UNITED STATES PATENT AND TRADEMARK OFFICE
`
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`
`APPLE INC.,
`Petitioner,
`
`v.
`
`FASTVDO LLC,
`Patent Owner.
`
`Patent No. 5,850,482
`
`
`Inter Partes Review No. _____________
`
`
`DECLARATION OF ANDREW LIPPMAN
`
`
`
`
`Apple Inc. Exhibit 1002 Page 1
`
`

`
`TABLE OF CONTENTS
`
`
`Page
`
`
`INTRODUCTION .......................................................................................... 1 
`I.
`II. QUALIFICATIONS ....................................................................................... 1 
`III. MATERIALS CONSIDERED ....................................................................... 4 
`IV. DEFINITIONS AND STANDARDS ............................................................. 6 
`V.
`BACKGROUND OF THE TECHNOLOGY ................................................. 8 
`A.
`Summary .............................................................................................. 8 
`B.
`Source Coding ...................................................................................... 9 
`C.
`Channel Coding .................................................................................. 14 
`D. Unequal Error Protection ................................................................... 16 
`E.
`Conclusion .......................................................................................... 17 
`VI. THE ’482 PATENT ...................................................................................... 18 
`VII. CLAIM CONSTRUCTION ......................................................................... 23 
`VIII. OBVIOUSNESS OF THE CLAIMS OF THE ’482 PATENT .................... 25 
`A. Obviousness of the Claims of the ’482 Patent Based on Kato .......... 25 
`1.
`Relevant Disclosures in Kato ................................................... 25 
`2.
`Obviousness in View of Kato .................................................. 29 
`B. Obviousness of the Claims of the ’482 Patent Based on Fiala in
`View of Fazel and Fazel ’622 ............................................................ 36 
`1.
`Relevant Disclosures in Fiala ................................................... 36 
`2.
`Relevant Disclosures in Fazel .................................................. 38 
`3.
`Relevant Disclosures in Fazel ’622 ......................................... 39 
`4.
`Obviousness Based on Fiala in View of Fazel and
`Fazel ’622 ................................................................................. 40 
`
`
`
`
`
`i
`
`
`
`Apple Inc. Exhibit 1002 Page 2
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`I, Andrew Lippman, hereby declare the following:
`
`I.
`
`INTRODUCTION
`1.
`
`I have been retained by counsel for Apple Inc. (“Petitioner”) as a
`
`technical expert in connection with the proceeding identified above. I submit this
`
`declaration in support of Apple Inc.’s Petition for Inter Partes Review of United
`
`States Patent No. 5,850,482 (“the ’482 patent”).
`
`2.
`
`I am being paid at an hourly rate for my work on this matter. I have
`
`no personal or financial stake or interest in the outcome of the present proceeding.
`
`II. QUALIFICATIONS
`3.
`I am currently a Senior Research Scientist at the Massachusetts
`
`Institute of Technology (“MIT”) and Associate Director of the MIT Media
`
`Laboratory, an approximately $50 Million per year research and teaching facility at
`
`MIT, which I helped establish in the early 1980s. I direct a special interest group
`
`called Ultimate Media, and am co-principal investigator of the Communications
`
`Futures Program, which unifies diverse research projects across MIT related to the
`
`technology, policy, and economics of communications over the Internet.
`
`4.
`
`At MIT, I have supervised over 50 Master’s and Ph.D theses in the
`
`Media Arts and Sciences program and have taught courses such as Digital Video
`
`and MIT’s freshman physics seminars. Through the course of my career, I have
`
`directed and served as principal investigator of research projects supported by the
`
`
`
`1
`
`Apple Inc. Exhibit 1002 Page 3
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`defense department (DARPA), the Office of Naval Research (ONR), The National
`
`Science Foundation (NSF), and over 50 industrial companies. I have never
`
`precisely calculated my net research volume, but it is in excess of $50 Million.
`
`5.
`
`I received my undergraduate degree in Electrical Engineering from
`
`MIT in 1971. I received a Master of Science degree from MIT in 1978 and a Ph.D
`
`in Electrical Engineering from the École Polytechnique Fédérale de Lausanne in
`
`1995. My thesis was on scalable video, a technique for representing visual data in
`
`a fluid and variable networking and processing environment, similar to what we
`
`call streaming today.
`
`6.
`
`I also designed data storage schemes to embed digital data in analogue
`
`video using optical videodiscs, which were analogue video storage devices used to
`
`distribute entertainment and interactive programming in the 1980s and 1990s. My
`
`Master’s student and I developed the channel coding parameters that allowed for
`
`reliable data storage on this optical medium. The results were published in Steve
`
`Yelick’s thesis “Authoring of Optical Videodiscs with Digital Data” in June 1982.
`
`7.
`
`In the early 1980s, I established a research program called “Movies of
`
`the Future,” a multi-sponsor program addressing image distribution, analysis, and
`
`interaction. In 1986, I established the “Television of Tomorrow” program to
`
`research digital and scalable video processing technology. The Television of
`
`Tomorrow program initially had nine sponsors, representing the television
`
`
`
`2
`
`Apple Inc. Exhibit 1002 Page 4
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`industry, consumer electronics industry, and a content company from North
`
`America, Asia-Pacific, and Europe, respectively. I co-authored an article called
`
`“Digital Television: A Perspective,” with Arun Netravali, which reported the ideas
`
`arising from this work. “Digital Television: A Perspective” was published as the
`
`lead article in the June, 1995 IEEE proceedings.
`
`8.
`
`I participated in the second meeting of the Motion Picture Experts
`
`Group, an ISO standards committee effort that defined the standards for storing
`
`and distributing MPEG Video. I co-wrote the paper that defined the requirements
`
`for the MPEG-2 standard with Okubo and McCann in 1995. My participation
`
`came after a presentation at the Torino Picture Coding Symposium where I
`
`presented work on asymmetric coding of video for low-rate channels.
`
`9.
`
`I served on the editorial board of the Image Communication Journal
`
`between 1989 and 2003.
`
`10.
`
`I am named as an inventor on six patents in the area of video and
`
`digital processing and have served on the advisory boards for technology
`
`companies in fields ranging from video conferencing to music analysis. I have
`
`authored or co-authored over 65 published papers in the fields of communications,
`
`video coding, and television, including articles in joint authorship with Bernd
`
`Girod and Edward Adelson, whose work is referenced in these proceedings. These
`
`articles address video coding, joint source/channel coding, video information
`
`
`
`3
`
`Apple Inc. Exhibit 1002 Page 5
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`analysis, HDTV systems, and radio communications. Many of these topics are
`
`contemporaneous with the matters under discussion here.
`
`11.
`
`I have served as an expert for several patent and arbitration cases in
`
`the field of video communications systems and the human interface, mobile
`
`systems, and music streaming. Based on my experiences described above, and as
`
`indicated in my CV (attached hereto as Exhibit 1), I am qualified to provide the
`
`following opinions with respect to the ’482 patent.
`
`III. MATERIALS CONSIDERED
`12.
`In preparing this declaration, I have reviewed the ’482 patent
`
`(Ex. 1001 to the Petition) and its prosecution history, as well as Apple’s Petition
`
`for Inter Partes Review of the ’482 patent. I have also considered the following
`
`materials:
`
` U.S. Patent No. 5,392,037 to Kato (“Kato”);
`
` U.S. Patent No. 5,218,622 to Fazel et al. (“Fazel ’622”);
`
` K. Fazel et al., Application of Unequal Error Protection Codes on Combined
`Source-Channel Coding of Images, International Conference on
`Communications, Including SuperComm Technical Sessions (IEEE),
`Atlanta, April 15-19, 1990, Vol. 3, pp. 898-903 (“Fazel”);
`
` E. R. Fiala & D. H. Greene, Data Compression with Finite Windows, 32
`Communications of the ACM 490, 490-505 (1989) (“Fiala”);
`
` Thomas M. Cover & Joy A. Thomas, ELEMENTS OF INFORMATION THEORY,
`(John Wiley & Sons 1991) (“Cover & Thomas”);
`
` Arun N. Netravali & Barry G. Haskell, DIGITAL PICTURES,
`REPRESENTATION, COMPRESSION, AND STANDARDS (Plenum Press 1988);
`
`
`
`4
`
`Apple Inc. Exhibit 1002 Page 6
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
` J. Modestino & D. Daut, Combined Source-Channel Coding of Images, 27
`IEEE TRANSACTIONS ON COMMUNICATIONS 1644, 1644-1659 (Nov. 1979);
`
` O. Rioul & M. Vetterli, Wavelets and Signal Processing, 8 IEEE SIGNAL
`PROCESSING MAGAZINE, Oct. 1991, at 14;
`
` C. E. Shannon, A Mathematical Theory of Communication, 27 BELL SYSTEM
`TECHNICAL J., July 1948, at 379, 379-423;
`
` W. F. Schreiber & A. B. Lippman, Reliable EDTV/HDTV Transmission in
`Low-Quality Analog Channels, 98 SMPTE J. 496, 496-503 (July 1989);
`
` Ramchandran et al, Multiresolution Broadcast for Digital HDTV Using Joint
`Source/Channel Coding, 11 IEEE J. ON SELECTED AREAS IN COMMUN. 6, 6-
`23 (Jan. 1993);
`
` U.S. Patent No. 5,289,501 to Seshadri et al.;
`
` J. Hagenauer, Rate-Compatible Punctured Convolutional Codes and their
`Applications, 36 IEEE TRANS. ON COMMUN. 389, 389-500 (April 1988);
`
` A. R. Calderbank & N. Seshadri, Multilevel Codes for Unequal Error
`Protection, 39 IEEE TRANS. ON INFORMATION THEORY 1234, 1234-48 (July
`1993);
`
` J. Modestino et al., Combined Source-Channel Coding of Images Using the
`Block Cosine Transform, 29 IEEE TRANS. ON COMMUN. 1261, 1261-74
`(Sept. 1981);
`
` S. W. Golomb, Run Length Encodings, 12 IEEE TRANS. ON INFORMATION
`THEORY 399, 399-401 (1966);
`
` Matoba et al, Still Image Transmission Using Unequal Error Protection
`Coding in Mobile Radio Channel, 79 ELECTRONICS AND COMMUN. IN JAPAN,
`PART 1 75, 75-84 (1996) (translated from the Denshi Joho Tsushin Gakkai
`Ronbunshi, Vol. 78-B-II, No 3, March, 1995, pp. 92-101);
`
` Wallace, The JPEG Still Picture Compression Standard, IEEE
`TRANSACTIONS ON CONSUMER ELECTRONICS, Vol. 38, No. 1, Feb. 1992; and
`
` Lin, Codes with Multi-Level Error-Correcting Capabilities, DISCRETE
`MATHEMATICS 83 (1990), pp. 301-14.
`
`
`
`5
`
`Apple Inc. Exhibit 1002 Page 7
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`IV. DEFINITIONS AND STANDARDS
`13.
`I have been informed and understand that claims are generally
`
`construed from the perspective of one of ordinary skill in the art at the time of the
`
`claimed invention. I understand that as a general rule, claim terms are given their
`
`plain and ordinary meaning, as understood by one of skill in the art at the time of
`
`the claimed invention.
`
`14.
`
`I understand that means-plus-function claiming occurs when a claim
`
`term is expressed as a means or step for performing a specified function without
`
`the recital of structure, material, or acts in support thereof. Such a claim term
`
`should be construed to cover the corresponding structure, material, or acts
`
`described in the specification and equivalents thereof. The use of the word
`
`“means” creates a rebuttable presumption that it is a means-plus-function claim
`
`term.
`
`15.
`
`I understand that construing a means-plus-function claim term is a
`
`two-step process. First, one identifies the claimed function. Then, one determines
`
`if any structure disclosed in the specification corresponds to the claimed function.
`
`Where there are multiple claimed functions, the patentee must disclose adequate
`
`corresponding structure to perform all of the claimed functions. If the function is
`
`performed by a general purpose computer or microprocessor, then the specification
`
`must also disclose the algorithm that the computer performs to accomplish that
`
`
`
`6
`
`Apple Inc. Exhibit 1002 Page 8
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`function. If the patentee fails to disclose adequate corresponding structure, the
`
`claim is indefinite. In particular, failure to disclose the corresponding algorithm
`
`for a computer-implemented means-plus-function term renders the claim
`
`indefinite, because its scope cannot be determined with reasonable certainty by one
`
`of ordinary skill in the art.
`
`16.
`
`I have also been informed and understand that the subject matter of a
`
`patent claim is obvious if the differences between the subject matter of the claim
`
`and the prior art are such that, to a person having ordinary skill in the art, the
`
`subject matter as a whole would have been obvious at the time the invention was
`
`made. I have also been informed that an obviousness determination requires
`
`consideration of the following factors: (1) the scope and content of the prior art;
`
`(2) the differences between the prior art and the claimed subject matter; (3) the
`
`level of ordinary skill in the art; and (4) any objective evidence of non-
`
`obviousness.
`
`17.
`
`I have been informed and understand that claimed subject matter
`
`would have been obvious to one of ordinary skill in the art if, for example, it
`
`results from (1) the combination of known elements according to known methods
`
`to yield predictable results, (2) the simple substitution of one known element for
`
`another to obtain predictable results, (3) the use of a known technique to improve
`
`similar devices in the same way, (4) applying a known technique to a known
`
`
`
`7
`
`Apple Inc. Exhibit 1002 Page 9
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`device ready for improvement to yield predictable results, or (5) pursuing known
`
`options within one’s technical grasp in response to a design need or market
`
`pressure to solve a problem. I have also been informed that the analysis of
`
`obviousness incorporates the logic, judgment, and common sense of a person of
`
`ordinary skill in the art, which does not necessarily require explication in any
`
`particular reference.
`
`18.
`
`In my opinion, a person of ordinary skill in the art pertaining to the
`
`’482 patent at the time of the alleged invention would include a person with a
`
`Bachelor’s degree in electrical engineering, computer engineering, or computer
`
`science, and 3-5 years of experience with data encoding. However, I recognize
`
`that someone with less technical education and more experience—and vice versa—
`
`could also meet this standard.
`
`V. BACKGROUND OF THE TECHNOLOGY1
`A.
`Summary
`19. Data encoding generally involves transforming data into a different
`
`form. The goals of data encoding are to reduce the amount of data that needs to be
`
`stored or transmitted and to ensure that the data is reproduced with acceptable
`
`accuracy even when there are errors in storage or transmission.
`
`
`1 Unless otherwise specified, all citations in this section are to the ’482 patent.
`
`
`
`8
`
`Apple Inc. Exhibit 1002 Page 10
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`20. Data encoding typically involves two steps—source coding and
`
`channel coding. I will discuss basic concepts of source coding and channel coding
`
`below. Although data encoding can be performed on many different types of data,
`
`I focus my discussion on the encoding of image data because that is the primary
`
`subject matter of the ’482 patent. Encoding of other data, such as audio data,
`
`involves similar concepts.
`
`B.
`Source Coding
`21. Source coding, sometimes referred to as “compression,” involves
`
`encoding data to reduce its size.
`
`22.
`
`Images are large data ensembles that are often too large to transmit or
`
`store efficiently. For example, a standard definition U.S. television image at the
`
`time of the alleged invention could represent about 210 megabits per second in
`
`studio formats. By contrast, the original CD-ROM had a data rate of
`
`approximately 1.5 megabits per second (with errors). Consequently, the MPEG
`
`standard was created to reproduce moving images at home VCR quality using the
`
`rate of a CD-ROM as a baseline. Likewise, JPEG, a standard primarily for still
`
`images, was designed to reduce the file size of still pictures.
`
`23. Normally, images are compressed by removing redundancy in the data
`
`and by removing detail that would not be important to an ordinary viewer. Image
`
`data contains quite a bit of redundancy. Lossless image coding, or “entropy
`
`
`
`9
`
`Apple Inc. Exhibit 1002 Page 11
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`coding,” reduces this redundancy. It is suitable for images with technical uses,
`
`such as X-rays and telescope pictures, or for archiving original images.
`
`24. For images that are intended for normal viewing, additional
`
`compression can be achieved by removing data that the human visual system may
`
`not detect, or is otherwise unimportant to a viewer. This is often called “lossy
`
`compression.” With lossy compression, some data is permanently lost in the
`
`compression process.
`
`25. At the time of the ’482 patent, most schemes for image compression
`
`involved first performing a transformation of the image, which concentrates most
`
`of the energy of the original picture into a smaller number of frequency
`
`coefficients that can be used to re-create the original. [’482 patent, 2:11-14.] This
`
`is a reversible process that does not itself introduce noise or error.
`
`26. Common transformations include wavelets, the Discrete Cosine
`
`Transform (DCT), subband transformations, and pyramid coding. These
`
`transformations typically operate on a region of the picture, as opposed to the
`
`entire picture. In general, such transformations produce a DC component, which
`
`represents the average brightness of a picture region, and sets of AC components,
`
`which represent the picture details such as edges, textures, and shading. Most of
`
`the energy is in the DC component and the energy in each coefficient drops in
`
`proportion to frequency.
`
`
`
`10
`
`Apple Inc. Exhibit 1002 Page 12
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`27. One major goal of transformation is energy compaction. Many
`
`transform coefficients are very small, and dropping them or encoding them
`
`coarsely would merely make a picture look softer or less detailed.
`
`28. During the step of quantization, the lowest energy transform
`
`coefficients are eliminated. As noted in the ’482 patent, a great many higher order
`
`coefficients are zero or near-zero and can be eliminated without significant
`
`degradation. [4:12-18.]
`
`29. The significant coefficients that remain are “quantized”—that is,
`
`represented with reduced precision. Quantization is the process of representing
`
`data with fewer bits. For example, all numbers within an interval between t1 and t2
`
`(t2>t1) may be represented as the same level l1 rather than its specific value
`
`between t1 and t2. As a result, fewer bits are needed to represent quantized data
`
`than to represent pre-quantized data. Quantization (and/or removing transform
`
`coefficients) is the only stage in lossy image compression where data is lost. [’482
`
`patent, 3:32.]
`
`30. Quantization precision can be varied in different parts of the image so
`
`that more bits are used for busier regions and fewer for quieter ones. The human
`
`visual system may also guide quantization; for example, people are generally less
`
`sensitive to noise in the busier regions of a picture. The combination of a
`
`
`
`11
`
`Apple Inc. Exhibit 1002 Page 13
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`transformation with quantization thus efficiently and effectively reduces the data in
`
`the picture, albeit at a cost of some error in the reconstruction.
`
`31. Finally, the resulting data is entropy coded to further reduce the data
`
`rate. Entropy is a measure of the average uncertainty in the random variable. It is
`
`the number of bits, on average, required to describe the random variable. When
`
`some values are more likely than others, data compression can be achieved by
`
`assigning short codes to the most frequent outcomes of the data source and
`
`necessarily longer codes to the less frequent outcomes. Samuel Morse intuitively
`
`did this for telegraphy by representing more common letters with shorter Morse
`
`codes and less common ones with longer codes. This greatly improved the overall
`
`speed of telegraphy.
`
`32. A code is called a “prefix code” if no code word is a prefix of another
`
`code word. To provide a simple example, if the code word for a first value is 111,
`
`then the code word for the second value could not be 1110, because 111 is a prefix
`
`of 1110. The code word for the second value could, however, be 1101, because
`
`111 is not a prefix of 1101.
`
`33. Two types of prefix coding schemes mentioned in the ’482 patent are
`
`Huffman coding and unary coding. [4:36-50, 15:12-24.] The Huffman algorithm
`
`can be used to construct an optimal (shortest expected length) prefix code for a
`
`given distribution.
`
`
`
`12
`
`Apple Inc. Exhibit 1002 Page 14
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`34. Entropy coding thus represents fixed sized numbers as strings of
`
`varying length, where the more common ones are shorter and the rarer ones are
`
`longer. Entropy codes are inherently variable length codes, and the converse is
`
`likewise almost always true. Therefore, the terms “entropy coding” and “variable
`
`length coding” are often used interchangeably. A Huffman code is an entropy
`
`code and a variable length code. Arithmetic coding is also an entropy code. There
`
`are multiple ways to implement Huffman or Arithmetic coding.
`
`35. Prefix codes are sometimes known as “instantaneous codes” because a
`
`decoder that receives a continuous stream of bits instantaneously knows when it
`
`has received a complete code word. Indeed, that is the reason that prefix codes are
`
`used—they allow a decoder that has received a continuous stream of code words to
`
`know when one code word ends and the next code word begins. The concept of
`
`“prefix codes” should not be confused with the different concept, which I discuss
`
`below, that code words can be divided into two parts (e.g., a prefix and a suffix).
`
`36. A known problem with relying on prefix codes to determine when a
`
`code word is complete is that an error may occur in that code word during storage
`
`or transmission. If so, the decoder will not only receive the wrong bits for that
`
`code word, it will read out too few or too many of the received bits for that code
`
`word. This will destroy the decoder’s ability to determine where within the
`
`
`
`13
`
`Apple Inc. Exhibit 1002 Page 15
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`received bits subsequent code words begin and end and thus prevent recovery of
`
`the data after the error.
`
`37. All of these concepts of source coding were well understood and in
`
`practical use prior to the ’482 patent. Prominent examples were JPEG for still
`
`pictures and MPEG for moving images.
`
`C. Channel Coding
`38. Channel coding involves formatting data to make it suitable for use in
`
`a particular channel and adding bits to the data stream to detect and correct errors
`
`that are introduced by that channel. There is a duality between the problems of
`
`data compression and data transmission. During compression, redundancy in the
`
`data is removed, whereas during data transmission, or channel coding, redundancy
`
`is added in a controlled fashion to combat errors in the channel.
`
`39. When we use the word “channel,” we typically mean a transmission
`
`channel such as wires or radio. Real channels introduce errors or noise into the
`
`signal they carry. We are generally familiar with the idea of noise—it is the
`
`unintended degradation introduced, for example, when radio signals are weak.
`
`However, storage media can be considered a channel as well—they introduce
`
`errors in the data, either by incorrectly reproducing the stored data or by dropping
`
`bits of data. Thus a storage device is just another channel.
`
`
`
`14
`
`Apple Inc. Exhibit 1002 Page 16
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`40.
`
`In 1948, Claude Shannon of Bell Labs introduced the notion of
`
`channel capacity, the idea that there is a maximum rate at which information can
`
`be sent through a channel when that channel is imperfect. One achieves this
`
`maximum rate by adding redundancy back into the data, which allows a receiver to
`
`correct for corruption or errors in storage or transmission. This is called “forward
`
`error correction,” since the added data is added in advance rather than requested
`
`after an error has been detected.
`
`41. The simplest method of forward error correction is to transmit a bit or
`
`a message several times and “vote” on the result. Then the most popular version is
`
`accepted as the correct one. The more times we repeat the message, the more votes
`
`we get, and the lower the likelihood of a reception error. This requires a
`
`significant amount of overhead that is equal to the same amount of data as the
`
`original message for each vote. Another method is to add “parity bits” to a
`
`message, which can reveal and correct some errors. But there are better ways.
`
`42. One of the first codes for doing this was developed by Richard
`
`Hamming in 1950. The idea was to add bits to a code that can detect a given
`
`number of errors and correct a (smaller) number. A sample Hamming code
`
`transmits seven bits to represent four bits of data, and it can correct any single bit
`
`error in the transmission. Note that this reliability entails almost 60% overhead—
`
`
`
`15
`
`Apple Inc. Exhibit 1002 Page 17
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`three added bits for every four-bit word. More complicated codes process a block
`
`of data and can correct more errors.
`
`43. A variety of channel models have been researched, including the
`
`binary symmetric channel, erasure channels, and fading channels (generally for
`
`radio). One normally optimizes the channel coder for the particular channel. For
`
`example, on an optical storage device such as a CD-ROM, a scratch can often
`
`remove a block of data rather than a single bit. Interleaving the data so that
`
`adjacent bits are physically farther apart on the disk can ameliorate this, but at
`
`some expense in the memory and delay of the coder/decoder. More sophisticated
`
`codes than Hamming codes process large blocks of data or more than one block at
`
`a time, and can correct more severe errors. For example, a block code called Reed-
`
`Solomon is used on the CD-ROM and can correct error sequences as long as 4,000
`
`bits [see Cover & Thomas]. These are all forward error correction schemes.
`
`D. Unequal Error Protection
`44.
`It was well known in the prior art that not all encoded data is equally
`
`important. For example, compressed image data typically comprises two kinds of
`
`information: parameter information describing how the data was encoded, and
`
`picture data representing image components. Often, the parameter information
`
`varies throughout an image in response to the desired file size or data rate, as well
`
`as the amount of detail in the picture. An error in the parameter data can destroy
`
`
`
`16
`
`Apple Inc. Exhibit 1002 Page 18
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`the reconstruction of the data. As another example, errors in some types of picture
`
`data are more noticeable to viewers than errors in other types of picture data.
`
`45. Because some data is more important than others, it was known in the
`
`prior art to apply unequal error protection to data and thereby save some of the
`
`overhead caused by forward error correction. I discuss two specific examples
`
`(Kato and Fazel) in my analysis below regarding the obviousness of the claims of
`
`the ’482 patent.
`
`E. Conclusion
`46. Above I have described the basic elements used in a picture coding
`
`system; these include the source coder, and its transform, quantization, and entropy
`
`coding of the resulting transform coefficients. I also considered forward error
`
`correction for the transmission channel or storage medium. And finally, I outlined
`
`advantageous ways to apply error protection to source encoded data. As noted,
`
`there are many variations of each of these elements. However, all of this was
`
`known and understood by persons skilled in the art by the time the application for
`
`the ’482 patent was filed on April 17, 1996.
`
`
`
`17
`
`Apple Inc. Exhibit 1002 Page 19
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`VI. THE ’482 PATENT2
`47. The ’482 patent discloses an image coding system in which more
`
`important encoded data is subject to unequal error protection. The stated goal is to
`
`“provide an improved error resilient method and apparatus for entropy coding of
`
`data which can utilize unequal error protection techniques of channel coding.”
`
`[6:33-36.]
`
`48. More specifically, the ’482 patent presents “split field coding”
`
`[Abstract, 13:50], whereby individual code words are divided into a prefix and a
`
`suffix [6:53-56]. The prefix is susceptible to bit errors because it can cause loss of
`
`decoder synchronization, but an error in the suffix is limited to the code word in
`
`which the error occurs, and the error is restricted to a range near the correct value.
`
`[6:56-65.] Therefore, the prefix fields are more important, so that they are encoded
`
`with a relatively “higher level of error protection” [7:2].
`
`49. Figure 1 of the ’482 patent (below) presents the context for the
`
`claimed entropy coder: a wavelet image coder with its associated quantizer,
`
`coupled to unequal error protection for the entropy coded data. Ultimately, the
`
`compressed data is either transmitted or stored.
`
`
`2 Unless otherwise specified, all citations in this section are to the ’482 patent.
`
`
`
`18
`
`Apple Inc. Exhibit 1002 Page 20
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`’482 patent, FIG. 1
`
`
`
`50. The ’482 patent describes an “unfortunate consequence of data
`
`compression” as “increased susceptibility to channel errors.” [5:47-49.]
`
`According to the ’482 patent, these errors can be “far-reaching” [5:54] in their
`
`effect on the reconstructed image, generally having wide-area or global
`
`consequences, such as streaks across the image [5:66-6:4], tearing of the image
`
`[6:12-16], and when it occurs on entropy-coded data, loss of code word
`
`synchronization that can persist past the word that itself is corrupted [6:17-21],
`
`which can have “catastrophic effects on a reconstructed image” [6:20-21].
`
`51. The encoding method of the ’482 patent is depicted in Figure 2 of the
`
`patent. Blocks 30-33 of this method, which are basic elements of wavelet coding,
`
`are shown in more detail in Figure 3. These components are described in the
`
`“Background of the Invention” at 2:35 through 4:20 of the ’482 patent.
`
`
`
`19
`
`Apple Inc. Exhibit 1002 Page 21
`
`

`
`Inter Partes Review of USP 5,850,482
`
`
`
`Docket No. 106840000558
`
`52. The image is first subjected to a wavelet transform that concentrates
`
`the image energy (or variance) into a small number of coefficients, most of which
`
`(with the exception of the DC component) are near zero or otherwise small. These
`
`are then gathered into a histogram [Fig. 2, block 31] from which a quantization
`
`interval will be calculated [block 32] and then analyzed for significance [block 33].
`
`This is depicted in Figure 5 of the ’482 patent. In general, most transform
`
`coefficients are near zero. The near-zero coefficients are deemed insignificant and
`
`set to zero, in essence a “dead zone” [’482 patent, 3:63-66].
`
`53. At this point, one can envision an array of transform coefficients,
`
`most of whi