1273779 五、發明説明（J) 發明背景 1 ·發明部份 (請先閲讀背面之注意事項再填寫本頁) 本發明係有關資料壓縮技術，且更明確言之，係有關在 不同型式之無損失編碼器中選擇之方法及裝置，以最佳化系 統性能。 2.有關技藝之說明 資料壓縮操作在減少用以儲存或發送資訊之位元數至最 少程度，並包含廣大範圍之軟體及硬體壓縮技術。顯然，視 欲壓縮之資料型式及若干其他因素而定，特定壓縮技術在壓 縮率及編碼速度上可提供明顯較優之性能。 經濟部智慈財產局貨工消費合作社印製 一般言之，資料壓縮包含取出一符號或片語流，並變換 其爲較原貪料爲小（在位兀長度上）之代碼。已知之壓縮技 術及演算法可分爲二主要家族，包括損失性及無損失。損失 性資料壓縮可用以大幅增加資料壓縮率；然而，壓縮之增加 以精確性之一些損失換來。結果，損失性壓縮普通僅實施於 一些資料損失爲可接受之情形。例如，當應用於數位語音信 號及圖晝影像上時，可有效使用損失性壓縮。另一方面，無 損失壓縮爲一資料壓縮家族，此使用經設計之技術，俾在壓 縮/解壓縮循環後，產生輸入資料流之確實複製本。當儲存 資料庫記錄，文字處理檔等時，需使用此型式之壓縮，在此 ’資訊之損失絕對不可接受。本發明目的在無損失資料壓縮 〇 當對一組特定符號提供一機率表時，一些無損失壓縮演 本紙張尺度適用中國國家標準（CNS ) A4規格（21〇X 297公I ) -4- 1273779 經濟部智慧財產局員工消費合作社印製 A7 B7五、發明説明（2 ) 算法使用資訊理論來產生可變長度碼。根據一模型對一特定 付疏或一*組付藏（即^1息）決疋$目(1出某一代號。該模型爲一 組法則，用以處理輸入資訊，並反應而決定輸出何代碼。一 演算法或程式使用該模型來分析符號（例如，決定有關該符 號之機率），及然後，根據該處理輸出一適當之代碼。有若 干方法來模造資料，所有此等可使用相同編碼技術來產生其 輸出。一般言之，爲有效壓縮資料，需選擇一模型，此以高 機率預測符號或片語，因爲具有局機率之符號或訊息具有低 資訊含量，且故此需要較少之位元來編碼。次一步驟爲使用 一特定之無損失編碼器對符號編碼。 一·般言之，無損失壓縮編碼器可依其是否實施統計模型 或字典基礎之模型分類。統計模型讀出一單個符號，並每次 使用字母之外觀之機率對其編碼，而字典基礎之模型則使用 單個代碼取代符號串。顯然，在字典基礎之模型，模型遠較 在統計基礎之模型中重要，因爲有關每一符號之編碼之問題 大爲降低。 知道一種統計資料表示爲Shannon-Fano ( S — F )編碼 法。S - F碼讀法發展來提供可變長度位元編碼法，俾使編 碼符號可確實具有（或密切接近於）訊息或符號所含之資訊 之位元數。S - F編碼法依賴知道一訊息中每一符號之出現 之機率。在決定機率後，製造一代碼表，每一代碼具有不同 之位元數（具有低機率之符號宜具有較多之位元）。此種編 碼技術之一問題爲其產生可變長度之代碼，此具有整數位元 ，即使欲編碼之資訊可能需要非整數之位元亦然。 (請先閱讀背面之注意事項再填寫本頁) •裝· 訂 線 本纸張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） - 5- 1273779 A7 B7 五、發明説明（3) (請先閱讀背面之注意事項再填寫本頁) 另一種編碼法，Huffman編碼法與S - F編碼法相似， 因爲此製造整數位元之可變長度代碼，但此使用完全不同之 演算法。一般言之，S — F及Huffman編碼法在性能上接近 ’但Huffman編碼法已確定恆至少等於s - F編碼法之效率 ’故爲較佳者，尤其是二演算法使用同量之處理功率。雖 Huffman編碼法較易實施，及在編碼及解碼上較經濟，但由 於與S - F編碼法同樣使用每代碼之整數位元，故效率不佳 。如確定一特別符號具有1 · 5位元之資訊含量（即資訊量 )，Huffman編碼器產生具有位元數爲一或二之一代碼。一 般言之，如一統計方法可指定9 0 %之機率給一特定之符號 ;則最佳之代碼大小可爲〇 · 1 5位元；然而，Huffman或 S - F編碼法可能指定一個位元之代碼給該符號，此較所需 者大六倍。 經濟部智慧財產局工消費合作社印製 鑒於有關使用整數位元之此問題，發展出算術編碼法。 算術編碼法由一單個浮點輸出數取代一輸入符號流，並繞過 由一特定代碼取代一輸入符號之步驟。由於當符號機率爲一 半之整數冪時，算術編碼法不限於僅最佳者（此最常並非如 此），此獲得欲編碼之符號之理論資訊量，故此對任何已知 之來源壓縮效率最大。換言之，如一特定字母之資訊量爲1 • 5位元，算術編碼法使用1 · 5位元來對該符號編碼，此 爲Huffman及Shan non-Fa no編碼法可能所不可能。雖算術編 碼法極爲有效，但在C P U功率及記憶上消耗較大量之計算 資源。此乃由於需建立需要大記憶量之複雜模型，且演算法 本身需要大量之計算操作之故。 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） -0 - 1273779 A7 B7 五、發明説明（4 ) (請先閱讀背面之注意事項再填寫本頁) 上式無損失編碼法之另一種稱爲取代或字典基礎之編碼 法，字典基礎之壓縮演算法由參考前發生之特定片語取代資 料流中發生之該等片語（即位元組群）。與以上系統（此等 由符號編碼爲位元串達成壓縮，使用較原符號爲少之位元） 不同者，字典基礎之演算法並不對單個符號編碼。而是，字 典基礎之壓縮技術對可變長度符號串作爲單個”記號”編碼 。此等記號形成一片語字典之索引。由於記號較其所取代之 片語小’故發生壓縮。二類主要之字典基礎之壓縮設計稱爲 L e m p e 1 - Z i v壓縮編碼器家族之L Z 7 7及L Z 7 8壓縮演算 法。顯然，字典基礎之編碼法廣泛使用於桌上型通用壓縮中 ，且由”計算服務資訊服務”實施，以對位元圖圖畫影像編 碼。例如，G I F格式使用一 L Z W變型來壓縮重複之序列 及螢幕影像。雖字典基礎之壓縮技術爲非常普遍之壓縮形態 ，但此演算法之缺點爲需要較複雜之資料結構來處理該字典 〇 經濟部智慧財產局員工消費合作社印製 總而言之，隨諸如網際網路等通訊媒體之擴張，資料壓 縮對資料之有效通訊仍極重要，由不同之壓縮演算法在特定 之區域中提供特定之優點。有多種資料壓縮方法在本藝中實 施，包括上述者及其他。而且，已發展出有關每種已知之壓 縮演算法之許多變型及許多改進。而且，視有關該系統之若 干因素及壓縮之資料之型式而定，各可宜於提供最佳之資料 編碼。 由於視各種操作因素，包括欲編碼之資料而定，不同之 已知編碼技術提供獨有之優點，故需要一無損失壓縮系統， 本纸張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） - 7- 1273779 經濟部智慧財產局員工消費合作社印製 Α7 Β7 五、發明説明（c) 5 此由不同型式之編碼器對資料作選擇性編碼。尤其是，通訊 工業需要一種系統，此實施不同型式之編碼器，尤其是當進 來之資料係自多個來源接收，此等提供不同型式之未知資料 ’即當資料流之不同部份由不同編碼技術最佳壓縮時爲然。 發明槪要 本發明係有關一種方法及裝置，此決定若干埋設之編碼 設計之何個最佳壓縮進來資料流之不同部份。較宜實施例之 方法設計應付具有不同資訊包特徵之資料（例如來自編碼器 未知之來源），各可具有不同之所屬統計。 依據較宜實施例之一第一方面，無損失壓縮資料流之一 方法包括提供多個無損失編碼器。該方法故此包括選擇無損 失編碼器之一來壓縮該資料流；及其後由所選之無損失編碼 窃¥彳δ亥貪料流編碼。 依據較宜實施例之另一方面，無損失壓縮資料流之方法 包括使用多個無損失編碼器來壓縮資料流之一測試部份。一 旦測試部份壓縮後，該方法決定有關每一無損失編碼器之性 能特性。然後，該方法包括根據決定步驟選擇無損失編碼器 之一，及由所選之編碼器對資料流之一第一部份編碼。其後 ，該方法包括對資料流之另一測試部份及一第二部份重覆該 使用，決定，選擇，及編碼步驟。顯然，重覆步驟可包括選 擇一不同之無損失編碼器。 依據較宜實施例之另一方面，無損失編碼器各使用（· 1 )一壓縮技術，及（2 )由選擇步驟所決定之每字之位元數 本纸張尺度適用中國國家標準（CNS ) Α4規格（210Χ 297公釐） ---------抑衣------、訂——— (請先閱讀背面之注意事項再填寫本頁)1273779 V. INSTRUCTIONS (J) BACKGROUND OF THE INVENTION 1. Inventive part (please read the note on the back and then fill out this page) The present invention relates to data compression technology, and more specifically, it relates to losslessness in different types. The method and device selected in the encoder to optimize system performance. 2. Description of the Skills The data compression operation reduces the number of bits used to store or transmit information to a minimum and includes a wide range of software and hardware compression techniques. Obviously, depending on the type of data to be compressed and a number of other factors, certain compression techniques provide significantly better performance in terms of compression ratio and coding speed. Printed by the Ministry of Economic Affairs, Zhici Property Bureau, the Consumer Cooperatives Co., Ltd. Generally speaking, data compression involves taking out a symbol or a stream of words and transforming it into a code that is smaller (in terms of length) than the original. The known compression techniques and algorithms can be divided into two main families, including loss and no loss. Loss of data compression can be used to significantly increase the data compression rate; however, the increase in compression is exchanged for some loss of accuracy. As a result, lossy compression is generally only implemented where some data loss is acceptable. For example, when applied to digital voice signals and graphics images, lossy compression can be effectively used. On the other hand, lossless compression is a family of data compressions, which uses a technique designed to produce a true copy of the input data stream after the compression/decompression cycle. This type of compression is required when storing database records, word processing files, etc., and the loss of information is absolutely unacceptable. The purpose of the present invention is to provide a probability table for a specific set of symbols when lossless data compression. Some lossless compression paper scales are applicable to the Chinese National Standard (CNS) A4 specification (21〇X 297 public I) -4- 1273779 Ministry of Economic Affairs Intellectual Property Bureau employee consumption cooperative printed A7 B7 V. Invention Description (2) The algorithm uses information theory to generate variable length codes. According to a model, a specific payment or a group of deposits (ie, ^1 interest) is determined by a certain number. The model is a set of rules for processing input information and reacting to determine the output. Code. An algorithm or program uses the model to analyze symbols (eg, determine the probability of the symbol), and then, output an appropriate code based on the process. There are several ways to model the data, all of which can use the same code. Technology to generate its output. In general, in order to effectively compress data, a model needs to be selected, which predicts symbols or phrases with high probability, because symbols or messages with a chance of having a low information content, and therefore require fewer bits. The second step is to encode the symbol using a specific lossless encoder. 1. In general, the lossless compression coder can be classified according to whether it implements a statistical model or a dictionary based model. A single symbol, and each time it is encoded with the probability of the appearance of the letter, and the dictionary-based model replaces the symbol string with a single code. Obviously, based on the dictionary Models, models are far more important than models based on statistics, because the problem of coding for each symbol is greatly reduced. Know a statistical representation of Shannon-Fano (S-F) coding. Development of S-F code reading To provide variable length bitwise encoding so that the encoded symbol can have (or is closely related to) the number of bits of information contained in the message or symbol. The S-F encoding relies on knowing the presence of each symbol in a message. Probability. After determining the probability, a code table is created, each code having a different number of bits (a symbol with a low probability should have more bits). One of the problems of this coding technique is that it produces a variable length. The code, which has integer bits, even if the information to be encoded may require a non-integer bit. (Please read the note on the back and then fill out this page) • Install and set the paper size to apply to Chinese national standards (CNS) A4 size (210X297 mm) - 5- 1273779 A7 B7 V. Invention description (3) (Please read the note on the back and fill out this page) Another coding method, Huffman coding method and S-F The code method is similar, because this makes a variable-length code of integer bits, but this uses a completely different algorithm. In general, the S-F and Huffman coding methods are close in performance but the Huffman coding method has been determined to be at least equal to The efficiency of the s-F coding method is preferred, especially the second algorithm uses the same amount of processing power. Although the Huffman coding method is easier to implement and economical in coding and decoding, it is encoded with S-F. The method also uses integer bits per code, so the efficiency is not good. If it is determined that a special symbol has an information content of 1.5 bits (ie, information amount), the Huffman encoder generates a code with one or two bits. In general, if a statistical method can specify a probability of 90% to a specific symbol; then the optimal code size can be 〇·1 5 bits; however, the Huffman or S-F encoding may specify a bit. The code gives the symbol, which is six times larger than the one required. Printed by the Ministry of Economic Affairs, the Intellectual Property Bureau, and the Consumer Cooperative. In view of the problem of using integer bits, the arithmetic coding method was developed. Arithmetic coding replaces an input symbol stream by a single floating point output number and bypasses the step of replacing an input symbol with a particular code. Since the arithmetic coding method is not limited to only the best one when the symbol probability is an integer power of one half, which is most often the case, this obtains the theoretical information amount of the symbol to be encoded, and thus has the greatest compression efficiency for any known source. In other words, if the amount of information for a particular letter is 1 • 5 bits, the arithmetic coding uses 1 · 5 bits to encode the symbol, which may be impossible for Huffman and Shan non-Fa no coding. Although arithmetic coding is extremely effective, it consumes a large amount of computing resources in C P U power and memory. This is due to the need to build complex models that require large amounts of memory, and the algorithm itself requires a large number of computational operations. This paper scale applies to China National Standard (CNS) A4 specification (210X297 mm) -0 - 1273779 A7 B7 V. Invention description (4) (Please read the note on the back and fill in this page) Another coding method, called substitution or dictionary basis, the dictionary-based compression algorithm replaces the words (ie, the tuple group) that occur in the data stream by the particular phrase that occurs before the reference. Unlike the above systems (these are encoded by symbol encoding as a bit string, using bits smaller than the original symbol), the dictionary-based algorithm does not encode a single symbol. Rather, the dictionary-based compression technique encodes a variable-length symbol string as a single "mark". These tokens form an index into a dictionary of words. Compression occurs because the token is smaller than the phrase it replaces. The second type of dictionary-based compression design is called L e m p e 1 - Z i v compression encoder family L Z 7 7 and L Z 7 8 compression algorithm. Obviously, the dictionary-based encoding method is widely used in desktop general compression and is implemented by the "computing service information service" to encode the bitmap image. For example, the G I F format uses an L Z W variant to compress repeated sequences and screen images. Although the dictionary-based compression technique is a very common compression form, the shortcoming of this algorithm is that it requires a more complicated data structure to process the dictionary. The Ministry of Economic Affairs, the Intellectual Property Office, the employee consumption cooperative, in general, with communications such as the Internet. With the expansion of the media, data compression is still very important for effective communication of data, and different compression algorithms provide specific advantages in specific areas. A variety of data compression methods are implemented in the art, including the above and others. Moreover, many variations and many improvements have been developed with respect to each of the known compression algorithms. Moreover, depending on the factors involved in the system and the type of data to be compressed, it is appropriate to provide the best data coding. Due to various operational factors, including the data to be encoded, different known coding techniques provide unique advantages, so a lossless compression system is required. This paper scale applies to the Chinese National Standard (CNS) A4 specification (210X297).厘) - 7- 1273779 Ministry of Economic Affairs Intellectual Property Bureau Staff Consumer Cooperative Printed Β7 Β7 V. Invention Description (c) 5 This code is selectively encoded by different types of encoders. In particular, the communications industry needs a system that implements different types of encoders, especially when incoming data is received from multiple sources, which provide different types of unknown data' ie when different parts of the data stream are encoded differently The best technology is when compressing. SUMMARY OF THE INVENTION The present invention is directed to a method and apparatus for determining which of the plurality of embedded coding designs is optimally compressed into different portions of the data stream. The method of the preferred embodiment is designed to cope with data having different packet characteristics (e.g., from sources unknown to the encoder), each having different statistics. According to a first aspect of the preferred embodiment, the method of compressing a data stream without loss comprises providing a plurality of lossless encoders. The method thus includes selecting one of the lossless encoders to compress the data stream; and then encoding the selected lossless code. According to another aspect of the preferred embodiment, the method of compressing the data stream without loss includes compressing one of the test portions of the data stream using a plurality of lossless encoders. Once the test is partially compressed, the method determines the performance characteristics of each lossless encoder. The method then includes selecting one of the lossless encoders in accordance with the determining step, and encoding the first portion of one of the data streams by the selected encoder. Thereafter, the method includes repeating the use, decision, selection, and encoding steps for another test portion and a second portion of the data stream. Obviously, the repeating step can include selecting a different lossless encoder. According to another aspect of the preferred embodiment, the lossless encoder uses (1) a compression technique, and (2) the number of bits per word determined by the selection step. The paper size applies to the Chinese national standard (CNS) Α4 specifications (210Χ 297 mm) ---------Suppressing clothes ------, order-- (please read the notes on the back and fill out this page)
-8 - 1273779 A7 B7 五、發明説明（6) 於編碼步驟中。而且’壓縮技術爲鼻術編碼法，Η11 f f m a η編 碼法，及L Z編碼法之一。 (請先閱讀背面之注意事項再填寫本頁) 依據較宜實施例之又另一方面，一種無損失資料壓縮之 裝置包含一介面，用以接收一資料流。而且，該裝置包含多 個無損失編碼器及一處理器。在操作中，每一無損失編碼器 分別壓縮資料流之一測試部份，且該處理器反應而決定有關 每一無損失編碼器之性能特性，及然後根據性能特性，選擇 無損失編碼器之一來對資料流之至少一第一部份編碼。 依據較宜實施例之又另一方面，性能特性包含對應無損 失編碼器之該測試部份之壓縮之壓縮率及持續時間之至少之 ~。而且，編碼器包含多個處理器，及每一無損失編碼器相 當於一處理器，且其中，無損失編碼器平行壓縮同一測試部 份。 經濟部智慧財產局員工消費合作社印製 自詳細說明及附圖，精於本藝之人士可明瞭本發明之此 等及其他目的，優點，及特色。然而，應明瞭詳細說明及附 Η雖指示本發明之較宜實施例，但僅以例解之方式提出，且 非限制。在本發明之範圍內可作許多更改及修改，而不脫離 其精神，且本發明包含所有此等修改。 附圖簡述 本發明之較宜實施例顯示於附圖中，其中，相之參考編 號表示所有相似部份，且其中： ® 1爲流程圖，顯示較宜實施例之方法之大體操作； 圖1 Α爲圖表，顯示圖1所示方法中所用之一列無損失 冬 本纸張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） 經.濟部智慈財產局員工消費合作社印製 1273779 A7 _ B7 五、發明説明（7) 編爾器； 圖2爲大體方塊圖，顯示較宜實施例之一編碼/解碼系 糸充；及 圖3爲槪要圖，顯示由圖2所不之系統編碼/解碼之資 料流。 主要元件對照 3 2 無損失編碼器 4〇 系統 4 1 編碼器 4 2 輸入介面 4 3 資料流 4 4 輸入/輸出線 4 6 數位信號處理晶片 4 8 P C I匯流排 5 0 電腦 5 2 媒體 6 4 資料包分部 較佳實施例之詳細說明 參考圖1 ，在步驟1 2初始化及開始後，方法1 〇在步 驟1 2輸入資料至系統。在步驟1 4所輸入之資料可爲同步 或異步資料。顯然，可自未指定之來源，諸如監_〜物件之 溫度，壓力等（軍事應用所收集之遙測資料），逝不斷發送 本紙張尺度適用中國國家標準（CMS ) A4規格（210X297公釐） 私衣------II--.------ (請先閲讀背面之注意事項再填寫本頁)-8 - 1273779 A7 B7 V. Description of invention (6) In the encoding step. Moreover, the 'compression technique is the nasal coding method, the Η11 f f m a η coding method, and one of the L Z coding methods. (Please read the note on the back and then fill out this page.) According to another aspect of the preferred embodiment, a lossless data compression device includes an interface for receiving a data stream. Moreover, the device includes a plurality of lossless encoders and a processor. In operation, each lossless encoder compresses one of the test portions of the data stream, and the processor reacts to determine the performance characteristics of each lossless encoder, and then selects a lossless encoder based on the performance characteristics. First, at least a first portion of the data stream is encoded. According to still another aspect of the preferred embodiment, the performance characteristic includes at least a compression ratio and a duration of compression of the test portion of the lossless encoder. Moreover, the encoder includes a plurality of processors, and each lossless encoder is equivalent to a processor, and wherein the lossless encoder compresses the same test portion in parallel. Printed by the Intellectual Property Office of the Ministry of Economic Affairs and the Consumers' Cooperatives. These and other objects, advantages and features of the present invention will become apparent to those skilled in the art. However, the detailed description and the accompanying drawings are intended to be construed as Many changes and modifications may be made without departing from the spirit and scope of the invention. BRIEF DESCRIPTION OF THE DRAWINGS Preferred embodiments of the invention are shown in the drawings, in which reference numerals indicate all like parts, and wherein: <RTI ID=0.0> 1 Α is a chart showing that one of the methods used in the method shown in Figure 1 has no loss. The winter paper size applies to the Chinese National Standard (CNS) A4 specification (210X297 mm). Printed by the Ministry of Finance, Zhici Property Bureau, Staff Consumer Cooperative, 1273779 A7 _ B7 V. Description of the invention (7) Arranger; Figure 2 is a general block diagram showing the encoding/decoding system of one of the preferred embodiments; and Figure 3 is a schematic diagram showing the structure of Figure 2 System data encoding/decoding data stream. Main component comparison 3 2 lossless encoder 4〇 system 4 1 encoder 4 2 input interface 4 3 data stream 4 4 input/output line 4 6 digital signal processing chip 4 8 PCI bus 5 0 computer 5 2 media 6 4 data DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT Referring to Figure 1, after initialization and start of step 12, Method 1 inputs data to the system in step 12. The data entered in step 14 can be synchronous or asynchronous. Obviously, from unspecified sources, such as the temperature of the equipment, pressure, etc. (telemetry data collected by military applications), the paper is continuously sent to the Chinese National Standard (CMS) A4 specification (210X297 mm). Clothing ------II--.------ (Please read the notes on the back and fill out this page)
-10- 1273779 A7 B7 五、發明説明（8) (請先閱讀背面之注意事項再填寫本頁) 讀數至較宜實施例之系統編碼器（以下說明）之感測器接收 資料流。未指定之資料需意爲有關該資料之統計爲隨機者， 且故此，與由單一型式之編碼器依據有關資料之統計之知識 執行壓縮之已知系統不同，較佳實施例能對不同型式之資料 所構成之資料流有效編碼。其他型式之應用包含醫院監視應 用，化學工廠，核子電廠，及其他，在此，此型式之隨機資 料可自多個來源發出。 當資料不斷輸入至該系統時，資料發送至通訊塊之一分 部’在此’方法1 〇在步驟1 6處理該資料，分割或裝框該 資料’俾供其進一步通訊之用。通訊塊上之該分部由方法1 0以普通方式執行。其次，在步驟1 8，預處理該資料，此 可包含產生一組織圖，指示有關在步驟1 6中裝框之資料之 統計。 經濟部智慧財產局員工消費合作社印製 一旦資料在步驟1 8中經預處理，方法1 0在步驟 2 0視需要加進同步及標頭碼，俾進一步處理及辨認資料流 中之資料位元。於完成步驟2 0後，資料發送至多個編碼器 ，此等提供無損失之壓縮。明確言之，在步驟2 2，方法1 〇由多個無損失編碼器對資料流之一測試部份編碼，並決定 有關每一編碼器之系統性能標準。用以在步驟 2 2中對資料之該部份編碼之編碼器顯示於圖1 a之圖表3 0中之3 2處。圖表之各欄顯示不同型式之無損失編碼技術 /演算法’此等可包含Huffm㈣編碼法，算術編碼法， Lempel-Z1V編碼法，以及此等之變體及其他已知之編碼技術 。顯然，該方法亦比較編碼技術之輸出及未編碼/壓縮之資 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） -11 - 1273779 A7 B7 9 五、發明説明（ 料流，因爲在一些情形，未壓縮之資料可爲最佳。 (請先閲讀背面之注意事項再填寫本頁) ~般言之，各欄包含無損失之編碼技術。各列包含可用 以對資料編碼之每字位元數（bpwl-m )之不同指定。例如， 有關介面之每字位元可設定於例如8位元，1 0位元等。結 果，在步驟2 2，方法1 〇由η X m數之無損失編碼器對資 料之一部份編碼。步驟2 2宜在一測試時間或資料量中執行 ’以決定何無損失編碼器先達成最佳系統性能。其後，對該 資料編碼（說明於下）。 在步驟2 2中由編碼器3 2所執行之測試壓縮宜平行實 施’以迅速編輯與每一無損失編碼器相對應之資料。由於電 腦功率如此低廉，致利益（例如，就編碼速度而言）大爲超 過成本，故測試資料可作平行編碼。而且，在另一實施例， 圖1 A所示之每一編碼器3 2可在指定之時間中對測試資料 依次編碼’以產生對應之性能資料。雖非較宜，但當電腦功 率有限時，可執行順次測試。 經濟部智慧財產局員工消費合作社印製 轉至以下表1，顯示在步驟2 2中所產生之九個不同之 無失編碼器（三不同字長度x三不同編碼技術）之性能標 準。先注意輸入位元率設定於一預定値，同時輸出位元率雖 宜依據所用之傳輸媒體設定，但可根據有關所用之無損失編 碼器之反饋資訊，不斷更新。輸出位元率最佳儘可能小。如 顯不於表1，在壓縮一測試量之資料後，對一指定之速度輸 入（干位元/秒）及速度輸出（千位元/秒）各決定輸出檔 之位兀組大小，壓縮率，及編碼時間。例如，在輸入檔具有 3 0 4， 1 8 〇， 9 2位元組，及當使用每字8位元時， 表紙張尺度適财_家椁公瘦） 12 1273779 A7 B7 五、發明説明（10)-10- 1273779 A7 B7 V. INSTRUCTIONS (8) (Please read the notes on the back and fill out this page.) Read the data stream from the sensor to the system encoder (described below) of the preferred embodiment. Unspecified information is intended to mean that the statistics relating to the data are random, and therefore, the preferred embodiment can be different from the known system in which compression is performed by a single type of encoder based on the knowledge of the statistics of the relevant data. The data stream formed by the data is effectively encoded. Other types of applications include hospital surveillance applications, chemical plants, nuclear power plants, and others, where random data of this type can be sent from multiple sources. When data is continuously entered into the system, the data is sent to a subsection of the communication block, where the method 1 processes the data in step 16 and divides or frames the data for further communication. This division on the communication block is performed by the method 10 in the usual manner. Next, the data is preprocessed in step 18. This may include generating an organization map indicating the statistics relating to the data framed in step 16. Printed by the Intellectual Property Office of the Ministry of Economic Affairs, the Consumer Cooperatives Once the data is pre-processed in step 18, the method 10 adds the synchronization and header codes as needed in step 2, and further processes and identifies the data bits in the data stream. . After completing step 20, the data is sent to multiple encoders, which provide compression without loss. Specifically, in step 22, method 1 encodes a portion of the data stream by a plurality of lossless encoders and determines system performance criteria for each encoder. The encoder used to encode the portion of the data in step 22 is shown at 3 2 of the graph 30 of Figure 1a. The columns of the graph show different types of lossless coding techniques/algorithms' which may include Huffm (four) coding, arithmetic coding, Lempel-Z1V coding, and variations of these and other known coding techniques. Obviously, the method also compares the output of the coding technique with the unencoded/compressed capital paper scale applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) -11 - 1273779 A7 B7 9 V. Description of the invention (flow, because In some cases, uncompressed data can be optimal. (Please read the note on the back and fill out this page.) ~ In general, each column contains lossless coding techniques. Each column contains each word that can be used to encode the data. The number of bits (bpwl-m) is differently specified. For example, each word bit of the interface can be set to, for example, 8 bits, 10 bits, etc. As a result, in step 2 2, method 1 〇 is determined by η X m The lossless encoder encodes one part of the data. Step 2 2 should be performed in a test time or data volume to determine the loss of the encoder to achieve the best system performance. Then, encode the data (description In the following step 2, the test compression performed by the encoder 32 in step 2 2 should be implemented in parallel to quickly edit the data corresponding to each lossless encoder. Since the computer power is so low, it is beneficial (for example, Coding speed In other words, the test data can be coded in parallel. Moreover, in another embodiment, each encoder 32 shown in FIG. 1A can sequentially encode the test data in a specified time to generate a corresponding Performance data. Although not suitable, when the computer power is limited, the sequential test can be performed. The Ministry of Economic Affairs Intellectual Property Bureau employee consumption cooperative prints to Table 1 below, showing the nine different ones generated in step 22. The performance standard of the lossless encoder (three different word lengths x three different encoding technologies). First note that the input bit rate is set at a predetermined time, and the output bit rate should be set according to the transmission medium used, but it can be used according to the relevant The feedback information of the lossless encoder is continuously updated. The output bit rate is optimally as small as possible. If it is not shown in Table 1, after compressing a test quantity data, a specified speed input (dry bit/second) And speed output (thousands per second) each determines the output file size, group size, compression ratio, and encoding time. For example, in the input file has 3 0 4, 1 8 〇, 9 2 bytes, and when use When Word 8 bits, the paper sheet suitable financial _ scale thin home coffin well) 12 1273779 A7 B7 V. invention is described in (10)
Huffman編碼法達成壓縮比率爲 (請先閱讀背面之注意事項再填寫本頁) 1 · 8 2 7 2，Lemp-Ziv編碼法達成比率爲2 · 5 0 5，及 算術編碼法達成比率爲2 . 7 7 2 4。而且，此等演算法之 測試資料之編碼時間分別爲1 2 8秒，5 2 2秒，及1，5 8 2秒。一旦產生每一無損失編碼器之性能標準後，方法1 0執行步驟2 4，以選出編碼器之一，在一預定時間量或特 定資料量上對資料編碼，壓縮。 顯然，在步驟2 4中所作之選擇普通並非單根據所達成 之壓縮率作成，而是，該選擇根據整個處理時間及壓縮比率 性能特性之聯合作成。例如，在表1中，算術編碼法對每字 8位元達成壓縮比率爲2 . 7 7 2 4，此大於Lemp-Ziv編 碼法所達成之壓縮比率2 . 5 0 5。然而，算術編碼法使用 較之Lemp-Ziv無損失編碼器之編碼多1 5分鐘。在此情形 ，方法1 0在步驟2 4中可能選擇Lemp-Ziv編碼器。然而 ，如由所有η X m無損失編碼器所達成之性能不滿足一臨限 程度，則方法1 0可決定發送不壓縮之資料。此決定視使用 者之需求等而定。 經濟部智慧財產局員工消費合作社印製 表1所示之輸入時脈率取決於發送資料之媒體（例如， 網際網路）及所實施之編碼演算法之型式。時間性能標準依 據以下等式產生， (等式1 ) 在等式1中，tp_essing包括有壓縮該資料之時間，系統延遲 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） -13- 1273779 A7 B7 五、發明説明（u) 等。而且t。爲發送資料之時間’且等於檔案大小除以壓縮 率及輸出速度，即位元率’並反映由壓縮資料所達成之時間 節省。壓縮率（c R )等於輸入檔案大小除以輸出檔案大小 (請先閱讀背面之注意事項再填寫本頁) 經濟部智慈財產局a(工消費合作社印製 本紙張尺度適用中國國家標準（CNS ) A4規格（210X 297公釐） 1273779 ΑΊ ΑΊ Β7 五、發明説明（12) 經濟部智慧財產局Μ工消費合作社印製 撇 CU9 卜19二 6寸1 寸.CNΟΝΙεοϋςΟΝΙΙ ooir 寸 uv L ocoe 99寸卜.I 寸s 二unml οοΓΟς m l<rcl 卜9寸9.i 9£rooIr 寸COI 馨« 96L CO69 二 oocn气 I ILIVI 寸 ος~οο66~ζ寸 Inz/cn 0095 m SU.I 寸0460:900.口 I 0ΟΟΓ6 S9CS! AofI CNIsrCNI99;6I Μ Z — APS 1The Huffman coding method achieves a compression ratio of (please read the back note and then fill out this page) 1 · 8 2 7 2, the Lemp-Ziv coding method achieves a ratio of 2 · 5 0 5, and the arithmetic coding method achieves a ratio of 2. 7 7 2 4. Moreover, the coding time of the test data of these algorithms is 1 2 8 seconds, 5 2 2 seconds, and 1,5 8 2 seconds. Once the performance criteria for each lossless encoder is generated, method 10 executes step 24 to select one of the encoders to encode and compress the data for a predetermined amount of time or a particular amount of data. Obviously, the choices made in step 24 are generally not made solely on the basis of the achieved compression ratio, but rather the selection is based on the combination of the overall processing time and the compression ratio performance characteristics. For example, in Table 1, the arithmetic coding method achieves a compression ratio of 2.8 bits per word of 2. 7 2 2 4, which is greater than the compression ratio of 2.55 5 achieved by the Lemp-Ziv coding method. However, the arithmetic coding method uses an encoding of 15 minutes longer than that of the Lemp-Ziv lossless encoder. In this case, method 10 may select a Lemp-Ziv encoder in step 24. However, if the performance achieved by all η X m lossless encoders does not meet a threshold, then method 10 may decide to send uncompressed data. This decision depends on the needs of the user and so on. Printed by the Ministry of Economic Affairs, Intellectual Property Office, Staff Consumer Cooperatives The input clock rate shown in Table 1 depends on the media (eg, the Internet) that sent the data and the type of coding algorithm implemented. The time performance standard is generated according to the following equation: (Equation 1) In Equation 1, tp_essing includes the time to compress the data, and the system delays the paper scale to apply the Chinese National Standard (CNS) A4 specification (210X297 mm) -13 - 1273779 A7 B7 V. Description of invention (u), etc. And t. The time to send the data 'is equal to the file size divided by the compression rate and the output speed, ie the bit rate' and reflects the time savings achieved by compressing the data. The compression ratio (c R ) is equal to the input file size divided by the output file size (please read the note on the back and fill out this page). The Ministry of Economic Affairs, Zhici Property Bureau a (the consumer consumption cooperative printed this paper scale applies to the Chinese national standard (CNS) A4 size (210X 297 mm) 1273779 ΑΊ ΑΊ Β7 V. Invention description (12) Ministry of Economic Affairs Intellectual Property Bureau Completion Consumer Cooperative Printed 撇CU9 Bu 19 2 6 inch 1 inch. CNΟΝΙεοϋςΟΝΙΙ ooir inch uv L ocoe 99 inch .I inch s two unml οοΓΟς m l<rcl 卜 9 inch 9.i 9£rooIr inch COI 馨« 96L CO69 二 oocn gas I ILIVI inch ος~οο66~ inch inch Inz/cn 0095 m SU.I inch 0460:900 .口I 0ΟΟΓ6 S9CS! AofI CNIsrCNI99;6I Μ Z — APS 1
ςςς m 二 VZLrl 96°.9uo;0I 6S°°.I 寸 sCNls 寸寸 roocsl寸二CN1I s 寸01 HOCTCI 002 ZLir\ 96CO0卜寸(99Ϊ smJJnH (急/ικ赵vf)M滕33i掛 (急/IK赵VI-)侧· Ki$ (#:έ 盤 姘liaf (圮5})ψ¥雜¾鑕呍赵π{ψ· (念/]^妇^)侧滕93饍 (念/IK^\f)M«YII (色题篮 lilis (1K5}rR ¥_33_呍5}01 蚱 _ (念/IR妇+)姻餿93i慨 (^/ικϊΐ)侧滕 Yi$ (sHml^ ,ί簡 (IK赵)七 ¥_丑 _1K5}co舴職 本纸張尺度適用中國國家襟準（CNS ) Α4規格（210Χ297公釐） (請先閱讀背面之注意事項再填寫本頁) 燊· -I— -15- 1273779 A7 B7 五、發明説明（13) (請先閱讀背面之注意事項存填寫本頁) 一旦在步驟2 4中選出一編碼器（η，m ) 3 2 (圖1 >時，方法1 0由所選之編碼器宜在預定之時間量中對資料 編碼。其後，程式回至步驟2 2，以對資料流之新測試部份 編碼，並選擇一最佳之編碼器對資料流之次一部份編碼。此 操作可需要實施圖表3 0所示之一不同無損失編碼器。 經濟部智慈財產局員工消費合作社印製 轉至圖2，用以執行方法1 〇之一系統4 0包含一編碼 器4 1，具有一輸入介面4 2，此包含一時脈輸入端 C 1 ，及一資料輸入端d 1用以接收同步或異步之資料流4 3 °介面4 2經輸入及輸出資料控制同步輸入/輸出線4 4 連接至一數位信號處理器（D S P )晶片4 6。顯然，D S P 4 6宜執行圖1所示之方法1 〇中之步驟1 6及1 8，以 裝框該資料，並準備其供壓縮之用。D S P 4 6之輸出經一 P C I匯流排4 8連接至電腦5 0，此連通裝框之資料至電 腦。電腦5 0宜加進適當之標頭碼於資料流中，以指示不同 之資料包，並用以由圖表3 0所示之每一無損失編碼器對測 試資料編碼/壓縮。如上述，電腦5 0可包含多個處理器， 各可爲由圖1 A之圖表3 0所實施之一對應無損失編碼器， 對資料編碼/壓縮。或且，可使用單個電腦5 0，在預定之 時間中依次實施每一無損失編碼器之測試壓縮。 電腦5 0亦可用以加進標頭碼於資料中，以確保檔案正 確解壓縮。經壓縮之資料然後經P C I匯流排4 8發送至D S P晶片4 6，依需要分割資料，供所實施之特定通訊系統 使用。此處理可包括由插入空塊及/或刪除現有塊來緩衝資 料。其後，可加進特定之同步碼，及發送資料流沿輸入/輸 本紙張尺度適用中國國家標準（CNS ) A4規格（210X 297公釐） -16- 1273779 A7 B7 五、發明説明（14) 出線回至介面4 2。該特定之介面碼設定包括指定位元/字 數，每框字數，同步碼，控制和等。介面 (請先閱讀背面之注意事項再填寫本頁) 4 2然後輸出該資料流於線d 2上，俾可發送於媒體5 2， 諸如網際網路上。輸出時脈率C 2宜由操作者設定，且取決 於所實施之媒體5 2之型式。 其次’系統4 0之解碼器5 3包含一介面5 4，具有一 資料輸入端D 3用以接收來自媒體5 3之在時脈率C 3上之 壓縮資料，此相當於介面4 2所輸出之時脈率C 2。顯然， 時脈C 2及C 3可選擇。介面5 4發送壓縮之資料流通過資 料控制同步線5 6，同時刪除由編碼器4 1所加進之P〇C 同步。其後，一 D S P晶片5 8偵測標頭碼，並移去資料流 中之空塊。經D S P晶片5 8處理之資料然後發送經P C I 匯流排6 0至電腦6 2。電腦6 2解壓縮該資料，且宜實施 普通控制和核對（C S C )比較技術。亦可由電腦6 2實施 額外偵錯或改錯編碼器，一 Reed-Solomon改錯編碼器爲通 訊網路用之標準，且宜包含。顯然，上述處理操作可由電腦 經濟部智慧財產局Μ工消費合作社印製 5 0，6 2，或D S P晶片4 6，4 8任一執行，但已說明 較宜之實施。壓縮之資料然後在時脈率C 4 = C 1上發送回 至介面5 4及至資料線D 4上。 圖1所述及由圖2所示裝置所執行之方法步驟之表示槪 要顯示於圖3，用於遙測資料流上。圖3之右方上標示A之 箭頭指示編碼處理，而沿圖3所示資料之左方上之箭頭B則 指示解碼處理。更明確言之，資料流4 3輸入至介面4 2 ( 圖2 )，及然後宜由D S P 4 6依預指定之方式裝框爲資料 -17 - 本紙張尺度適用中國國家標準（CNS ) Α4規格（210X 297公釐） 1273779 A7 B7 五、發明説明（15) 包分部6 4 ’ 6 6 (宜爲數千位元組，例如二8 k分部）。 其後，分部6 4，6 6壓縮爲例如一 4 · 5 k塊6 8及一 4 • 3 k塊7 0。然後，一標頭7 3，7 5加於資料包塊中（ 連同上述之介面資訊），以分別製成塊7 2，7 4。然後， 各塊經緩衝’以構成經緩衝及壓縮之資料包7 6，如需要， 此等可再分割，以製造資料流7 8。然後，由D S P晶片4 6加進P〇C同步，及新資料流8 0可發送經例如網際網路 5 2 (圖2 )而至解碼器5 3，在此如上述解碼。 在本發明範圍內可作許多更改及修改’而不脫離其精神 。本發明範圍內之其他更改及修改顯示於後附之專利申請中 (請先閲讀背面之注意事項再填寫本頁) 經濟部智慈財產局0(工消費合作社印製 - 18- 本纸張尺度適用中國國家標準（CNS ) A4規格（21〇>< 297公釐）Σςς m two VZLrl 96°.9uo;0I 6S°°.I inch sCNls inch inch roocsl inch two CN1I s inch 01 HOCTCI 002 ZLir\ 96CO0 卜 inch (99Ϊ smJJnH (emergency / ικ赵 vf) M Teng 33i hanging (emergency / IK Zhao VI-) Side · Ki$ (#:έ 姘 姘 姘 姘 姘 姘 } 杂 杂 杂 杂 杂 杂 杂 π π π π π π π π π π π π π 念 念 念 念 念 93 93 93 93 93 93 93 93 93 93 93 93 93 93 )M«YII (color basket lilis (1K5}rR ¥_33_呍5}01 蚱 _ (Nian / IR women)) Marriage 93i gene (^ / ικϊΐ) side Teng Yi$ (sHml^, ίjian (IK Zhao) Seven ¥ _ ugly _1 K5} co 舴 本 纸张 纸张 纸张 纸张 纸张 纸张 纸张 纸张 纸张 CN CN CN CN CN CN CN 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格 规格15- 1273779 A7 B7 V. INSTRUCTIONS (13) (Please read the note on the back and fill in this page) Once you have selected an encoder (η, m ) 3 2 in step 2 4 (Figure 1 > The selected encoder should encode the data for a predetermined amount of time. Thereafter, the program returns to step 2 2 to encode the new test portion of the data stream and select an optimal encoder pair data. The second part of the stream is encoded. This operation may require the implementation of one of the different lossless encoders shown in Figure 30. The Ministry of Economic Affairs Zhici Property Bureau employee consumption cooperative printed to Figure 2 to perform the method 1 系统 One system 40 includes an encoder 4 1 with an input interface 4 2, which includes a clock input C 1 And a data input terminal d 1 for receiving a synchronous or asynchronous data stream. The 3 3 interface 4 is controlled by the input and output data. The synchronous input/output line 4 4 is connected to a digital signal processor (DSP) chip 46. Obviously, the DSP 46 should perform the steps 16 and 18 of the method 1 shown in Figure 1 to frame the data and prepare it for compression. The output of the DSP 4 6 is via a PCI bus 4 8 Connect to the computer 50, this connects the framed data to the computer. The computer 50 should add the appropriate header code in the data stream to indicate different data packets and use each of the data shown in Figure 30. The loss encoder encodes/compresses the test data. As described above, the computer 50 may include a plurality of processors, each of which may be a lossless encoder implemented by the chart 30 of Fig. 1A, encoding/compressing the data. Or, a single computer 50 can be used, which is sequentially implemented in a predetermined time. Test compression of a lossless encoder. The computer 50 can also be used to add a header code to the data to ensure that the file is properly decompressed. The compressed data is then sent to the DSP chip via the PCI bus 4 4 . Information needs to be segmented for use by the particular communication system being implemented. This processing may include buffering the data by inserting empty blocks and/or deleting existing blocks. Thereafter, a specific synchronization code can be added, and the data stream can be sent along the input/input paper size. The Chinese National Standard (CNS) A4 specification (210X 297 mm) -16-1273779 A7 B7 5. Invention description (14) Go back to interface 4 2 . The specific interface code setting includes the specified bit/word number, the number of words per frame, the sync code, the control sum, and the like. Interface (Please read the note on the back and fill out this page) 4 2 Then output the data on line d 2, which can be sent to media 5 2, such as the Internet. The output clock rate C 2 is preferably set by the operator and depends on the type of media 5 2 being implemented. Next, the decoder 4 3 of the system 40 includes an interface 5 4 having a data input terminal D 3 for receiving compressed data from the media 5 at the clock rate C 3 , which is equivalent to the output of the interface 4 2 . The clock rate is C 2 . Obviously, the clocks C 2 and C 3 are selectable. The interface 54 sends the compressed data stream through the data control sync line 5 6 while deleting the P〇C sync added by the encoder 41. Thereafter, a D S P chip 58 detects the header code and removes the empty block in the data stream. The data processed by the D S P chip 58 is then sent through the P C I bus 60 to the computer 62. The computer 6 2 decompresses the data and should implement a common control and verification (C S C ) comparison technique. An additional debug or error correcting encoder can also be implemented by the computer 62. A Reed-Solomon error correcting encoder is a standard for the communication network and should be included. Obviously, the above processing operations can be performed by any of the 50, 62, or D S P wafers 4, 4 8 printed by the Intellectual Property Office of the Ministry of Computer and Economy, but the implementation has been described. The compressed data is then sent back to interface 5 4 and to data line D 4 at clock rate C 4 = C 1 . The representation of the method steps described in Figure 1 and performed by the apparatus of Figure 2 is shown in Figure 3 for use in telemetry data streams. The arrow labeled A on the right side of Fig. 3 indicates the encoding process, and the arrow B on the left side of the data shown in Fig. 3 indicates the decoding process. More specifically, the data stream 4 3 is input to the interface 4 2 (Fig. 2), and then should be framed as a data by the DSP 4 6 in a pre-specified manner. - This paper scale applies to the Chinese National Standard (CNS) Α 4 specification. (210X 297 mm) 1273779 A7 B7 V. Description of invention (15) Package division 6 4 ' 6 6 (should be thousands of bytes, for example, two 8 k divisions). Thereafter, the segments 64, 6 6 are compressed into, for example, a 4 · 5 k block 6 8 and a 4 • 3 k block 70. Then, a header 7 3, 7 5 is added to the data block (along with the interface information described above) to form blocks 7 2, 7 4 respectively. The blocks are then buffered' to form a buffered and compressed data packet. If desired, these can be subdivided to produce a data stream 78. Then, P 〇 C sync is added by the D S P chip 46, and the new data stream 80 can be transmitted, for example, via the Internet 5 2 (Fig. 2) to the decoder 53 where it is decoded as described above. Many changes and modifications may be made without departing from the spirit of the invention. Other changes and modifications within the scope of the present invention are shown in the attached patent application (please read the note on the back and then fill out this page) Ministry of Economic Affairs Zhici Property Bureau 0 (Working Consumer Cooperative Printing - 18- Paper Size Applicable to China National Standard (CNS) A4 specification (21〇>< 297 mm)