`
`Paper No. ___
`Filed: November 21, 2017
`
`
`
`
`
`
`
`
`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`_____________________________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`_____________________________
`
`
`APPLE INC.,
`Petitioner,
`
`v.
`
`CALIFORNIA INSTITUTE OF TECHNOLOGY,
`Patent Owner.
`_____________________________
`
`Case IPR2017-00700
`Patent No. 7,421,032
`
`
`
`_____________________________
`
`
`
`PATENT OWNER’S RESPONSE
`PURSUANT TO 37 C.F.R. § 42.120
`
`
`
`TABLE OF CONTENTS
`
`Dr. Davis’s evasiveness during his deposition undermines his
`
`Petitioner fails to identify parity bits that are determined “in
`
`STATEMENT OF PRECISE RELIEF REQUESTED .................................. 1
`I.
`INTRODUCTION AND OVERVIEW OF ARGUMENT ............................ 1
`II.
`III. OVERVIEW OF THE ART AND CITED REFERENCES .......................... 4
`A. MacKay (EX1002) .............................................................................. 6
`B.
`Ping (EX1003) .................................................................................... 7
`C.
`Divsalar (EX1017) .............................................................................. 9
`IV. WEIGHT TO BE GIVEN RESPECTIVE EXPERT TESTIMONY ............ 10
`A. Dr. Davis’s testimony includes basic errors demonstrating a
`lack of credibility .............................................................................. 10
`B.
`Dr. Davis’s testimony is not independent .......................................... 12
`C.
`credibility.......................................................................................... 12
`V.
`CLAIM CONSTRUCTION ........................................................................ 14
`“irregular” ......................................................................................... 14
`A.
`B.
`“Tanner Graph” ................................................................................ 15
`VI. GROUND 1: PING, MACKAY AND DIVSALAR DO NOT
`RENDER CLAIMS 11, 12, AND 14-16 OBVIOUS ................................... 17
`A.
`Legal Principles ................................................................................ 18
`B.
`accordance with the [claimed] Tanner graph” ................................... 19
`C.
`irregular repetition ............................................................................ 21
`D. MacKay does not teach nonuniform row weights.............................. 25
`E.
`MacKay ............................................................................................ 26
`1. Ping is already irregular as defined by MacKay ............................. 26
`2. The proposed modification would eliminate Ping’s stated
`improvement ................................................................................. 31
`3. Petitioner’s additional arguments regarding motivation to
`combine fail .................................................................................. 34
`
`Petitioner fails to establish that either Ping or MacKay discloses
`
`A POSA would not be motivated to modify Ping in view of
`
`-i-
`
`
`
`
`Petitioner fails to Provide a Rationale to Further Modify Ping
`
`4. Dr. Davis’s claim that MacKay’s irregularity is ill-defined
`indicates a lack of motivation to combine ..................................... 38
`F.
`Petitioner inadequately defines its proposed modification ................. 40
`G. Modifying Ping in view of MacKay would not be expected to
`succeed. ............................................................................................ 44
`H.
`and MacKay in view of Divsalar ....................................................... 49
`VII. GROUND 2: THE COMBINATION OF PING, MACKAY,
`OBVIOUS .................................................................................................. 53
`VIII. OBJECTIVE INDICIA OF NONOBVIOUSNESS ..................................... 54
`A. Nexus between the Objective Evidence and the Claims .................... 55
`B.
`Long-felt need and failure of others .................................................. 58
`C.
`Industry Praise .................................................................................. 61
`D. Unexpected Results........................................................................... 63
`E.
`Commercial Success ......................................................................... 64
`IX. CONCLUSION .......................................................................................... 66
`X. APPENDIX ................................................................................................ 68
`
`DIVSALAR, AND LUBY97 DOES NOT RENDER CLAIM 13
`
`-ii-
`
`
`
`
`I.
`
`STATEMENT OF PRECISE RELIEF REQUESTED
`
`Apple, Inc. (“Petitioner”) filed a petition for inter partes review of various
`
`claims of U.S. Patent No. 7,421,032 (the “’032 patent”, EX1001). The patent
`
`owner (“Caltech”) hereby requests that the Board now issue a final written
`
`decision confirming that claims 11-16 are not unpatentable.
`
`II.
`
`INTRODUCTION AND OVERVIEW OF ARGUMENT
`
`The ’032 patent is one of four Caltech patents that resulted from research
`
`performed by the inventors, Dr. Jin, Dr. Khandekar, and Dr. McEliece, in 1999-
`
`2000. The patents claim inventions directed to a revolutionary class of error-
`
`correction codes, dubbed “irregular repeat and accumulate codes,” or “IRA codes,”
`
`which surpassed the performance of the best known codes at that time. One of the
`
`features that made IRA codes superior to other known codes, however, was their
`
`capability of being encoded and decoded with linear complexity, a critical
`
`requirement for most practical applications. No other code known at the time
`
`could boast linear encoding, linear decoding, and performance near the theoretical
`
`Shannon limit.
`
`In arguing that the instituted claims are unpatentable, Petitioner relies chiefly
`
`on three prior art references: the MacKay reference, which discloses randomly
`
`generated parity-check matrices (which are “irregular” in the sense that 11 of 12
`
`columns are weight 3 and 1 of 12 columns are weight 9), the Ping reference, which
`
`-1-
`
`
`
`
`describes a method of improving random parity-check matrices of the type
`
`described by MacKay by imposing certain structural constraints to the matrix, and
`
`the Divsalar reference, which describes an altogether different kind of code: a
`
`simple “turbo-like” code created for the purpose of proving a mathematical
`
`conjecture.
`
`Petitioner’s obviousness challenges are lacking in many respects. In
`
`ascribing motivation to combine the asserted references, Petitioner attempts to take
`
`MacKay’s teachings about nonuniform column weights in a full parity-check
`
`matrix and apply it to only a part of Ping’s parity-check matrix. Yet nothing in
`
`MacKay teaches the propriety of applying a general aspect of a full matrix to
`
`merely a part of a matrix in a different code. Indeed, Ping’s parity-check matrix as
`
`a whole is already “irregular” (in fact, more “irregular”) according to MacKay’s
`
`teachings, and neither reference provides any motivation to add more irregularity
`
`to part of the matrix, as Petition proposes. To the contrary, Petitioner’s proposed
`
`combination ignores and destroys fundamental constraints of Ping’s codes imposed
`
`explicitly for performance reasons. Ping’s code is presented as an improvement
`
`over random parity-check matrices like those in MacKay, and modifying it in light
`
`of MacKay would have been viewed as a step backwards. There would simply be
`
`no motivation to modify Ping in light of the fact it already achieves what MacKay
`
`-2-
`
`
`
`
`teaches, and the proposed modification would eliminate the very improvements
`
`Ping proposes.
`
`The petition is further flawed in its proposal to modify Ping’s Hd submatrix
`
`by “setting some columns to weight 9 and others to weight 3.” Pet. 42. Aside
`
`from the fact that MacKay does not teach such a modification, Petitioner fails to
`
`specifically describe how such a modification would be accomplished for
`
`generating a workable code. Among other things, Petitioner provides no guidance
`
`as to which columns should be modified, how many should be modified, and how
`
`such a modification would maintain the constraints taught by Ping and MacKay.
`
`Petitioner’s failure to provide any meaningful detail regarding its proposed
`
`modification underscores the lack of any plausible motivation to combine with a
`
`reasonable expectation of success. A POSA would have known that error-
`
`correcting codes were an unpredictable field of endeavor and that merely
`
`combining elements from different codes could not be expected to succeed, yet
`
`nowhere in the petition is expectation of success addressed. Petitioner ignores this
`
`critical requirement of an obviousness inquiry under Graham v. John Deere, and
`
`the challenge can be rejected for this reason alone.
`
`Furthermore, objective indicia of nonobviousness weigh heavily in favor of
`
`nonobviousness. The IRA encoding and decoding methods and systems described
`
`and claimed in the ’032 patent were a groundbreaking development in the field of
`
`-3-
`
`
`
`
`coding theory. The invention overcame longstanding issues with previously
`
`known error correcting codes in a way that was unexpected, has been widely
`
`praised since its introduction, and has experienced commercial success by others
`
`through adoption in numerous information transmission standards.
`
`For these reasons, all of the remaining grounds of challenge must be denied.
`
`III. OVERVIEW OF THE ART AND CITED REFERENCES
`
`Modifying or constructing error-correction codes was a highly unpredictable
`
`endeavor. Since it was effectively impossible to mathematically prove the
`
`performance of a code, researchers were forced to engage in extensive trial-and-
`
`error and experimentation to determine whether new codes led to an improvement.
`
`Even when improvements occurred, the reasons for improved performance were
`
`typically not well-understood. EX2004, ¶46. As Petitioner’s expert conceded
`
`during cross-examination:
`
`What you would really like to be able to do is a formal
`
`mathematical analysis of the strength of the codes that
`
`you are working with, but that’s often really hard. So
`
`often what the engineers in particular would do is … take
`
`a variety of different [codes], run simulations and … then
`
`I will get a general sense of what the [mathematical]
`
`analysis would have shown me. … [I]t might even be
`
`impossible to do the mathematical analysis.
`
`-4-
`
`
`
`
`EX2033 at 256:21-257:12 (emphasis added). Caltech’s expert, Dr. Mitzenmacher,
`
`likewise explains that discoveries had to be made via guesswork and
`
`experimentation. EX2004, ¶46. As a result, it was rarely the case that a researcher
`
`could reasonably predict that a particular modification would result in an
`
`improvement in the performance of a code. Id.
`
`Experiments showed that the performance of a code was highly dependent
`
`on the specific properties and constraints of the code. These codes did not have
`
`interchangeable parts, where a property of a performant code could simply be
`
`incorporated into other codes to improve them. Instead, such chimeras could end
`
`up being nonfunctioning. Id.
`
`One example of this unpredictability is illustrated in the discovery of the
`
`classes of codes found in the cited prior art references: turbo codes and low-density
`
`parity-check codes. Turbo codes were discovered by Claude Berrou in 1993, and
`
`his discovery was met with skepticism because he could not explain why his code
`
`performed. Many believed he had made an error in his initial experiments. Yet
`
`after his results were independently confirmed, research and use of his turbo codes
`
`became widespread. (EX2004, ¶¶47-51)
`
`LDPC codes, which are often viewed as a competitor to turbo codes, also
`
`had a modest beginning. LDPC codes were first discovered by Dr. Robert
`
`Gallager in 1963, but largely remained ignored for over 35 years. That such codes
`
`-5-
`
`
`
`
`performed well was surprising to those skilled in the art. Dr. Gallager himself
`
`explains the unpredictability of this field with regard to his code, “I had a little bit
`
`of an inkling [they could be good], but I also had a suspicion that they well might
`
`not be. And I spent a long time trying to resolve whether they were or not.”
`
`(EX2004, ¶¶52-56)
`
`A. MacKay (EX1002)
`
`MacKay highlights the unpredictability in the field at the time and the
`
`corresponding need for experimentation to
`
`identify functioning codes. MacKay
`
`investigated performance of irregular Gallager
`
`codes, a class of codes defined by randomly
`
`generating low-density parity-check matrices that were “irregular,” which
`
`according to MacKay meant that the columns of the parity-check matrices had
`
`nonuniform weights. EX1002, p. 1449. Specifically, MacKay’s irregular codes
`
`had 1/12 of the columns had a weight of nine, and the remaining 11/12 columns
`
`had weight of 3. EX1002, Table 1; see also Fig. 2 (excerpt shown). MacKay
`
`randomly generated several parity-check matrices using patterns with the above
`
`constraints and according to a generalized Poisson distribution. See e.g., EX1002,
`
`p. 1451 (“The edges are placed ‘completely at random’.”). MacKay then ran
`
`simulations to test performance of the constructions. Such experimentation was
`
`-6-
`
`
`
`
`necessary because mathematical analysis would not have been able to predict
`
`whether the constructions improved performance. (EX2004, ¶57)
`
`MacKay’s codes were divided into sub-classes (i.e., Poisson, sub-Poisson,
`
`super-Poisson patterns) and MacKay, based on its testing, reported that “super-
`
`Poisson” patterns performed better than the other patterns. MacKay noted that a
`
`number of the randomly generated codes exhibited high error floors and had to be
`
`discarded. EX1002, p. 1452 (“We discard the two codes with error floors…”).
`
`MacKay explains that such error floors were the result of “cycles of length 4,”
`
`whose avoidance “is not so easy to enforce in irregular Gallager codes with high
`
`weight columns.” Id., p. 1454; see also, id. p. 1449 (acknowledging that
`
`performance is “sensitive to the distribution of column weights.”). MacKay did
`
`not provide any of the actual codes used in its evaluation of Poisson patterns.
`
`(EX2004, ¶58)
`
`B.
`
`Ping (EX1003)
`
`Ping describes a method of improving Gallager codes with random parity-
`
`check matrices, such as those described by MacKay, by introducing specific non-
`
`random structural constraints to the parity-check matrix, because encoding codes
`
`with random matrices was “costly in terms of both memory and the operations
`
`involved.” EX1003, p. 38. (EX2004, ¶59)
`
`-7-
`
`
`
`
`Ping instructs that, first, the columns of the parity-check matrix H be
`
`constructed as two submatrices: submatrix Hp and submatrix Hd, each of which
`
`have a specifically defined structure. Id. With regard to Hp, Ping teaches it is a
`
`deterministic square matrix populated as follows:
`
`
`
`With regard to Hd, Ping instructs that it be subdivided into “t equal sub-blocks,”
`
`where t is “a preset integer constrained by (i) t divides n-k and (ii) n-k divides kt”
`
`(n is the length of the codeword and k is the number of information bits). Id. Hd
`
`appears as follows:
`
`
`
`For each of these sub-blocks, there is exactly “one element 1 per column and kt/(n-
`
`k) 1s per row.” Id. This means that the submatrix Hd has both uniform column
`
`and row weight, and the 1s are evenly distributed within the submatrix. Ping
`
`explains that these constraints for Hd are necessary because they “best increase the
`
`recurrence distance of each bit in the encoding chain … and, intuitively, reduce[]
`
`the correlation during the decoding process.” Id. In other words, Ping specifically
`
`identifies its uniform distribution of Hd as an improvement over random parity-
`
`-8-
`
`
`
`
`check matrices like those found in MacKay. See also, id., p. 39 (“Conclusion: It
`
`has been shown that a semi-random approach to LDPC code design can achieve
`
`essentially the same performance as the existing method with considerably reduced
`
`complexity.”) (EX2004, ¶¶60-63)
`
`C. Divsalar (EX1017)
`
`The Divsalar reference describes the work of Dr. Dariush Divsalar, along
`
`with two of the inventors of the ’032 patent (Dr. McEliece and Dr. Jin), in
`
`developing a repeat accumulate (RA) code. EX2031 ¶ 16-32.
`
`RA codes as taught in Divsalar are nonsystematic codes, meaning that only
`
`the encoded codeword bits (parity bits) are transmitted. RA codes always perform
`
`regular repetition of information bits and every repeated bit in an RA code is
`
`separately accumulated to generate a new parity bit.1 At a rate of 1/q (where q is
`
`the amount of repetition), RA codes are impractically slow. Indeed, Dr. Divslar
`
`explains that the codes were never intended to be competitive error-correction
`
`codes, nor would they have been mistaken as such—they were designed as a
`
`research tool for learning about certain characteristics of turbo codes and similar
`
`codes. EX2031 ¶ 27, 28, 32; EX2004 ¶ 58.
`
`
`1 In contrast, subsets of information bits in IRA codes are summed, and the
`
`sums are then accumulated. As a result of these differences, IRA codes exhibit
`
`significantly better performance than RA codes.
`
`-9-
`
`
`
`
`Additionally, Divsalar did not analyze RA codes using parity-check matrices
`
`or Tanner graphs, because at the time of the invention of the ’032 patent, such an
`
`analysis would not have been common: turbo codes and LDPC codes were viewed
`
`as two distinct types of codes using different approaches to code design and
`
`analysis. EX2031 ¶26; EX2004 ¶¶ 54-56.
`
`IV. WEIGHT TO BE GIVEN RESPECTIVE EXPERT TESTIMONY
`
`A. Dr. Davis’s testimony includes basic errors demonstrating a lack
`of credibility
`
`Basic technical errors are an important clue to witness credibility. See, e.g.,
`
`Merck & Co. v. Teva Pharm. USA, Inc., 347 F.3d 1367, 1371 (Fed. Cir. 2003)
`
`(noting chemist made errors that those in the art would have considered basic).
`
`Here, Dr. Davis could not answer basic questions about Berrou, the seminal paper
`
`on turbo codes, without rereading the entire article. EX2033, 54:17-60:3. He
`
`could not give an opinion on what “irregular” meant in the field, and implied such
`
`information was unhelpful to the Board. Id., 87:7-89:16.
`
`With regard to Ping and MacKay, Dr. Davis inaccurately testified that
`
`MacKay’s “super-Poisson” pattern was consistent with Ping’s teaching of
`
`“increas[ing] the recurrence distance.” Ping teaches that its Hd submatrix is
`
`designed to “best increase the recurrence distance of each bit” (EX1003, p. 38),
`
`which means “spread[ing] your 1s around.” EX2033, 233:7. Yet Dr. Davis
`
`incorrectly testified that “MacKay teaches in this same direction, when he is
`
`-10-
`
`
`
`
`talking about sub-Poisson, super-Poisson, Poisson ways of constructing parity-
`
`check matrices.” Id., 233:14-17. MacKay actually teaches the opposite, that its
`
`best-performing “super-Poisson constructions” have a “distribution of high weight
`
`columns per row [with] greater variance” (EX1002, p. 1451)—in other words, the
`
`1s in MacKay’s super-Poisson construction are more clustered instead of spread
`
`out. See infra, Section V.E; EX2004, ¶128.
`
`His unfamiliarity with the actual teachings of cited references,2 as well as
`
`with the actual knowledge in the relevant art, compels him to use hindsight to
`
`define irregularity in terms of Caltech’s claims. The testimony of Dr. Davis should
`
`be discounted accordingly.3
`
`
`2 As explained in Caltech’s POR in IPR2017-00210, Dr. Davis’s testimony
`
`includes basic errors with many of the references he assessed.
`
`3 Dr. Davis’s unfamiliarity with the key references should also be considered in
`
`view of his admission that none of his publications related to repeat-accumulate
`
`codes, low-density parity-check codes, turbo codes, or irregular codes in general.
`
`EX2033, 27:4-28:9. Unsurprisingly, he never attended the Allerton Conference on
`
`Communication, Control and Computing because “the work that I do in coding
`
`theory wasn’t being presented at that conference.” Id., 32:14-22.
`
`-11-
`
`
`
`
`B. Dr. Davis’s testimony is not independent
`
`While the petition and expert declaration are expected to be consistent,
`
`expert testimony that simply tracks and repeats the petition is entitled to little
`
`weight. Wowza Media Sys., LLC v. Adobe Sys., Inc., IPR2013-00054, Paper 16, 4
`
`(2013). The petition and the Davis declaration are essentially identical in language.
`
`For example, the sections discussing Ground 1 are nearly identical. Compare Pet.,
`
`39-64, with EX1004, ¶¶111-171. This significantly undercuts the independence
`
`and objectivity of Dr. Davis’s testimony.
`
`C. Dr. Davis’s evasiveness during his deposition undermines his
`credibility
`
`In 10X Genomics, Inc. v. Univ. of Chicago, the Board explained that expert
`
`evasiveness or unresponsiveness during cross examination would reduce the
`
`weight of the expert’s direct testimony. IPR2015-01157, Paper 30, 2 (2016). Dr.
`
`Davis evaded straightforward questions about the art that an artisan of the time
`
`should have been able to give straightforward answers. For example, he evaded
`
`questions on whether Berrou’s Figure 5 showed a relationship between bit error
`
`rate and signal-to-noise ratio despite the axes being clearly labeled as such. Id.,
`
`56:19-57:6, 58:19-59:3.
`
`Regarding “irregular,” a key term in this trial, he evaded answering whether
`
`his definition of irregular was the conventional meaning in the field of error-
`
`correction codes. Id., 66:10-68:4. He evaded answering where the prior art
`
`-12-
`
`
`
`
`provided a definition of “irregular” that was the same as his definition. Id., 72:17-
`
`75:18. He avoided answering what definition of “irregular” he would use in the
`
`field of error-correction codes generally. Id., 78:18-81:12. He avoided answering
`
`whether his definition of “irregular” was consistent with the definition used with
`
`Tanner graphs. Id., 83:21-87:6. His unresponsiveness during cross-examination on
`
`this pivotal term warrants that his direct testimony be given little or no weight.
`
`He also avoided answering where MacKay expressly discloses irregular
`
`repetition of information bits (it does not). Id., 249:2-251:21. And he was evasive
`
`on the self-evident question of whether Ping depicts a Tanner graph (it does not).
`
`Id., 269:21-272:12.
`
`The contrast between cross-examination and redirect is also striking.
`
`Redirect occurred after a break during which Dr. Davis had a “discussion about the
`
`substance of the testimony and the general nature of the redirect” with Apple’s
`
`counsel. Id., 275:9-13. This discussion enabled Dr. Davis to be far more responsive
`
`and direct for Apple’s counsel. This witness behavior is precisely the sort of
`
`behavior the Board has condemned in decisions like 10X Genomics. The
`
`appropriate response is to accord little or no weight to the direct and redirect
`
`testimony of Dr. Davis.
`
`-13-
`
`
`
`
`V. CLAIM CONSTRUCTION
`
`A claim subject to inter partes review receives the broadest reasonable
`
`construction or interpretation in light of the specification of the patent in which it
`
`appears (“BRI”). See 37 C.F.R. § 42.100(b); Cuozzo Speed Techs., LLC v. Lee, 136
`
`S. Ct. 2131, 2142-45 (2016). However, the Board may not construe a term “so
`
`broadly that its constructions are unreasonable under general claim construction
`
`principles.” Microsoft Corp. v. Proxyconn, Inc., 789 F.3d 1292, 1298 (Fed. Cir.
`
`2015).
`
`A.
`
`“irregular”
`
`Claim 13 requires a coder “configured to perform an irregular repeat on
`
`message bits.” Petitioner proposes that the term “irregular” be construed as “the
`
`notion that different message bits or groups of message bits contribute to different
`
`numbers of parity bits.” Pet. 24. Caltech does not believe the term needs to be
`
`construed, as the plain and ordinary meaning of irregular repetition is clear. That
`
`message bits contribute in differing numbers to parity bits is made clear in the
`
`claim language. (EX2004, ¶69)
`
`In addition, one should be aware that the prior art’s use of the term
`
`“irregular” differs from Petitioner’s proposed construction. The prior art defines
`
`an “irregular code” as having “parity-check matrices [that] have nonuniform
`
`weight per column.” EX1002, p. 1449; see also EX1009, p. 249 (“[R]andom
`
`-14-
`
`
`
`
`irregular bipartite graphs, which we call irregular codes.”). The prior art is
`
`referring to a relationship between codeword bits (which might not even include
`
`message bits) and parity-check equations, whereas the proposed construction refers
`
`to a relationship between message bits and parity bits. As discussed below, regular
`
`repetition can lead to nonuniform parity-check matrices, and irregular repetition
`
`can lead to (nearly) uniform parity-check matrices. An “irregular code” as
`
`understood in the prior art is distinct from the “irregular” repeat Petitioner seeks to
`
`construe. (EX2004, ¶70)
`
`B.
`
`“Tanner Graph”
`
`Claim 11 recites “parity bits in accordance with the following Tanner
`
`graph”:
`
`-15-
`
`
`
`
`
`
`Petitioner proposes that this graph be construed as a “graph representing an
`
`[irregular repeat accumulate (“IRA”)] code as a set of parity-checks where every
`
`message bit is repeated, at least two different subsets of message bits are repeated a
`
`different number of times, and check nodes, randomly connected to the repeated
`
`message bits, enforce constraints that determine parity bits” and that “a parity bit is
`
`determined as a function of both information bits and other parity bits as shown by
`
`the configuration of nodes and edges of the Tanner graph.” Pet. 25-26.
`
`Caltech does not believe the Tanner graph needs to be construed in the
`
`manner proposed, because a POSA would have readily understood how to encode
`
`bits according to the Tanner graph in the claims and in view of the specification.
`
`-16-
`
`
`
`
`Moreover, Petitioner’s construction is unhelpful in that it introduces unnecessary
`
`verbosity and then merely refers back to the very same Tanner graph. This circular
`
`logic underscores why it is unnecessary to construe the Tanner graph as proposed.
`
`(EX2004, ¶73)
`
`VI. GROUND 1: PING, MACKAY AND DIVSALAR DO NOT RENDER
`CLAIMS 11, 12, AND 14-16 OBVIOUS
`
`Petitioner fails to demonstrate that claims 11, 12, and 14-16 would have
`
`been obvious in view of the combination of Ping, MacKay, and Divsalar for at
`
`least the following reasons. First, Petitioner fails to show how the combination
`
`generates parity bits in accordance with the claimed Tanner graph. Second,
`
`Petitioner fails to demonstrate that Ping and MacKay teach irregular repetition.
`
`Third, for claim 12, MacKay does not teach nonuniform row weights. Fourth,
`
`Petitioner fails to establish that a POSA would have been motivated by MacKay to
`
`incorporate nonuniform weights into Ping’s submatrix because MacKay’s
`
`teachings are only applicable to full parity-check matrices and Ping’s full parity-
`
`check matrix is already as nonuniform as MacKay’s irregular codes. And Ping
`
`expressly teaches modifying the random codes of the type disclosed in MacKay so
`
`as to impart the specific structure the proposed modification seeks to destroy. Fifth,
`
`Petitioner’s proposed modification is not taught anywhere in MacKay. Sixth,
`
`Petitioner’s proposed modification lacks specificity. Seventh, Petitioner fails to
`
`address, let alone establish, that its proposed combination of Ping and MacKay
`
`-17-
`
`
`
`
`would have any reasonable expectation of succeeding, a requirement when proving
`
`obviousness. Eighth, a POSA would not have been motivated to further modify
`
`Ping in view of Divsalar, and Petitioner fails to explain how that modification
`
`would be made and whether there would be a reasonable expectation of success.
`
`For these reasons, Ground 1 should be rejected. (EX2004, ¶¶74-147)
`
`A. Legal Principles
`
`In order to establish that a patent claim is obvious under 35 U.S.C. § 103,
`
`one must first determine (1) the scope of the prior art, (2) differences between the
`
`prior art and the claims at issue, and (3) the level of ordinary skill in the art—
`
`“Against this background, the obviousness or nonobviousness of the subject matter
`
`is determined,” with additional “secondary considerations” given to certain indicia
`
`of nonobviousness. KSR Intern. Co. v. Teleflex Inc., 550 U.S. 398, 404 (2007)
`
`(citing Graham v. John Deere Co., 383 U.S. 1, 17-18 (1950)). Those challenging a
`
`claim must provide some articulated reasoning that includes identifying “a reason
`
`that would have prompted a person of ordinary skill in the relevant field to
`
`combine the elements in the way the claimed new invention does.” Id.
`
`Importantly, it is also a petitioner’s burden to show that at the time of the
`
`invention there was a “reasonable expectation of success” for the proposed
`
`combination. Intelligent Bio-Sys. v. Illumina Cambridge, 821 F.3d 1359, 1367-68
`
`(Fed. Cir. 2016) (challenger’s “burden to demonstrate … that the skilled artisan
`
`-18-
`
`
`
`
`would have had a reasonable expectation of success in [combining the
`
`references].”); see also DePuy Spine, Inc. v. Medtronic Sofamor Danek, Inc., 567
`
`F.3d 1314, 1326 (Fed.Cir.2009); MPEP § 2143.2.I (“Obviousness requires a
`
`reasonable expectation of success”). Thus, merely identifying elements in the prior
`
`art is not sufficient to establish obviousness—a POSA must have reasonably
`
`expected that the combination would have succeeded for its intended purpose.
`
`Based on these principles, the Board must deny obviousness challenges
`
`when a petitioner, as is the case here, fails to explain or provide evidence as to how
`
`the proposed combination would predictably result in the improvement that
`
`allegedly motivated the combination. JTEKT Corp. v. GKN Automotive, Ltd.,
`
`IPR2016-00046, Paper No. 27 at 28-29.
`
`B.
`
`Petitioner fails to identify parity bits that are determined “in
`accordance with the [claimed] Tanner graph”
`
`Claim 11 requires generating parity bits “in accordance with the following
`
`Tanner graph.” According to Petitioner’s construction of the Tanner graph, parity
`
`bits must be determined “as shown by the configuration of nodes and edges of the
`
`Tanner graph.” Petitioner fails to prove its combination discloses this limitation.
`
`(EX2004, ¶¶75-77)
`
`None of Ping, MacKay or Divsalar disclose a Tanner graph that is anything
`
`like the one recited in the claim. Ping and Divsalar do not contain any Tanner
`
`graphs. EX2004, ¶76; EX2033, 272:10-12; EX2031, ¶26. MacKay discloses part
`
`-19-
`
`
`
`
`of a bipartite graph only to illustrate the “cycles of four” that causes error floors in
`
`a code. EX1002, p. 1452 (topology examples (1) and (2)); EX2033, 273:19-274:3.
`
`However, MacKay discarded those codes in its analysis. Id. (“We discard the two
`
`codes with error floors.”). Thus, none of the references provide a Tanner graph
`
`that teach a POSA a way of generating parity bits. (EX2004, ¶76)
`
`In addition, Petitioner fails to show that the proposed combination generates
`
`parity bits in accordance with the Tanner graph. Petitioner only concludes that
`
`“Ping teaches that ‘a parity bit [such as the third parity bit, p3] is determined as a
`
`function of both information bits and other parity bits.’” Pet. 50. Yet Petitioner is
`
`completely silent on whether and how the process for generating parity bits
`
`matches up with the other encoding requirements represented by the nodes and
`
`edges of the Tanner graph. (EX2004, ¶77)
`
`-20-
`
`
`
`
`
`
`C.
`
`Petitioner fails to establish that either Ping or MacKay discloses
`irregular repetition
`
`Claim 11 as construed by Petitioner requires that it have “irregular repeat”
`
`and “at least two different subsets of message bits are repeated a different number
`
`of times.” Petitioner does not establish that either reference discloses irregular
`
`repetition. (EX2004, ¶¶78-87)
`
`Petitioner acknowledges that Ping fails to disclose irregular repetition. See,
`
`e.g., Pet. 51 (“Ping’s outer LDPC coder is regular”). (EX2004, ¶79)
`
`Similarly, Petitioner fails to show where MacKay discloses irregular
`
`repetition. In Petitioner’s analysis of the claim language, they claim that MacKay
`
`provides a “specific example where some information bits contribute to nine parity
`
`-21-
`
`
`
`
`bits and others contribute to three parity bits.” Pet. at 51 (citing to EX1002