Godo Kaisha IP Bridge 1

Patent Trials and Appeals Board

Oct 29, 2020

IPR2020-00867 (P.T.A.B. Oct. 29, 2020)

Trials@uspto.gov Paper 11
571-272-7822 Date: October 29, 2020

UNITED STATES PATENT AND TRADEMARK OFFICE
____________________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________________

UNIFIED PATENTS, LLC,
Petitioner,
v.
GODO KAISHA IP BRIDGE 1,
Patent Owner.
____________________

IPR2020-00867
Patent 7,305,035 B2
____________

Before JUSTIN T. ARBES, DAVID C. MCKONE,
and MELISSA A. HAAPALA, Administrative Patent Judges.

MCKONE, Administrative Patent Judge.

DECISION
Denying Institution of Inter Partes Review
35 U.S.C. § 314

IPR2020-00867
Patent 7,305,035 B2
2
I. INTRODUCTION
A. Background and Summary
Unified Patents, LLC (“Petitioner”) filed a Petition (Paper 2, “Pet.”)
requesting inter partes review of claim 1 of U.S. Patent No. 7,305,035 B2
(Ex. 1001, “the ’035 patent”). Pet. 2. Godo Kaisha IP Bridge 1 (“Patent
Owner”) filed a Preliminary Response (Paper 8, “Prelim. Resp.”).
We have authority to determine whether to institute an inter partes
review. See 35 U.S.C. § 314 (2016); 37 C.F.R. § 42.4(a) (2019). The
standard for instituting an inter partes review is set forth in 35 U.S.C.
§ 314(a), which provides that an inter partes review may not be instituted
“unless . . . there is a reasonable likelihood that the petitioner would prevail
with respect to at least 1 of the claims challenged in the petition.” For the
reasons explained below, we decline to institute an inter partes review of the
’035 patent.

B. Related Matters
The parties indicate that the ’035 patent is not involved in any
litigation or other judicial proceeding. Pet. 1; Paper 4, 1. Pending U.S.
Patent Application No. 16/043,862 claims priority to the ’035 Patent. Pet. 1.

C. The ’035 Patent
The ’035 patent describes a variable length coding technique for
coding and decoding coefficients in blocks of moving picture data.
Ex. 1001, 1:7–12. In coding a moving picture, compression of information
volume is usually performed by using redundancies in the information.
Id. at 1:16–22. Conventionally (e.g., in the MPEG-4 moving picture coding
standard), such data undergo a transformation into the frequency domain,
IPR2020-00867
Patent 7,305,035 B2
3
and blocks of the frequency information (e.g., 4X4 blocks of pixels) are
quantized to generate coefficients. Id. at 1:19–28. The quantized blocks of
coefficients are scanned in the order of low frequency components to high
frequency components to generate pairs of values about each coefficient,
each pair indicating (1) the number of zero coefficients subsequent to the
current coefficient (a Run, or “R” value); and (2) the value (Level or “L”
value) of the current coefficient. Id. at 1:29–35. Each (R, L) pair is
transformed into a code number using a predetermined code table, and the
code number, in turn, is encoded into a Variable Length Code (“VLC”)
using a VLC table. Id. at 1:35–47.
The ’035 patent describes the following problem with this method:
However, using the existing method engenders a decrease in
coding efficiency since the code length gets longer as the
number of consecutive zero coefficients R and a coefficient
value L get larger. Usually, the decrease in coding efficiency is
obvious when a low frequency component value is coded since
the coefficient value L as a low frequency component value is
large.
Namely, as a result of assigning a single VLC table according
to the occurrence probability and a single unique variable
length code according to a pair of R and L, the coefficient value
L indicating a large value is transformed into a variable length
code having a very long code length. Even when coding L
separately from R (one-dimensional coding of L) using a single
VLC table, the same problem occurs as in the case of coding R
and L as a pair.
Id. at 1:48–62.
In the solution described in the ’035 patent, an RL sequence
generation unit first converts the quantized frequency coefficients into a
sequence of R and L pairs (“RL values”), as shown in Figures 3A, 3B, and
4A. Figures 3A and 3B are reproduced below:
IPR2020-00867
Patent 7,305,035 B2
4

Figure 3A is a diagram showing quantized frequency coefficients in a block,
with upper left frequency coefficients denoting low frequency components
(e.g., direct-current components) and with frequency components becoming
larger toward the lower right. Id. at 6:43–49. Figure 3B, using arrows,
shows a progression (from low frequency to high frequency coefficients) of
scanning the coefficients shown in Figure 3A. Id. at 6:49–54. Figure 4A is
reproduced below:

Figure 4A is a listing of RL values generated by the RL sequence generation
unit, with EOB (End of Block) indicating that all subsequent coefficient
values are zero. Id. at 6:55–61.
IPR2020-00867
Patent 7,305,035 B2
5
The sequence of RL values is then reordered (id. at 6:64–67), as
shown in Figure 4B, reproduced below:

Figure 4B is a listing of RL values reversed in order from that of Figure 4A,
except the EOB value is not reordered. Id. at 6:66–67.
A code table correlates the RL values with code numbers. Id. at 7:4–
6, Fig. 5. Multiple VLC tables correlate the code numbers with variable
length codes. Id. at 7:6–8, Fig. 6. Figure 6 is reproduced below:
IPR2020-00867
Patent 7,305,035 B2
6

Figure 6 is a diagram showing an example of a VLC table. Id. at 7:31. In
this example,
The VLC table 1 is used for the first RL value. In this case, the
code number for the first RL value is “2,” therefore, the
variable length code is “011.” The conversion of the code
numbers into the variable length codes is performed
subsequently, and when an absolute value of L exceeds a
threshold value, the VLC table 2 is used for the following RL
values. Assume that a threshold value of the absolute value of
L is “2,” the absolute value of L exceeds the threshold value at
the fourth RL value (0, 3). Therefore, the VLC table 1 is used
for the first through the fourth RL values and the VLC table 2 is
used for the fifth RL value and thereafter.
IPR2020-00867
Patent 7,305,035 B2
7
Id. at 7:45–55. In this example, “[t]he RL values are converted into the
variable length codes in an order reverse to the order for scanning.” Id. at
8:33–35. According to the ’035 patent,
The absolute value of L usually becomes larger in the low-
frequency domain, therefore, the absolute value of L become[s]
larger when the RL values are converted into the variable length
codes in an order reverse to the order in which the RL values
are generated by scanning the coefficients from the low-
frequency domain toward the high frequency domain.
Therefore, when the absolute value of L gets larger after the
absolute value of L has exceeded the threshold value, that is, by
using the VLC table in which the variable length code becomes
shorter as the code number gets larger, the total amount of code
can be reduced. Namely, the total code amount for L can be
reduced also by coding L and R separately, and also, by using
plural VLC tables.
Id. at 8:47–60.
Claim 1, reproduced below, is the only challenged claim:
1. A decoding method for performing variable-length
decoding on coded data obtained by performing variable-length
coding on coefficients of a block which are obtained by
performing frequency transformation on picture data of the
block that has a predetermined sixe of pixels, the method
comprising:
performing variable-length decoding on coded data from
a high frequency component toward a low
frequency component to obtain decoded
coefficients; and
inverse scanning the decoded coefficients into two-
dimensional coefficients of the block;
wherein said performing variable-length decoding
comprises:
decoding a first coefficient using a first variable
length code table of a plurality of variable
length code tables;
IPR2020-00867
Patent 7,305,035 B2
8
switching to a second variable length code table of
the plurality of variable length code tables if
an absolute value of the first coefficient
exceeds a first threshold;
decoding a second coefficient following the first
coefficient using the second variable length
code table;
switching to a third variable length code table of
the plurality of variable length code tables if
an absolute value of the second coefficient
exceeds a second threshold, the second
threshold being greater than the first
threshold; and
decoding a third coefficient following the second
coefficient using the third variable length
code table.

D. Evidence
Petitioner relies on the references listed below.
Reference Date Exhibit
No.
Luna US 6,298,087 B1 Oct. 2, 2001 1004
Silva E.A. da Silva & M. Ghanbari, A DCT-
Based Aliasing Cancellation Method in
Subband Coding, VOL. 3, NO. 5, IEEE
TRANSACTIONS ON CIRCUITS AND
SYSTEMS FOR VIDEO TECHNOLOGY
384–87
Oct. 1993 1005
VCEG-
M51r1
M. Karczewicz, Improved Intra coding,
VCEG-M51, ITU –
TELECOMMUNICATIONS
STANDARDIZATION SECTOR, STUDY
GROUP 16 QUESTION 6, VIDEO CODING
EXPERTS GROUP (VCEG)
Mar. 21,
2001
1006

IPR2020-00867
Patent 7,305,035 B2
9
Petitioner also relies on the Declaration of Lina J. Karam, Ph.D.
(Ex. 1003).

E. The Asserted Ground of Unpatentability
Petitioner contends that claim 1 would have been obvious, under pre-
AIA 35 U.S.C. § 103(a), over Luna, Silva, and VCEG-M51r1. Pet. 4.

II. ANALYSIS
A. Claim Construction
We construe claims “using the same claim construction standard that
would be used to construe the claim in a civil action under 35 U.S.C. 282(b),
including construing the claim in accordance with the ordinary and
customary meaning of such claim as understood by one of ordinary skill in
the art and the prosecution history pertaining to the patent.” 37 C.F.R.
§ 42.100(b) (2019); see also Phillips v. AWH Corp., 415 F.3d 1303
(Fed. Cir. 2005) (en banc).
Petitioner contends that “first coefficient,” “second coefficient
following the first coefficient,” and “third coefficient following the second
coefficient,” as recited in claim 1, “are used to distinguish different elements
of the claim, and do not require the coefficients to be adjacent” in a sequence
of decoded coefficients. Pet. 12. Patent Owner agrees with this construction
“so far as the ordinal terms ‘first,’ ‘second,’ and ‘third’ should not be read to
denote that the elements are immediate neighbors.” Prelim. Resp. 19. We
agree with Patent Owner, however, that none of these terms is a source of
controversy in this proceeding. Id.
Patent Owner proposes construing “switching to a [second/third]
variable length code table . . . if an absolute value of the [first/second]
IPR2020-00867
Patent 7,305,035 B2
10
coefficient exceeds a [first/second] threshold,” as recited in claim 1, to be
necessary conditions for switching, and that the thresholds be predetermined
values. Prelim. Resp. 21–25. Nevertheless, regardless of whether we accept
this construction, we conclude that Petitioner is not reasonably likely to
demonstrate the unpatentability of claim 1. Accordingly, it is not necessary
to construe these terms. See Nidec Motor Corp. v. Zhongshan Broad Ocean
Motor Co., 868 F.3d 1013, 1017 (Fed. Cir. 2017) (noting that “we need only
construe terms ‘that are in controversy, and only to the extent necessary to
resolve the controversy’”) (quoting Vivid Techs., Inc. v. Am. Sci. & Eng’g,
Inc., 200 F.3d 795, 803 (Fed. Cir. 1999)).
Based on the record before us, we do not find it necessary to provide
express claim constructions for any other terms.

B. Obviousness of Claim 1 over Luna, Silva, and VCEG-M51r1
Petitioner contends that claim 1 would have been obvious over Luna,
Silva, and VCEG-M51r1. Pet. 29–71. For the reasons given below,
Petitioner has not made a sufficient showing.
A claim is unpatentable under 35 U.S.C. § 103 if the differences
between the claimed subject matter and the prior art are “such that the
claimed invention as a whole would have been obvious before the effective
filing date of the claimed invention to a person having ordinary skill in the
art to which the claimed invention pertains.” We resolve the question of
obviousness on the basis of underlying factual determinations, including
(1) the scope and content of the prior art; (2) any differences between the
claimed subject matter and the prior art; (3) the level of skill in the art; and
IPR2020-00867
Patent 7,305,035 B2
11
(4) objective evidence of nonobviousness, i.e., secondary considerations.1
See Graham v. John Deere Co., 383 U.S. 1, 17–18 (1966).

1. Level of Skill in the Art
Petitioner, relying on the testimony of Dr. Karam, contends that a
person of ordinary skill in the art “would have had a bachelor’s degree in
electrical or computer engineering, or a closely related scientific field such
as physics or computer science, and two years of work experience with
video compression.” Pet. 10 (citing Ex. 1003 ¶¶ 38–42). Patent Owner
“generally agrees with [Petitioner’s] definition” of a skilled artisan. Prelim.
Resp. 25. Petitioner’s proposal is consistent with the technology described
in the Specification and the cited prior art. For purposes of this Decision, we
adopt Petitioner’s proposed level of skill.

2. Scope and Content of the Prior Art
a) Overview of Luna
Luna relates to variable length decoding and decompression of a video
signal. Ex. 1004, 1:6–9. In its challenge, Petitioner relies primarily on
Luna’s description of the background art. Pet. 29–34, 36, 44, 51 (citing
Ex. 1004, 1:7–10, 1:32–33, 1:51–55, 1:66–2:13).
Luna’s background discusses video compression in the context of the
MPEG-2 standard. Ex. 1004, 1:32–44. In this example, video data are
transformed, using a 2-dimensional discrete cosine transform (DCT)
algorithm, into an 88 matrix of coefficients, which are then quantized,

1 The record does not include allegations or evidence of objective indicia of
nonobviousness.
IPR2020-00867
Patent 7,305,035 B2
12
resulting in many higher frequency coefficients being set to zero. Id. at
1:50–62. The coefficients are then converted to a linear stream of run/level
pairs (RL values), which are, in turn, coded using a VLC. Id. at 1:62–2:2.
Luna describes decoding such data as including “the steps of VLC
decoding, inverse scanning to reconstruct the 88 2-dimensional quantised
DCT coefficient matrix, inverse quantising to recover a truncated set of DCT
coefficients, and inverse DCT.” Id. at 2:3–7.

b) Overview of Silva
Silva is an “Express Letter” published in the journal IEEE
Transactions on Circuits and Systems for Video Technology. Petitioner
offers the testimony of Sylvia Hall-Ellis, Ph.D., an Adjunct Professor in the
School of Information at San José State University in San José, California
(Ex. 1008), as well as the testimony of Dr. Karam, as evidence that Silva
was publicly accessible prior to the critical date of the ’035 patent and,
therefore, is prior art. Pet. 18–22. Patent Owner does not contest the prior
art status of Silva at this stage of the proceeding. On the current record, we
are persuaded that Petitioner has established a reasonable likelihood that
Silva is prior art to the ’035 patent.
Silva describes techniques for DCT-based subband coding of images.
Ex. 1005, 384. In Silva’s example, a transformed image is decomposed, or
partitioned, into four subbands, as shown in Figure 2, reproduced below:
IPR2020-00867
Patent 7,305,035 B2
13

Figure 2 is a diagram of a four-band subband decomposition, with partition 1
including the lowest frequency coefficients and partition 4 including the
highest frequency coefficients. Id. According to Silva,
The lowest band of a subband coded picture is similar to the
original picture. However, due to subsampling, aliasing is
introduced, causing some disturbances. If the components from
the high bands that cancel the aliasing in the lowest band could
be added to this band, a smooth low-pass version of the image
would be obtained. Thus, when the subband decomposition is
employed as part of an image coding scheme, it is desirable to
identify those components from the high bands that cancel the
aliasing components from the lowest band.
Id.
To minimize aliasing, Silva teaches concentrating scanning on the
portions of the partitions that have the highest energy. This is shown in
Figure 3, reproduced below.
IPR2020-00867
Patent 7,305,035 B2
14

Figure 3 is four pictures (a–d) representing partitions 1–4, respectively, of
Figure 2. Silva explains that, to minimize aliasing, scanning of the partitions
should focus on those areas of highest energy, which vary from partition to
partition: “a smooth low-pass version of the image can be reconstructed
with the dense shaded parts shown in Fig. 3, i.e., the whole of band 1, parts
1-2, 1-3, and the intersection of 2-4 and 3-4. In addition, these are the
frequencies of each band with the highest energies.” Id. at 384–85.
Silva teaches that, to concentrate on the coefficients of each partition
with the highest energies, a different scanning direction should be used for
each partition, as shown in Figure 4, reproduced below:

Figure 4 includes 4 pictures (a–d) of scanning directions for partitions 1–4 of
Figure 2, respectively. According to Silva,
IPR2020-00867
Patent 7,305,035 B2
15
if the scanning directions for the 88 DCT coefficients for
each band are the ones shown in Fig. 4, the coefficients with the
highest energy values will be, in general, scanned first. This
will lead to shorter zero-runs and earlier end of block codes
(EOB codes), and, as a consequence, lower bitrates than if the
usual zigzag scanning is employed in all the bands.
Id. at 385 (endnote omitted).

c) Overview of VCEG-M51r1
VCEG-M51r1 is a submission to a working group of a standard
setting body. The parties dispute whether VCEG-M51r1 was publicly
accessible as of the critical date of the ’035 patent and, thus, whether it is
prior art to the ’035 patent. Pet. 23–29; Prelim. Resp. 45–48. Because
Petitioner has not shown a reasonable likelihood of success, even if VCEG-
M51r1 is prior art, we need not reach this issue.
VCEG-M51r1 proposes “using additional VLC tables derived from” a
first VLC table (“UVLC”) to arrive at an improved method of encoding
transform coefficients. Ex. 1006, 1. VCEG-M51r1 teaches calculating R
and L indices “using the run (run_previous) and level (level_previous)
information of the last decoded coefficient.” Id. at 2. These indices are used
to access one of two table Ms (reproduced below), with an entry of the
appropriate table M (M(l,r)) indicating which of multiple VLC tables should
be used.

IPR2020-00867
Patent 7,305,035 B2
16
The pictures above are two instances of Table M, the first showing table
entries for indices r and l when QP2 is less than 24, and the second showing
table entries for indices r and l when QP is greater than or equal to 24. Id.
The entries (1, 2, 3) in the table indicate which VLC table (UVLC, UVLC2,
UVLC3) will be used. Id.

3. Analysis of Claim 1
Petitioner contends that Luna teaches the preamble of claim 1.
Pet. 29–32. Specifically, Petitioner contends that Luna’s description of
decoding performed pursuant to section 7 of the MPEG-2 standard, along
with the steps of VLC decoding, teaches a “decoding method for performing
variable-length decoding,” as recited in claim 1. Id. at 30 (citing Ex. 1004,
1:7–10, 2:3–13). Petitioner further argues that Luna’s description of
transforming video data into the frequency domain using a DCT algorithm,
followed by quantisation and linearization, teaches that the decoding is on
“coded data obtained by performing variable-length coding on coefficients
of a block which are obtained by performing frequency transformation on
picture data of the block,” as recited in claim 1. Id. at 30–31 (citing
Ex. 1004, 1:51–58, 1:66–2:2). Petitioner also argues that Luna’s description
of dividing an input picture into 88 pixel blocks teaches a block “that has a
predetermined size of pixels,” as recited in claim 1. Id. at 31–32 (citing

2 According to Dr. Karam, QP is a “quantization parameter” that is “set
during the encoding process that influences the compression rate of the
compressed video, thereby influencing the resulting bit rate of the video. In
the H.263 and MPEG video compression standards, the quantization
parameter can range from a value of 1 to 31.” Ex. 1003 ¶ 125.
IPR2020-00867
Patent 7,305,035 B2
17
Ex. 1004, 1:32–33, 1:51–55). Patent Owner does not dispute these
allegations.
The parties dispute whether the art teaches “performing variable-
length decoding on coded data from a high frequency component toward a
low frequency component to obtain decoded coefficients,” as recited in
claim 1 (emphasis added). Petitioner cites Luna for “performing variable-
length decoding on coded data . . . to obtain decoded coefficients,” but
concedes that “Luna does not describe the order in which the coefficients are
scanned for linearization.”3 Pet. 32–34 (citing Ex. 1004, 1:51–2:12).
Petitioner contends that Silva describes scanning in a zig-zag pattern
from a bottom-right corner to a top-left corner (from a high frequency
component toward a low frequency component), and that a skilled artisan
would have recognized that this teaching could be applied to Luna’s
description of linearization of a quantized coefficient matrix. Id. at 35–36,
39–40 (citing Ex. 1005, Fig. 4(d)).
Petitioner contends that a skilled artisan would have combined the
teachings of Luna and Silva because the references are analogous art, and
because “a [person of ordinary skill in the art] would have recognized

3 Patent Owner argues that, in fact, Luna does describe the scanning order.
Prelim. Resp. 28–32. As Patent Owner points out, Luna (Ex. 1004, 2:3–12)
expressly describes performing video decoding according to Section 7 of the
ISO/IEC 13818-2 MPEG-2 standard. Id. at 28–29. Patent Owner introduces
evidence, by way of the ITU-T H.262 and H.263 Recommendations
documents, that this standard decodes coefficients in the order of low
frequency to high frequency. Id. at 29–30 (citing Ex. 2001, 18–19;
Ex. 2002, 25–26; Ex. 2004, 10). The H.262 documents note that “[t]he
identical text is also published as ISO/IEC International Standard 13818-2.”
Ex. 2001, 2; see also Ex. 2002, 4. Thus, through its reference to the
standard, Luna appears to expressly describe a scanning order from low
frequency to high frequency.
IPR2020-00867
Patent 7,305,035 B2
18
Silva’s description of known scanning techniques that are shown to be
‘efficient’ as pertinent to the problem of improving coding efficiency.”
Pet. 41 (quoting Ex. 1005, 386; citing Ex. 1003 ¶ 92). According to
Petitioner, Silva “explicitly provides a reason to incorporate its teachings to
other video processing references like Luna” by stating that “if the scanning
directions for the 88 DCT coefficients for each band are the ones shown in
Fig. 4, the coefficients with the highest energy values will be, in general,
scanned first. This will lead to shorter zero-runs and earlier end of block
codes (EOB) codes), and, as a consequence, lower bitrates than if the
usual zigzag scanning is employed in all the bands.” Id. at 42 (quoting
Ex. 1005, 385; citing Ex. 1003 ¶ 93). Further, citing to Dr. Karam’s
testimony, Petitioner argues that “researchers in the field of video coding
were noting the benefits of ordering coefficients from a high frequency to a
low frequency, specifically in the context of variable length coding
techniques.” Id. (citing Ex. 1003 ¶¶ 97–99). Petitioner argues that “Luna
does not explicitly disclose the direction in which its blocks are scanned
before the coding process occurs,”4 but that a skilled artisan “would have
recognized that [there existed] any number of options for scanning,
including scanning from top-left to bottom-right, or the reverse, as in Silva.”
Id. at 42–43.
In response, Patent Owner argues that Petitioner has not shown that a
skilled artisan would have combined the cited teachings of Luna and Silva.
As noted above, we are persuaded by Patent Owner’s argument that Luna
teaches a scanning order from low frequency to high frequency. Prelim.

4 As noted above, we disagree with Petitioner that Luna does not disclose the
direction in which its blocks are scanned, given Luna’s express citation to
Section 7 of the MPEG-2 standard.
IPR2020-00867
Patent 7,305,035 B2
19
Resp. 28–30. According to Patent Owner, the ’035 patent, the prior art, and
textbooks describe good reasons for doing so. Id. at 31–32 (citing Ex. 1001,
6:61–64 (“[B]y performing scanning starting at the low-frequency domain
toward the high frequency domain, it is possible to reduce the amount of
information included in the RL sequence.”); Ex. 1004, 1:60–62 (“As most of
the higher frequency coefficients are set to zero, linearization produces long
runs of zeros.”); Ex. 1005, 385 (“[I]f the scanning directions for the 88
DCT coefficients for each band are the ones shown in Fig. 4, the coefficients
with the highest energy values will be, in general, scanned first. This will
lead to shorter zero-runs and earlier end of block codes (EOB codes), and, as
a consequence, lower bitrates than if the usual zigzag scanning is employed
in all the bands.”); Ex. 2006,5 199 (“For natural images, the low-frequency
DCT coefficients tend to have a lot of energy while the high frequency
coefficients tend to be very small.”)).
Patent Owner further argues that “[t]he Petition . . . improperly
divorces the Silva-cited scanning order from its specific subband context,
and omits any explanation or evidence as to how or why it would have been
obvious to implement this subband-specific scanning order in the
standardized H.262 transform coefficient block of Luna.” Prelim. Resp. 27.
Patent Owner argues that Silva’s reverse-scanning description is tied to a
specific atypical scenario in which images are decomposed into four
subbands and that the reverse scanning is limited to the highest frequency
subband. Id. at 32–33 (citing Ex. 1005, 384–85, Figs. 2, 4). In this scenario,
Silva explains, higher energy coefficients appear in the bottom right of the

5 L. Guan et al., MULTIMEDIA IMAGE AND VIDEO PROCESSING 197–202 (2d
ed. 2012).
IPR2020-00867
Patent 7,305,035 B2
20
highest frequency subband. Ex. 1005, 385, Fig. 3(d); Prelim. Resp. 34.
Patent Owner argues that “[t]his is the opposite of the standard coefficient
block of Luna, where the highest energy coefficients are on the top-left and
the low energy/zero-value coefficients are on the bottom-right.” Prelim.
Resp. 34 (citing Ex. 1004, 1:55–58). Patent Owner argues that Silva
describes scanning in reverse order in subband 4 for the same reason Luna
describes scanning from top left to bottom right, namely, to scan coefficients
with the highest energy values first, leading to shorter zero-runs and EOB
codes. Id. at 34–35 (citing Ex. 1004, 1:55–58, 1:60–62; Ex. 1005, 385).
As to Petitioner’s alleged specific teaching in Silva (e.g., Pet. 41–42;
Ex. 1005, 385 (“This will lead to shorter zero-runs and earlier end of block
codes (EOB codes), and, as a consequence, lower bitrates than if the usual
zigzag scanning is employed in all the bands.”) (endnote omitted)), Patent
Owner argues that this teaching would not provide such a benefit if used
with Luna’s technique. Prelim. Resp. 37–39. Instead, Patent Owner argues,
“using Silva’s bottom-right to top-left scanning order in Luna would scan
‘the coefficients with the highest energy values’ last, NOT first.” Id. at 38
(citing Ex. 1004, 1:55–62). This is because Luna’s coefficients are ordered,
per the H.262 standard, from high energy (low frequency) to low energy
(high frequency). Thus, reversing the scan order would result in low energy
coefficients being scanned first, contrary to Silva’s teaching. Id. at 39–40.
We agree with Patent Owner. Petitioner’s proposed reasons to
combine Luna and Silva are conclusory and contrary to the teachings of the
references. Petitioner points to Silva’s statements that its techniques
improve efficiency and will lead to shorter zero-runs and earlier EOB codes.
Pet. 41–42 (citing Ex. 1005, 386). But Petitioner ignores the context in
which Silva describes achieving these benefits, namely an atypical situation
IPR2020-00867
Patent 7,305,035 B2
21
in which a high frequency subband of an image has coefficients arranged
from low energy to high energy, rather than the more typical high energy to
low energy. Ex. 1005, 384–85. Dr. Karam’s testimony largely repeats the
arguments in the Petition and also fails to account sufficiently for these
teachings of Silva. Ex. 1003 ¶¶ 92–96. Thus, this testimony lacks
credibility and is entitled to little weight. Petitioner has not shown
sufficiently that a skilled artisan would have had a reason, with rational
underpinning, to combine Luna and Silva.6
The parties also dispute whether the prior art teaches
switching to a second variable length code table of the plurality
of variable length code tables if an absolute value of the
first coefficient exceeds a first threshold
. . . [and]
switching to a third variable length code table of the plurality of
variable length code tables if an absolute value of the
second coefficient exceeds a second threshold, the
second threshold being greater than the first threshold
as recited in claim 1. Petitioner concedes that Luna does not teach multiple
VLC tables, but argues that a skilled artisan “would have had reason to

6 Petitioner further alleges that “researchers in the field of video coding were
noting the benefits of ordering coefficients from a high frequency to a low
frequency, specifically in the context of variable length coding techniques,”
citing to Dr. Karam’s testimony, but does not otherwise explain its argument
or discuss the evidence. See Pet. 42 (citing Ex. 1003 ¶¶ 97–99). This is an
improper incorporation by reference of analysis presented in Dr. Karam’s
testimony on the relevance of four references not discussed in the Petition,
Exhibits 1024–1027, and will not be considered. See 37 C.F.R. § 42.6(a)(3)
(“Arguments must not be incorporated by reference from one document into
another document.”). In any case, Dr. Karam does not explain sufficiently
why these additional teachings would have suggested that a skilled artisan
would have had a reason to reorder the scanning in Luna.
IPR2020-00867
Patent 7,305,035 B2
22
implement alternative VLC table implementations, such as the VLC tables
provided in VCEG-M51r1.” Pet. 51. Essentially, Petitioner argues that, if
QP is kept above 24 and a sequence of RL values is chosen such that R is 2
or less, then according to table M in VCEG-M51r1, when the absolute value
of the previous value of L is 1, the VLC table will be switched from UVLC
to UVLC2. Id. at 46–48, 51–63. Petitioner identifies L=1 as the “first
threshold.” Id. at 47–48. Petitioner makes a similar argument for switching
from UVLC2 to UVLC3, with L=3 as the “second threshold.” Id. at 48–49,
66–69.
Petitioner contends that “incorporating the teachings of VCEG-M51r1
with Luna’s teachings would have been no more than the use of a known
technique to improve a similar method in the same way,” and that a skilled
artisan “would have further recognized that VCEG-M51r1 discloses an
improvement to [the] decoding techniques” of Luna. Id. at 57 (citing
Ex. 1003 ¶ 133). Petitioner points to VCEG-M51r1’s characterization of its
technique as an “[i]mproved method of encoding transform coefficients” and
providing “coding improvements.” Id. (quoting Ex. 1006, 1, 3). Petitioner
contends that “contemporaneous discussion of the VCEG-M51r1 proposal
indicated that the proposed method resulted in” bit rate savings, and that a
skilled artisan would have recognized bit rate savings as advantageous.7

7 As Patent Owner points out, Prelim. Resp. 18 n.3, the contemporaneous
discussion cited by Petitioner characterizes any such advantages as
“limited.” Ex. 1011, 10. Nevertheless, we need not resolve whether the
technique of VCEG-M51r1 actually would have provided improvements to
Luna, as Petitioner argues, because Petitioner’s overall proposed
combination relies on the scanning order of Silva, rather than that of Luna,
as discussed below.
IPR2020-00867
Patent 7,305,035 B2
23
Id. at 57–58 (citing Ex. 1011, 108; Ex. 1003 ¶ 134). Petitioner argues that
there would have been a reasonable expectation of success in incorporating
VCEG-M51r1’s teachings into the decoding method of Luna because a
skilled artisan “would have recognized that incorporating the teachings of
VCEG-M51r1 would have required no more than a modification of Luna’s
prior art decoding process with different VLC tables disclosed by VCEG-
M51r1.” Id. at 58 (citing Ex. 1003 ¶ 135). Dr. Karam’s testimony, cited in
support of these arguments, does little more than repeat the arguments in the
Petition, without adding to them materially. Ex. 1003 ¶¶ 134–135.
In response, Patent Owner contends that Petitioner “fails to account
for the disparate methods of the Luna-Silva combination and VCEG-
M51r1,” and that Petitioner’s “obviousness analysis is with reference to
Luna, and not Luna as modified by Silva.” Prelim. Resp. 67. In particular,
Patent Owner argues, Petitioner has failed to account for whether VCEG-
M51r1’s technique would have been an improvement to the Luna/Silva
combination asserted by Petitioner in which the scanning and decoding order
proceeds in reverse. Id. at 67–68. According to Patent Owner, VCEG-
M51r1 achieves its improvements in the context of scanning that proceeds
from low frequency coefficients to high frequency coefficients, “the VLC
table determination and switching proposed by VCEG-M51r1 is explicitly
dependent on the decoding order,” and “[t]o apply the VLC table switching
technique of VCEG-M51r1 to the reverse scanning order (as proposed by the
Petition’s combination of Luna and Silva) would fundamentally change the
variable length coding of VCEG-M51r1.” Id. at 68–69. Patent Owner
argues that “the Petition attempts to apply the technique to a substantively

8 Here, we use the page numbering added by Petitioner to the exhibit.
IPR2020-00867
Patent 7,305,035 B2
24
different method and in a critically different way, i.e., to an undisclosed
sequence of increasing coefficient value order with different tables and bits
used to code the same data,” and that “[t]he bitrate results of using such a
divergent variable length coding and ensuing bit sequence would not have
been predictable, let alone predictably successful.” Id. at 72.9
We agree with Patent Owner. Claim 1 recites switching to different
VLC tables based on the values of coded coefficients. Those coefficients are
coded, according to claim 1, “from a high frequency component toward a
low frequency component.” As detailed above, Petitioner’s theory of
obviousness is that a skilled artisan would have recognized VCEG-M51r1 as
providing an improvement to Luna’s technique, which codes coefficients
from a low frequency component toward a high frequency component.
Petitioner provides no persuasive evidence or explanation as to why VCEG-
M51r1’s teachings would have provided an improvement if the scanning
order were reversed, as it would have been with Petitioner’s combination of
Luna and Silva. Accordingly, Petitioner has not articulated a reason, with
rational underpinning, for combining the teachings of VCEG-M51r1 with
those of Luna and Silva.
Because Petitioner does not advance reasons, with rational
underpinning, for combining Luna, Silva, and VCEG-M51r1, Petitioner does
not establish a reasonable likelihood that claim 1 would have been obvious
over Luna, Silva, and VCEG-M51r1.

9 Patent Owner makes other arguments as to VCEG-M51r1. Prelim. Resp.
48–66. However, as we deny the Petition on other grounds, it is not
necessary to address these additional arguments.
IPR2020-00867
Patent 7,305,035 B2
25
III. CONCLUSION
For the foregoing reasons, we determine that the information
presented in the Petition does not establish a reasonable likelihood that
Petitioner would prevail in showing that claim 1 of the ’035 patent is
unpatentable.

IV. ORDER
In consideration of the foregoing, it is hereby:
ORDERED that the Petition is denied as to the challenged claim of
the ’035 patent, and no inter partes review is instituted.

IPR2020-00867
Patent 7,305,035 B2
26

FOR PETITIONER:
Raghav Bajaj
Ashraf Fawzy
Roshan Mansinghani
David L. McCombs
Jonathan R. Bowser
HAYNES AND BOONE, LLP
raghav.bajaj.ipr@haynesboone.com
afawzy@unifiedpatents.com
roshan@unifiedpatents.com
david.mccombs.ipr@haynesboone.com
jon.bowser.ipr@haynesboone.com

FOR PATENT OWNER:
William H. Mandir
Fadi Kiblawi
John F. Rabena
SUGHRUE MION, PLLC
wmandir@sughrue.com
fkiblawi@sughrue.com
jrabena@sughrue.com