Unified Error Correction IP for 6G and storage
在4G/5G/6G數位通訊中資料流經常會發生錯誤。要將數百萬bits從一端完整地搬運到其他地方，需要每一次錯誤更正碼的檢查。這保護資料的成本花費包括計算時間與重傳時間成本、計算硬體成本和驗證ECC成本。容錯率高正比於更正能力好，迭代解碼較快收斂有利縮短latency，重傳次數機率降低，基地台信號覆蓋率較廣，更留有餘裕做化簡信號處理的硬體設計可節省功耗。因此，錯誤更正碼是通訊基頻技術的基石，沒有它，通訊就不可能可靠地進行。
通訊基頻(Baseband)的錯誤更正碼（ECC/FEC）需要進一步對齊到儲存產品使用的驗證品質。然而，現有通訊標準的錯誤更正碼無法完全解決訊號中的錯誤。在2016制定5G通訊標準的驗證水準常年在1e-3到1e-5的水準，亦即換算在1000bits的碼字模擬中僅用100 Mega Bytes資料量做前期驗證，我有展示一個YT影片說明如何用30秒完成5G的驗證程度。每個模式僅用30秒的驗證程度來決定5G超過數億元的通訊產業是很誇張的事情。 這品質遠遠不夠在高速應用的環境。例如，當今4K視訊串流中60秒內超過100MB就容易出現一個或以上的錯誤造成畫面閃爍，目前解法僅靠著緩衝非即時顯示，這是遠遠不夠用於高速環境的高可靠度，更何況要做到URLLC去低延遲做遠端醫療，遠端運輸，智慧駕駛等等的事情。
優秀的錯誤更正碼目前無其他選擇，只有兩種壁壘分明的方法：LDPC code和polar code。LDPC碼目前已應用在所有產品，但如何相容於各種長度去應用做到低複雜度，以及如何公平的比較並作加壓驗證證明各種設計的可靠程度是目前存在的議題。polar code有其限制由華為主推僅有5G使用，特性如循序解碼不利於平性高速解碼，長碼更正能力薄弱而無吸引力。 ECC/FEC在學術的觀點被認為發paper的永動機，主要調侃他們注重數學式與新觀點的描述，但不注重壓力測試的程度，所以被形容成在象牙塔裡的研究。對於各巨頭科技公司在通訊技術注重於專利堡壘的利益維護，而公司裡面的工程師領個差不多的薪水，無心於提出不足之處加壓驗證來搞糟自己，一個小時可完成的任務不會拉長到一百個小時這並不划算，與其做工程不如寫專利會更符合公司利益與想法。所以，可預期6G FEC的品質討論將跟5G差不多遲滯不前。
我的研究注重在LDPC碼，它基於校驗矩陣parity matrix的建立，校驗矩陣是由0跟1的排列組成，其中排列的規則、個數與多模的長度相容性就是它的重點。我們的IP設計使用用最少的1(degree)，譬如說每個行僅用四個1(degree-4)來符合各種長度，並做到每個模式都能經歷Tera Bytes的驗證僅錯幾個的程度，相比於現存的應用研究需要超過100至1000倍的時間驗證。 這題目最真實的考驗設計者用最少的代價做出最通用的品質，然後要經歷過最長時間的加壓驗證。我根據我的理解與通用性寫成digital IP的code generator，可即時做儲存與通訊的應用。目前有一個通用外觀的矩陣樣式專利(SEP)在NTHU。
一流人定標準，二流人玩品牌，三流人做產品。目前台灣需要由我代表參與3GPP，在制定6G時期拿出最好的ecc品質去引領6G 基頻電路baseband的競爭。一個經典的歷史是華為5G在3gpp制定的時期運用它各式的影響力去強加入polar code。如果品質如宣稱地達到高品質後，錯誤更正碼技術的獲利模式可以收專利費和ip授權費。我們相信這類智財技術可以綁定國內的科技代工與晶圓代工等等模式，能對社會大眾做最有利的利益回饋。 通過這技術的進步，可以提高即時視訊醫療的服務品質，讓其可靠地進行。因此，我需要繼續推進錯誤更正碼技術的發展，讓它能夠更好地服務於社會大眾。總之，錯誤更正碼是通訊技術的基石，它的發展能深遠地影響社會大眾。
In 4G, 5G, and 6G digital communications, data streams frequently encounter errors. Transmitting millions of bits from one end to another accurately requires error correction code (ECC) checks at every step. The cost of protecting data encompasses computation time, retransmission time, hardware costs, and the cost of validating ECC. Higher fault tolerance equates to better correction capability, with faster convergence in iterative decoding reducing latency, fewer retransmissions, broader base station signal coverage, and simplified signal processing hardware design that conserves power. Thus, ECC is the cornerstone of communication baseband technology; without it, reliable communication would be impossible. Error correction codes (ECC/FEC) for baseband processor need to align with the validation quality used in storage products. However, current communication standards' ECC cannot completely resolve signal errors. In 2016, the validation standard for 5G communication was set at 1e-3 to 1e-5, which means that for a 1000-bit codeword, only 100 megabytes of data were used for initial validation. I have a YouTube video demonstrating how to achieve this level of validation in 30 seconds. It is astonishing that 30 seconds of validation for each mode could determine the standards for a multi-billion-dollar industry like 5G. This quality is far from sufficient for high-speed applications. For example, in today's 4K video streaming, more than 100MB within 60 seconds can easily result in one or more errors causing screen flicker. The current solution relies on buffering for non-real-time display, which is inadequate for high-reliability, high-speed environments, let alone for URLLC in remote healthcare, remote transportation, and smart driving.
There are currently no other options for excellent error correction codes, only two distinct methods: LDPC codes and Polar codes. LDPC codes are widely used in all products, but issues remain on how to apply them with various lengths to achieve low complexity and how to fairly compare and validate the reliability of different designs. Polar codes, heavily promoted by Huawei and used only in 5G, face limitations such as sequential decoding, which is not conducive to parallel high-speed decoding, and weak error correction capability for long codes. From an academic perspective, ECC/FEC is often seen as a perpetual motion machine for publishing papers, focusing on mathematical formulas and new viewpoints rather than the extent of stress testing. Consequently, it is described as research done in an ivory tower. In contrast, major tech companies focus on protecting their SEP patent portfolios, with engineers receiving similar salaries and having no motivation to highlight deficiencies and stress test thoroughly. Therefore, it is unlikely that the quality of 6G FEC will see significant improvement over 5G.
My research focuses on LDPC codes, based on constructing parity matrices consisting of 0s and 1s. The key is the arrangement rules, the number of 1s, and the compatibility with various lengths. Our IP design uses the minimal number of 1s (degree), for instance, each column having only four 1s (degree-4) to fit various lengths, ensuring that each mode can endure terabytes of validation with only a few errors. This requires 100 to 1000 times more validation than existing applications. This topic tests the designer's ability to achieve the highest quality at the lowest cost and endure the longest stress testing. Based on my understanding and versatility, I have written a digital IP code generator that can be used in both storage and communication applications. Currently, there is a universal matrix pattern patent (SEP) at Tsinghua University.
First-class people set standards, second-class people build brands, and third-class people make products. Taiwan needs representation in 3GPP to present the best ECC quality during the 6G standard-setting phase, leading the competition in 6G baseband circuits. A classic example is Huawei's influence in incorporating Polar codes during the 5G standard-setting process in 3GPP. If the claimed high quality is achieved, the profit model for error correction codes can include patent fees and IP licensing fees. We believe this intellectual property can bind to domestic technology and semiconductor fab models, providing the greatest benefits to society. Advancements in this technology can improve the quality of real-time video medical services, ensuring reliable operation. Therefore, I need to continue advancing ECC technology to better serve society. In summary, ECC is the cornerstone of communication technology, and its development can profoundly impact the public.