Uniloc 2017 LLC

Patent Trials and Appeals Board

Jun 17, 2020

IPR2019-00054 (P.T.A.B. Jun. 17, 2020)

Trials@uspto.gov Paper No. 19
571-272-7822 Date: June 17, 2020

UNITED STATES PATENT AND TRADEMARK OFFICE
____________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________

APPLE, INC.,
Petitioner,

v.

UNILOC 2017 LLC
Patent Owner.
____________

IPR2019-00054
Patent 6,304,612 B1
____________

Before KALYAN K. DESHPANDE, JOHN F. HORVATH, and
GARTH D. BAER, Administrative Patent Judges.

HORVATH, Administrative Patent Judge.

JUDGMENT
Final Written Decision
Determining All Challenged Claims Unpatentable
35 U.S.C. § 318(a)

IPR2019-00054
Patent 6,304,612 B1

2
I. INTRODUCTION
A. Background
Apple, Inc., (“Petitioner”) filed a Petition requesting inter partes review
of claims 1–11 (“the challenged claims”) of U.S. Patent No. 6,304,612 B1
(Ex. 1001, “the ’612 patent”). Paper 2 (“Pet.”). Uniloc 2017 LLC (“Patent
Owner”), filed a Preliminary Response. Paper 6 (Prelim. Resp.). Upon
consideration of the Petition and Preliminary Response, we instituted inter
partes review of all challenged claims on all grounds raised. Paper 7 (“Dec.
Inst.”).
Patent Owner filed a Response to the Petition (Paper 9, “PO Resp.”),
Petitioner filed a Reply (Paper 10, “Pet. Reply”), and Patent Owner filed a
Sur-Reply (Paper 11, “PO Sur-Reply”). An oral hearing was held on April
6, 2020, and the hearing transcript is included in the record. See Paper 18
(“Tr.”).
We have jurisdiction under 35 U.S.C. § 6(b). This is a Final Written
Decision under 35 U.S.C. § 318(a) and 37 C.F.R. § 42.73. For the reasons
set forth below, we find Petitioner has shown by a preponderance of
evidence that claims 1–11 of the ’612 patent are unpatentable.
B. Related Matters
Petitioner and Patent Owner indicate that they are unaware of any
other matters that can affect or be affected by this proceeding. See Pet. 65;
Paper 3 (2).
IPR2019-00054
Patent 6,304,612 B1

3
C. Evidence Relied Upon1
References Effective Date Exhibit
Seshardi US 5,208,816 May 4, 1993 1004
Chennakeshu US 5,349,589 Sept. 20, 1994 1005

D. Instituted Ground of Unpatentability
Claim(s)
Challenged 35 U.S.C. § Reference(s)/Basis
1–11 103(a) Seshardi, Chennakeshu
II. ANALYSIS
A. The ’612 Patent
The ’612 patent relates to a transmission system for transmitting
“sequences of source symbols in a reliable way to a receiver.” Ex. 1001,
1:19–20. The source symbols “are encoded into sequences of channel
symbols by a channel encoder,” such as a convolutional encoder. Id. at
1:25–29. A convolutional encoder clocks “the sequence of source symbols
. . . into a shift register with length v,” and the “state of the convolutional
encoder is defined by the content of the shift register.” Id. at 10:9–12. Thus,
for a binary v-stage convolutional encoder, “the number of possible states of
the convolutional encoder is equal to 2v.” Id. at 10:12–14. The channel
symbols are obtained from the source symbols that are clocked into the
convolutional encoder “by combining several [source] symbols . . . at

1 Petitioner also relies upon the Declaration of Dr. Charles D. Creusere
(Ex. 1003).
IPR2019-00054
Patent 6,304,612 B1

4
different taps of the shift register by using modulo-two operations.” Id. at
10:14–16.
Convolutional encoders are initialized to a predetermined state. Id. at
1:27–29. Conventionally, the predetermined state is zero, but this “reduces
the number of source symbols [that] can be transmitted in one sequence of
channel symbols.” Id. at 1:29–34. To remedy this, “tail-biting” encoders
use a predetermined state that is “defined by the last source symbols in the
sequence of source symbols.” Id. at 1:35–43. However, this increases the
complexity of decoding the channel symbols because the initial or
predetermined state of the encoder is no longer known. Id. at 1:53–56.
Nonetheless, it was known to reduce this complexity by “decoding the
channel signal repeatedly in a circular fashion.” Id. at 1:57–59.
Channel decoders use a trellis to decode convolutionally encoded
data, i.e., to determine “the sequences of source symbols from the sequences
of [transmitted] channel symbols.” Id. at 10:4–6. The trellis allows channel
decoders to estimate state sequences “as they are present during the
encoding process” by “determining candidate state sequences on [the] basis
of a likelihood measure, further to be referred to as a path metric.” Id. at
10:16–20. The “path metric is determined from the channel signal and said
candidate state sequences.” Id. at 10:20–22.
The channel decoder disclosed in the ’612 patent uses a trellis to
decode source symbols encoded with a tail-biting convolutional encoder by
“extending, on [the] basis of a cyclically extended channel signal, the state
sequence beyond a number of states equal to the number of source symbols
in a sequence of source symbols.” Id. at 2:24–28. The channel decoder
selects “the final state sequence on [the] basis of the terminating states of the
IPR2019-00054
Patent 6,304,612 B1

5
extended state sequences.” Id. at 2:28–31.
Figure 3 of the ’612 patent, reproduced below, discloses an extended
trellis used to decode convolutionally encoded data when the convolutional
encoder has been initialized to the final state (i.e., state 206 or 01) of the
encoded data.

Figure 3 is “an example of a trellis to be used in the channel decoders 28 and
29 for determining the sequences of source symbols from the sequences of
channel symbols” processed by the decoder. Id. at 10:4–6. The channel
decoders can be Viterbi decoders. Id. at 7:58–61.
The decoding process begins at trellis stage t = 0 by considering each
of the possible initial states (i.e., 00, 10, 01, and 11) of the channel encoder
to be equally likely. Id. at 10:24–27. For each channel signal received (e.g.,
at trellis stages t = 1, v . . . N . . . N+v) extended candidate sequences are
generated by appending to an “originating candidate sequence . . . one of the
possible new states” of the channel encoder. Id. at 10:27–32. The path
metric of an extended candidate sequence “is calculated from the path metric
of the originating state and a branch metric determined from the channel
signal and the channel symbols corresponding to the [trellis state] transition
IPR2019-00054
Patent 6,304,612 B1

6
between the originating state and the new state.” Id. at 10:32–37. For
example, when the two-state candidate sequence 01-10 terminating at trellis
stage t = 1 is extended to the three-state candidate sequence 01-10-01
terminating at trellis stage t = v, the path metric for the three-state candidate
sequence is determined by adding to the path metric for the two-state
candidate sequence the branch metric representing the difference, if any,
between the received channel symbols and the channel symbols expected
from encoding source symbol “1,” which is the source symbol needed to
extend the two-state candidate sequence 01-10 to the three-state candidate
sequence 01-10-01. See id. at 10:27–37, Fig. 3.
This channel symbol decoding process is continued “until the trellis
has been extended N times, in which N is the number of source symbols.”
Id. at 10:41–44. At t = N, the extended candidate sequence having the best
path metric “is used as [a] starting point for a trace back operation to find
a[n] earlier state . . . at the vth extension of the state sequence.”2 Id. at 10:44–
47. The candidate sequences are then further extended from trellis stage
t = N to trellis stage t = N+v, at which point a candidate sequence that
terminates in the same state as the earlier state (i.e., 01 or state 204 at stage
t = v), and having the best path metric is selected as the final state sequence.3
Id. at 10:49–53. The selected candidate sequence is traced back to state 204

2 The vth extension refers to the candidate sequence terminating in trellis
stage t = v. As shown in Figure 3, all candidate sequences at trellis stage
t = v terminate in state 204 (i.e., 01).
3 That is, a candidate sequence is selected as the final state sequence when it
(a) terminates in state 206 at trellis stage t = N+v, where state 206 is the
same state (i.e., 01) as state 204 at trellis stage t = v, and (b) has the best path
metric of all candidate sequences terminating in state 206.
IPR2019-00054
Patent 6,304,612 B1

7
to determine the source symbols required to create the state transitions along
the trellis path. Id. at 10:53–54. The source symbols recovered in this
manner are out of order, but can be reordered by shifting them back over v
symbols to restore the correct order. Id. at 10:54–58.
Figure 5 of the ’612 patent, reproduced below, discloses an algorithm
for determining the source symbols from the channel symbols by navigating
the trellis shown in Figure 3.

IPR2019-00054
Patent 6,304,612 B1

8
Figure 5 is a “flow diagram of a program for a programmable processor for
implementing a channel decoder according to the present invention.” Id. at
3:41–43.
A source symbol pointer i is incremented from values i = 0 to N+ε.4
Id. at 11:12–20, 11:47–66, Fig. 5 (steps 220–240). For each value of pointer
i, branch metrics for the possible trellis state transitions are determined from
the received channel symbols, indexed by pointer j. Id. at 11:12–29, 11:47–
12:57, Fig. 5 (steps 220–228). If pointer i is outside the range of mmdstart
and mmdstop, an ACS (Add, Compare, Select) operation adds the branch
metrics of trellis state transitions to the path metrics of the candidate state
sequences that are extended by those transitions, compares the path metrics
for the extended candidate state sequences, and selects “the path having the
largest path metric.” Id. at 12:6–18, Fig. 5 (steps 230, 234). If pointer i is
within the range of mmdstart and mmdstop, the same ACS operation to find
and select the best extended candidate state sequence is performed, and an
MMD (minimum metric difference) is calculated for that sequence. Id. at
12:22–23, Fig. 5 (steps 230, 232). MMD represents the “difference between
the two path metrics of the competing paths ending in a new state,” and is “a
good measure for the transmission quality.” Id. at 12:23–29. When
i = N+v, the trellis “state having the largest path metric is selected as the
best final state,” i.e., state s_max is selected at trellis stage t = N+v. Id. at
12:53–57, Fig. 5 (steps 240, 242).

4 The ’612 patent discloses that “a suitable value for ε is v.” Ex. 1001,
12:48–49. Thus, pointer i represents a stage in the trellis at time t, where t
ranges from t = 0 to t = N+v.
IPR2019-00054
Patent 6,304,612 B1

9
An N-level trace-back operation is then performed along the best path
from final state s_max to an earlier state s_0 at trellis stage t = v. Id. at
12:58–63, Fig. 5 (step 244). If states s_max and s_0 are the same state (e.g.,
01), the source symbols that produce the trellis transitions along this best
path are output together with the MMD value of the best path. Id. at 12:65–
13:7, Fig. 5 (steps 246, 250). However, if s_max and s_0 are not the same
state, s_0 is chosen to be the final state at trellis stage t = N+v, and a path is
traced back N levels from final state s_0 at t = N+v to earlier state s_0 at
t = v. Id. at 13:1–7, Fig. 5 (steps 246, 248). The source symbols that
produce the trellis state transitions in this new final state sequence are output
together with the MMD value of the new final state sequence. Id. at 13:1–7,
Fig. 5 (step 250).
B. Illustrative Claims
Of the challenged claims, claims 1, 7, and 9–11 are independent,
claims 2–6 depend directly or indirectly from claim 1, and claim 8 depends
directly from claim 7. Claim 1 is reproduced below.
1. A transmission system comprising:
a transmitter comprising a channel encoder for
encoding sequences of source symbols into
sequences of channel symbols and being arranged
for transmitting a channel signal representing the
channel symbols to a receiver,
a receiver comprising a channel decoder for
deriving the sequences of source symbols from a
channel signal representing the channel symbols by
keeping track of a plurality of state sequences with
a corresponding likelihood measure representing
the likelihood measure of the state sequences,
IPR2019-00054
Patent 6,304,612 B1

10
wherein the channel decoder determines the final
state sequence by comparing the likelihood
measures of a plurality of candidate state sequences
terminating in said state and selecting the state
sequence with the largest likelihood measure, and
the channel decoder further determines the
respective difference measures between the
likelihood measure of the selected candidate
sequence and the likelihood measures of the
rejected candidate sequences;
whereby the minimum difference of likelihood
measures determined between the selected
candidate sequence and the rejected candidate
sequences is a criterion for at least measuring the
reliability of the final state sequence.
Ex. 1001, 13:9–33. Claim 7 recites a receiver that includes the channel
decoder recited in claim 1. Compare id. at 13:61–14:14; with id. at 13:9–33.
Claim 9 recites the channel decoder recited in claim 1. Compare id. at
14:26–47; with id. at 13:9–33. Claim 10 recites a method for transmitting a
channel signal having channel symbols encoded by the transmission system
of claim 1. Compare id. at 14:48–67; with id. at 13:9–33. Claim 11 recites a
method for decoding the channel signal transmitted by the transmission
method of claim 10. Compare id. at 15:1–16:7; with id. at 14:48–67.
C. Level of Skill in the Art
Petitioner, relying on the testimony of Dr. Creusere, argues a person
of ordinary skill in the art would have had “a bachelor’s degree in electrical
engineering, computer science, or the equivalent and three years of
experience working with digital communications systems or in network
engineering” or would have had a master’s degree in these fields with an
emphasis on digital communications systems or network engineering. Pet. 6
IPR2019-00054
Patent 6,304,612 B1

11
(citing Ex. 1003 ¶ 33). Patent Owner does not dispute this definition, and
does not offer an alternative. PO Resp. 3.
Because we find Dr. Creusere’s definition of a person of ordinary skill
in the art to be reasonable and consistent with the types of problems and
solutions disclosed in the patent and prior art of record, we adopt it as our
own. See, e.g., In re GPAC Inc., 57 F.3d 1573, 1579 (Fed. Cir. 1995).
D. Claim Construction
In an inter partes review filed before November 13, 2018, claim terms
of an unexpired patent are given their broadest reasonable interpretation in
light of the specification of the patent in which they appear. 37 C.F.R.
§ 42.100(b) (2018); 83 Fed. Reg. 51,340. Under the broadest reasonable
interpretation standard, claim terms are generally given their ordinary and
customary meaning, as would have been understood by one of ordinary skill
in the art, in the context of the entire disclosure. In re Translogic Tech., Inc.,
504 F.3d 1249, 1257 (Fed. Cir. 2007). Only claim terms in controversy need
to be construed and only to the extent necessary to resolve the controversy.
See Nidec Motor Corp. v. Zhongshan Broad Ocean Motor Co., 868 F.3d
1013, 1017 (Fed. Cir. 2017).
Petitioner proposes express constructions for the terms “receiver” and
“channel signal.” See Pet. 8–10. In its Preliminary Response, Patent Owner
argued “the Board need not construe any claim term in a particular manner.”
Prelim. Resp. 9. Nonetheless, Patent Owner argued that the claims required
a particular ordering of steps, namely, “that the ‘final state sequence’ is first
determined and then difference measures are determined.” Id. at 9–10.
Patent Owner repeats that claim construction argument in its Response. See
PO Resp. 5 (“[T]he determination of [a final state sequence] must occur
IPR2019-00054
Patent 6,304,612 B1

12
before the determination of [difference measures]” because “[i]n the absence
of having previously completed the determination of [a final state sequence]
there simply would be no ‘selected candidate sequence’ and, consequently,
there would be no basis from which the ‘respective difference measures’
could be determined.”).5
In our Institution Decision, we agreed with Patent Owner that no
claim terms require express construction, but disagreed that the claims
require the difference measures to be determined after the final state
sequence is determined. Dec. Inst. 10–11, 32–34. In particular, we found
the limitations to “determine[] the final state sequence” and to “further
determine[] . . . difference measures” meant to determine difference
measures “in addition to” the final state sequence rather than “after” it. Id.
at 32–34 (emphasis added). This is because Figure 5 of the ’612 patent
discloses determining the difference measures before determining the final
state sequence. Id. In particular, Figure 5 discloses determining a final state
sequence at trellis stage i = N+ε (step 242) but determining the difference
measures for all candidate state sequences between trellis stages mmdstart
and mmdstop (steps 230–232), i.e., before selecting one of those candidate
state sequences as the final state sequence at trellis stage i = N+ε. Id.

5 Petitioner labels as 1(c) the claim 1 limitation that the decoder “determines
the final state sequence” and labels as 1(d) the limitation that the decoder
“determines . . . difference measures.” See Pet. 28, 30. Patent Owner,
following this convention, refers to limitations 1(c) and 1(d) in the Patent
Owner Response and Sur-Reply. We replace these references throughout
this Decision with references to the “final state sequence” and “difference
measures,” respectively, or similar variations.
IPR2019-00054
Patent 6,304,612 B1

13
Patent Owner argues we erred in our Institution Decision by
preliminarily construing the claims in a manner that does not require the
difference measures to be determined after the final state sequence. See PO
Resp. 4–12; see also PO Sur-Reply 1–6. Petitioner disagrees. See Pet.
Reply 5–19. Moreover, Petitioner argues that even if the claims do require
determining the difference measures after determining the final state
sequence, the challenged claims are unpatentable because “[i]n Seshardi, the
most likely path [final state sequence] has already been determined when
the difference in the metrics of the most likely and second most likely paths
is compared.” Id. at 2–3.
Upon consideration of all of the evidence and arguments presented by
the parties, we conclude that we do not need to resolve whether the claims
require determining the difference measures after determining the final state
sequence, as Patent Owner contends, because Petitioner demonstrates by a
preponderance of evidence that Seshardi determines difference measures
after the final state sequence for the reasons discussed in § II.F.2, infra.
Accordingly, for the reasons discussed above, we do not expressly
construe any claim terms in the ’612 patent, and do not determine whether
the claims are limited to determining difference measures after determining
the final state sequence.
E. Overview of the Prior Art
1. Seshardi
Seshardi discloses “a family of generalized Viterbi algorithms
(GVA).” Ex. 1004, 3:19–20. The GVAs are applied “to the soft or hard
decoding of convolutional codes . . . or other trellis-based structures.” Id. at
IPR2019-00054
Patent 6,304,612 B1

14
3:20–24. Seshardi’s system for performing convolutional encoding and
decoding is illustrated in Figure 1, which is reproduced below.

Figure 1 of Seshardi illustrates “a system for communicating
information from a data source 100 to a receiving location through a
communication channel 130.” Id. at 5:12–14. Optional block encoder 110
IPR2019-00054
Patent 6,304,612 B1

15
adds “appropriate redundancy, i.e., parity check bits, to the output of data
source 100 prior to presenting such data to the convolutional encoder 120.”
Id. at 5:18–21. Convolutional encoder 120 is any standard encoder chosen
based on “data rate and transmission channel characteristics.” Id. at 5:26–
29. Likewise, modulator 125 and demodulator 140 are of a standard design
“suited to transmission channel 130,” and demodulator 140 “performs
standard demodulation in a manner complementary to the modulation
provided by modulator 125.” Id. at 5:32–33, 5:43–45. “The output from
demodulator 140 is provided to decoder 145,” which may include “an
optional block decoder [that] is complementary to [optional] block encoder
110.” Id. at 5:45–48. Decoder 145 includes GVA decoder 150, which
“provides a maximum likelihood candidate for the data sequence actually
transmitted by data source 100,” including “an indication (or flag) to
indicate whether or not that candidate is significantly more likely than
another possible candidate available at the decoder.” Id. at 5:54–62. The
flag indicates if the decoded data sequence “is reliable . . . or unreliable” and
is “based on the measured likelihood of correctness for the most likely
sequence [through the trellis] as compared with the likelihood of the second
most likely sequence and, optionally, sequences of successively lower
likelihood.” Id. at 3:36–43.
Seshardi discloses both parallel and serial versions of the GVA. Id. at
6:26–27. In the parallel version, “identifying the L most likely candidates is
achieved in one pass through a [modified] trellis,” and in the serial version,
the L most likely candidates are obtained “using successive passes through a
[conventional] trellis structure.” Id. at 6:27–36. Figure 4 of Seshardi,
IPR2019-00054
Patent 6,304,612 B1

16
reproduced below, is a trellis decoder for decoding symbols encoded per
Seshardi’s GVA algorithm.

Figure 4 of Seshardi illustrates “the complete trellis diagram” for the
convolutional encoder illustrated in Figure 2. Id. at 4:16–17. “Because the
encoder of FIG. 2 is a binary convolutional encoder,” the trellis has four
rows representing the four possible trellis states S0 through S3 “associated
with the respective bit patterns 00, 01, 10, and 11.” Id. at 7:3–13.
Seshardi encodes an example 7-bit input or source sequence 0000000
using the encoder of Figure 2 to generate the 14-bit output or channel
sequence 00 00 00 00 00 00 00. Id. at 7:21–26. The Figure 4 trellis
illustrates the possible paths for decoding the output, all of which start at
trellis stage j = 0 and end at trellis stage j = 7 in known starting/ending state
S0 or 00. See id. at 7:19–21. For each node in the trellis, “[a]t each stage i
[sic j] the upper branch leaving a [trellis] state at time i [sic j] corresponds to
an [encoder] input of 0, while the lower branch corresponds to an input of
1.” Id. at 7:16–19. To decode the encoded sequence (i.e., 00 00 00 00 00
00), “the Viterbi algorithm calculates the log-likelihood function P(r|v)
IPR2019-00054
Patent 6,304,612 B1

17
known as the metric associated with the path v in the trellis.” Id. at 7:43–50.
This metric is calculated “as the sum of the individual branch metrics log
P(r|v) corresponding to the branches of the path v.” Id. at 7:51–53.
Seshardi discloses that in a conventional Viterbi decoder, “the metrics
of all paths entering each state [of the trellis] are compared, and the path
with the largest metric (the survivor) is stored, as is its metric.” Id. at 7:60–
66. By contrast, Seshardi’s parallel GVA decoder “find[s] the globally L
best decoded candidates” by finding and retaining “the L best candidates
into each [trellis] state at every level.” Id. at 8:30–33. Thus, to find “the
two best global paths” that traverse the entire trellis, “the two best paths into
each state at every trellis level” are retained. Id. at 8:34–38. Seshardi
discloses that regardless of the paths found and retained, the “trellis will
terminate in a known state [i.e., S0 at j = 7] and the two surviving paths that
terminate in that state are released as the most likely and the second most
likely candidates (with path metrics in that order).” Id. at 9:11–15.
In addition to finding the L best paths through the trellis and their
respective path metrics, Seshardi discloses performing “[e]rror detection . . .
using the GVA by comparing the difference in the metrics of the most likely
and the second most likely paths, and declaring the most likely path to be
unreliable if the difference is less than a preset threshold T.” Id. at 12:45–
49.
2. Chennakeshu
Chennakeshu discloses a “convolutional error-correcting code
combined with an error-detecting code to achieve improved digital
transmission results.” Ex. 1005, 2:14–18. Chennakeshu describes the
following data transmission method:
IPR2019-00054
Patent 6,304,612 B1

18
A frame of data to be transmitted is separated into key bits,
critical bits, and unprotected bits. Parity bits created from the key
bits, the key bits and critical bits are convolutionally encoded
using a tail-biting scheme, and are then merged with the
unprotected bits and transmitted to a receiver. The received data
is split into the unprotected bits and convolutionally encoded bits
which are trellis decoded into a predetermined number of paths,
each having a metric, a plurality of parity bits, and a plurality of
key bits. A path is chosen having the lowest metric and no errors
in the key bits. If no such path can be found, a fatal decoding
error message is created and a predetermined path is used. The
final path is decoded into digital information which is used in a
digital device such as a speech synthesizer.
Id. at 2:18–32.
Chennakeshu discloses that a conventional convolutional encoder
includes a shift register such that “the initial state, S(0), of the shift register
(level 0 of the trellis) is zero S(0) = 00.” Id. at 3:63–64. Moreover, after a
data word is sent, “a number of zero input bits are sent to clear the shift
registers.” Id. at 3:66–4:1. This “causes the state of the shift register to be
zero” and results in the final state of the trellis being “S(8) = 00.” Id. at 4:1–
5. In this way, “the initial and final states of the shift register are known in
conventional convolutional encoding before any decoding begins.” Id. at
4:7–9. Thus, to recover the sent data using a trellis, all that is needed “are
the intervening convolutional encoder shift register states S(k) where k = 2
[through] 7.” Id. at 4:10–13.
Chennakeshu further discloses that “[t]he decoder of choice for
convolutional decoding is the Viterbi algorithm (VA),” which performs “an
efficient search of the best path in the trellis of all possible transmitted
sequences.” Id. at 4:50–53. Moreover, “[t]he generalized Viterbi algorithm
(GVA) enhances the standard Viterbi algorithm by utilizing the second best
IPR2019-00054
Patent 6,304,612 B1

19
path, third best path, etc., to decode the convolutionally encoded bits.” Id. at
4:64–67. In the Viterbi algorithm, bit errors are estimated by comparing
received bits to trellis state transition bits, and “[t]he estimated bit errors,
called a metric, are saved for a set of transitions through the trellis.” Id. at
4:21–34. “The set of these transitions and associated metrics is called a
‘path,’” and “[t]he path having the lowest metric and the correct initial and
final states signifies the correct path.” Id. at 4:34–36, 4:47–49.
Chennakeshu discloses that “[t]he bits used to zero the convolutional
shift register after a data word is sent are called tail bits” and “are a source of
inefficiency [that] can be eliminated by using tail-biting.” Id. at 5:6–9. In a
tail-biting convolutional encoder, L bits of data are encoded by initializing
“the shift register . . . with the first K–1 information bits,” encoding bits K
through L in a conventional manner, and then encoding the first K–1 bits.
Id. at 5:17–24. Doing so “returns the shift register to the same state as
before.” Id. at 5:27. Although tail-biting eliminates the inefficiency of
padding the encoded data with zero bits to ensure the shift register’s return
to the zero state, it increases the complexity of the decoder because “the
initial state of the trellis is not known when tail-biting is used.” Id. at 5:56–
57. Thus, the trellis must be “searched for each possible starting state using
the standard Viterbi algorithm.” Id. at 5:57–59.
Chennakeshu discloses two algorithms for decoding L bits of data
encoded using a tail-biting convolutional encoder, noting that a tail-biting
encoder “treat[s] the information bits as an infinite repetition of the same
block of bits,” and thus results in a convolutionally encoded “output [that]
must also be periodic.” Id. at 6:48–52. Thus, “the initial state of the trellis
IPR2019-00054
Patent 6,304,612 B1

20
will be the same as the final state,” as in a conventional decoder, but will
represent the initial bits of data rather than the zero state. Id. at 6:1–5.
Chennakeshu’s first algorithm searches the trellis “for the path with
the best metric” and “backtrack[s] the best path from state S(L – 1)” to state
S(0). Id. at 6:20–24. Due to the expected periodicity in trellis states when a
tail-biting encoder is used, the initial state of the trellis, S(0), is expected to
“also be the final state of the encoder, i.e., S(0) = S(L–1) for an error free
channel.” Id. at 6:26–28. If S(0) does not equal S(L–1), Chennakeshu
knows an error has occurred, assumes S(0) is the correct initial trellis state,
and uses it “to produce the first few information bits” along the best path.
Id. at 6:29–32. Subsequently, a trellis path ending in a state S(0) at trellis
stage L–1 is backtracked “[t]o determine the remaining information bits.”
Id. at 6:32–36.
Chennakeshu recognizes two problems with this algorithm, namely,
“the last few decoded bits are not very reliable,” and the “estimate of the
initial bits of the sequence [are] rather poor.” Id. at 6:41–47. Therefore,
Chennakeshu proposes a second algorithm that extends the trellis “from the
usual L levels to L+E levels.” Id. at 6:52–53. “The trellis is searched for
L+E levels to find the best path,” and then backtracked L levels to trellis
level E. Id. at 6:56–57. “The trellis state at that point [i.e., S(E)] is an
estimate of the state at level L+E,” and is used to determine “what some of
the information bits were.” Id. at 6:58–60. It is also used “as the final state”
(i.e., S(L+E) is set to S(E)), and the trellis is “backtrack[ed] L levels again to
tell us what the remaining [information] bits were.” Id. at 6:60–62.
IPR2019-00054
Patent 6,304,612 B1

21
F. Patentability of Claims 1–11 over Seshardi and Chennakeshu
Petitioner argues claims 1–11 are unpatentable as obvious over the
combination of Seshardi and Chennakeshu. See Pet. 16–64; Pet. Reply 2–
28. Patent Owner disagrees. See PO Resp. 3–24; PO Sur-Reply 2–15. For
the reasons discussed below, we find Petitioner has demonstrated by a
preponderance of evidence that claims 1–11 are unpatentable over Seshardi
and Chennakeshu.
1. Reasons to Combine Seshardi and Chennakeshu
Petitioner argues Seshardi and Chennakeshu are from the same field
of endeavor, namely, “decoding convolutionally encoded data.” Pet. 59
(citing Ex. 1004, 3:19–4:6; Ex. 1005, 2:14–33). Moreover, Petitioner
argues, a person of ordinary skill in the art would have combined the
references by modifying Seshardi’s “data transmission system” to include
Chennakeshu’s transmitter and receiver because Seshardi “recognizes that a
transmitter is an expected component of the system” and a skilled artisan
would have known that “decoding is performed at a receiver that receives
the transmitted data.” Id. (citing Ex. 1004, code (57); Ex. 1003 ¶¶ 153–154).
Petitioner further argues that “Seshardi’s express teachings of transmitting
data and receiving data . . . provides an express teaching, suggestion, or
motivation that would have led a [person skilled in the art] to include in the
Seshardi system a transmitter and receiver” as taught by Chennakeshu. Id.
at 61 (citing Ex. 1003 ¶ 153). Lastly, Petitioner argues that a skilled artisan
would have had a reasonable expectation of success modifying Seshardi’s
system to include Chennakeshu’s transmitter and receiver because these “are
well understood components that would be easily integrated . . . without
IPR2019-00054
Patent 6,304,612 B1

22
undue effort or special expertise.” Id. at 60–61 (citing Ex. 1003 ¶¶ 155–
156).
Petitioner further argues that a person skilled in the art would have
modified Seshardi’s convolutional encoder/decoder to implement the
tailbiting encoding/decoding technique of Chennakeshu because “Seshardi
recognizes that zero-valued ‘tail bits’ are commonly added to the source
input block in a convolutional encoder” and “Chennakeshu recognizes that
these tail bits create inefficiency that can be eliminated via tailbiting.”
Pet. 61–62 (citing Ex. 1004, 7:22–26; Ex. 1005, 5:6–32; Ex. 1003 ¶ 157).
Petitioner argues that “Chennakeshu includes an express teaching,
suggestion, or motivation” to modify Seshardi’s encoder/decoder to perform
tail-biting because “Seshardi teaches a decoder using the GVA,” and
“Chennakeshu states ‘[t]he present invention combines the tail-biting
scheme . . . with the generalized Viterbi algorithm [GVA].” Id. at 63–64
(citing Ex. 1004, 5:8; quoting Ex. 1005, 6:65–67) (first alteration in
original).6 Lastly, Petitioner argues that a skilled artisan would have had a
reasonable expectation of success modifying Seshardi’s encoder/decoder to
use tail-biting because doing so “would entail simple and well-known
programming of the convolutional encoder, with the result of achieving the
same benefits as the tailbiting process employed in Chennakeshu.” Id. at
62–63 (citing Ex. 1003 ¶ 159).
Patent Owner argues that Petitioner has failed to articulate sufficient
reasoning to explain why a person skilled in the art would have combined

6 Petitioner mistakenly refers to Chennakeshu as Exhibit 1004. We correct
that error here, as we did in our Institution Decision. See Dec. Inst. 27 n.4.
IPR2019-00054
Patent 6,304,612 B1

23
the teachings of Seshardi and Chennakeshu. See PO Resp. 17–19; PO Sur-
Reply 9–11. First, Patent Owner argues, Petitioner has failed to provide any
reasoning that is “specifically directed to the claim [language]” requiring the
channel decoder to determine “the final state sequence” and “the difference
measures.”7 PO Resp. 18; PO Sur-Reply 10. Second, Patent Owner argues,
“the proposed modification of Chennakeshu based on Seshardi would
impermissibly change [the] basic principles under which Chennakeshu is
designed to operate.” PO Resp. 18–19. Specifically, Patent Owner argues
that Chennakeshu selects a best path “having the lowest metric,” Seshardi
selects a best path “corresponding to the largest accumulated metric,” and
“Petitioner fails to explain how modifying Chennakeshu allegedly to make a
final determination based [on] an entirely different metric would not change
the basic principles under which Chennakeshu is designed to operate.” Id. at
19 (quoting Ex. 1004, 12:55–61; citing Ex. 1005, 4:33–39, 8:1–3).
Petitioner responds that Patent Owner’s arguments are inapposite
because they are “directed to a ground not present in the Petition.”
Pet. Reply 20. Petitioner argues that the Petition relies on “Seshardi as the
primary reference and Chennakeshu as the secondary reference” and states
that “it would have been obvious to modify Seshardi with Chennakeshu[]” by
“combining Chennakeshu’s teaching of a receiver and transmitter with
Seshardi’s system” and “using Chennakeshu’s tailbiting techniques to
improve Seshardi’s convolutional encoder.” Id. at 20–21 (citing Pet. 7, 59–
64). Thus, Petitioner argues, the Petition only relies on Chennakeshu for
teaching a transmitter/receiver and a tailbiting encoder/decoder rather than

7 See n.5, supra.
IPR2019-00054
Patent 6,304,612 B1

24
the path metric used to select a best path. Id. at 21 (citing Pet. 14, 59–64).
Petitioner further argues that the Petition relies on Seshardi alone for
teaching a channel decoder that determines the “final state sequence” and
“difference measures” and, therefore, does not need to provide a reason to
combine the teachings of Chennakeshu with Seshardi with respect to these
limitations.8 Id.
Upon consideration of the evidence and arguments presented, as
summarized above, we are persuaded for the following reasons that
Petitioner has articulated sufficient reasoning with a rational underpinning to
modify Seshardi based on the teachings of Chennakeshu in the manner
Petitioner proposes.
First, the Petition suggests modifying Seshardi’s system to include
Chennakeshu’s transmitter, receiver, and tail-biting encoding/decoding
method. See Pet. 59 (“combining Chennakeshu’s teaching of a receiver and
transmitter with Seshardi”); see also id. at 61 (“include in the Seshardi
system a transmitter and receiver”) (“use Chennakeshu’s known techniques
relative to tailbiting to improve Seshardhi’s convolutional encoder”). Thus,
the Petition relies on Seshardi, not Chennakeshu, for defining the path metric
and selecting the path having the highest path metric as the final state
sequence. See Pet. 28–30 (citing Ex. 1004, 3:34–36, 5:56–59, 7:60–63,
9:10–16, 12:55–61). For this reason, we are not persuaded by Patent
Owner’s argument that Petitioner’s proposed combination is improper
because Chennakeshu’s decoder would not work as intended.9

8 See n.5, supra.
9 We also note that Chennakeshu uses a path metric that counts “bit errors”
and selects the path “having the lowest metric” or fewest bit errors.
IPR2019-00054
Patent 6,304,612 B1

25
Second, Petitioner relies on Seshardi for teaching the “final state
sequence” and “difference measure” limitations. See Pet. 28–31. Thus, we
are not persuaded by Patent Owner’s argument that Petitioner has failed to
provide sufficient reasoning to combine the teachings of Seshardi and
Chennakeshu because they have failed to provide “any argument specifically
directed to the claim [language] recited in [the final state sequence] and
[difference measure limitations].”10 PO Resp. 18; see, e.g., KSR Int’l Co. v.
Teleflex Inc., 550 U.S. 398, 418–19 (2007) (finding an obviousness analysis
“need not seek out precise teachings directed to the specific subject matter of
the challenged claims” and that “neither the particular motivation nor the
avowed purpose of the patentee controls”).
Third, and finally, we find Petitioner’s reasoning for combining the
teachings of Seshardi and Chennakeshu has a rational underpinning. See
Pet. 59–64. Seshardi discloses “a system for communicating information
from a data source 100 to a receiving location through a communication

Ex. 1005, 4:33–37, 4:46–49. By contrast, Seshardi uses a path metric that
counts bits that are correctly received. This is evident from Seshardi’s
formula for the incremental path metric ∆λ = ri1 ∙ xi1 + ri2 ∙ xi2. Ex. 1004,
8:63. For S0 to S0 state transitions, the expected symbols xi1 and xi2 are both
+1. Id. at 6:58–60, 8:65–9:2. This reduces the path metric to ∆λ = ri1 + ri2.
Id. at 8:64. When the received symbols ri1 and ri2 are the same as the
expected symbols (i.e., ri1 = ri2 = +1) the path metric increases by ∆λ = +2.
When one of the received symbols differs from an expected symbol (e.g., ri1
= +1 and ri2 = –1) the path metric remains the same because ∆λ = 0. When
both received symbols differ from the expected symbols (i.e., ri1 = ri2 = –1)
the path metric decreases by ∆λ = –2. Thus, when Seshardi selects the path
with the largest path metric, Seshardi selects the path with the most correctly
received bits, i.e., the path with the fewest bit errors, just like Chennakeshu.
10 See n.5, supra.
IPR2019-00054
Patent 6,304,612 B1

26
channel 130.” Ex. 1004, 5:12–14, Fig. 1. The system includes a two-stage
convolutional encoder 120 and a GVA decoder 145. Id. at 5:26–31, 5:45–
46, Figs. 1, 2. Encoder 120 “starts with state S0 and returns to S0,” and
includes “M = 2 zero-valued ‘tail’ bits to complete [a] frame.” Id. at 7:3–8,
7:21–26, Figs. 1–3. GVA decoder 145 uses a trellis to decode received
symbols that “terminate[s] in a known state [S0] and the two surviving paths
that terminate in that state are released as the most likely [best] and second
most likely [second best] candidate [paths].” Id. at 7:14–21, 9:10–16, Figs.
1, 4.
Chennakeshu teaches an “apparatus for digital radio communications”
that “employs [a] convolutional error-correcting code combined with an
error-detecting code.” Ex. 1005, 2:14–16. The apparatus includes
convolutional encoder 34, transmitter 38, and receiver 40 that includes
convolutional decoder 52. Id. at 7:23–8:27, Figs. 5, 6. Chennakeshu teaches
the tail bits used to initialize or zero conventional convolutional encoders,
such as Seshardi’s encoder 120, “are a source of inefficiency which can be
eliminated by using tail-biting.” Id. at 5:6–9. Therefore, Chennakeshu
proposes “a tail-biting scheme” in which the convolutional encoder’s “shift
register is initialized with the first K–1 information bits” rather than with all
zeros. Id. at 5:17–19, Fig. 3a. This increases the complexity of the decoder,
however, because “the initial state of the trellis is not known when tail-biting
is used.” Id. at 5:56–57. Nonetheless, due to the “periodic nature of the
coded stream,” a tail-biting decoder can be constructed by “extend[ing] the
trellis from the usual L levels to L + E levels” to find the best path through
IPR2019-00054
Patent 6,304,612 B1

27
the trellis.11 Id. at 6:48–57. Chennakeshu further teaches “combining [this]
tail-biting scheme . . . with the generalized Viterbi algorithm,” i.e., with the
GVA decoding algorithm disclosed by Seshardi. Id. at 6:65–67. That is,
Chennakeshu expressly suggests Petitioner’s proposed combination.
2. Claim 1
Claim 1 recites a transmission system, and requires the system to have
a transmitter that (a) includes a channel encoder to encode sequences of
source symbols into sequences of channel symbols, and that (b) transmits a
channel signal representing the channel symbols to a receiver. Ex. 1001,
13:9–13.
Petitioner demonstrates by a preponderance of evidence that Seshardi
discloses this limitation. See Pet. 16–21. Seshardi discloses “a system for
communicating information from a data source 100 to a receiving location
through a communication channel 130.” Ex. 1004, 5:12–17. The system
“cod[es] and decod[es] digital information for transmission over [the]
communication channel” using “parallel- and serial-generalized Viterbi
decoding algorithms (GVA) that produce a ranked ordered list of the L best
candidates after a trellis search.” Id. at 1:9–11, code (57). Seshardi’s
system includes convolutional encoder 120, which encodes symbols from
data source 100 (a sequence of source symbols) into convolutionally
encoded data symbols (a sequence of channel symbols), and transmits the
channel symbols over channel 130 to a receiving location (receiver). Id. at
5:12–14, 6:46–7:2, 7:43–44, Figs. 1, 2. Patent Owner does not dispute

11 Here, “L” is the length of the message to be encoded and “E” is a periodic
extension (e.g., 3K or 4K) related to the number of information bits “K–1”
that are used to initialize the encoder. See Ex. 1005, 5:17–24, 6:53–55.
IPR2019-00054
Patent 6,304,612 B1

28
Petitioner’s contentions regarding this limitation. See PO Resp. 12–17; PO
Sur-Reply 6–9.
Petitioner further demonstrates by a preponderance of evidence that
Chennakeshu discloses the transmitter and encoder limitations of claim 1 to
the extent Seshardi does not. See Pet. 21–23. Chennakeshu discloses a
convolutional encoder 34 that encodes data from a speech encoder 3 (a
sequence of source symbols) to generate convolutionally encoded speech
data (a sequence of channel symbols) that is transmitted by digital
transmitter 38 to receiver 40 as RF signal 7 (a channel signal). Ex. 1005,
7:23–35, 7:42–53, Figs. 5, 6. Patent Owner does not dispute these
contentions, either. See PO Resp. 12–17; PO Sur-Reply 6–9.
Claim 1 further requires the system to have a receiver that includes a
channel decoder to derive the sequences of source symbols from the channel
signal by keeping track of a plurality of state sequences with corresponding
likelihood measures. Ex. 1001, 13:14–19. Relying on the testimony of Dr.
Creusere, Petitioner argues that a person skill in the art would have readily
recognized that the ’612 patent supports this limitation by describing “the
well-known estimation of the best path, i.e., the most likely sequence of
source symbols input into the convolutional encoder, using the Viterbi
algorithm.” Pet. 24 (citing Ex. 1003 ¶ 75). Patent Owner does not dispute
this contention. See PO Resp. 12–17; PO Sur-Reply 6–9.
Petitioner demonstrates by a preponderance of evidence how Seshardi
teaches using the Viterbi algorithm to derive sequences of source symbols
from the channel signal by keeping track of a plurality of state sequences
with corresponding likelihood measures. See Pet. 25–28. Seshardi discloses
a receiving location (receiver) that includes decoder 150 executing the
IPR2019-00054
Patent 6,304,612 B1

29
generalized Viterbi algorithm (GVA) on a received signal to identify “a
maximum likelihood candidate for the data sequence actually transmitted by
data source 100.” Ex. 1004, 5:12–14, 5:54–62, Fig. 1. GVA decoder 150
identifies the maximum likelihood candidate by calculating a path metric for
each path v through a trellis (i.e., a plurality of candidate sequences with
corresponding likelihood measures), and selecting the path having the
highest path metric. Id. at 5:8–6:4, 7:14–16, 7:43–67, 8:52–9:16, Fig. 4.
Patent Owner does not dispute Petitioner’s contentions regarding this
limitation. See PO Resp. 12–17; PO Sur-Reply 6–9.
Clam 1 further requires the channel decoder to (a) determine the final
state sequence by comparing the likelihood measures of a plurality of
candidate state sequences terminating in said state and selecting the state
sequence with the largest likelihood measure, and (b) further determine the
respective difference measures between the likelihood measure of the
selected candidate sequence and the likelihood measures of the rejected
candidate sequences.12 Ex. 1001, 13:20–28.
Notwithstanding Patent Owner’s arguments to the contrary, discussed
infra, Petitioner demonstrates by a preponderance of evidence how Seshardi
teaches these limitations. See Pet. 28–31. Seshardi teaches GVA decoder
150 “releases the maximum likelihood (ML) sequence” and selects “a

12 Petitioner correctly notes that the term “said state” appearing in this
limitation lacks antecedent basis. See Pet. 28; see also Ex. 1001, 13:20–23.
Thus, we interpret “said state” to mean “a state” in the trellis, including a
state in the final stage of the trellis where a final state sequence is chosen.
We made the same interpretation in our Institution Decision, and neither
party disputes that interpretation. See Dec. Inst. 33 n.5; PO Response 4–12;
Pet. Reply 2–28.
IPR2019-00054
Patent 6,304,612 B1

30
maximum likelihood candidate for the data sequence actually transmitted by
data source 100.” Ex. 1004, 3:34–36, 5:56–59. Seshardi does this by
traversing a trellis in such a way that at each state (e.g., S0) and each stage j
of the trellis:
Out of the four extensions, the two paths with the highest
accumulated metrics are chosen and retained for further
extension from S0(j). Similar calculations are performed at all
other states. Finally, the trellis will terminate in a known state
and the two surviving paths that terminate in that state are
released as the most likely and the second most likely
candidates (with path metrics in that order). They are called the
best and second best path.
Id. at 9:8–16. Seshardi teaches “[t]he path corresponding to the largest
accumulated metric value is . . . the most likely path, also referred to as the
best path.” Id. at 12:55–58. Similarly, “[t]he path with the second largest
metric is called the second best path” and “paths with progressively smaller
metrics are referred to as the third best, fourth best, etc.” Id. at 12:58–61.
Thus, Seshardi teaches selecting the path having the largest path metric from
among a plurality of paths at the final stage of the trellis as the “best path” or
“final state sequence” as required by claim 1.
Seshardi further teaches determining difference measures between the
candidate sequence selected to be the best path and the rejected candidate
sequences (e.g., second, third, and fourth best paths) “by comparing the
difference in the metrics of the most likely and the second most likely paths”
or between “the likelihood of correctness for the most likely sequence as
compared with the likelihood of the second most likely sequence and,
optionally, sequences of successively lower likelihood.” Id. at 3:33–43,
12:45–49.
IPR2019-00054
Patent 6,304,612 B1

31
Patent Owner argues Petitioner “fails to prove that ‘Seshardi
calculates difference measures before selecting or rejecting a candidate
sequence as the final or best candidate sequence.’” PO Resp. 17. Patent
Owner argues this is so because Seshardi discloses “comparing the
difference in the metrics of the most likely and the second most likely paths,
and declaring the most likely path to be unreliable if the difference is less
than a preset threshold.” Id. at 13 (quoting Ex. 1004, 12:45–49) (emphases
omitted). Thus, according to Patent Owner, “the ‘most likely path’ . . . is not
a previously selected final state sequence” because “it is still subject to
rejection due to error control.” Id. at 14. As a result, Patent Owner argues,
Seshardi determines the difference measures between candidate state
sequences “before completing any alleged selection/rejection of candidate
state sequences” as the final state sequence. Id. at 13.
We are not persuaded by Patent Owner’s arguments for several
reasons. First, according to claim 1, the “final state sequence” is the
sequence having “the largest likelihood measure” that is selected at the
terminating stage of the trellis.13 Ex. 1001, 13:20–24. Seshardi’s “best”
sequence is, therefore, “a final state sequence” because it is the candidate
sequence having the largest likelihood measure selected at the terminating
trellis stage. See Ex. 1004, 9:11–15 (disclosing “the trellis will terminate in
a known state and the two surviving paths that terminate in that state are
. . . the most likely and the second most likely candidates”), id. at 12:55–58
(“The path corresponding to the largest accumulated metric value is . . . the
most likely path, also referred to as the best path.”).

13 See n.12, supra.
IPR2019-00054
Patent 6,304,612 B1

32
Second, claim 1 is open-ended, and recites a system that “comprises”
a channel decoder that determines a “final state sequence” and a “difference
measure” that measures the reliability of the final state sequence. See
Ex. 1001, 13:9–33. Nothing in claim 1 requires the system to use the “final
state sequence” for any purpose, and nothing in claim 1 prohibits the system
from rejecting the “final state sequence” if the “difference measure”
indicates it is unreliable. Id. Seshardi’s “best” sequence is a “final state
sequence” for the reason discussed above—it is the sequence selected at the
terminating trellis stage having the largest likelihood measure. See
Ex. 1004, 9:11–15, 12:55–58. That fact remains true of the “best” sequence
even if it is subsequently rejected in favor of the “second best” or “third
best” sequence because it has been flagged as unreliable. As noted above,
nothing in open-ended claim 1 prohibits the system from rejecting the “final
state sequence” in favor of an alternative state sequence.
Third, even if we were to agree with Patent Owner that the “best”
sequence is not “the final state sequence” when it is rejected for the “second
best” sequence,14 Seshardi does not always reject the “best” sequence as
unreliable. Rather, Seshardi only rejects the “best” sequence as unreliable
when it is flagged because the difference measure between the “best” and

14 An argument we categorically reject for the reasons explained, supra.
Seshardi’s “best” sequence is by definition “the final state sequence”
because it is the sequence having the largest path metric at the last trellis
stage. Rejecting this sequence for the “second best” sequence would not
make the “second best” sequence the “final state sequence” because the
“second best” sequence does not have the largest path metric at the last
trellis stage. Only the “best” sequence has the largest path metric. And this
is true regardless of whether the reliability flag is set or not set.
IPR2019-00054
Patent 6,304,612 B1

33
“second best” sequence is less than a threshold. See, e.g., id. at 6:5–7
(“When the test indicated at block 160 yields a ‘no flag set’ result, the
candidate state sequence (the decoded frame of speech information) is
accepted.”), 6:14–17 (“In like manner, when non-speech information is
being processed and no flag is set when the test indicated by block 185 is
performed, then the frame of non-speech information is accepted.”), Fig. 1.
Thus, using Patent Owner’s own logic, the “best” sequence is “the final state
sequence” at least during those times when it is not rejected as unreliable.
This is sufficient to demonstrate that Seshardi teaches this limitation. See
Hewlett-Packard Co. v. Mustek Systems, Inc., 340 F.3d 1314, 1326 (Fed.
Cir. 2003) (finding “a prior art product that sometimes, but not always,
embodies a claimed method nonetheless teaches that aspect of the
invention.”); see also Unwired Planet, LLC v. Google Inc., 841 F.3d 995,
1002 (Fed. Cir. 2016) ("It is enough that the combination would sometimes
perform all the method steps. . . . Because the [prior art method] would
sometimes meet the . . . claim[] elements, the Board was correct to conclude
that the proposed combination taught all of the elements of claim 1.").
Finally, Seshardi teaches determining the reliability of the sequence
selected as the “best” sequence by determining difference measures between
the “best” and “second best” sequences after selecting the “best” and
“second best” sequences. Thus, Seshardi teaches determining the “final
state sequence” and the “difference measures” in the order in which those
limitations are recited in claim 1 (assuming that order is required by the
claims, as Patent Owner contends).
For all of the reasons discussed above, Petitioner has demonstrated by
a preponderance of evidence that Seshardi teaches GVA decoder 150
IPR2019-00054
Patent 6,304,612 B1

34
“determines the final state sequence” and “further determines the respective
difference measures” limitations of claim 1 in that order.
Lastly, claim 1 requires the minimum difference of likelihood
measures determined between the selected and rejected candidate sequences
be a criterion for measuring the reliability of the final state sequence.
Ex. 1001, 13:29–33.
Petitioner demonstrates by a preponderance of evidence how Seshardi
teaches this limitation. See Pet. 32–34. Seshardi discloses calculating a
reliability measure “based on the measured likelihood of correctness for the
most likely sequence as compared with the likelihood of the second most
likely sequence and, optionally, sequences of successively lower likelihood,”
and declaring the most likely sequence unreliable when “the difference in
the metrics of the most likely and the second most likely paths . . . is less
than a preset threshold T.” Ex. 1004, 3:36–43, 5:59–62, 12:45–49.
According to the unrebutted testimony of Dr. Creusere, a person of
ordinary skill in the art would have known that the difference measure
between Seshardi’s most likely and second most likely paths is the minimum
difference of likelihood measures between the selected (“best”) and rejected
(“second best,” “third best,” etc.) candidate state sequences because “[a]ny
lesser paths are going to have path metrics smaller than the best or second
best paths. Therefore, the smallest or minimum difference of likelihood
measures between the best [selected] path and any other [rejected] path is
the difference between the best path and the second best path.” Ex. 1003
¶ 114. Patent Owner does not dispute these contentions. See PO Resp. 12–
17; PO Sur-Reply 6–9.
IPR2019-00054
Patent 6,304,612 B1

35
Accordingly, for the reasons explained above, we find Petitioner has
articulated sufficient reasoning with a rational underpinning to combine the
teachings of Seshardi and Chennakeshu, and has demonstrated by a
preponderance of evidence that the combination teaches all of the limitations
recited in claim 1. Therefore, Petitioner has demonstrated by a
preponderance of evidence that claim 1 is unpatentable over the combination
of Seshardi and Chennakeshu.
3. Claim 2
Claim 2 depends from claim 1, and further requires the channel
decoder to extend, on the basis of a cyclically extended channel signal, the
state sequence beyond a number of states equal to the number of source
symbols in a sequence of source symbols. Ex. 1001, 13:34–40. Claim 2
further requires the channel decoder to select the final state sequence on the
basis of the terminating states of the extended state sequences. Id.
Petitioner demonstrates that Chennakeshu teaches these limitations.
See Pet. 41–43. Chennakeshu discloses receiving a channel signal generated
by encoding L source symbols treated “as an infinite repetition of the same
block of bits,” and “extend[ing] the trellis from the usual L levels to L+E
levels,” where the “extra E levels are just a periodic extension of the parity
bits.” Ex. 1005, 5:19–30, 6:48–64. Chennakeshu discloses searching this
extended trellis “for L+E levels to find the best path” and using state S(L+E)
“as the final state and backtrack[ing] L levels . . . to tell us what the
remaining bits were.” Id. at 6:56–64.
Patent Owner does not dispute Petitioner’s contentions, but argues
that claim 2 is patentable for the same reasons as claim 1. See PO Resp. 20–
IPR2019-00054
Patent 6,304,612 B1

36
24. We are not persuaded by Patent Owner’s arguments for the reasons
discussed in § II.F.2, supra.
Accordingly, for the reasons discussed above, Petitioner has
demonstrated by a preponderance of evidence that claim 2 is unpatentable
over the combination of Seshardi and Chennakeshu.
4. Claim 3
Claim 3 depends from claim 2, and further requires the channel
decoder to determine a suitable state sequence when the state sequences
comprise a number of states equal to the number of source symbols in a
sequence of source symbols.15 Ex. 1001, 13:41–45. Petitioner demonstrates
that Chennakeshu teaches or suggests this limitation. See Pet. 44–45. For
example, Chennakeshu discloses encoding an L-bit source sequence (e.g.,
where L = 5), and traversing L trellis stages to identifying a state sequence
that recovers the L-bit source sequence. Ex. 1005, 5:22–25, 5:34–39, 6:6–
10, 6:20–28, Fig. 4.
Patent Owner does not dispute Petitioner’s contentions, but argues
that claim 3 is patentable for the same reasons as claim 1. See PO Resp. 20–
24. We are not persuaded by Patent Owner’s arguments for the reasons
discussed in § II.F.1, supra.

15 Claim 3 recites “determining the suitable state sequence.” Ex. 1001,
13:42–43 (emphasis added). Petitioner correctly notes that this phrase lacks
antecedent basis. See Pet. 44. The ’612 patent discloses “selecting a
suitable state sequence after the channel signal has been processed at least
once,” where “the suitable state sequence is preferably the most likely state
sequence. This state has a likelihood measure indicating a largest
likelihood.” Ex. 1001, 2:1–5. 2:18–20. Petitioner interprets claim 3
consistently with these disclosures, as do we.
IPR2019-00054
Patent 6,304,612 B1

37
Accordingly, for the reasons discussed above, Petitioner has
demonstrated by a preponderance of evidence that claim 3 is unpatentable
over the combination of Seshardi and Chennakeshu.
5. Claim 4
Claim 4 also depends from claim 2, and requires the channel decoder
to determine a suitable state sequence from terminating states of the
extended state sequence.16 Ex. 1001, 13:46–49. Petitioner demonstrates
how Chennakeshu teaches this limitation. See Pet. 46. Chennakeshu
discloses searching an extended trellis “for L+E levels to find the best path”
and using state S(L+E) “as the final state and backtrack[ing] L levels . . . to
tell us what the remaining bits were.” Ex. 1005, 6:56–64.
Patent Owner does not dispute Petitioner’s contentions, but argues
that claim 4 is patentable for the same reasons as claim 1. See PO Resp. 20–
24. We are not persuaded by Patent Owner’s arguments for the reasons
discussed in § II.F.2, supra.
Accordingly, for the reasons discussed above, Petitioner has
demonstrated by a preponderance of evidence that claim 4 is unpatentable
over the combination of Seshardi and Chennakeshu.

16 Claim 4 also recites “the suitable state sequence.” Ex. 1001, 13:47–48
(emphasis added); see also n.15, supra. As with claim 3, we interpret this
phrase to mean the state sequence selected after the channel signal has been
processed at least once having the largest likelihood measure. See n.15,
supra. Claim 4 also recites “the channel encoder,” despite depending from
claim 2 and reciting additional functionality of the channel decoder of claim
2. Ex. 1001, 13:46–47 (emphasis added). We treat this as a typographical
error, as Petitioner has, and interpret claim 4 as referring to the channel
decoder of claim 2. See Pet. 45.
IPR2019-00054
Patent 6,304,612 B1

38
6. Claim 5
Claim 5 depends from claim 1, and further requires the channel
decoder to keep track of the minimum difference measure for each of the
state sequences. Ex. 1001, 13:50–52.
Notwithstanding Patent Owner’s argument to the contrary, discussed
infra, Petitioner demonstrates how the combination of Seshardi and
Chennakeshu teaches this limitation. See Pet. 30–35, 47–48. For example,
Seshardi discloses GVA decoder 150 determines a best sequence that is a
maximum likelihood sequence and difference measures between the best
sequence and next best (second, third, fourth best) sequences. See Ex. 1004,
12:45–49 (disclosing “comparing the difference in the metrics of the most
likely and the second most likely paths”), 3:39–43 (disclosing the “reliability
determination is based on the measured likelihood correctness for the most
likely sequence as compared with the likelihood of the second most likely
sequence and, optionally, sequences of successively lower likelihood.”).
According to the unrebutted testimony of Dr. Creusere, a person skilled in
the art would have known that to determine and compare these difference
measures, Seshardi would have had to have kept track of them. Ex. 1003
¶ 134.
Moreover, Chennakeshu discloses a convolutional decoder that uses a
data buffer 54 to keep track of paths through the trellis and their associated
path metrics. Ex. 1005, 7:61–8:6. According to the unrebutted testimony of
Dr. Creusere, a person skilled in the art would have known that Seshardi’s
GVA decoder 150 would also have kept track of paths, path metrics, and
difference measures to accurately perform decoding. Ex. 1003 ¶ 135.
IPR2019-00054
Patent 6,304,612 B1

39
Patent Owner first argues claim 5 is patentable for the same reasons as
claim 1. See PO Resp. 20. We are not persuaded by this argument for the
reasons discussed in § II.F.2, supra.
Patent Owner next argues that claim 5 is patentable because “keeping
track of the minimum difference measure,” is “an additional requirement,
separate and apart from the requirement in claim 1 that the channel decoder
be configured to have previously determined the ‘minimum difference’
value.” PO Resp. 20. Patent Owner argues the Petition improperly treats
“the ‘keeping track’ requirement of claim 5” as reciting “nothing more than
what is already necessary ‘to accomplish the reliability determination.’” Id.
at 21 (quoting Pet. 48).
We are not persuaded by this argument. Petitioner treats the “keeping
track” limitation separately from the determining “difference measure”
limitation by relying on Dr. Creusere’s opinion that “it would have been
obvious to . . . keep[] track of the disclosed difference measures, as a
decoder keeping track of decoding values is known.” Pet. 48 (citing
Ex. 1003 ¶¶ 134–135). According to Dr. Creusere, a person skilled in the art
“would reasonably understand that [Seshardi’s] GVA decoder 150 is
keeping track of the disclosed difference measures or that such is otherwise
obvious.” Ex. 1003 ¶ 133. “[I]f the decoder is not ‘keeping track’ of the
difference measures,” Dr. Creusere opines, “there is not a practical way for
the decoder to perform the disclosed comparison of the difference
measures.” Id. ¶ 134.
We find Dr. Creusere’s opinion credible because it is supported by
Chennakeshu’s disclosure of “keeping track” of paths and path metrics by
storing them in a data buffer during a decoding process. See Ex. 1005, 7:61–
IPR2019-00054
Patent 6,304,612 B1

40
8:6 (teaching storing encoded bits in a “decoder buffer 51” and “process[ing]
the convolutionally encoded bits in convolutional decoder buffer 51 to
produce a number of paths with an associated . . . metric. The paths and
metrics are stored by the convolutional decoder in the path buffer 54.”). We
agree with Dr. Creusere that this teaching “supports [Dr. Creusere’s]
position that Seshardi teaches or it is otherwise obvious that the Seshardi
GVA decoder 150 keeps track of the difference measures in order to
accomplish [Seshardi’s] reliability determination.” Ex. 1003 ¶ 135.
Patent Owner next argues that Chennakeshu’s teaching to store paths
and path metrics in path buffer 54 does not cure the deficiency in Seshardi
because “claim 5 does not recite storing paths or metrics . . . in a path
buffer” but instead recites “keeping track of the minimum difference
measure for each of the state sequences.” PO Resp. 21.
We are not persuaded by this argument for the reasons explained
above. Petitioner does not rely on Chennakeshu for teaching keeping track
of the minimum difference measures limitation of claim 5. Rather,
Petitioner relies on Seshardi for teaching or suggesting this limitation
because Seshardi’s decoding process includes determining difference
measures, and Chennakeshu teaches that it was known to store “decoding
values . . . for accurately performing the decoding process.” Pet. 48.
“Nonobviousness cannot be established by attacking references individually
where the [challenge] is based upon the teachings of a combinations of
references.” In re Merck & Co., 800 F.2d 1091, 1097 (Fed. Cir. 1986).
Finally, Patent Owner argues that Petitioner “fails to point to any
teaching in Seshardi, or any statement in the supporting Declaration, that the
minimum difference measure is tracked for each of the state sequences.” PO
IPR2019-00054
Patent 6,304,612 B1

41
Sur-Reply 11–12 (emphasis omitted). We disagree. Petitioner states
“Seshardi further teaches determining difference measures for each of the
state sequences,” and a person skilled in the art would have found it obvious
that “the Seshardi GVA decoder . . . keeps track of the disclosed difference
measures.” Pet. 47. Dr. Creusere testifies that “Seshardi teaches that its
GVA decoder 150 keeps track of the difference measures for each state
sequence,” in part because “if the decoder is not ‘keeping track’ of the
difference measures, then there is no practical way for the decoder to
perform the disclosed comparison of the difference measures.” Ex. 1003
¶¶ 133–134.
Accordingly, for the reasons discussed above, Petitioner has
demonstrated by a preponderance of evidence that claim 5 is unpatentable
over the combination of Seshardi and Chennakeshu.
7. Claim 6
Claim 6 depends from claim 1, and further requires the channel
decoder to select a suitable state sequence after the channel signal has been
processed at least once, to determine from said suitable state sequence an
earlier state in said suitable state sequence, and to select as the final state
sequence the state sequence terminating in a state corresponding to said
earlier state. Ex. 1001, 13:53–60.
Notwithstanding Patent Owner’s arguments to the contrary, discussed
infra, Petitioner demonstrates how the combination of Seshardi and
Chennakeshu teaches these limitations. See Pet. 51–52. Chennakeshu
teaches decoding L information bits by “extend[ing] the trellis from the
usual L levels to L+E levels,” and searching the trellis “for L+E levels to
find the best path” that ends at state S(L+E) (selecting a suitable state
IPR2019-00054
Patent 6,304,612 B1

42
sequence after processing the channel signal at least once). Ex. 1005, 6:52–
57. Chennakeshu then backtracks L levels along this best path to trellis level
E to find “[t]he trellis state at that point [which] is an estimate of the state at
level L+E” (determining an earlier state from said suitable state sequence).
Id. at 6:57–59. Chennakeshu then uses this earlier state (i.e., state S(E) at
trellis level E) “as the final state [S(L+E)] and backtrack[s] L levels again to
tell us what the remaining bits were” (select as the final state sequence the
state sequence terminating in a state corresponding to said earlier state). Id.
at 6:60–62.
Patent Owner first argues claim 6 is patentable for the same reasons as
claim 1. See PO Resp. 20. We are not persuaded by this argument for the
reasons discussed in § II.F.2, supra.
Patent Owner next argues that Petitioner’s analysis of claim 6 is
deficient because it “fails to recognize that claim 6 requires two distinct
selections, with the latter ‘selecting’ expressly referring back to the
antecedent ‘selecting’ and ‘determining’ requirements.” PO Resp. 23.
Petitioner responds that its analysis recognizes two distinct selecting steps,
namely, by mapping the first step of selecting a suitable state sequence to
“Chennakeshu’s teaching of searching the trellis for L+E levels to find the
best path” and mapping the second step of selecting a final state sequence to
Chennakeshu’s teaching of “determining an earlier state by backtracking L
levels” and “selecting as the final state sequence the state sequence
terminating in a state correspond to the earlier state, i.e., S(E).” Pet. Reply
25–26 (citing Pet. 51–52; Ex. 1005, 6:52–64). We agree with Petitioner, and
are not persuaded by Patent Owner’s argument for the reasons stated in
Petitioner’s Reply.
IPR2019-00054
Patent 6,304,612 B1

43
Finally, Patent Owner argues that Petitioner’s analysis of claim 6 is
deficient because Chennakeshu teaches that “only ‘some’ of the information
bits are identified” at trellis state S(E), and this “confirms that the alleged
‘selecting a suitable state sequence’ is not performed ‘after the channel
signal has been processed at least once.” PO Resp. 23. Petitioner responds
that it mapped “processing the channel signal at least once” to
Chennakeshu’s disclosure of extending a received sequence “along the trellis
to a state S(L+E).” Pet. Reply 27 (citing Pet. 51; Ex. 1005, 6:52–60).
We agree that the Petition maps “processing the channel signal at least
once” to Chennakeshu’s teaching to extend the trellis to L+E levels in order
to find the best path, and that this teaching reads on the claimed limitation.
Chennakeshu teaches encoding L bits of data by placing “L information bits
in a circular buffer” and “treat[ing] the information bits as an infinite
repetition of the same block of bits.” Ex. 1005, 5:28–30, 6:49–51.
Chennakeshu then teaches “extend[ing] the trellis from the usual L levels to
L+E levels” and searching the trellis “for L+E levels to find the best path.”
Id. at 6:51–57. Petitioner identifies Chennakeshu’s selection of the “best
path” at trellis level L+E as the selection of a “suitable state sequence.” See
Pet. 51–52. Because this “best path” is selected after an L+E level trellis
search to recover the L information bits encoded in the channel signal, it
reads on the claimed limitation of “selecting a suitable state sequence after
the channel signal has been processed at least once.”
Accordingly, for the reasons discussed above, Petitioner has
demonstrated by a preponderance of evidence that claim 6 is unpatentable
over the combination of Seshardi and Chennakeshu.
IPR2019-00054
Patent 6,304,612 B1

44
8. Claims 7 and 8
As discussed in § II.B, supra, claim 7 is an independent claim that
recites a receiver that includes the channel decoder of the transmission
system recited in claim 1. Consequently, Petitioner’s analysis of claim 7
largely relies on its analysis of claim 1 to demonstrate how the combination
of Seshardi and Chennakeshu teach each of the limitations recited in claim 7.
See Pet. 53–55. Patent Owner does not argue for the patentability of claim
7. See PO Resp. 12–24.
Claim 8, depends from claim 7, and requires the channel decoder in
the receiver of claim 7 to perform the same selection process performed by
the channel decoder in the receiver of the transmission system recited in
claim 6. Compare Ex. 1001, 14:17–24; with id. at 13:53–60. Petitioner
relies on its analysis of claim 6 to demonstrate how Chennakeshu teaches the
limitations of claim 8. See Pet. 55. Although Patent Owner argues that
“each of the challenged dependent claims depend from one of the challenged
independent claims” and is “tainted by the same deficiencies addressed
above,” this argument is not an argument for the patentability of dependent
claim 8 because Patent Owner does not argue for the patentability of
independent claim 7. See PO Resp. 20; see also id. at 12–19.
Accordingly, for the reasons discussed above,17 Petitioner has
demonstrated by a preponderance of evidence that claims 7 and 8 are
unpatentable over the combination of Seshardi and Chennakeshu.

17 Assuming Patent Owner meant to argue that claims 7 and 8 are patentable
for the same reasons as claim 1, and claim 8 is patentable for the same
reasons as claim 6, we are not persuaded for the reasons discussed in
§§ II.F.1, II.F.2, and II.F.7, supra.
IPR2019-00054
Patent 6,304,612 B1

45
9. Claims 9–11
As discussed in § II.B, supra, claim 9 is an independent claim that
recites the channel decoder in the receiver of the transmission system recited
in claim 1, claim 10 is an independent claim that recites a method for
transmitting a signal using the transmission system recited in claim 1, and
claim 11 is an independent claim that recites a method for decoding the
signal transmitted by the transmission system of claim 1. Consequently,
Petitioner’s analysis of claims 9–11 largely relies on its analysis of claim 1
to demonstrate how the combination of Seshardi and Chennakeshu teach
each of the limitations recited in claims 9–11. See Pet. 55–58.
Patent Owner, recognizing that “[t]he Petition relies exclusively on its
challenge to independent claim 1 in further challenging independent claims
9–11,” argues that claims 9–11 are patentable for the same reasons as claim
1. See PO Resp. 20. We are not persuaded by these arguments for the
reasons discussed in § II.F.2, supra. Accordingly, we find Petitioner has
demonstrated by a preponderance of evidence that claims 9–11 are
unpatentable over the combination of Seshardi and Chennakeshu.
G. Constitutionality of Inter Partes Reviews
Patent Owner argues that “[t]he importance placed on review of the
decisions of Court of Criminal Appeals Judges in Edmond v. US, 520 U.S.
651 (1997), is inconsistent with Arthrex’s[18] determination that invalidation
of statutory limitations on the removal of APJs is sufficient to render APJs
inferior officers.” PO Sur-Reply 15. Therefore, Patent Owner argues, “only
Congress can fix the IPR statutory scheme, and this case must be

18 Arthrex, Inc. v Smith & Nephew, Inc., 941 F.3d 1320 (Fed. Cir. 2019).
IPR2019-00054
Patent 6,304,612 B1

46
dismissed.” Id.
We decline to address the merits of Patent Owner’s constitutional
arguments.19
III. CONCLUSION
We have reviewed the Petition, Patent Owner Response, Petitioner
Reply, and Patent Owner Sur-Reply. We have considered all of the
evidence and arguments presented by Petitioner and Patent Owner, and have
weighed and assessed the entirety of the evidence as a whole.
We determine, on this record, that Petitioner has demonstrated by a
preponderance of evidence that claims 1–11 of the ’612 patent are
unpatentable over Seshardi and Chennakeshu.20

19 We note that Patent Owner’s constitutional arguments were not raised in
Patent Owner’s Response, and are not responsive to any argument raised in
Petitioner’s Reply as required by our rules. See 37 C.F.R. § 42.23(b).
20 Should Patent Owner wish to pursue amendment of the challenged
claims in a reissue or reexamination proceeding subsequent to the
issuance of this decision, we draw Patent Owner’s attention to the April
2019 Notice Regarding Options for Amendments by Patent Owner
Through Reissue or Reexamination During a Pending AIA Trial
Proceeding. See 84 Fed. Reg. 16,654 (Apr. 22, 2019). If Patent Owner
chooses to file a reissue application or a request for reexamination of the
challenged patent, we remind Patent Owner of its continuing obligation to
notify the Board of any such related matters in updated mandatory
notices. See 37 C.F.R. § 42.8(a)(3), (b)(2).

Claims 35 U.S.C. § Reference(s)/Basis
Claims
Shown
Unpatentable
Claims
Not Shown
Unpatentable
1–11 103 Seshardi,
Chennakeshu
1–11
IPR2019-00054
Patent 6,304,612 B1

47
IV. ORDER
It is hereby:
ORDERED that claims 1–11 of the ’612 patent are unpatentable under
35 U.S.C. § 103(a) as obvious over Seshardi and Chennakeshu; and
FURTHER ORDERED that this Decision is final, and a party to this
proceeding seeking judicial review of the Decision must comply with the
notice and service requirements of 37 C.F.R. § 90.2.

IPR2019-00054
Patent 6,304,612 B1

48
PETITIONER:
Adam P. Seitz
Paul R. Hart
Jennifer C. Bailey
ERISE IP, P.A.
Adam.Seitz@eriseip.com
Paul.Hart@eriseip.com
Jennifer.Bailey@eriseip.com
PTAB@eriseip.com

PATENT OWNER:
Ryan Loveless
Brett Mangrum
James Etheridge
Jeffrey Huang
ETHERIDGE LAW GROUP
ryan@etheridgelaw.com
brett@etheridgelaw.com
jim@etheridgelaw.com
jeff@etheridgelaw.com