https://ir.lib.ncu.edu.tw/handle/987654321/103 Search the Channel s https://ir.lib.ncu.edu.tw/simple-search https://ir.lib.ncu.edu.tw/handle/987654321/110742 title: 基於磁滯效應之鎳氫電池荷電狀態估測與剩餘放電時間預測;基於磁滯效應之鎳氫電池荷電狀態估測與剩餘放電時間預測 abstract: 本論文針對鎳氫電池（Nickel-Metal Hydride， NiMH）於動態負載條件下之荷電狀態（State of Charge， SOC）估測問題進行研究。由於NiMH電池具有開路電壓（Open Circuit Voltage， OCV）平坦區特性與明顯之充放電磁滯效應（Hysteresis），使傳統SOC估測方法於動態操作環境下容易產生估測誤差累積與收斂能力下降等問題。因此，本研究提出一套結合磁滯補償、自適應量測雜訊調整功能之三維擴展式卡爾曼濾波器（Three-Dimensional Extended Kalman Filter， 3D-EKF）估測架構。在研究方法方面，本研究以一階RC等效電路模型及七階多項式OCV-SOC關係模型為基礎，建立包含SOC、極化電壓與磁滯電壓之三維狀態空間模型。相較於傳統二維擴展式卡爾曼濾波器（2D-EKF），本研究進一步導入動態磁滯補償模型，以描述充放電切換過程中之電壓路徑差異；同時提出基於OCV斜率特性之自適應量測雜訊協方差（Adaptive Measurement Noise Covariance， ( )）調整機制，使演算法能於電壓平坦區降低量測雜訊干擾，提高估測穩定性。此外，本研究結合SOC估測結果與60秒移動平均電流濾波，建立剩餘運行時間（Time-to-Cutoff， TTC）與剩餘能量（Remaining Energy， ( )）之估測機制，並設計結合時間門檻與預測電壓門檻之雙重防護策略。在系統實現方面，本研究建構由實體電池、動態負載、ESP32/ADS1115資料擷取及PC端MATLAB線上估測所組成之實體電池線上估測平台，以ESP32微控制器搭配16位元ADS1115類比數位轉換器進行即時電壓與電流量測，並透過UDP通訊將量測資料傳輸至MATLAB平台執行3D-EKF估測運算。為驗證所提出方法之有效性，本研究設計純電阻負載、馬達動態負載及頻繁充放電切換等三組實驗情境，每組納入五次正式實驗，共十五筆資料，並刻意設定約50個百分點之初始SOC偏差，以驗證閉迴路估測器之初始狀態修正能力。所有RMSE均以量測起始10分鐘後之資料計算。在庫倫效率設定為0.95之模擬測試中，CC、2D-A及3D-A之RMSE分別為7.25%、3.55%及2.80%。三組代表性Run 1中，Case I之CC、2D-F、2D-A、3D-F及3D-A RMSE分別為3.02%、4.66%、4.67%、3.01%及3.00%；Case II分別為2.86%、2.22%、1.39%、1.80%及2.16%；Case III分別為7.36%、4.94%、4.37%、4.73%及4.02%。五次實驗平均RMSE方面，Case I分別為3.84%、4.93%、4.99%、3.15%及3.11%；Case II分別為3.17%、1.84%、1.67%、2.62%及2.82%；Case III分別為5.82%、2.95%、2.79%、3.03%及2.74%。結果顯示，Case I與Case III以3D-A具有較低之跨實驗平均RMSE，Case II則以2D-A最低，說明磁滯狀態與自適應量測權重之效益會隨SOC工作區域、負載軌跡及資料條件而改變。所有EKF之五次平均prior-NIS均低於自由度1之95%卡方門檻3.841，顯示目前Q與R設定較為保守。此外，本研究亦使用指定之Case II資料進行TTC與雙重截止保護驗證；預測端電壓條件與5分鐘TTC條件分別於約114分鐘及約120分鐘首次成立。;This thesis investigates the state-of-charge (SOC) estimation of nickel-metal hydride (NiMH) batteries under dynamic loading conditions. The flat region of the open-circuit-voltage–SOC (OCV–SOC) relationship reduces the voltage sensitivity to SOC， while charge–discharge hysteresis introduces a path-dependent voltage difference. These characteristics may reduce the correction capability of a conventional two-dimensional extended Kalman filter (2D-EKF). To address these issues， a three-dimensional extended Kalman filter (3D-EKF) incorporating a hysteresis-voltage state and an adaptive measurement-noise covariance mechanism is developed. The proposed estimator is based on a first-order RC equivalent-circuit model and a seventh-order OCV–SOC polynomial identified for the tested battery. The three state variables are polarization voltage， SOC， and hysteresis voltage. The adaptive measurement-noise covariance is determined from the local OCV–SOC slope using a three-region mapping. The covariance is set to its upper value in the flat OCV region， decreases linearly and continuously between the two slope thresholds， and returns exactly to the baseline value in the steep region. This mechanism reduces the influence of low-information voltage residuals without artificially increasing the observability of SOC. An experimental platform consisting of a NiMH battery， dynamic loads， an ESP32 microcontroller， ADS1115 analog-to-digital converters， UDP communication， and a MATLAB-based estimator was constructed. The evaluation included a simulation with fixed Gaussian-noise sequences and three experimental cases involving a resistive load， a motor load with a single charge–discharge transition， and a motor load with frequent operating-mode transitions. The EKF estimators were initialized at 50% SOC， whereas the reference SOC was initialized at 100%. All root-mean-square errors (RMSEs) were calculated using data collected after the first 10 minutes. With a coulombic-efficiency setting of 0.95， the simulation RMSEs of coulomb counting， 2D-A， and 3D-A were 7.25%， 3.55%， and 2.80%， respectively. Fifteen formal experimental runs were evaluated， with five runs for each of Cases I， II， and III. After excluding the first 10 min， the five-run mean RMSEs for CC， 2D-F， 2D-A， 3D-F， and 3D-A were 3.84%， 4.93%， 4.99%， 3.15%， and 3.11% in Case I; 3.17%， 1.84%， 1.67%， 2.62%， and 2.82% in Case II; and 5.82%， 2.95%， 2.79%， 3.03%， and 2.74% in Case III， respectively. Thus， 3D-A achieved the lowest cross-run mean RMSE in Cases I and III， whereas 2D-A performed best in Case II. The mean one-step-prior NIS values of all EKF variants remained below the 95% chi-square threshold of 3.841 for one degree of freedom， indicating conservative covariance settings. The designated Case II dataset was also used to demonstrate time-to-cutoff， remaining-energy， and dual-threshold protection functions. The one-step-prior voltage condition and the 5-min TTC condition were first triggered at 114.017 min and 120.778 min， respectively. <br> Tue, 20 Oct 2026 08:22:48 GMT https://ir.lib.ncu.edu.tw/handle/987654321/110739 title: 基於匹配追蹤之排序連續干擾消除演算法於 OFDM 雷達偵測：弱目標遮蔽效應之抑制;基於匹配追蹤之排序連續干擾消除演算法於 OFDM 雷達偵測：弱目標遮蔽效應之抑制 abstract: 隨著第五代行動通訊技術與整合式感測與通訊（Integrated Sensing and Communication， ISAC）演進，正交分頻多工（Orthogonal Frequency Division Multiplexing， OFDM）波形在未來車聯網（Vehicle-to-Everything， V2X）之雷達感測應用中備受關注。然而，在實際的交通場景中，目標本體的雷達截面積（Radar Cross Section， RCS）差異，以及距離所引發之自由空間路徑損耗（Free Space Path Loss， FSPL），加以強弱目標間懸殊之功率差異，導致接收端獲取之目標回波訊號能量產生極大的落差。 二維快速傅立葉變換（2D-FFT）偵測方法極度依賴循環前綴（Cyclic Prefix， CP）以維持子載波正交性；目標回波之延遲若超過 CP 長度，將引發符號間干擾（Inter-Symbol Interference， ISI）破壞訊號的正交性，致使距離資訊無法正確估計。儘管導入匹配追蹤（Matching Pursuit）之壓縮感測演算法能有效克服CP 長度之距離偵測限制，然因高能量強目標之旁瓣（Side-lobe）能量往往會遮蔽低能量弱目標之主瓣（Main-lobe），仍存在弱目標無法被偵測的問題。為有效解決此強弱目標混合場景中之遮蔽效應，提出基於匹配追蹤之有序逐次干擾消除（Matching Pursuit-based Ordered Successive Interference Cancellation）演算法。本研究所提之演算法架構分為初始偵測與迭代修正兩個階段。第一階段透過匹配追蹤並結合有序逐次干擾消除機制進行初始偵測，依據回波能量由強至弱之順序初步估計目標延遲；接著，為避免強目標估測偏差引發之誤差傳遞（Error Propagation）效應，本演算法於第二階段根據初始偵測的結果，將個別目標從原始訊號中隔離，進行無干擾之訊號重建與估計。 以 5G NR FR2 參數為基礎，並建構V2X 場景進行模擬驗證。實驗結果顯示，所提演算法成功偵測所有目標，有效抑制強目標旁瓣。相較於 2D-FFT 與匹配追蹤方法，本演算法在弱目標偵測完整性方面展現顯著優勢，驗證了將 OSIC 機制引入 OFDM 雷達感測領域之可行性與有效性。;With the advancement of fifth-generation (5G) mobile communication technologies and the evolution toward Integrated Sensing and Communication (ISAC)， Orthogonal Frequency Division Multiplexing (OFDM) waveforms have attracted considerable attention for radar sensing applications in future Vehicle-to-Everything (V2X) systems. However， in realistic traffic scenarios， differences in target Radar Cross Section (RCS) and the Free Space Path Loss (FSPL) induced by varying target ranges collectively result in a large dynamic range of received echo signal power across multiple targets. The two-dimensional Fast Fourier Transform (2D-FFT) detection method relies critically on the Cyclic Prefix (CP) to maintain subcarrier orthogonality. When the propagation delay of a target echo exceeds the CP length， Inter-Symbol Interference (ISI) is introduced， destroying signal orthogonality and preventing accurate range estimation. Although Matching Pursuit (MP)-based compressed sensing algorithms can overcome the range detection limitation imposed by the CP length， the Side-lobe energy of high-power strong targets tends to mask the Main-lobe of low-power weak targets， leaving weak targets undetectable. To effectively address this masking effect in mixed strong-and-weak target scenarios， this study proposes a Matching Pursuit-based Ordered Successive Interference Cancellation (MP-Based OSIC) algorithm. The proposed algorithm operates in two stages: initial detection and refined estimation. In the first stage， Matching Pursuit combined with the ordered successive interference cancellation mechanism sequentially estimates target delays in descending order of echo energy. Subsequently， to mitigate error propagation caused by estimation bias of strong targets， the second stage isolates each individual target from the original received signal based on the initial detection results， enabling interference-free signal reconstruction and parameter re-estimation. Simulations are conducted based on 5G NR FR2 system parameters under a constructed V2X scenario. Experimental results demonstrate that the proposed algorithm successfully detects all targets and effectively suppresses strong target sidelobes. Compared with the 2D-FFT and Matching Pursuit methods， the proposed algorithm exhibits significant advantages in weak target detection completeness， validating the feasibility and effectiveness of incorporating the OSIC mechanism into OFDM radar sensing. <br> Tue, 20 Oct 2026 08:22:43 GMT https://ir.lib.ncu.edu.tw/handle/987654321/110735 title: 基於二次多項式迴歸之OFDM雷達能量域模型重建與距離估計;Energy-Domain Model Reconstruction and Range Estimation for OFDM Radar Based on Quadratic Polynomial Regression abstract: 隨著無線通訊系統朝向高頻寬與高度整合發展，如何在既有通訊架構下延伸環境 感測功能，已成為通訊與感測整合(Joint Communication and Sensing， JCS)領域之重要研究方向。正交分頻多工(Orthogonal Frequency Division Multiplexing， OFDM)波形廣泛應用於現代無線通訊系統，於系統驗證與量測過程中所蒐集之接收端 IQ 資料，本質上包含完整之通道頻率響應資訊，除可評估通訊品質外，亦隱含多徑延遲與距離相關特徵，使既有通訊訊號具備延伸應用於雷達感測之潛力。 在各式 OFDM 雷達距離估計方法中，能量估計法(energy-based estimation)僅利用子載波能量分佈即可進行距離推估，具有不依賴精確相位同步、計算複雜度低且高度相容於既有通訊硬體等優勢。然而於實際量測環境中，射頻前端頻率響應、天線增益不一致與硬體非理想效應將造成子載波能量呈現低頻乘性趨勢失真，破壞理想週期性結構，導致距離估計結果產生顯著誤差與不穩定現象。 為解決上述問題，本研究從能量域訊號模型出發，將通道失真建模為乘性低頻包 絡(multiplicative energy envelope)，並提出一套基於二次多項式迴歸之趨勢建模與振幅等化方法(quadratic polynomial regression-based detrending and equalization)。透過最小平方準則估計子載波能量背景趨勢並進行等化補償，可有效移除低頻失真，同時保留距離相關之高頻週期特徵。距離估計階段則採用可分離式四參數弦波估計法，透過 Zero-crossing 初始化與可分離式非線性最小平方估計相位斜率，再由相位斜率換算目標的估計距離。 實測結果顯示，在 5.8GHz/160MHz、5.8GHz/320MHz、7.2GHz/160MHz、 7.2GHz/320MHz 四組頻帶與頻寬條件、1~15m 連續距離量測下，PA 預補償與二次多項式包絡補償同時啟用時，整體平均絕對誤差(MAE)分別為 0.088m、0.082m、0.172m與0.157m，顯著提升估計準確度與穩定性。上述成果驗證既有通訊量測設備在無額外硬體成本下即可延伸為近距離雷達感測平台，展現 OFDM 通訊與雷達硬體共用架構之實務應用可行性。;With the rapid evolution of wireless communication systems toward higher bandwidth and tighter integration， extending environmental sensing capabilities within existing communication infrastructures has become a key research topic in joint communication and sensing (JCS). Orthogonal Frequency Division Multiplexing (OFDM) waveforms are widely adopted in modern wireless standards， and the in-phase and quadrature (IQ) measurements collected at the receiver inherently contain complete channel frequency response (CFR) information. Beyond communication quality evaluation， such measurements also embed multipath delay and distance-related characteristics， enabling communication signals to be reused for radar sensing. Among various OFDM-based ranging techniques， energy-based estimation determines target distance directly from subcarrier energy distributions and offers advantages including low computational complexity， no requirement for precise phase synchronization， and high compatibility with existing communication hardware. However， in practical measurement environments， radio-frequency front-end responses， antenna gain mismatches， and hardware non-idealities introduce low-frequency multiplicative distortions across subcarriers. These distortions disrupt the ideal periodic energy structure and significantly degrade ranging accuracy and stability. To address this issue， this study models the channel distortion as a multiplicative low-frequency energy envelope and proposes a quadratic polynomial regression-based detrending and equalization approach. By estimating the background energy trend using a least-squares criterion and compensating the measured energy sequence accordingly， the proposed method effectively suppresses low-frequency distortions while preserving the high-frequency periodic features associated with target distance. The range estimation stage employs a Separable four-parameter sinusoidal estimation method， in which the phase slope is first initialized by zero-crossing analysis and then refined using separable nonlinear least squares. The target range is subsequently obtained from the estimated phase slope. Experimental results demonstrate that， under four frequency-band and bandwidth conditions (5.8GHz/160MHz， 5.8GHz/320MHz， 7.2GHz/160MHz， and 7.2GHz/320MHz) over a continuous 1-15m distance range， the mean absolute error (MAE) achieved with both PA precompensation and quadratic polynomial envelope compensation enabled is 0.088m， 0.082m， 0.172m， and 0.157m， respectively. This substantial improvement verifies that existing communication measurement platforms can be directly extended to short-range radar sensing without additional hardware cost， highlighting the practical feasibility of OFDM-based hardware sharing between communication and radar systems. <br> Tue, 20 Oct 2026 08:22:28 GMT https://ir.lib.ncu.edu.tw/handle/987654321/110731 title: 基於IQxel-MX實現IEEE 802.11be 規範之 OFDM Radar量測;Realization and Verification of IEEE 802.11be OFDM Characteristics Based Radar Sensing Using IQxel-MX abstract: 正交分頻多工（Orthogonal Frequency Division Multiplexing， OFDM）具備高頻譜效率與明確的子載波結構，適合延伸至通訊與感測整合應用。本研究參考 IEEE 802.11be（Wi-Fi 7）之主要 OFDM 參數，設計並實現一套自定義 OFDM 雷達量測系統，並以 LitePoint IQxel-MX 綜合測試儀作為波形發射、回波擷取與演算法驗證平台。系統採用 160 MHz 與 320 MHz 頻寬，於 5.8 GHz 與 7.2 GHz 載波頻段進行靜態單目標距離量測。 為處理實體收發時的幀起始位置不確定性，本研究以前導 Zadoff-Chu（ZC）序列完成時序同步與資料幀定位；完成 OFDM 符號切片、循環前綴移除及頻域映射後，再以符號共同相位估測降低殘餘載波頻率偏移與本地振盪器相位漂移所造成的符號間相位旋轉，並利用最小平方法整合多個 OFDM 符號之通道頻率響應。此流程建立了由 MATLAB 波形生成、IQxel-MX 射頻發射與接收、I/Q 資料回傳至 MATLAB 後端分析的完整量測鏈。 本研究所使用之 IQxel-MX 綜合測試儀器涵蓋 400 MHz 至 7.3 GHz 的射頻操作範圍，並提供 480 MHz 即時分析頻寬。接收資料分別採用 FFT-based 距離估測（距離向 IFFT 法）與能量估計法進行分析。前者由通道頻率響應之子載波相位斜率形成距離響應，並搭配共同相位校正、頻域加窗、中央零填補與三點拋物線內插；後者則利用直達路徑與主要反射路徑所形成的頻域通道能量週期估測超額延遲，再依準單基地幾何換算目標距離。 實驗於 1 m 至 15 m 範圍內完成四組條件、共 600 筆重複量測。能量估計法於 5.8 GHz／BW160、5.8 GHz／BW320、7.2 GHz／BW160 與 7.2 GHz／BW320 條件下之平均絕對誤差分別為 0.071 m、0.100 m、0.169 m 與 0.137 m；FFT-based 距離估測之平均絕對誤差分別為 0.164 m、0.170 m、0.174 m 與 0.174 m。結果證實，本研究之 IQxel-MX 架構可作為參考 IEEE 802.11be 主要 OFDM 參數之高頻寬靜態單目標測距與演算法驗證平台；惟目前尚未執行慢時間都卜勒 FFT，因此實測結果不代表完整距離－都卜勒成像能力。 此外，本研究另以 IEEE 802.11be EHT 標準 PPDU 進行延伸測距驗證，於 5 m 與 10 m、BW160 與 BW320 條件下，透過標準封包偵測、載波頻率偏移校正與 EHT-LTF 通道估測取得 CFR，並分別導入能量估計法與 FFT-based 距離估測。IQxel-MX 所量得之 OFDM Composite EVM 約介於 −35.935 dB 至 −39.196 dB，結果顯示所建立之量測平台可由自定義 OFDM waveform 進一步延伸至 IEEE 802.11be EHT PPDU 之通道量測與距離估測。;Orthogonal frequency-division multiplexing (OFDM) provides high spectral efficiency and a well-defined subcarrier structure， making it suitable for integrated communication and sensing applications. This study develops a custom OFDM radar measurement system parameterized according to key IEEE 802.11be (Wi-Fi 7) settings. A LitePoint IQxel-MX wireless connectivity tester is used as the platform for waveform transmission， echo acquisition， and algorithm verification. Measurements are conducted at carrier frequencies of 5.8 GHz and 7.2 GHz with bandwidths of 160 MHz and 320 MHz for static single-target ranging. To address uncertainty in the received frame start， a Zadoff-Chu (ZC) preamble is used for timing synchronization and frame localization. After OFDM symbol segmentation， cyclic-prefix removal， and frequency-domain mapping， symbol-wise common-phase estimation is applied to reduce inter-symbol phase rotation caused by residual carrier-frequency offset and local-oscillator phase drift. The channel frequency response is then integrated across multiple OFDM symbols using a least-squares estimator. This procedure establishes a complete measurement chain from MATLAB waveform generation and IQxel-MX RF transmission/reception to MATLAB-based I/Q post-processing. The IQxel-MX instrument configuration used in this study covers 400 MHz to 7.3 GHz and provides 480 MHz of real-time analysis bandwidth. The received data are processed using two range-estimation approaches: FFT-based range estimation using a range-domain IFFT， and an energy-based estimation method. The FFT-based method applies common-phase correction， frequency-domain windowing， centered zero-padding， and three-point parabolic interpolation. The energy-based method estimates the excess delay from the periodic channel-energy pattern produced by the direct and dominant reflected paths and converts the delay to range using a quasi-monostatic geometric approximation. A total of 600 repeated measurements are collected over target distances from 1 m to 15 m. For the 5.8 GHz/160 MHz， 5.8 GHz/320 MHz， 7.2 GHz/160 MHz， and 7.2 GHz/320 MHz conditions， the mean absolute errors of the energy-based method are 0.071 m， 0.100 m， 0.169 m， and 0.137 m， respectively， whereas those of the FFT-based method are 0.164 m， 0.170 m， 0.174 m， and 0.174 m. The results demonstrate the feasibility of the IQxel-MX platform for high-bandwidth static single-target ranging and algorithm verification using an OFDM waveform parameterized according to key IEEE 802.11be settings. Because no slow-time Doppler FFT is performed in the experiments， the reported results do not constitute full range-Doppler imaging. In addition， IEEE 802.11be EHT PPDUs are employed for extended ranging experiments at target distances of 5 m and 10 m with 160 MHz and 320 MHz bandwidths. CFRs obtained through standard packet detection， carrier-frequency-offset correction， and EHT-LTF channel estimation are further processed using the energy-based and FFT-based ranging methods. The measured OFDM composite EVM ranges from −35.935 dB to −39.196 dB， demonstrating that the proposed IQxel-MX platform can be extended from custom OFDM waveforms to channel measurement and ranging based on IEEE 802.11be EHT PPDUs. <br> Tue, 20 Oct 2026 08:22:08 GMT