`
`Paper No. ___
`Filed: April 11, 2018
`
`
`
`
`
`
`
`
`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`_____________________________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`_____________________________
`
`
`APPLE INC.,
`Petitioner,
`
`v.
`
`CALIFORNIA INSTITUTE OF TECHNOLOGY,
`Patent Owner.
`_____________________________
`
`Case IPR2017-00701
`Patent No. 7,421,032
`
`
`
`_____________________________
`
`
`
`PATENT OWNER’S SURREPLY
`
`
`
`TABLE OF CONTENTS
`
`Petitioner’s new argument that MacKay discloses nonuniform
`
`I.INTRODUCTION ................................................................................................ 1
`II.ARGUMENT ...................................................................................................... 1
`A.
`column weights for information bits should be rejected ...................... 1
`B.
`LDGM ................................................................................................ 2
`C.
`No explanation for how to modify Ping to be non-systematic ............. 2
`D. No motivation to combine Ping and MacKay...................................... 3
`E.
`Reasonable expectation of success was never addressed ..................... 5
`F.
`corresponding experimental data should be rejected ........................... 5
`G. New attorney-generated Tanner graphs and block diagram ................. 8
`III.CONCLUSION .................................................................................................. 8
`
`Petitioner failed to explain how Ping discloses Claim 6’s
`
`The newly proposed Hd submatrix modifications and
`
`-i-
`
`
`
`
`I.
`
`INTRODUCTION
`
`In view of new argument and evidence submitted in Petitioner’s Reply
`
`briefing, the Board (Paper 43) authorized a short sur-reply but prohibited
`
`submission of rebuttal evidence. As illustrated in further detail below, the Reply
`
`materials are replete with untimely and improper new argument and evidence—
`
`including submission of newly generated experimental data, attorney-generated
`
`Tanner graphs and block diagrams, and a declaration from a new witness. The
`
`Reply provides no justification for replacing Dr. Davis with a new witness. Dr.
`
`Davis was aware of his Fulbright commitment since at least February 2017 and he
`
`testified he remains available for deposition in the U.S. EX1173, ¶3. Accordingly,
`
`the Reply materials should be disregarded and given no weight.
`
`II. ARGUMENT
`
`A.
`
`Petitioner’s new argument that MacKay discloses nonuniform
`column weights for information bits should be rejected
`
`As the POR explained, the petition failed to provide any evidence that
`
`MacKay discloses non-uniform column weights for information bits. POR 18-21.
`
`Having realized the flaws in its petition, Petitioner now relies on MacKay’s
`
`Figures 5 and 6 to pivot to a new theory that MacKay discloses information bits
`
`appearing in a variable number of subsets. Reply 3-4. This is improper and should
`
`be rejected, not least because Caltech will not have an opportunity to rebut the
`
`argument with expert evidence. Dell Inc. v. Acceleron, LLC, 818 F.3d 1293, 1301
`
`-1-
`
`
`
`
`(Fed. Cir. 2016). Even then, Petitioner’s new argument does not explain why
`
`Figures 5 and 6 would motivate a POSA to modify Ping’s Hd submatrix (they
`
`would not). MacKay presents Figures 5 and 6 as a way to achieve “fast encoding”
`
`by applying a “lower triangular structure” already found in Ping. EX1102 1453;
`
`EX1103 38. Moreover, MacKay’s fast-encoding codes perform worse than the
`
`“ordinary-encoding codes” described earlier in the paper. EX1102, 1454, Fig. 7.
`
`B.
`
`Petitioner failed to explain how Ping discloses Claim 6’s LDGM
`
`As explained in the POR (22), the petition does not provide any explanation
`
`for how Ping’s Hd submatrix would meet the definition of a generator matrix.
`
`Having recognized that deficiency, the Reply (6-7) attempts to explain how Hd
`
`could be used as a generator matrix. This argument should be rejected as untimely.
`
`Moreover, it does not explain why a POSA would use Hd as a generator matrix to
`
`implement Ping’s equations.
`
`C. No explanation for how to modify Ping to be non-systematic
`
`POR (22-23) pointed out that the petition provided no rationale for
`
`modifying Ping’s code to be non-systematic in view of Divsalar. The Reply does
`
`not address this deficiency. Instead, the Reply (8) now claims making Ping’s code
`
`non-systematic would have been simple, citing to Dr. Frey’s declaration (EX1165)
`
`at ¶29. Yet Dr. Frey’s description is not a simple one, and he provides zero
`
`explanation for why a POSA would be motivated to make such changes.
`
`-2-
`
`
`
`
`Moreover, the Reply does not address the POR’s argument that making Ping non-
`
`systematic would destroy Ping’s code by removing the Hd submatrix from the
`
`parity-check matrix. POR 23.
`
`D. No motivation to combine Ping and MacKay
`
`There is no motivation to modify Ping at least because its parity-check
`
`matrix is already irregular and MacKay does not teach selective application of
`
`uneven column weights to a submatrix. POR 24-29. The Reply’s (8) response is
`
`that this argument should be rejected “for at least the reasons in the Petition and
`
`DI.” But while the petition does not address the fact that Ping’s parity-check
`
`matrix is already irregular (see POR 24-26), the Reply (9) admits that Ping’s
`
`parity-check matrix already has nonuniform column weights of, e.g., 4, 2, and 1.
`
`The Reply does not dispute that setting Ping’s “t” value to 9 shows a parity-
`
`check matrix that is more irregular than MacKay’s. Rather, the Reply (9) falsely
`
`asserts that this example is “contrived,” but Caltech’s example of Hd having
`
`column weights 9 was based on Petitioner’s proposal to use column weights of 3
`
`and 9 for Ping’s Hd. Pet. 40; see also EX2033 229:4-9 (“[A]ny positive integer is a
`
`possibility”). PO’s example simply adopts one of the weights proposed by
`
`Petitioner, while maintaining Hd’s uniform column weight, as instructed by Ping.
`
`The Reply (10) absurdly asserts that it is improper to compare Ping’s H
`
`matrix with MacKay’s parity-check matrices. As Ping’s H matrix is its parity-
`
`-3-
`
`
`
`
`check matrix, it is the only thing properly compared with MacKay’s parity-check
`
`matrices. Hd and Hp are indisputably not parity-check matrices. EX2033, 218:3-5.
`
`The Reply (10-11) incorrectly asserts that the only way to obtain MacKay’s
`
`benefits gained from nonuniform column weights is to modify Hd. The easiest way
`
`to obtain MacKay’s nonuniform column weights is to do nothing to Ping, because
`
`Ping’s parity-check matrix already has nonuniform column weights.
`
`The Reply (11-13) next relies on purported Tanner graph depictions of Ping
`
`and MacKay, which should be rejected as untimely, as discussed below in Section
`
`II.G. In addition, Petitioner’s description of the exhibits is misleading. The Reply
`
`(12) claims the “open circles on the left” are “message nodes,” and incorrectly
`
`claims that “Ping’s message nodes all have degree four.” But EX1148’s right
`
`nodes are message nodes (because they correspond to parity bits in the codeword),
`
`and have degrees less than four.
`
`Moreover, both graphs depict a misleadingly identical “Random
`
`Permutation,” but as Dr. Mitzenmacher noted, these permutations are constrained
`
`in very different ways. EX2038, 426:11-428:2. Ping distributes its edges evenly to
`
`“best increase the recurrence distance” (EX1103, 38), whereas MacKay’s 93y does
`
`the opposite of “distribut[ing] high weight columns per row [with] greater variance”
`
`-4-
`
`
`
`
`(EX1102, 1451), which involves more clumping of edges. POR 48-49.1
`
`The Reply (16-17) also admits that its modification breaks Ping’s constraints,
`
`but claims the combination does not “prevent[] the ones from still being distributed
`
`and randomly placed.” To the contrary, Ping’s teachings prevent this combination
`
`in requiring “exactly one element 1 per column” to “best increase the recurrence
`
`distance of each bit” and “reduce[] the correlation during the decoding process.”
`
`EX1103, 38. The Reply does not address how its combination would maintain
`
`these benefits. Indeed, the very codes Dr. Frey tested do precisely the opposite. See
`
`Section II.F.
`
`E. Reasonable expectation of success was never addressed
`
`The POR (16-17, 42-47) points out that the petition wholly lacks discussion
`
`of reasonable expectation of success or the (now conceded) unpredictability in the
`
`field. As such, any discussion of reasonable expectation of success presented in the
`
`Reply (e.g., 17-20) is improperly new and should be ignored. Office Patent Trial
`
`Practice Guide, 77 Fed. Reg. 48767 (Aug. 14, 2012).
`
`F.
`
`The newly proposed Hd submatrix modifications and
`corresponding experimental data should be rejected
`
`
`1 The Reply (13) next mischaracterizes Caltech’s argument (a problem common
`
`throughout Petitioner’s Reply briefs) as claiming “Ping’s encoding is not
`
`performed in two steps.” Caltech’s actual argument is presented at POR 32-34.
`
`-5-
`
`
`
`
`Petitioner relies on newly proposed modifications to Ping’s submatrix Hd
`
`and related experimental data generated by substitute expert Dr. Frey. Reply 17-20,
`
`citing EX1168. The data should be disregarded for a number of reasons, not least
`
`of which is that the modifications are found nowhere in the petition and the
`
`experimental data has not been shown to have any relevance to Petitioner’s case.
`
`First, Dr. Frey modifies Ping’s code in a way never proposed in the petition.
`
`In fact, an entire section of the POR (38-42) is dedicated to discussing a critical
`
`lack of specificity in the proposed modification. Whereas the petition only makes
`
`a general assertion to modify Ping’s Hd submatrix to have column weights of 9 and
`
`3, the experiment cited in Petitioner’s reply instead applies specific weight
`
`distributions that require graphical depictions, one of which does not even have
`
`column weights of 9 and 3. EX1165 ¶¶50, 54. Dr. Frey provides no explanation
`
`how he arrived at these distributions and they are not taught in MacKay or Ping.
`
`Notably, these weight distributions break Ping’s constraints of sub-blocks
`
`containing “one element 1 per column” and uniform row weights (the latter of
`
`which is also a constraint in MacKay), and do not at all maximize the recurrence
`
`distance. EX1103, 38. Petitioner now attempts to belatedly cure a fundamental
`
`defect in the petition, which should be rejected at this stage of the proceeding.
`
`In addition, the experimental data materials are not contemporaneous with
`
`the “relevant time” and the understanding of a POSA. It is irrelevant what Dr. Frey
`
`-6-
`
`
`
`
`claims he could do in the year 2018 when armed with Caltech’s disclosures, Dr.
`
`Jin’s original coding work, contemporary resources,2 and some 18 years of post-
`
`filing date knowledge. Further demonstrating use of improper hindsight, Dr. Frey
`
`admits that his decoder is “like the one described in the ’032 patent.” EX1165, ¶48.
`
`His testimony presents zero reflection of the environment in 1999-2000, and
`
`provides no information as to why a POSA would make the proposed modification
`
`18 years ago or reasonably expected success at that time.
`
`If surreply evidence was permitted, Caltech’s witnesses would have been
`
`able to provide rebuttal testimony explaining its unreliability—including numerous
`
`technical flaws, apparent cherry-picking of parameters, and selection of a weight
`
`distribution that in no way flows from any prior art reference at issue in this case.
`
`For example, while Dr. Frey claims that his nonuniform column weight simulation
`
`files “operate[] identically” to the version corresponding to the unaltered Ping code
`
`(EX1165 ¶¶51,55), examining the code files tells a different story. Dr. Frey altered
`
`a number of parameters beyond the Hd matrix, including the Gaussian noise levels
`
`applied to each simulation, the number of blocks per noise level, and even the
`
`number of decoding iterations (using twice as many iterations for the irregular
`
`codes). EX1168 1, 3, 6 (“Parameters” sections). These differences alone are
`
`2 E.g., Dr. Frey (¶42) used Matlab, a software program that received over 35
`
`version updates since May 2000. See https://en.wikipedia.org/wiki/MATLAB.
`
`-7-
`
`
`
`
`sufficient to preclude meaningful comparison between the purported simulations,
`
`yet are unexplained, casting doubt on the entire methodology. The new data is
`
`untimely, unreliable and fails to comply with 37 C.F.R. §42.65.
`
`G. New attorney-generated Tanner graphs and block diagram
`
`Petitioner relies on new exhibits 1171 and 1172, purported to be block
`
`diagrams of implementations of Ping’s encoder, and exhibits 1148 and 1149,
`
`purported to be Tanner graph representations of Ping and MacKay codes,
`
`respectively. Reply 12, 14-15. Yet these figures are not in the petition. Reliance on
`
`them now should be rejected as untimely. Moreover, Petitioner’s argument that
`
`these exhibits show a motivation to combine (Reply 11-15) is logically flawed.
`
`Simply because Petitioner’s lawyers generated similarly-styled graphs in preparing
`
`the Reply no way demonstrates that the a POSA would have motivation to
`
`combine the codes 18 years ago, or expected success in improving error-correction.
`
`Furthermore, the new exhibits are erroneous and tainted with impermissible
`
`hindsight. Prior to Caltech’s IRA patents, Tanner graphs were conventionally
`
`presented as bipartite graphs depicting the relationship between a codeword and
`
`parity-check equations. See, e.g., EX2004 ¶40; EX1104 ¶¶54-55; EX1109 258.
`
`III. CONCLUSION
`
`Accordingly, the Reply materials should be given no weight.
`
`-8-
`
`
`
`
`
`
`Date: April 11, 2018
`
`Respectfully submitted,
`
`
`
`/ Michael T. Rosato /
`Michael T. Rosato, Lead Counsel
`Reg. No. 52,182
`
`
`
`-9-
`
`
`
`
`CERTIFICATE OF SERVICE
`
`I certify that the foregoing Patent Owner’s Surreply was served on this 11th
`
`day of April, 2018, on the Petitioner at the electronic service addresses of the
`
`Petitioner as follows:
`
`
`Richard Goldenberg
`Dominic Massa
`Michael H. Smith
`James M. Dowd
`Mark D. Selwyn
`Arthur Shum
`WILMER CUTLER PICKERING HALE AND DORR LLP
`richard.goldenberg@wilmerhale.com
`dominic.massa@wilmerhale.com
`michaelh.smith@wilmerhale.com
`james.dowd@wilmerhale.com
`mark.selwyn@wilmerhale.com
`arthur.shum@wilmerhale.com
`
`
`
`
`
`
`Date: April 11, 2018
`
`
`
`
`
`
`
`
`
`
`Respectfully submitted,
`
`
`/ Michael T. Rosato /
`Michael T. Rosato, Lead Counsel
`Reg. No. 52,182
`
`
`
`-10-
`
`
`