DSpace community: 通訊工程學系碩士在職專班

適用於112 Gbps接收機之基於峰值偵測與能量估測穩健性VSS-NLMS等化器設計;Design of Peak-Detection and Energy-Estimation Based Robust VSS-NLMS Equalizer for 112 Gbps Receivers

Fri, 06 Mar 2026 10:47:54 GMT

title: 適用於112 Gbps接收機之基於峰值偵測與能量估測穩健性VSS-NLMS等化器設計;Design of Peak-Detection and Energy-Estimation Based Robust VSS-NLMS Equalizer for 112 Gbps Receivers abstract: 本研究旨在分析與驗證基於峰值偵測之可變步階演算法之延伸探討，在 112 Gbps 四階脈衝振幅調變 (PAM-4) 之高速串列傳輸 (SerDes) 系統中的效能。在 112G PAM-4 低訊雜比(SNR)環境下的高速鏈路中，標準 LMS 演算法面臨收斂速度與穩態誤差之間的關鍵權衡(Trade-off)問題；VSS-NLMS 演算法透過動態調整步階(Step-size)，旨在同時實現快速收斂與低穩態均方根誤差(RMSE)。本論文建立一完整的 SerDes 端對端模擬平台，包含由 S 參數定義的業界標準通道模型、前饋等化器(FFE)與決策回饋等化器(DFE)之組合架構。透過詳盡的學習曲線 (Learning Curve)與位元錯誤率(BER)分析，本研究將量化比較 VSS-NLMS、歸一化最小均方與標準 LMS 演算法，在高速 PAM-4 通道下的收斂特性與穩態效能。本研究驗證 VSS-NLMS 演算法在解決傳統 LMS 權衡問題上的優勢，並評估其作為下一代 SerDes 系統中先進等化器核心演算法的可行性。此外，實際高速互連在系統層級可能受到突發性電磁干擾所致之脈衝/突波雜訊(例如 ESD 與 EFT/Burst 等瞬態事件)，其非高斯且具厚尾特性易造成自適應等化器步長估測與權重更新被離群誤差主導。故本研究進一步於步長估測與權重收斂路徑導入基於 M-估計量之影響函數，以提升演算法面對突波干擾時之收斂穩健性。 ;This research aims to analyze and validate the performance of the Variable Step-Size Normalized Least Mean Square (VSS-NLMS) algorithm, specifically focusing on its peakdetection and energy-estimation extensions, within a 112 Gbps PAM-4 (Four-Level Pulse Amplitude Modulation) high-speed Serializer/De-serializer (SerDes) system. In high-speed links, the standard LMS algorithm faces a critical trade-off between convergence speed and steady-state error. The VSS-NLMS algorithm, by dynamically adjusting its step-size, seeks to achieve both fast convergence and low steady-state Root Mean Squared Error (RMSE) simultaneously. This thesis establishes a complete end-to-end SerDes simulation platform, incorporating an industry-standard channel model defined by S-parameters and a combined Feed-Forward Equalizer (FFE) and Decision Feedback Equalizer (DFE) architecture. Through detailed analysis of learning curves and Bit Error Rate (BER), this study quantitatively compares the convergence characteristics and steady-state performance of VSS-NLMS, Normalized LMS (NLMS), and standard LMS algorithms under a high-speed PAM-4 channel. This work validates the advantages of the VSS-NLMS algorithm in resolving the traditional LMS tradeoff and evaluates its feasibility as an advanced equalizer core for next-generation SerDes systems. And high-speed interconnects may be subject to sudden electromagnetic interference(ESD) at the system level, resulting in pulse/surge noise(such as ESD and EFT/Burst transient events). This non-Gaussian, thick-tailed noise can easily cause outlier errors to dominate the adaptive equalizer step size estimation and weight updates. Therefore, this study further incorporates an influence function based on the M-estimater into the step size estimation and weight convergence path to improve the algorithm′s convergence robustness in the face of surge interference.

自動化有機育苗系統

Fri, 06 Mar 2026 10:47:41 GMT

title: 自動化有機育苗系統 abstract: 近年我國農業人口因都市化與少子化而持續下降，且農業勞動力高度高齡化，導致育苗等基礎農務面臨人力不足的挑戰。在此背景下，導入自動化技術以降低人力依賴並提升作業穩定性已成為農業發展的重要方向，因此本研究開發一套以潮汐灌溉（Ebb and Flow）為核心之自動化有機育苗系統，期望提供一致可控的育苗環境，並蒐集後續數位孿生模型所需之環境與生長資料。本系統整合育苗箱、水位控制機構、感測器模組，以及用於記錄幼苗生長影像之影像紀錄設備，可自動進行進排水並回收水資源，同時量測光照、土壤濕度與環境溫濕度等參數，並擷取幼苗生長影像以支援數位孿生建立。實驗結果顯示，本系統能有效促進種子發芽並維持幼苗初期穩定生長，驗證潮汐灌溉應用於育苗之可行性，然而，育苗箱深度設計、設備清潔維護流程及幼苗生長一致性仍需改善，以提升整體系統性能與實務應用價值。;The agricultural workforce has been declining due to urbanization and low birth rates, while rapidly aging demographics have intensified labor shortages in fundamental tasks such as seedling production. Addressing this issue requires the adoption of automation technologies to reduce labor dependency and improve operational stability. Motivated by this need, this study develops an automated organic seedling cultivation system based on the Ebb and Flow irrigation method, aiming to provide consistent growth conditions and generate environmental and growth data for future digital-twin applications. The system integrates seedling boxes, a water-level control mechanism, sensor modules, and an image-recording unit for capturing seedling growth. It automates water inflow and drainage with water recycling, measures parameters such as light intensity, soil moisture, and water level, and records growth images. Experimental results indicate that the system effectively supports seed germination and stable early seedling development, demonstrating the feasibility of applying Ebb and Flow irrigation to seedling cultivation. Nevertheless, improvements in box-depth design, maintenance procedures, and seedling uniformity are needed to enhance system performance and practical applicability.

ProjectionFormerBEV：深度感知且時序穩定的體素交叉注意力三維物體檢測框架;ProjectionFormerBEV: A Depth-Aware and Temporally Stable Framework for 3D Object Detection with Voxel Cross-Attention

Fri, 06 Mar 2026 10:47:27 GMT

title: ProjectionFormerBEV：深度感知且時序穩定的體素交叉注意力三維物體檢測框架;ProjectionFormerBEV: A Depth-Aware and Temporally Stable Framework for 3D Object Detection with Voxel Cross-Attention abstract: 在自動駕駛中，僅透過相機來實現可靠的三維物體檢測仍然具有挑戰，主要原因包括深度不確定、感測器安裝的偏差以及運動模糊等問題。本研究提出了ProjectionFormerBEV，一種基於相機的 BEV 表示框架。我們的方法首先將圖像特徵轉換為偽體素特徵，並通過體素交叉注意力（Voxel Cross-Attention）模組將它們融合，從而改進空間對齊。此外，我們還在不同幀之間引入了自注意力機制，使模型能夠根據場景變化調整預測，並且減輕運動模糊的影響。我們在 nuScenes-mini 數據集上對該方法進行了評估。ProjectionFormerBEV 在NDS 與 mAP 上分別達到 0.0517 和 0.0071，相較於可比基線，分別提升了 11.7% 和 74.6%。在部分幀遺失、輸入影像短暫停滯或相機位置略微偏移的情況下，模型仍表現出穩定的性能。定性分析結果也顯示，模型預測的邊界框在 BEV 與相機視圖中均更接近真實標註的位置。綜合以上結果，可見 ProjectionFormerBEV 為基於相機的三維感知提供了一種可行的方案，並具備在實際駕駛場景中應用的潛力。;For autonomous driving, achieving reliable 3D object detection with cameras alone remains difficult due to challenges such as uncertain depth, sensor setup deviations, and motion blur. In this work, we introduce ProjectionFormerBEV, a camera-only BEV framework. Our approach first converts image features into pseudo-voxels and uses a Voxel Cross-Attention module to fuse them, improving spatial alignment. Additionally, we included a self-attention mechanism across frames, allowing the model to adjust its predictions according to scene changes and to reduce the effects of motion blur. We evaluated the approach on the nuScenes-mini benchmark. ProjectionFormerBEV obtained an NDS of 0.0517 and an mAP of 0.0071, corresponding to 11.7% and 74.6% relative improvements over comparable baselines. The model also showed consistent performance when some frames were missing, inputs froze temporarily, or camera positions shifted slightly. Our qualitative study further showed that the predicted bounding boxes tended to align more closely with the ground truth in both BEV and camera views. Taken together, these findings indicate that ProjectionFormerBEV offers a practical approach for camera-based 3D perception and could be employed in real driving conditions.

分段式Leaky-LMS 演算法之高速串列通訊系統等化器設計與效能分析;Design and Performance Analysis of Equalizer for High-Speed Serial Communication Systems Based on Segmented Leaky-LMS Algorithm

Fri, 06 Mar 2026 10:47:14 GMT

title: 分段式Leaky-LMS 演算法之高速串列通訊系統等化器設計與效能分析;Design and Performance Analysis of Equalizer for High-Speed Serial Communication Systems Based on Segmented Leaky-LMS Algorithm abstract: 隨著資料中心流量的爆炸性增長，112 Gbps PAM-4 高速串列通訊系統對等化器的性能要求日益嚴苛。傳統 LMS 演算法雖然運算簡單，但在決策導向模式下面臨權重漂移與過度擬合的問題；而全域式 Leaky-LMS 演算法雖能改善權重穩定性，卻因對主游標施加洩漏約束而導致嚴重的性能衰退。本研究提出一種創新的分段式 Leaky-LMS 適應性等化演算法，其核心設計理念為「保護主游標、分段施加洩漏」。演算法依據抽頭與主游標的距離將洩漏因子劃分為三個區域：中心區對主游標不施加洩漏（λ_center = 0），近端區施加中等洩漏（λ_near），遠端區施加較強洩漏（λ_far）。此策略旨在保護主游標權重不受約束，同時選擇性地壓縮不必要的遠端抽頭，兼顧等化性能與權重稀疏化。透過完整的 112 Gbps PAM-4 SerDes 系統模擬平台進行驗證，本研究得出以下關鍵成果：（1）分段式 Leaky-LMS 成功將 FFE 權重稀疏性從 52.38% 提升至 71.43%，提升幅度達 19.05%；（2）MSE 性能衰退僅 0.22 dB，相較於全域式 Leaky-LMS 的 2.43 dB 衰退，性能改善達 2.21 dB；（3）計算複雜度降低 66% 至 91%，有效減少硬體實現成本；（4）演算法在全 SNR 範圍（20-40 dB）均展現穩定的性能表現。本研究成果為 112 Gbps PAM-4 及更高速率的下一代高速串列通訊系統提供了一種兼具理論創新與實用價值的適應性等化解決方案。 ;With the explosive growth of data center traffic, 112 Gbps PAM-4 high-speed serial communication systems impose increasingly stringent requirements on equalizer performance. While the conventional LMS algorithm offers computational simplicity, it suffers from weight drift and overfitting issues in decision-directed mode. The global Leaky-LMS algorithm, though improving weight stability, causes severe performance degradation due to leakage constraints applied to the main cursor tap. This research proposes an innovative segmented Leaky-LMS adaptive equalization algorithm, with the core design philosophy of "protecting the main cursor while applying segmented leakage." The algorithm partitions leakage factors into three regions based on tap distance from the main cursor: the center region applies zero leakage to the main cursor (λ_center = 0), the near-end region applies moderate leakage (λ_near), and the far-end region applies stronger leakage (λ_far). This strategy protects the main cursor weight from constraints while selectively suppressing unnecessary far-end taps, achieving both equalization performance and weight sparsity. Through comprehensive validation on a complete 112 Gbps PAM-4 SerDes system simulation platform, this research achieves the following key results: (1) The segmented Leaky-LMS successfully increases FFE weight sparsity from 52.38% to 71.43%, representing a 19.05% improvement; (2) MSE performance degradation is only 0.22 dB, compared to 2.43 dB degradation with global Leaky-LMS, yielding a 2.21 dB performance improvement; (3) Computational complexity is reduced by 66% to 91%, effectively lowering hardware implementation costs; (4) The algorithm demonstrates stable performance across the full SNR range (20-40 dB). This research provides a theoretically innovative and practically valuable adaptive equalization solution for 112 Gbps PAM-4 and next-generation higher-speed serial communication systems.