合成孔徑雷達透過裝載在衛星或是飛機上，能夠不受天氣、日夜影響的遙測地表，生成二維的影像。在載體飛行的過程中，雷達與目標物的距離變化會產生都普勒效應，精確的都普勒參數對於測距都普勒演算法的跨耦合補償相當重要。基於相位演算法和多視互乘頻率演算法能分別應用於基頻都普勒質心估測與都普勒歧異值估測。而估測前的區塊切割，能透過品質篩選刪除雜訊大的區塊，只選擇可靠的估測結果。我們提出相位離散度作為品質篩選，多視互乘頻率演算法的估測成功率能從69%提升至75%。在硬體設計上，將都普勒參數和測距都普勒演算法做整合設計，提出了共用測距壓縮資料的流程圖，並針對於硬體運算做簡化，同時考量運算延遲、產出率與量化後之效能。為了減少硬體用量，在三次測距壓縮的硬體實作上，使用運算排程和共用技巧，與二次測距壓縮的部分硬體共用，在少量硬體的增加下，最終成像能達到36dB的量化訊雜比。對於都普勒參數估測的硬體實作，透過排程減少複數乘法器、累加器和位元反轉記憶體，比原本設計共減少28%的乘法器和25%的記憶體。而在文獻比較上，我們的設計有快於4倍的處理速度。為了支援1.6秒內的即時運算，設計滿足51Gbps存取速度的測距都普勒演算法高頻寬記憶體介面，處理速度比近期文獻快至少2.93倍；同時設計記憶體空間配置，控制PCIe介面在電腦與FPGA間高速傳輸。另一方面在電腦端的處理流程，透過混合式編譯加速資料格式轉換，完成影像的傳輸與處理。最後，我們將所有操作整合於使用者圖形化介面，完成方便於使用者操作的成像加速處理器。;Synthetic Aperture Radar (SAR) as a payload on satellites or aircraft can remotely sense the Earth’s surface regardless of weather conditions. During sensing, the movement between the radar and the target results in the Doppler effect. Accurate Doppler parameters are crucial for cross-coupling compensation in the range Doppler algorithm (RDA). The phase-based algorithm and the Multilook Beat Frequency Algorithm (MLBF) can be applied to baseband Doppler centroid estimation and Doppler ambiguity estimation, respectively. Prior to estimation, block partition can eliminate noisy blocks through quality measurement. We propose phase spread as a quality measurement index, which can improve the success rate of estimation in MLBF from 69% to 75%. We then integrate the function of Doppler parameter estimation into the RDA processor. The range compressed data from RDA are shared to simplify hardware computation. The computation latency as well as throughput and performance of quantization are all investigated. To reduce hardware utilization in the implementation of third-order secondary range compression to support larger-squint angles, scheduling and sharing techniques are employed. For hardware implementation of Doppler parameter estimation, we totally save 28% multipliers and 25% memory, and the processing speed is four times faster than the prior work and thus can deal with real-time processing for a SAR image of size 32K×8K. In order to support real-time operations for SAR imaging within 1.6 seconds, an RDA interface to high bandwidth memory (HBM) is realized with a 51Gbps throughput. Finally, we design the graphical user interface which can benefit users to operate the imaging acceleration processor easily and provide configurability.
