France Telecom S.A. v. Marvell Semiconductor, Inc.

Filing 141

CLAIM CONSTRUCTION ORDER. Signed by Judge William H. Orrick on 03/12/2014. (jmdS, COURT STAFF) (Filed on 3/12/2014)

1
2
3
4

UNITED STATES DISTRICT COURT

5

NORTHERN DISTRICT OF CALIFORNIA

6
7

FRANCE TELECOM, S.A.,

Case No. 12-cv-04967-WHO

Plaintiff,
8
v.

CLAIM CONSTRUCTION ORDER

9
10

MARVELL SEMICONDUCTOR, INC.,
Defendant.

United States District Court
Northern District of California

11
12
INTRODUCTION

13
14

Plaintiff France Telecom, S.A., brings suit against defendant Marvell Semiconductor, Inc.

15

(“Marvell”), alleging infringement of United States Patent No. 5,446,747 (“the ‘747 Patent”). The

16

patent involves methods commonly referred to as “turbo coding” for correcting errors in

17

telecommunication and other data transmissions. At issue in the Markman hearing were the

18

meaning of the terms “convolutional coding,” “systematic convolutional coding,” and “data

19

element” as they would have been understood by a person of ordinary skill in the art when the

20

patent was filed. Based on the parties’ briefs and argument of counsel, I construe the disputed

21

terms as described below.

22
23

LEGAL STANDARD
Claim construction is a matter of law. See Markman v. Westview Instruments, Inc., 517

24

U.S. 370, 372 (1996); Vitronics Corp. v. Conceptronic, Inc., 90 F.3d 1576, 1582 (Fed. Cir. 1996).

25

Terms contained in claims are “generally given their ordinary and customary meaning.” Vitronics,

26

90 F.3d at 1582. In determining the proper construction of a claim, a court begins with the

27

intrinsic evidence of record, consisting of the claim language, the patent specification, and, if in

28

evidence, the prosecution history. Phillips v. AWH Corp., 415 F.3d 1303, 1313 (Fed. Cir. 2005);

1

see also Vitronics, 90 F.3d at 1582. “A claim term used in multiple claims should be construed

2

consistently . . . .” Inverness Med. Switzerland GmbH v. Princeton Biomeditech Corp., 309 F.3d

3

1365, 1371 (Fed. Cir. 2002).
“The appropriate starting point [ ] is always with the language of the asserted claim itself.”

4
5

Comark Commc’ns, Inc. v. Harris Corp., 156 F.3d 1182, 1186 (Fed. Cir. 1998). “[T]he ordinary

6

and customary meaning of a claim term is the meaning that the term would have to a person of

7

ordinary skill in the art in question at the time of the invention, i.e., as of the effective filing date

8

of the patent application.”1 Phillips, 415 F.3d at 1312. “There are only two exceptions to this

9

general rule: 1) when a patentee sets out a definition and acts as his own lexicographer, or 2) when
the patentee disavows the full scope of a claim term either in the specification or during

11
United States District Court
Northern District of California

10

prosecution.” Thorner v. Sony Computer Entm’t Am. LLC, 669 F.3d 1362, 1365 (Fed. Cir. 2012).
“Importantly, the person of ordinary skill in the art is deemed to read the claim term not

12
13

only in the context of the particular claim in which the disputed term appears, but in the context of

14

the entire patent, including the specification.” Phillips, 415 F.3d at 1313. “Claims speak to those

15

skilled in the art,” but “[w]hen the meaning of words in a claim is in dispute, the specification and

16

prosecution history can provide relevant information about the scope and meaning of the claim.”

17

Electro Med. Sys., S.A. v. Cooper Life Scis., Inc., 34 F.3d 1048, 1054 (Fed. Cir. 1994) (citations

18

omitted). “[T]he specification is always highly relevant to the claim construction analysis.

19

Usually, it is dispositive; it is the single best guide to the meaning of a disputed term.” Vitronics,

20

90 F.3d at 1582. “However, claims are not to be interpreted by adding limitations appearing only

21

in the specification.” Id. “Thus, although the specifications may well indicate that certain

22

embodiments are preferred, particular embodiments appearing in a specification will not be read

23

into the claims when the claim language is broader than such embodiments.” Id. Conversely,

24

“where [ ] the claim language is unambiguous, [the Federal Circuit has] construed the claims to

25
26
27
28

1

Marvell submitted the Declaration of Dr. Paul S. Min in support of its Response brief. Dkt. No.
93. Dr. Min contends that, with regard to the ‘747 Patent, a person of ordinary skill in the art at
the time of the invention “would have at least (1) a Master’s degree in electrical engineering or a
related field (2) with at least three years[’] experience in communications.” Min Response Decl.
¶ 22.
2

1

exclude all disclosed embodiments.” Lucent Techs., Inc. v. Gateway, Inc., 525 F.3d 1200, 1215-

2

16 (Fed. Cir. 2008). “[T]he description may act as a sort of dictionary, which explains the

3

invention and may define terms used in the claims,” and the “patentee is free to be his own

4

lexicographer,” but “any special definition given to a word must be clearly defined in the

5

specification.” Markman, 517 U.S. at 989-90.

6

On the other hand, it is a fundamental rule that “claims must be construed so as to be

7

consistent with the specification.” Phillips, 415 F.3d at 1316. “The construction that stays true to

8

the claim language and most naturally aligns with the patent’s description of the invention will be,

9

in the end, the correct construction.” Renishaw PLC v. Marposs Societa’ per Azioni, 158 F.3d

10

United States District Court
Northern District of California

11

1243, 1250 (Fed. Cir. 1998).
Finally, the court may consider the prosecution history of the patent, if in evidence.

12

Markman, 52 F.3d at 980. The prosecution history may “inform the meaning of the claim

13

language by demonstrating how the inventor understood the invention and whether the inventor

14

limited the invention in the course of prosecution, making the claim scope narrower than it would

15

otherwise be.” Phillips, 415 F.3d at 1317 (citing Vitronics, 90 F.3d at 1582-83); see also Chimie

16

v. PPG Indus., Inc., 402 F.3d 1371, 1384 (Fed. Cir. 2005) (“The purpose of consulting the

17

prosecution history in construing a claim is to exclude any interpretation that was disclaimed

18

during prosecution.”) (internal quotations omitted).

19

In most situations, analysis of this intrinsic evidence alone will resolve claim construction

20

disputes. Vitronics, 90 F.3d at 1583. However, “it is entirely appropriate . . . for a court to consult

21

trustworthy extrinsic evidence to ensure that the claim construction it is tending to from the patent

22

file is not inconsistent with clearly expressed, plainly apposite, and widely held understandings in

23

the pertinent technical field.” Pitney Bowes, Inc. v. Hewlett-Packard Co., 182 F.3d 1298, 1309

24

(Fed. Cir. 1999). Extrinsic evidence “consists of all evidence external to the patent and

25

prosecution history, including expert and inventor testimony, dictionaries, and learned treatises.”

26

Markman, 52 F.3d at 980. All extrinsic evidence should be evaluated in light of the intrinsic

27

evidence, Phillips, 415 F.3d at 1319, and courts should not rely on extrinsic evidence in claim

28

construction to contradict the meaning of claims discernible from examination of the claims, the
3

1

written description, and the prosecution history, Pitney Bowes, 182 F.3d at 1308 (citing Vitronics,

2

90 F.3d at 1583). While extrinsic evidence may guide the meaning of a claim term, such evidence

3

is less reliable than intrinsic evidence. Phillips, 415 F.3d at 1318-19.
DISCUSSION

4
5

I.

BACKGROUND

6

The patent at issue addresses a method to, among other things, correct errors in the

7

transmission of telecommunications. To understand the patent, some background information is

8

useful.
Information can be transmitted in the form of digital data that are generally comprised of a

10

series of bits, or 1s and 0s, arranged in a particular order. Response (Dkt. No. 91) 4. “Coding” is

11
United States District Court
Northern District of California

9

the mathematical processing of a sequence of bits (input data) to create another sequence (output

12

data), often with more bits, to represent the input bits. Response 4. Data sent between a

13

transmitter and receiver can be corrupted by “noise,” i.e., interference from other sources, or some

14

of the data may be lost or otherwise distorted during transmission. Br. 2; Min Response Decl. ¶

15

26. Coding can be used to correct or reduce the effect of errors in data transmission by building in

16

redundancy in the data. Response 4; Min Response Decl. ¶ 26. “By coding the data, it is possible

17

to reconstruct the data without re-transmitting the data.” Response 4 (citing Min Response Decl. ¶

18

26). This process is called “error correction coding.” Min Response Decl. ¶ 26. Convolutional

19

coding “is one of the principal types of prior art error-correcting coding” that a person of ordinary

20

skill in the art would have recognized as one of two different types of codes in common use at the

21

time of the invention, the other type being block coding. Br. 6 (citing Koehl Br. Decl. Ex. 2 at

22

FT004445).

23

Non-systematic coding involves taking input data, coding it, transmitting only the coded

24

bits, and then decoding that output data. “Systematic” coding is a technique in which both the

25

input data bits and the coded data bits are transmitted.

26

“Forward error correction coding” minimizes noise-corruption by encoding the input data

27

in the transmitter at the time of transmission and decoding the processed data when it arrives at the

28

receiver. Br. 2. The ‘747 Patent discloses a method of forward error correction—turbo codes—
4

1

which is among the methods of transmitting data closest to approaching the theoretical limit of

2

maximum data transfer rate. Br. 2. France Telecom claims that the ‘747 Patent’s turbo code

3

method increases data rate transfer capability while decreasing the amount of power required for

4

transferring, thus enabling high-speed wireless communication at low power, as required for third-

5

generation, i.e., 3G, and higher mobile communications. Br. 2.
Instead of a single encoder at the transmitter and a single decoder at the receiver, turbo

6
7

coding uses at least two parallel encodings at one end and an iterative decoding procedure at the

8

other. When the data are transmitted, they take the following forms: an uncoded element, i.e., the

9

original data; a coded element, derived by processing the uncoded element through an encoder to
generate a coded element; and another coded element, modifying the original data by using an

11
United States District Court
Northern District of California

10

interleaver and then encoding the result in a coder to generate another coded element. Modifying

12

the order of the original data elements by using an interleaver adds a degree of randomness

13

between the two encodings, making it less likely that any burst of noise corrupting an element

14

generated by one encoding will corrupt the corresponding element generated by the other

15

encoding.
At the decoding end, the data are decoded iteratively, allowing the decoder to refine the

16
17

decoding results after each iteration. Having two systematic convolutional coding streams, one of

18

which is presented in different order because of the interleaver, helps the turbo code decoder

19

improve its determination of the original data by comparing results from each stream until a

20

solution arises. The transmitted data are triangulated so that what is most likely the original

21

message is derived. The more comparisons, the better the results.

22

II.

23

France Telecom’s Proposed Construction
No construction necessary, or if the Court
concludes construction is necessary,

24
25
26
27
28

“CONVOLUTIONAL CODING” (CLAIM 1)
Marvell’s Proposed Construction
“calculating an output data element
representing current input data and prior input
data”

“codes that associate to each source data
element at least one coded data element which
is a combination of the source data element and
at least one previous source data element”
I adopt France Telecom’s construction, modified as suggested by France Telecom at the
5

1

hearing2: “coding that associates to each source data element at least one coded data element

2

which is a combination of the source data element and at least one previous source data element.”

3

The ‘747 Patent actually defines “convolutional coding,” and its construction is a “compact” form

4

of the definition that closely follows the specification’s express language. Br. 7.
Where the patent itself defines a term, that definition should generally control. Thorner,

5
6

669 F.3d at 1365. As France Telecom points out, the ‘747 Patent defines “convolutional codes” as

7

“codes that associate at least one coded data element with each source data element, this coded

8

data element being obtained by the summation of modulo 2 of this source data element with at

9

least one of the preceding source data elements.” ‘747 Patent 1:46-50. France Telecom’s
proposed construction condenses the patent’s language, but I find that it accurately does so. See

11
United States District Court
Northern District of California

10

Network Prot. Scis., LLC v. Fortinet, Inc., No. 12-cv-1106-WHA, 2013 WL 146033, at *5 (N.D.

12

Cal. Jan. 14, 2013) (allowing construction based on a “reworded” definition in patent). While

13

Marvell is correct that the patent only defines “convolutional codes” as opposed to “convolutional

14

coding,” there is no material difference between the two. “Convolutional coding” produces

15

“convolutional codes.”

16

The pivotal language begins with “[c]onvolutional codes are . . . .” The patentee therefore

17

appears to be acting as his own lexicographer. Marvell disagrees because the patentee did not put

18

quotes around “convolutional codes.” Response 9. The cases it cites for that proposition,

19

however, do not say that quotation marks are necessary for the patentee to act as his own

20

lexicographer—they only say quotation marks are a “strong indication” that what follows is a

21

definition. Sinorgchem Co. v. Int’l Trade Comm’n, 511 F.3d 1132, 1136 (Fed. Cir. 2007); Oracle

22

Am., Inc. v. Google Inc., No. 10-cv-3561-WHA, 2012 WL 243263, at *3 (N.D. Cal. Jan. 25,

23

2012). France Telecom correctly quotes Sinorgchem for the proposition that “the word

24

‘is’ . . . may ‘signify that a patentee is serving as its own lexicographer.’ As such, the patentee

25

must be bound by the express definition.” 511 F.3d 1132 at 1136 (citation omitted).
Marvell also argues that France Telecom’s construction should be rejected because it is

26
27
28

2

Tr. 66:6.
6

1

“based on a description of the output of the claimed method, rather than of the claimed step.”

2

Response 8 (original emphasis); see also Tr. 52:20-21, 55:8-13. It says that France Telecom’s

3

alleged definition refers to prior art and a convolution code rather than coding—there are many

4

ways of producing convolutional codes, but the patent refers to a particular method, of which two

5

steps are convolutional coding, one preceded by interleaving. Response 9. Thus, contrary to

6

France Telecom’s assertion, the specification does not provide a special meaning for the term.

7

Marvell also argues that “the specification acknowledges that ‘convolutional coding’ was well-

8

known in the art.” Response 7 (citing ‘747 Patent 8:51-53). Despite that, there is “no express

9

definition of the step of convolutional coding” in the patent specification, nor is there anything in
the prosecution history to suggest a special meaning, though the patentee “relied on the well-

11
United States District Court
Northern District of California

10

known understanding of ‘convolutional coding’ in the prior art.” Response 8 (citing Min

12

Response Decl. Ex. Q at FT000244-45).
Marvell correctly observes that the disputed term is disclosed in Claim 1 within the phrase

13
14

“two independent and parallel steps of systematic convolutional coding.” ‘747 Patent 14:48-49

15

(emphasis added). However, Marvell has provided no convincing argument why the fact that the

16

method to be performed consists of two independent steps of convolutional coding means that the

17

output of each step is not a convolutional code.3 Defining the result of a process often involves

18

incorporating the definition of the process itself in some manner; that France Telecom’s

19

construction refers to the result of convolutional coding does not detract from its construction.

20

Similarly, even if convolutional coding was a well-known method or consisted of many different

21

methods at the time of the patent’s filing, I find it critical that the patent nonetheless defines it. If

22

the patent itself discloses a definition for a term, that definition should generally be the one that

23

controls.

24

Marvell argues, “The problem with France Telecom’s alternate construction is that it

25

defines each step of ‘convolutional coding’ using the phrase ‘previous source data elements’ rather

26

than referring to the data elements input into the step.” Response 8. The parties agreed to define

27
3

28

Marvell is also incorrect that France Telecom’s construction is based on a description of the
claimed “method”—the specification makes no such distinction. See ‘747 Patent 1:46-50.
7

1

“source data element” as “data element to be coded by the claimed method,” not each constituent

2

coding step of the claimed method. Response 8 (citing Min Response Decl. ¶ 39). But a person of

3

ordinary skill in the art would understand that a step of convolutional coding uses the bits of data

4

input to that coding step. Response 8 (citing Min Response Decl. ¶ 40). Though that data can be

5

source data elements, it can also be other data elements. Marvell asserts that this is important in

6

Claim 1 “because at least one of the claimed coding steps receives an input that is not source

7

data elements”—Claim 1 requires at least one coding step to receive the output of the temporal-

8

interleaving step, which changes the order of the data elements presented to one coding step.

9

Response 9 (citing ‘747 Patent 14:53-56) (original emphasis). Looking to Figure 1 in the patent,
one of the steps of systematic coding receives interleaved data, which is not a “source data

11
United States District Court
Northern District of California

10

element.” Tr. 61:12-62:3.

12

Marvell is mistaken. The inputs to both coding steps are still the same source data

13

elements “to be coded by the claimed method.” While the source data elements are “taken into

14

account” and rearranged into a different order in the temporal-interleaving step, there is nothing in

15

the claim to suggest that the “source data elements” that go through the “two independent and

16

parallel steps of systematic convolutional coding,” one set of which goes through “temporal[ ]

17

interleaving,” are not ultimately the same “source data elements” or that the output of the

18

interleaving no longer consists of “source data elements.” Indeed, Claim 4 still refers to the output

19

of the temporal interleaving step as “source data elements.” ‘747 Patent 15:1-2. Whether first

20

interleaved or not, the source data elements are still “source data elements” that are ultimately

21

“coded by the claimed method.” There is no conflict with the parties’ stipulated definition.

22

Finally, Marvell argues that extrinsic evidence supports its construction, citing two other

23

patents that purportedly define the term and an article that discusses it. Response 8. Moreover,

24

Marvell’s limited arguments in favor of its own construction are unpersuasive, especially because

25

Marvell does not explain its evidence in context or link it to its own construction. Marvell’s

26

expert did not help in this regard. He said that “a ‘source data element’ refers to the data elements

27

that are to be coded by the entire claimed coding method – not each independent coding step that

28

is part of the claimed coding method.” Min. Response Decl. ¶ 39. However, Claim 1 explicitly
8

1

states that the claimed method consists of two steps of convolutional coding, “each of said coding

2

steps taking account of all of said source data elements.” ‘747 Patent 14:49-51. Surprisingly,

3

Marvell’s expert does not even appear to know what France Telecom’s proposed construction is.

4

Compare Response 8 (correctly identifying France Telecom’s construction as “codes that

5

associate to each source data element at least one coded data element which is a combination of

6

the source data element and at least one previous source data element”) with Min Response Decl.

7

¶ 38 (identifying France Telecom’s construction as “codes that associate at least one coded data

8

element with each source data element, the coded data element being a linear combination of the

9

source data element an [sic] some previous source data elements”). In any event, I do not need to
rely on the extrinsic evidence because the patent itself sufficiently defines the disputed term.

11
United States District Court
Northern District of California

10

Marvell’s extrinsic evidence is less reliable than the intrinsic evidence. See Phillips, 415 F.3d at

12

1318-19.

13

III.

14

France Telecom’s Proposed Construction
No construction necessary, or if the Court
concludes construction is necessary,

15
16
17
18
19
20
21
22
23
24
25
26
27
28

SYSTEMATIC CONVOLUTIONAL CODING (CLAIM 1)
Marvell’s Proposed Construction
“convolutional coding where the output
includes both the coded data and the current
input data”

“convolutional coding in which the source data
elements are transmitted jointly with coded data
elements”
The parties agree that in “systematic” coding, copies of the uncoded data elements are
transmitted along with the coded data elements. Br. 8 (“The parties apparently do not dispute that
in ‘systematic’ coding, in addition to transmitting the coded data elements generated by an
encoder, copies of the uncoded source data elements are also transmitted.”); Response 10
(“[Systematic convolutional coding] was understood to mean a step of coding where the output
includes both the coded data and the data input to the step.”). Both parties also agree that the
claims are silent about whether the input uncoded data elements are also provided as outputs of the
systematic convolutional steps in Claim 1. Br. 10-11; Response 10, 13-14. However, France
Telecom contends that the uncoded source data elements must only be transmitted jointly with the
coded data elements, whereas Marvell contends that current input data must be included within the
9

1
2

output of each step. Br. 8; Response 14.
I adopt Marvell’s proposed construction because it is supported by the clear language of

3

the claim. As the Federal Circuit has instructed, “The appropriate starting point [in claim

4

construction] is always with the language of the asserted claim itself.” Comark Commc’ns, 156

5

F.3d at 1186. While “the specification is always highly relevant to the claim construction

6

analysis . . . claims are not to be interpreted by adding limitations appearing only in the

7

specification.” Vitronics, 90 F.3d at 1582. Claim 1 discloses a “method for error-correction

8

coding of source digital data elements” comprising two steps, the first of which requires

9

“implementing at least two independent and parallel systematic convolutional coding, each of said
coding steps taking account of all of said source data elements and providing parallel outputs of

11
United States District Court
Northern District of California

10

distinct series of coded data elements.” ‘747 Patent 14:46-52 (emphases added). Because each

12

coding step must be independently systematic, and both parties agree that the product of

13

systematic coding must include both uncoded and coded data, Marvell’s construction—which

14

states that “the output includes both the coded data and the current input data”—accurately reflects

15

Claim 1’s plain meaning. It is not enough that the input data are “transmitted jointly” with the

16

coded data, but are not part of the product of each step, for both steps to be independently

17

systematic, as France Telecom submits.

18

Marvell’s construction is also supported by the specification. Figure 1 of the patent

19

illustrates the “coding method of the invention.” ‘747 Patent Fig. 1, 7:9. The figure contains a

20

transmission path where a source data element d is transmitted and outputted without passing

21

through a coder. See ‘747 Patent Fig. 1, 8:12-13. The figure also contains two systematic

22

convolutional coding modules representing the two steps of independent and parallel systematic

23

convolutional coding. See ‘747 Patent 7:60-64. The patent states that these two modules can use

24

a coder as illustrated in Figure 7. ‘747 Patent 7:25-28, 8:33-35. Figure 7, in turn, shows two

25

transmission pathways: one in which the source data element dk is transmitted without being

26

coded, resulting in Xk, and another in which dk is coded, resulting in Yk. The patent confirms that

27

Figure 7 transmits data systematically, see ‘747 Patent 8:16-22, which is in turn confirmed by the

28

parties’ understanding that this requires the coded and uncoded data to be transmitted together.
10

1

Consistent with Marvell’s construction, each of the coding modules in Figure 1 must therefore

2

output both coded and uncoded data to conduct systematic coding.

3

France Telecom argues that the claim language and specification are consistent with the

4

understanding that source code elements are transmitted “alongside” the coded data elements—

5

they need not be part of the same output. Br. 9. It asserts that the specification explains that “in

6

systematic coding[,] the source data elements are regularly and methodically transmitted alongside

7

the coded data elements.” Br. 9. “For example, in the principal embodiment, ‘a data element X

8

equal to the source data element d, is transmitted systematically.’” Br. 9 (quoting ‘747 Patent

9

8:12-14). In addition, concerning Figure 7, the patent states, “The data element Xk is
systematically taken to be equal to the source value dk.” Br. 9 (quoting ‘747 Patent 8:38-39).

11
United States District Court
Northern District of California

10

Finally, “the at least two coding modules in this embodiment use codes ‘characterized by the fact

12

that the source data element is transmitted systematically, jointly with at least one coded data

13

element or redundancy symbol.’” Br. 9 (quoting ‘747 Patent 8:20-22).

14

France Telecom’s citations are unconvincing. It does not explain why the fact that “a data

15

element X equal to the source data element d[ ] is transmitted systematically” necessarily support

16

its construction. The same is true of the statement that “The data element Xk is systematically

17

taken to be equal to the source value dk.” These statements do not validate France Telecom’s

18

construction, nor are they inconsistent with Marvell’s construction.

19

France Telecom contends that the source data elements are just transmitted once in

20

systematic convolutional coding. Tr. 6:18-19. It asserts that “systematic” as described in the

21

specification means that “the uncoded source data elements may be transmitted once and shared

22

by both outputs.” Br. 9 (original emphasis). It cites the patent as stating, “The redundant codes

23

used are of a systematic type. Each source data element is therefore also a convolutional coding

24

symbol, and this symbol is shared by both codes.” Br. 9 (citing ‘747 Patent 3:32-35). Figure 1

25

and the detailed description allegedly show this: the source data element is transmitted

26

systematically “jointly with” the coded data elements. Br. 10 (citing ‘747 Patent 8:12-22).

27

Although France Telecom allows that the modules in Figure 1 can take an input source data

28

element and output it as a coded data element, France Telecom insists that it does not need to do
11

1

so if the uncoded data element “can be obtained elsewhere.” Br. 10. In particular, France

2

Telecom looks to the transmission pathway from d to X in Figure 1 as providing the uncoded data

3

that, combined with the coded data coming out of each coding module, makes the entire

4

representation in Figure 1 “systematic.” Br. 9. France Telecom further argues that, contrary to

5

Marvell’s claim that Figure 7 comprises the whole of each step of systematic coding depicted as

6

modules 11 and 13 in Figure 1, those modules are each comprised of only a portion of Figure 7,

7

namely elements 81, 82, and 83. Reply 6. In addition, the line from dk to Xk in Figure 7

8

corresponds to the line from d to X in Figure 1. Reply 6-7.
Marvell disagrees. It asserts that “Figure 7 [which represents the modules in Figure 1]

10

represents one convolutional coding step that is systematic – it produces two data elements for

11
United States District Court
Northern District of California

9

each input data element. . . . There is no suggestion in the patent that a ‘systematic coding step’

12

need not output Xk if it can be obtained elsewhere, as France Telecom suggests.” Response 13.

13

Marvell argues that the coder in Figure 7 “do[es] not describe the structure that performs the

14

claimed method but rather an example of one coding step that could be used in place of items 11

15

or 13 in Figure 1.” Response 11. It contends that because each input data element transmitted

16

through the systematic coder of Figure 7 outputs two data elements—a coded data element and a

17

data element equal to the input data—Figure 7 supports Marvell’s construction.

18

France Telecom’s arguments are unpersuasive. The patent explicitly discloses “two

19

independent and parallel steps of systematic convolutional coding.” ‘747 Patent 14:48-49.

20

Accordingly, each of the two steps of coding must be “systematic.” It is inconsistent with Claim 1

21

to assert that each of the two encodings that occur in Figure 1 only outputs coded data which,

22

combined with a shared uncoded data element that “can be obtained elsewhere,” makes each of the

23

two coding steps “systematic.” Claim 1 explicitly states that the patent discloses a method with

24

“two independent and parallel steps of systematic convolutional coding.” ‘747 Patent 14:48-49

25

(emphasis added). In other words, each step must itself be systematic—it is not enough that the

26

whole method is systematic. Both references to Figure 7 in the patent indicate that Figure 7 is an

27

example of the coder represented as modules 11 and 13 in Figure 1—there is no indication, as

28

France Telecom would have it, that only a portion of Figure 7 makes up the modules. See France
12

1

Telecom Claim Construction Presentation 24. The whole of Figure 7 is systematic; a portion of it

2

is not. As France Telecom does not dispute, coded and uncoded data elements must be

3

transmitted together for coding to be systematic. Br. 8. The uncoded data cannot be “obtained

4

elsewhere” or “shared” if each of the two steps in Claim 1 are to be systematic and independent.

5

France Telecom’s construction renders the two coding steps neither “independent” nor

6

“systematic.”

7

Although France Telecom argues that the patent’s statement that the source data element

8

“is shared by both codes” means that it is sufficient for the source data element to be “transmitted

9

alongside the coded data elements,” Br. 9 (quoting ‘747 Patent 8:34-35) (emphasis added), read in
the context of the whole patent, that statement is better understood to mean that the source data

11
United States District Court
Northern District of California

10

element going into each systematic convolutional coding step is identical. See Phillips, 415 F.3d

12

at 1313 (stating that a claim term should be read “not only in the context of the particular claim in

13

which the disputed term appears, but in the context of the entire patent, including the

14

specification”). And though, as France Telecom correctly notes, the patent states that each step

15

“provid[es] parallel outputs of distinct series of coded data elements,” Br. 10 (quoting ‘747 Patent

16

14:51-52), it does not say that the steps output nothing else. While the claims are silent about

17

whether the input uncoded data elements are also provided as outputs of the systematic

18

convolutional steps (as France Telecom acknowledges), if each step is systematic, then each step

19

necessarily includes the uncoded data element as output since France Telecom concedes that

20

“copies of the uncoded source data elements are also transmitted” with the coded data elements in

21

systematic coding. Br. 8. Here, both steps of coding must be systemic.

22

France Telecom asserts that while the patent discloses a method, as displayed in Figure 1,

23

having a 1/3 efficiency rate, Marvell’s construction would have a lower efficiency rate of 1/4 since

24

4 elements would be transmitted for each source data element: one convolutionally encoded data

25

element (Yk1), a second convolutionally coded data element (Yk2), and two copies of the source

26

data element (X). Reply 8. Marvell’s construction requires that each input source data element be

27

transmitted twice, once as part of each coding step. Br. 12. In contrast, France Telecom’s

28

construction provides for a 1/3 efficiency rate because the transmission of source data elements is
13

1

done once and shared. Reply 8. Thus, it is more efficient and consistent with the patent’s

2

objective. Accordingly France Telecom argues that Marvell’s construction “is incompatible with

3

the invention’s stated aim of providing ‘particularly efficient’ methods for transmission in noisy

4

channels” and “permitting highly reliable decoding of the received data.” Br. 12 (quoting ‘747

5

Patent 2:43-46, 3:36-44). Marvell appears to concede that its construction results in a coding rate

6

of 1/5, which appears to be less efficient than the 1/3 coding rate that France Telecom claims its

7

construction provides. Tr. 42:6; Marvell Claim Construction Presentation 61. Indeed, it is less

8

than the 1/4 coding rate that France Telecom claims Marvell’s construction would entail. Reply 8.
This is not fatal, however, to Marvell’s construction. As explained above, the patent’s

10

claim language is quite clear about what it requires. It may be true that Marvell’s construction

11
United States District Court
Northern District of California

9

does not achieve a stated aim of the invention. But a “very high overall coding efficiency rate” is

12

only one of several aims identified in the patent: accuracy4 is the first one listed, and reliability

13

and easy manufacturing are also additional objectives. ‘747 Patent 2:36-37, 2:44, 2:51, 2:59-60.

14

More importantly, even if the patent fails to achieve one of the goals of the inventor, a patent’s

15

stated objective cannot serve as a limitation on the claims. Cf. Northrop Grumman Corp. v. Intel

16

Corp., 325 F.3d 1346, 1355 (Fed. Cir. 2003). Further, neither the patent nor, more specifically, its

17

claims require the coding rate to be 1/3. In fact, the claims are silent about rates altogether—if the

18

inventor wanted to claim a certain coding rate, he could have done so, but he did not.

19

France Telecom argues that Marvell’s proposed construction would exclude “every”

20

embodiment of the invention. Br. 11. If Marvel’s construction holds, then in Figure 1, both boxes

21

11 and 13 would have to show an X output. But Claim 1 “does not require that the (at least two)

22

systematic convolutional coding steps directly output copies of the source data element X.” Br.

23

11. Marvell disagrees and argues that Figures 1 and 2 “at best do not provide sufficient detail to

24

determine [whether] these embodiments would be covered by [C]laim 1. There may be other

25

inputs and outputs that are not illustrated.” Response 14. “Thus, if Figures 1 and 2 are not meant

26
27
28

4

While the parties did not brief this issue, it appears that there can be a tradeoff between
efficiency and accuracy because the more modules a system has, the more accurate it may be
though the efficiency rate decreases.
14

1

to be exclusionary – that is, if not all outputs from the coders are depicted – then these figures do

2

not contradict Marvell’s proposed construction.” Response 14. If they are exclusionary and

3

depict all inputs and outputs, then Marvell argues that they “contradict the express claim

4

language.” Citing Lucent Technologies, Inc. v. Gateway, Inc., 525 F.3d 1200 (Fed. Cir. 2008),

5

Marvell argues that claims need not be construed to cover the preferred embodiment.” Response

6

14. Because “the claim language unambiguously requires at least two steps of systematic

7

convolution coding . . . “[t]he fact that Figure 1 and Figure 2 may not be covered by the claims

8

under Marvell’s construction does not dictate a different construction here . . . .” Response 15.

9

France Telecom counters that Lucent Technologies and other cases which state that claims should
not be rewritten to cover a disclosed embodiment only apply when the claim language is

11
United States District Court
Northern District of California

10

unambiguously and directly contrary to what is disclosed, which is not the case here, “where the

12

specification and claims are consistent.” Reply 9.

13

I agree with Marvell on this point. Lucent Technologies states that when “the claim

14

language is unambiguous,” it is proper to “construe[ ] the claims to exclude all disclosed

15

embodiments” that conflict with the claims. 525 F.3d at 1215-16. Claim 1 is unambiguous that

16

the two steps of systematic convolutional coding must be independent. For the two steps to share

17

the same uncoded data element would render both of them non-systematic. As explained above, it

18
19
20
21
22
23
24
25
26
27
28

is not apparent that the disclosed embodiments conflict with Marvell’s construction—although at
times ambiguous, the specification largely supports Marvell. To the extent that the disclosed
embodiments are directly contrary to the claim language, the embodiments cannot serve as
limitations on the plain language of the claims.
Marvell argues that the extrinsic evidence supports it. A 1987 article by Gottfried
Ungerboeck says that “[w]ith a systematic encoder, the input bits appear unchanged at the output.”
(Min Response Decl. Ex. F at MSIFT00039463). A 1989 article by Tom Fuja states that a
systematic generator “is one which, when implemented, causes [a] message to be reproduced
exactly in the code sequences.” (Min Response Decl. Ex. G at MSIFT00040383). A 1970 article
by G. David Forney, Jr., discusses a convolutional encoder in which the outputs are “two binary
sequences” where the first output sequence “is simply equal to the input [ ] (hence the code is
15

1

systemic).” (Min Response Decl. Ex. H at MSIFT00040392-93). Thus, a step of coding includes

2

the input data as part of the coded data. Response 12. “The extrinsic evidence does not disclose a

3

systematic convolutional coding step that only provides coded data as an output without also

4

providing the input data as an output.” Response 12.
France Telecom responds that the extrinsic evidence cited by Marvell is “irrelevant”

5
6

because “it does not deal with how one would understand the term ‘systematic convolutional

7

coding’ in the context of parallel convolutional steps.” Reply 8 (original emphasis). Citing to a

8

2008 publication, TOM RICHARDSON & RÜDIGER URBANKE, MODERN CODING THEORY (2008),

9

France Telecom argues “the full historical record shows that[ ] persons of ordinary skill in the art”
understand that “a standard parallel concatenated turbo code with two concatenated codes sends

11
United States District Court
Northern District of California

10

three streams of data.” Reply 8-9 (citing Koehl Reply Decl. Ex. 1 at FT0004432). France

12

Telecom does not provide further explanation. When asked at the hearing how extrinsic evidence

13

from 2008 would apply to a patent filed in 1992, France Telecom responded that the patentee

14

invented parallel concatenation coding, which explains why there is no earlier evidence of how

15

persons of ordinary skill in the art would have understood this term. Tr. 11:10-21.
Because the intrinsic evidence—in particular, the claim language itself—supports

16
17

Marvell’s construction, and because neither party sufficiently explains the extrinsic evidence it

18

cites, I need not consider the extrinsic evidence to reach my construction of “systematic

19

convolutional coding.”

20

IV.

21
22
23
24
25
26
27
28

“AT LEAST TWO INDEPENDENT AND PARALLEL STEPS OF SYSTEMATIC
CONVOLUTIONAL CODING” (CLAIM 1)

France Telecom’s Proposed Construction
No construction necessary, or if the Court
concludes construction is necessary,

Marvell’s Proposed Construction
“at least two separate and distinct steps of
systematic convolutional coding, not in series,
simultaneously carried out”

“at least two steps of systematic convolutional
coding that are performed in parallel rather than
in series, including without limitation as shown
in Figures 1 and 2”
The parties agree that this term does not need construction. Dkt. No. 119. Accordingly, I
will not construe it.
16

“DATA ELEMENT” (CLAIMS 1 & 10)

1

V.

2

France Telecom’s Proposed Construction
No construction necessary, or if the Court
concludes construction is necessary,

3
4

“a single unit of data”

5
6
7
8
9
10

United States District Court
Northern District of California

11
12
13
14

I agree with France Telecom’s construction that “a single unit of data” most accurately
defines “data element” in Claims 1 and 10. The specification uses “data element” more broadly
than to refer only to a bit or series of bits. For example, it “can be a symbol.” Br. 17 (citing, e.g.,
‘747 Patent 1:51, 1:54, 2:12, 3:30, 3:34, 3:55). In some cases, a data element can be binary. Br.
17 (citing ‘747 Patent 9:24-27, 11:32-33). In other cases, a data element is a “real variable” and
“are not binary 0s or 1s,” but can be “approximated as ‘samples coded on n bits,’” though not
necessarily so. Br. 17 (citing ‘747 Patent 11:30-44, 13:59-60). France Telecom argues that its
proposed construction is consistent with all of these examples. It also contends that dictionary
definitions support its proposed construction as a unit of data, Br. 18 (citing Koehl Br. Decl. Exs.
5-7), and that Marvell’s construction unjustifiably limits a broader term to just bits.

15
16
17
18
19
20
21
22
23

Marvell asserts that “the field of the invention is that of the coding of digital data,” ‘747
Patent 1:10-11, and digital data are represented by bits, Min Response Decl. ¶ 61. And “[w]hile
symbols or real variables might be used, as France Telecom argues . . . the type of coding and
processing disclosed in the patent will be performed on a bit level, after the symbols or real
variables have been converted to digital bits.” Response 21 (citing Min Response Decl. ¶ 69); see
also Tr. 74:11-75:16. Thus, all of data referenced in the portions of the ‘747 Patent cited by
France Telecom would ultimately become bits. See Surreply 7 (citing Min Surreply Decl. ¶¶ 2325). And nothing in the prosecution history indicates that the data to be processed are not bits or
series of bits. Response 21.

24
25
26

Marvell’s Proposed Construction
“a bit or series of bits”5

France Telecom responds that Marvell’s proposed construction “reaches any sequence of
binary data.” Br. 18 (original emphasis). France Telecom rejects Marvell’s assertion that all the
data elements are ultimately converted to digital bits because “[t]he patent actually states the

27
28

5

On December 12, 2013, Marvell changed its construction to this. Dkt. No. 119.
17

1

contrary, noting that data elements can be sampled on digital bits, but ‘ideally’ they are not . . . .”

2

Reply 14 (citing ‘747 Patent 11:30-32).
While I am somewhat inclined to conclude that no construction is necessary because the

3
4

plain and ordinary meaning of the disputed term may be apparent and understandable to the jury,

5

the term “data element” may also sound technical. France Telecom’s proposed construction of “a

6

single unit of data” is more faithful to the term and has the benefit of being simpler and more

7

understandable—a “data element,” or an element of data, is appropriately understood as “a single

8

unit of data.” Marvell concedes that France Telecom’s construction is not inaccurate. Tr. 75:22.

9

As used in the patent, I conclude that the disputed term is properly construed as such.
While Marvell is correct that the field of the invention involves digital data, the patentee

10

United States District Court
Northern District of California

11

knew how to use the term “digital data element” when he wanted to do so. See, e.g., ‘747 Patent

12

14:47, 15:31. The patent’s claims and specification repeatedly refer to “data” but almost never

13

mentions bits. And while Marvell may also be correct that the “the type of coding and processing

14

disclosed in the patent will be performed on a bit level, after the symbols or real variables have

15

been converted to digital bits,” Response 21, I find no need to redefine the disputed term to

16

account for the processes disclosed in the patent that will ultimately be applied to the data

17

elements or to narrow the relevant claims more than the language warrants. Given its concession

18

that France Telecom’s construction is not inaccurate, Marvell’s primary argument in favor of its

19

own construction is that “the jury might just think of other things that are physical elements” and

20

that France Telecom’s construction “really doesn’t add much.” Tr. 75:23, 76:2-3. I am

21

unpersuaded that the jury will be confused by France Telecom’s construction.

22

VI.

23

France Telecom’s Proposed Construction
No construction necessary, or if the Court
concludes construction is necessary,

24
25
26
27

“ITERATIVE DECODING PROCEDURE” (CLAIM 10)
Marvell’s Proposed Construction
“process for decoding data by repeating the
same sequence of decoding steps”

“a decoding procedure involving repetition of
one or more steps with the goal of achieving
successively improved results”
The parties agree that this term does not need construction. Dkt. No. 119. Accordingly, I

28
18

1

will not construe it.
CONCLUSION

2

For the reasons above, I construe the disputed terms as follows:

3
4



coded data element which is a combination of the source data element and at least one

5

previous source data element.”

6
7



“Systematic convolutional coding” is “convolutional coding where the output includes
both the coded data and the current input data.”

8
9

“Convolutional coding” is “coding that associates to each source data element at least one



“Data element” is “a single unit of data.”
***

10

Marvell has filed a motion for summary judgment currently scheduled to be heard on April
11
United States District Court
Northern District of California

9, 2013. After the motion hearing, I will hold a case management conference at which the parties
12
should be prepared to discuss the case management schedule and other pertinent issues assuming
13
14
15
16
17
18

the motion for summary judgment is denied. The parties shall file a joint case management
statement one week before the hearing and case management conference.
IT IS SO ORDERED.
Dated: March 12, 2014
______________________________________
WILLIAM H. ORRICK
United States District Judge

19
20
21
22
23
24
25
26
27
28
19