund will be an important factor affecting social stability.
This study uses a financial network model to establish an investment portfolio to explore the investment performance in the Taiwan stock market and to provide investment suggestions for the labor pension fund. We use the constituent stocks of "Yuanta/P-shares Taiwan Top 50 ETF (0050.TW)" and "Yuanta/P-shares Mid-Cap 100 ETF (0051.TW)" as the empirical analysis data for this study. We construct a financial network of stock returns using the Minimum Spanning Tree and Planar Maximally Filtered Graph and determine the centrality of stocks accordingly. Stock selection uses methods based on high and low centrality, \ucf-dependent, and partial \ucf-dependent; weight allocation considers equal weight method, tangent portfolio weights from modern investment theory, and weights determined by hierarchical risk parity method, Sortino ratio, and upside-downside risk ratio. The performance of various investment portfolios in the backtesting period is evaluated using cumulative returns and empirical mode decomposition residuals, thereby dynamically adjusting the investment portfolio adopted in each investment cycle. Empirical results indicate that, from July 1, 2005, to November 9, 2023, the network investment portfolio strategy utilizing a Minimum Spanning Tree built on correlation, with stock selection based on \ucf-dependent, and allocation of weights determined through the Tangency portfolio method, yielded the highest cumulative return. Compared with the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX), although it carries higher risk, it also yields higher returns and its Sharpe ratio is superior to TAIEX. Moreover, the annualized return rate of the financial network investment strategy proposed in this paper is far superior to that of the Labor Pension Fund, TWII, and ETF 0056 over the past five years (2018~2022). The empirical results support the stability and good performance of the financial network investment strategy proposed in this study, which can provide a reference for future investment strategies for the labor pension fund
A Research on Video Game Sales Forecast using Text Mining Techniques
No full textGame has become an indispensable part of the global entertainment industry, with vast business opportunities accompanied by increasingly fierce competition. Enhancing sales is crucial for the industry. Prior to the release of a game, developers release promotional content and pre-order campaigns to stimulate players to purchase in advance. Players can read related information or evaluate through media reviews to see if it meets their expectations. During the pre-order period, information asymmetry led to a significant gap between expectations and reality, accelerating the game's decline and ultimately affecting the game's sales volume and brand image.
This research combines topic model and sentiment analysis to analyze data from 2018 Metacritic reviews and four weeks of sales volume data from VGChartz. Each review is categorized based on game elements, and predictive analysis is used to determine whether online reviews affect game sales volumes, exploring differences between media reviews and player reviews.
This research found that datasets rich in textual perform better, with early sales influenced by media reviews and later sales by player engagement and word of mouth, with gameplay being the most important topic. Media sentiment tends to be uniform, while player feedback is highly individualistic
Using Deep Learning with Image and Infrared Thermography for Crack Identification
No full textIn recent years, with the rapid development of deep learning models has resulted in their, use in image segmentation tasks pertaining to identification of building cracks. However, most deep learning models are single-input variants models based on visible light images to perform crack segmentation tasks. Therefore, this dissertation adopts a dual-input crack segmentation model, Ucrack, which takes both visible light and infrared images as model inputs. Ucrack consists of two parts: a visible light segmentation model and an infrared segmentation model. In the visible light segmentation model, two sub-models are formed integrally to yield a good performance than that from using the two sub-models to perform prediction individually. Since infrared images can provide different feature information from visible light images, the infrared segmentation model is leveraged to assist the visible light segmentation model to achieve better segmentation results. A public crack data set is used for performance evaluation. As the publicly available crack data set based on visible light and infrared images are small and fragmented, a new crack data set is constructed in this study
Tourist Check-In Behavior: A Case Study Comparing New and Existing Taitung Land Art Installations
No full textThe rise of social media has made 'check-in' a common behavior and its influence has also extended to tourism. This research examines this phenomenon by considering the effect of fear of missing out (FOMO) on check-in motivation and intention, travel motivation, and anticipated regret, and how these constructs collectively affect intention to visit Taitung Land Art Installations, both new and existing installations. Data analysis via SmartPLS on 199 and 217 valid survey responses for new and existing installations revealed varying effect of FOMO on check-in and travel motivation. For new installations, FOMO positively influences check-in and travel motivation, altruism and perceived enjoyment check-in motivations affect check-in intention, and check-in intention and travel motivation (arts) further affect visit intention positively. The influence of FOMO on check-in motivation and travel motivation is more limited for existing installations. While check-in intention affects visit intention, arts and relaxation travel motivation also affect visit intention
The inhibitory effects of marine-derived antimicrobial peptide Pardaxin on ovarian cancers
No full textOvarian cancer poses a significant challenge in the field of gynecological oncology. The five-year survival rate for advanced ovarian cancer typically falls below 50%. The current standard chemotherapy approach for ovarian cancer involves the use of a combination of paclitaxel and carboplatin. However, this drug combination does not prove effective for all patients. This study investigates the cytotoxic effects of marine teleost-derived antimicrobial peptide Pardaxin on both ovarian teratoma cells (PA-1) and ovarian epithelial adenocarcinoma cells (SKOV3). The MTT, TUNEL, Annexin V&PI, and Western blot assays collectively indicate that Pardaxin can induce cytotoxicity in ovarian cancer cells through the apoptotic process. Furthermore, an overproduction of mitochondrial and intracellular reactive oxygen species (ROS) is observed, which is associated with mitochondrial dysfunction following Pardaxin administration. Additionally, the study reveals an increase in the expression levels of autophagy-related proteins, such as beclin, p62, and LC3, demonstrating Pardaxin's ability to induce autophagy in ovarian cancers. Finally, it is found that the apoptosis and autophagy induced by Pardaxin can be mitigated by the antioxidant NAC. These results suggest that Pardaxin induces cell death through elevated intracellular ROS levels, involving mechanisms of both apoptosis and autophagy. In summary, Pardaxin increases ROS levels in ovarian cancer cells, thereby activating mitochondrial autophagy and apoptotic pathways, offering potential as a candidate drug for the treatment of ovarian cancer