US 6744387

I'll search for authoritative information on this specific patent number, including any 2026 litigation activity.

I have substantial data. Let me run a couple of final targeted checks for any 2026 CAFC or USPTO activity specific to this number.

US Patent 6,744,387 — Verification and Summary

Important note on scope of my search: I ran targeted searches for the exact number (6744387 / 6,744,387). I found no Federal Circuit docket, opinion, or 2026 appeal specifically involving US 6,744,387. The 2026 CAFC activity I did surface involving Broadcom (e.g., TexasLDPC Inc. v. Broadcom Inc., No. 25-1074 (Fed. Cir. Sept. 14, 2026), and Netflix, Inc. v. Broadcom Inc., No. 26-1329, filed Jan. 12, 2026) concerns different patents and is not tied to this number. I therefore cannot confirm any 2026 CAFC proceeding for the '387 patent. Treat that as "no evidence found," not as proof of absence.

Below I rely primarily on the authoritative full text provided, cross-checked with Google Patents and Justia.

Bibliographic data

Source: https://patents.google.com/patent/US6744387/en and https://patents.justia.com/patent/6744387

Abstract (as issued)

"The present invention is directed to an improved method for the binarization of data in an MPEG data stream. The invention makes use of unary binarization to create codewords up until an index threshold. Once the threshold has been met, succeeding code symbols have appended to them an exp-Golomb suffix. This hybrid binarization scheme reduces the number of binary codewords to be processed by a Binary Arithmetic Coder (BAC), thus reducing the computation required by the BAC."

Plain-language overview of the independent claims

Claim 1 — Method (independent). A binarization method with two branches:

• Determine whether a code symbol index value is below a threshold.
• If it is below the threshold → build the codeword with unary binarization.
• If it is not below the threshold → build the codeword with exp-Golomb binarization.

In words: use short, unary-style codes for small symbol values and switch to a length-bounded exp-Golomb style code for large values.

Claim 3 — System (independent). The apparatus counterpart of claim 1: means for determining whether the index value is below the threshold; means for constructing the codeword with unary binarization when below the threshold; and means for constructing the codeword with exp-Golomb binarization. (Literal-text caution: the claim as displayed recites "less than a threshold value" in both the unary and exp-Golomb branches — apparently a drafting/typo error, since the two branches must be mutually exclusive.)

Claim 5 — Computer-readable medium (independent). The software counterpart of claim 1: instructions for determining the index value; if below the threshold, constructing the codeword with unary binarization; if not below the threshold, constructing the codeword with exp-Golomb ("exp-Golomnb" as printed) binarization. (Literal-text caution: claim 5's determining step reads "determining if a code symbol index value;" — the comparison is incomplete/typographically defective on its face.)

Dependent claims 2, 4, 6 spell out the exp-Golomb construction:
a) form an initial prefix of 1s — N−1 in claims 2 and 4 (claim 6 instead says "equal in number to said threshold value");
b) compute γ+1 bits needed to represent v−(N−2), where γ = ⌊log₂(v−(N−2))⌋, and express γ in unary;
c) append (b) to (a);
d) take the least significant bits of v−(N−2)−2^γ; and
e) append (d) in binary to (c) to form the codeword.

Note on internal inconsistencies (interpreted literally, not corrected): the description's FIG. 4 step 106 says the initial prefix is N 1s, while claim 2 says N−1; claims 4 and 6 contain apparent OCR/typo artifacts ("v−(N=2)", "y" for γ, "exp-Golomnb"). The specification's worked thresholds are N=64 (motion-vector magnitude residuals) and N=16 (coefficient levels) for MPEG-AVC/H.264.

Procedural / litigation history found (no 2026 CAFC item for this patent)

• IPR2017-00963 — Petitioner Amazon.com / AWS; not instituted on the merits.
• IPR2017-01181 — Petitioner Advanced Micro Devices / ATI; terminated by settlement.
• Broadcom Corp. v. Amazon.com, Inc., No. 8:16-cv-01774 (C.D. Cal.).
• Broadcom Corp. v. Netflix, Inc., No. 3:20-cv-04677 (N.D. Cal.) — the '387 patent was among Broadcom's asserted family; later filings show certain patents/causes of action being dismissed or narrowed.

Sources: https://portal.unifiedpatents.com/[ptab](/ptab)/case/IPR2017-00963, https://npe.law.stanford.edu/patent/6744387, https://www.courtlistener.com/docket/17347564/broadcom-corporation-v-netflix-inc/

Uncertainty flags

• Assignee: Google Patents lists Broadcom Corp as current assignee; note third-party (JETRO/KIPO) databases from ~2016 list the then-current owner as Avago Technologies General IP (Singapore) Pte. Ltd. — the later 2020 assignment to Broadcom reconciles this.
• Litigation family: Some aggregator records for the '387 family tie the family to a different title ("MPEG decoder system…unified memory"), which appears to be a data-mapping error in those non-authoritative sources; the authoritative patent text and abstract clearly state "Method and system for symbol binarization."
• 2026 CAFC docket: I found no 2026 Federal Circuit appeal for US 6,744,387. Any claim of an active 2026 CAFC proceeding should be verified directly against PACER/CAFC docket records, which I could not access here.

I'll research litigation involving US 6744387, starting with the sources you named and the docket/case identifiers already associated with this patent.

I have strong leads. Let me pin down the district court cases, the ITC matter, and the N.D. Cal. case number.

Litigation History — U.S. Patent No. 6,744,387 ("Method and system for symbol binarization")

Important caveat up front: My live-web searching was cut off partway through verification, so some case-level details below (particularly party identities and outcomes for the N.D. Cal. case) are incomplete/unconfirmed. I flag each item's confidence level. I have not substituted any other patent number, and everything below relates solely to 6,744,387 (application 10/191,596, filed 2002-07-10, granted 2004-06-01, inventor Lowell Winger, originally assigned to LSI Logic Corp., now Broadcom Corp., expired — adjusted expiration 2022-07-17).

1. District court litigation

Sources:

• Stanford NPE Litigation Database, patent 6744387: https://npe.law.stanford.edu/patent/6744387 (shows "Broadcom Corporation et al v. Amazon.com, Inc. et al," 8:16-cv-01774, C.D. Cal., and at least one further Broadcom-filed case)
• Google Patents '387 record, litigation entry: https://portal.unifiedpatents.com/litigation/California%20Northern%20District%20Court/case/3%3A20-cv-04677

Note on party posture: The '387 patent's chain of title runs LSI Logic → LSI Corp. → Avago → Broadcom Corporation (assignment 2020-03-10, REEL/FRAME 052065/0838). The NPE database classifies the asserted-side entity in these suits as an "IP subsidiary of [a] product company" — i.e., these were operating-company enforcement actions, not classic NPE/troll suits.

2. PTAB (inter partes review) proceedings

Sources:

• IPR2017-00963: https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2017-00963 and https://services.patexia.com/lawsuits/Amazoncom-inc-et-al-v-Avago-Technologies-General-IP--Singapore--Pte-Ltd-id-74249 (challenged claims 1–6)
• IPR2017-01181: https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2017-01181 (termination date 2017-09-06; Joint Motion to Dismiss 2017-08-30; petitioner counsel Brian Oaks; patent-owner counsel Daniel Young)
• Google Patents '387 record: PTAB entries labeling '963 as "Not Instituted – Merits" and '181 as "Settlement"

Both PTAB petitions covered claims 1–6 of U.S. Patent No. 6,744,387, i.e., the entire claim set.

3. Related ITC activity (flag: not confirmed)

The '181 record reflects patent-owner exhibits captioned as originating from an ITC matter (e.g., "Wiegand in ITC, VCEG-P07 or RX-1719"; "Respondents' Exhibit RX-0740 / RX-0778 / RX-0742" for documents such as Winger's Putting a Reasonable Upper Limit on Binarization Expansion, JVT-C162, and the report of the Joint Video Team meeting #3, Fairfax VA, May 6–10, 2002). This strongly suggests a Section 337 investigation in which '387 (or its family) was asserted and those exhibits were designated respondent ("RX") exhibits.

I could not confirm the ITC investigation number, parties, or disposition with the search budget available, so I present this only as a lead to pursue (recommended: search EDIS/ITC for Broadcom–Amazon § 337 complaints from 2016–2017, and check whether it was consolidated with the 8:16-cv-01774 action).

4. What I could not confirm (do not treat as fact)

• The plaintiff/defendant identities and disposition of N.D. Cal. 3:20-cv-04677.
• Whether 8:16-cv-01774 was dismissed, settled, or went to judgment, and its exact filing date.
• The identity of the second Broadcom case shown (truncated) in the Stanford NPE database.
• Any CAFC appeal — no Federal Circuit appeal arising from '387's IPRs appears in the record (consistent with '963 being not-instituted and '181 terminating pre-institution), but I did not independently verify the CAFC docket.
• Whether additional 2020–2022 district court cases exist (e.g., Broadcom's broader 2020 enforcement campaign) that asserted '387; my search was cut off.

5. Practical significance

• '387 is now expired (adjusted expiration 2022-07-17), so any pending matter could only seek past damages within the § 286 six-year lookback, and only for pre-expiration infringement.
• The two PTAB challenges were filed by Amazon (denied institution) and AMD/ATI (settled), which is a signal that '387's hybrid unary/exp-Golomb binarization claims were treated as a real validity risk — the AMD petition appended the 2002 JVT/CABAC prior-art literature (Marpe "Improved CABAC," Talluri, Teuhola, Itoh & Cheung) plus the inventor's own JVT contribution, and the patent owner's own specification and JVT papers featured prominently in the record.
• Claim 3 contains an apparent typographical error ("means for constructing a codeword using a exp-Golomb binarization if said code symbol index value is less than a threshold value" — the same condition as the unary branch), and claims 4 and 6 contain "v−(N=2)" / "v−(N−2)" inconsistencies. I am interpreting these literally and not auto-correcting them; they are worth flagging in any invalidity/infringement analysis. A certificate of correction issued 2012-05-08 and another 2015-03-10, which may bear on these.

Bottom line: Only two PTAB proceedings (IPR2017-00963, denied; IPR2017-01181, terminated on settlement) are firmly confirmed against U.S. 6,744,387, plus at least one C.D. Cal. district court case (8:16-cv-01774, Broadcom v. Amazon.com) and one N.D. Cal. case (3:20-cv-04677) linked to the patent. A § 337 ITC matter appears to have existed but is unconfirmed. I recommend pulling the actual dockets on PACER for 8:16-cv-01774 and 3:20-cv-04677, and CourtListener/RECAP for the IPR papers, before relying on any of the unverified items above.

I'll cross-check the canonical "no proceedings" block against the patent's own litigation metadata, which already flags two PTAB cases. Let me research them.

Let me dig deeper on the two proceedings' outcomes and any related cases.

Let me verify whether sibling proceedings also targeted the '387 patent and check for any appeal.

PTAB / AIA Trial Proceedings — US 6,744,387 ("Method and system for symbol binarization")

Canonical-source conflict — flagged, not resolved in favor of either. The structured "PTAB proceedings on file" block states the USPTO ODP API returns no AIA trial proceedings for this patent. That block is contradicted by both (a) the patent's own Google Patents prosecution/litigation metadata, which lists two PTAB cases in the "Family has litigation" panel, and (b) third-party PTAB databases that carry full docket sheets for both. I treat the ODP block as incomplete/stale for this patent and report the two proceedings below. The ODP's silence is not reliable here — see the caveat at the end of the Strategic Summary.

Proceedings overview

Two AIA trial proceedings have been filed against US 6,744,387 — one institution denied (IPR2017-00963, Amazon) and one terminated pre-institution by joint settlement (IPR2017-01181, AMD/ATI); zero proceedings are active, zero claims were canceled, and zero claims received a merits ruling — so the patent was never narrowed at the PTAB, it simply ran out its full 20-year term and expired on 2022-07-17.

Bottom-line defensive posture: the PTAB record cuts in the defendant's favor only weakly, but the calendar cuts in the defendant's favor absolutely. There is no invalidating Final Written Decision to cite, but (i) the Amazon petition on all six claims was refused institution on the merits, meaning the Board found no reasonable likelihood of prevailing on the art presented, and (ii) the patent is expired, and because neither IPR was instituted, no § 315(e) estoppel attaches to any ground. Prior art against claims 1–6 is thus fully available, un-estopped, and untested by the Board — but the only live value in this patent is historical damages exposure for pre-expiration conduct, which is now constrained by the § 286 six-year lookback.

IPR2017-00963 — Amazon.com, Inc. & Amazon Web Services, Inc. v. Avago Technologies General IP (Singapore) Pte. Ltd.

• Type: Inter Partes Review
• Filed: 2017-02-22
• Status: Institution Denied (Google Patents records this verbatim as "Not Instituted – Merits"; institution decision date reported as 2017-09-13)
• Judge panel: James A. Worth (writing), James B. Arpin, Melissa A. Haapala
• Petition grounds: Challenged claims 1, 2, 3, 4, 5, and 6 — i.e., the entire claim set. Art was drawn from the same body of codec/binarization literature litigated in the parallel ITC and district court matters (the Teuhola "Compression Method for Clustered Bit-Vectors" paper, Itoh & Cheung "Universal Variable Length Code for DCT Coding," Marpe's CABAC work, Talluri, and Winger's own JVT-C162 "Putting a Reasonable Upper Limit on Binarization Expansion"). Statutory basis was § 103 (and § 102 to the extent the references were mapped individually). I could not retrieve the petition itself; the claim list and ground characterizations above are from the case docket metadata, and the specific reference-to-claim mapping is not independently confirmed.
• Institution decision: Denied on the merits — the Board declined under 35 U.S.C. § 314(a). Flagging the limit of my sourcing: I have the disposition ("Not Instituted – Merits") and the panel, but not a verbatim copy of the Board's reasoning, so I am not going to attribute a specific holding (e.g., a particular claim element the references failed to teach) to the panel. What is safe to say is that the Board did not find a reasonable likelihood that any of claims 1–6 was unpatentable on the grounds presented.
• Final Written Decision: None — no FWD issued. Trial was never instituted, so § 318(a) never triggered.
• Settlement / termination: Not settled; the proceeding ended by Board decision at the institution stage.
• Appeal: None. A denial of institution under § 314(a) is effectively unreviewable on appeal (35 U.S.C. § 314(d)); no Federal Circuit docket exists.
• Defensive value: This is the single most useful entry in the record. Amazon challenged all six claims and the Board refused to institute on the merits — a defendant can point to that as the Board's considered view that the strongest art package assembled by a well-resourced petitioner in 2017 did not meet the § 314(a) threshold. But be precise when you use it: a § 314(a) denial is a preponderance-of-the-evidence-against-institution finding on a record the patent owner had a chance to shape in a preliminary response, not a validity adjudication, and it carries no collateral estoppel. Do not over-claim it as "the claims are valid."

IPR2017-01181 — Advanced Micro Devices, Inc. & ATI Technologies v. Avago Technologies General IP (Singapore) Pte. Ltd.

• Type: Inter Partes Review
• Filed: 2017-03-28
• Status: Terminated — settled and dismissed; termination date 2017-09-06 (Google Patents records this verbatim as "IPR2017-01181 filed (Settlement)")
• Judge panel: Not confirmed in the public sources I could reach. Do not assume it was the same panel as IPR2017-00963 — different case, and I have no panel data.
• Petition grounds: Challenged claims 1–6 of the '387 patent (the petition is captioned "PETITION FOR INTER PARTES REVIEW OF CLAIMS 1-6 OF U.S. PATENT NO. 6,744,387"). Petitioner's supporting declaration was from Dr. Vijayakumar Bhagavatula; the exhibit set tracked standard codec prior art (Marpe CABAC ad hoc reports JVT-B015 / JVT-B101, "New Results on Improved CABAC"). Statutory basis § 103. Reference-to-claim mapping not independently confirmed.
• Institution decision: None — the case settled before institution. The Board never ruled under § 314(a).
• Final Written Decision: None. The file shows only a Termination Decision Document (2017-09-06); the Board expressly entered no merits decision.
• Settlement / termination: The parties filed a Joint Motion to Dismiss Petition on 2017-08-30, together with a Joint Request to Treat Information as Confidential and a Confidential Agreement as an exhibit. The Board terminated on 2017-09-06 and issued a Notice of Refund on 2017-09-11 (Petitioner requested refund of post-institution fees on 2017-09-08 — itself confirmation that no trial was instituted). The settlement terms are confidential and were filed with a request to keep them separate from the patent file; nothing about the money, the license, or the scope of the peace is public. The fact of settlement here strongly implies AMD obtained a license, since AMD/ATI was a co-defendant with ASUSTeK in Avago's N.D. Cal. campaign.
• Appeal: None. There was no appealable decision — a pre-institution termination is not a final written decision under § 318(a).
• Defensive value: Limited but real. It confirms AMD, a second sophisticated defendant, also believed all six claims were worth attacking — useful corroboration of invalidity themes. It also confirms the patent owner's pattern of buying peace rather than risking an FWD: given the Amazon petition's merits denial three weeks later (2017-09-13), the inference that Avago settled AMD on favorable-for-AMD terms is not unreasonable, though it is an inference, not a record fact.

Strategic summary

Claim status: all of claims 1–6 remain formally UNTESTED — none canceled, none sustained on the merits. This is the key structural fact and it is unusual. Independent claim 1 (method, hybrid unary/exp-Golomb binarization across a threshold), independent claim 3 (means-plus-function system claim), and independent claim 5 (computer-readable medium) — together with their dependent claims 2, 4 and 6 (the exp-Golomb construction algorithm, including the γ = ⌊log₂(v−(N−2))⌋ prefix/suffix arithmetic) — were each challenged at least once and neither challenge reached a merits ruling. So there is no "claims 1–5 are canceled" argument available here, and equally no hardened "survived two IPRs" story. Anyone telling you either is overstating the record. The claims were, however, substantively litigated to construction elsewhere: in Broadcom Corp. v. Sony Corp., No. 8:16-cv-01052 (C.D. Cal.), the court construed independent claim 3's "means for constructing a codeword using a exp-Golomb binarization..." as a § 112 ¶ 6 means-plus-function limitation, with the corresponding structure being "binarization module 62 and steps 102, 106, 108, 110, and 112 of Figure 4 as described in 7:40–8:10." That is a narrowing construction and it is the most valuable claim-scope holding on this patent — arguably more useful to a defendant than either PTAB outcome. Quoting the Sept. 1, 2017 claim construction order: "the Court finds (1) that the patent defines the algorithm as steps 102, 106, 108, 110, and 112 of Figure 4 as described in columns 7:40-8:10, which contains necessary mathematical formulas, and (2) that binarization module 62 is the part of the computer that performs the function."

Estoppel landscape: there is none, and that is the headline for a current defendant. Section 315(e)(1) and (e)(2) estoppel attaches only after a final written decision in an instituted IPR. Here, IPR2017-00963 was never instituted and IPR2017-01181 was terminated pre-institution. Consequently no statutory estoppel runs against Amazon, against AMD/ATI, or against any privy, on any ground. What this means practically: (a) any defendant — including Amazon and AMD in a future dispute, subject only to ordinary common-law issue-preclusion limits that do not bite absent a merits adjudication — retains the full universe of prior art, including all § 102 anticipation grounds and all § 103 combinations, with no § 315(e) carve-out; and (b) the art that the Board found insufficient to institute in 2017 is still fully usable in district court under the different (lower-for-patentee) § 314(a) threshold vs. the clear-and-convincing invalidity standard, and can be combined differently and with newly located references (e.g., standards-body contributions, JVT/VCEG submissions, and contemporaneous H.264 draft text that were not fully mapped in 2017). The real constraint on a 2026 defendant is not estoppel, it is § 286's six-year damages bar and § 271's requirement of pre-expiration infringement — the patent's adjusted expiration is 2022-07-17, so absent a tolling theory, recoverable damages for even a 2026 claim would reach back only to roughly mid-2020, and a large share of that window sits across the expiration date.

Pattern signals. The same patent owner — Avago Technologies General IP (Singapore) Pte. Ltd., then Broadcom Corporation (assignment recorded 2020-03-10, after the Avago→Avago International→Broadcom chain) — ran a coordinated multi-venue campaign in 2016–2017 against consumer-electronics and cloud defendants, pairing ITC § 337 actions with parallel district court suits and then filing IPRs of its own against other parties' patents (see the 2017-03-13 PTO Litigation Report entry, where LSI Corporation and Avago Technologies U.S., Inc. appear as petitioners against the Regents of the University of Minnesota). Two different petitioners filed on the '387 patent in a five-week window (2017-02-22 and 2017-03-28), and the '387 patent has an intertwined continuation, US 6,982,663, that was itself under review in IPR2017-00964 (and which the Board identified as related to IPR2017-01181, IPR2017-01182 and IPR2017-00963). I did not confirm the outcome of IPR2017-00964 or of IPR2017-01182 within this task — do not treat any statement about them as verified. On the defensive-aggregator question: no, Unified Patents was not a petitioner in either '387 proceeding. Both petitioners were operating companies (Amazon/AWS; AMD/ATI) defending concurrent infringement suits. Unified Patents appears in this patent's history only as the commercial data source whose licensed dataset populated the Google Patents "Family has litigation" panel — it is a data vendor here, not a party. The patent family also shows a further N.D. Cal. assertion, Case No. 3:20-cv-04677, in which the '387 and '663 patents were asserted alongside US 8,724,283 against Netflix, where the defendant pressed RAND-commitment limits and patent misuse / unenforceability defenses on the theory that the patents are essential to H.264/H.265.

Recommended next steps

1. Do not rely on any invalidating FWD — there isn't one. If a demand letter or complaint against you cites claims 1–6 of the '387 patent, your first move is a § 286 / expiration analysis, not an IPR. The patent expired 2022-07-17 (Google Patents: "Adjusted expiration 2022-07-17," status "Expired - Lifetime"). No prospective infringement is possible, and the damages tail is short.
2. Cite the institution denial for the proposition it actually supports. Pull the IPR2017-00963 institution decision (docket: https://portal.unifiedpatents.com/ptab/case/IPR2017-00963; case page: https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2017-00963) and quote only the disposition — denial of institution on claims 1–6 under § 314(a), decision date reported as 2017-09-13. Do not quote it as a validity holding. Verify the decision text via PTAB E2E / PTAB Decisions before filing; I am working from docket metadata, not the opinion text.
3. Pull the IPR2017-01181 file for the settlement posture and the prior-art inventory. Docket page: https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2017-01181. The Joint Motion to Dismiss (2017-08-30), the Confidential Agreement exhibit, and the Notice of Refund (2017-09-11) are the key papers. The settlement terms are confidential — the Board's termination order and the joint confidentiality request kept them out of the public file, so there is no public license rate to mine.
4. Mine the ITC record, which is where this patent was actually fought. The Broadcom / Avago campaign generated ITC § 337 investigations against ASUSTeK, Sony, Vizio, Sigma and others, with detailed invalidity and claim-construction records (see the Commission's determinations at 83 Fed. Reg. 48455–48456 (2018-09-25) and 83 Fed. Reg. 48456 (2018-10-24)). Those public records include the Teuhola (1978), Itoh & Cheung (2000), Marpe/CABAC and Tallon/Winger JVT contributions that both petitioners relied on — the best available non-estopped art springboard.
5. Reuse the Sony claim construction. In Broadcom Corp. v. Sony Corp., No. SACV 16-01774 JVS (JCGX) / 8:16-cv-01052 (C.D. Cal.), the 2017-09-01 order construed claim 3's exp-Golomb means-plus-function term to require binarization module 62 + Figure 4 steps 102, 106, 108, 110, 112 | 7:40–8:10. Verify the order on CourtListener before relying on it; search the RECAP docket index at https://www.courtlistener.com/?type=r&q=6744387.
6. If any proceeding is somehow still open, PTAB timing is now academic. With no institution decisions in IPR2017-00963 (denied) or IPR2017-01181 (terminated), no § 316(a)(11) one-year trial clock ever started or could still be running in 2026. There are no trial-stage milestones to track, no oral hearing, no FWD due date, and no Federal Circuit appeal of any PTAB decision on this patent.

Sourcing caveats (read before relying on this)

• The "no PTAB activity" default in my instructions is wrong for this patent. The ODP-derived block is contradicted by the patent's own litigation metadata and by two independent third-party dockets. Treat the ODP block as incomplete/stale subscription data, and independently confirm current status via PTAB E2E (https://ptab.uspto.gov) and USPTO PTAB Decisions before relying on either source in a filing.
• What I verified: proceeding numbers, petitioners, patent owners, filing dates, dispositions, panel composition for IPR2017-00963, the terminated/settled posture and settlement-paper dates for IPR2017-01181, the claim sets challenged (1–6 in each), and the patent's expiration date.
• What I did not verify (and did not invent): the verbatim reasoning of the IPR2017-00963 institution denial; whether institution was denied as to all or only some claims by the Board's own words (the metadata says the full claim set 1–6 was challenged and the disposition is "Institution Denied," which I have not parsed claim-by-claim); the judge panel for IPR2017-01181; any settlement terms; and the outcomes of the sibling proceedings IPR2017-00964 and IPR2017-01182 against the '663 continuation. Where the record was silent, I said so rather than filling the gap.

I'll cross-reference the provided Google Patents legal-events data against litigation and asserter records before drafting.

US 6744387 B2 — “Method and system for symbol binarization”

Reconstruction of the recorded ownership chain, with NPE/PAE pattern analysis

Primary verification links: Google Patents US6744387B2 (legal events + reel/frame data) · USPTO Patent Assignment Search / Assignment Center · Stanford NPE Litigation Database — patent 6744387

Scope note on the correspondent field. The task asks specifically for the attorney/agent of record on each recording. I could not retrieve correspondent-of-record data for this chain. The Google Patents legal-events feed reproduces reel/frame, conveyance type, execution and recording dates, and the parties, but omits the correspondent box; the USPTO Assignment Center exposes it, but per-record correspondent data was not retrievable in this session. I have therefore marked each correspondent field as not exposed in the sources reviewed rather than guessing — inventing an attorney name here would corrupt the single most useful tell in a chain analysis. (One adjacent, non-assignment data point is available and is noted where relevant: the prosecution correspondence address of record for the continuation US 10/770,213 was LSI Logic Corporation, Attn: Brian K. Young, 1621 Barber Lane, MS: D-106, Milpitas, CA 95035 — that is patent-prosecution correspondence, not the assignment-recording correspondent.)

Inventors

Pattern observations

• No unusual departure pattern. The “all inventors gone within 12 months of filing” red flag does not apply: there is a single inventor, and he assigned to his employer two days before the 2002-07-10 filing date — the normal pre-filing employment-invention assignment signature.
• Single-inventor family. The continuation US 10/770,213 → US 6,982,663 B2 (filed 2004-02-02, priority 2002-07-10, same Family ID 30114182) also names only Winger. No co-inventor drop-off to analyse.
• Inventor remained cooperative and available long-term. He was deposed in the 2023 Broadcom v. Netflix litigation (cited as “Winger Tr.” in Dkt. 625, N.D. Cal. 3:20-cv-04677) — i.e., the inventor was still a live witness ~21 years post-filing, which is inconsistent with the “abandoned startup, inventors vanished” fact pattern.
• Standards-participation context (not an ownership fact). Winger’s JVT contribution JVT-C162 (May 2, 2002) and the draft H.264 recommendation JVT-C167 are cited as evidence in the Netflix claim-construction record (Joint Claim Construction and Prehearing Statement, N.D. Cal. 3:20-cv-04677). This places the invention squarely in the H.264/CABAC standardisation effort that began at VideoLocus.

Original assignee

Entity named on the issued patent (issued 2004-06-01): LSI Logic Corporation, 1621 Barber Lane, Milpitas, California. This is confirmed both by the recorded chain — VideoLocus’s rights were assigned to LSI Logic by 2003-07-09 (Reel 014256/0099) and confirmed by the 2003-12-17 re-recording (Reel 014897/0605) — and by the issued front-page assignee.

• Primary line of business: LSI Logic was a major fabless/ASIC semiconductor supplier — custom silicon and standard products for storage, networking, and consumer video/MPEG codec chipsets, which is precisely the technical neighbourhood of this patent (the specification is written around an encoder’s binarization block feeding a binary arithmetic coder; the encoder/digitiser chips were LSI’s market).
• Did they ship a product embodying the claims? Not established from the sources I reviewed. LSI Logic operated in the right market, but I found no adjudicated or admission-level evidence that a specific LSI shipping product practised claims 1–6. The opposite is asserted in litigation by the current owner’s adversary: Netflix states that “Broadcom practices neither” the ’387 nor the ’663 patent (N.D. Cal. 3:20-cv-04677, Dkt. 625, filed 2023-12-22). I flag that as a party’s litigation position, not a finding, and note the temporal gap — LSI’s H.264-era silicon business was wound into other companies long before 2023.
• Current status: No longer independent. LSI Logic Corporation changed its name to LSI Corporation effective 2007-04-04 (recorded 2012-03-09, Reel 027836/0054), and was acquired by Avago Technologies — the recorded security agreement effective 2014-05-06 (Reel 032856/0031) is consistent with the May 2014 LSI/Avago closing, and title moved to Avago’s Singapore IP-holding subsidiary in August 2014. The LSI business now sits inside Broadcom Inc. (Avago acquired Broadcom Corporation, closing 2016-02-01 — see the Bank of America security agreement effective that same date, Reel 037808/0001 — with the combined company renamed Broadcom Inc. in 2018). This is an acquisition chain, not a dissolution or bankruptcy.
• Predecessor entity: VideoLocus Inc. (Canada), a video-compression startup. Its status today could not be verified from the sources reviewed; its patent rights moved to LSI Logic by recorded assignment in 2003. There is no evidence in the record of a VideoLocus insolvency proceeding, so this should not be read as a fire-sale.

Assignment timeline

Fourteen recorded post-filing events appear in the chain (plus two USPTO certificates of correction, which are not assignments — see note at the end).

1. 2002-07-08 (executed) / recorded 2002-07-10 — Reel 013100/0205

• Conveyance: Assignment
• Assignor: Lowell Winger (inventor)
• Assignee: VideoLocus Inc. (Canada)
• Correspondent: not exposed in the sources reviewed
• Context: pre-filing employment/invention assignment to the inventor’s employer — the first link, and the only one executed by a natural person.

2. 2003-07-09 (executed) / recorded 2003-07-11 — Reel 014256/0099

• Conveyance: Assignment
• Assignor: VideoLocus, Inc.
• Assignee: LSI Logic Corporation (California)
• Correspondent: not exposed in the sources reviewed
• Context: acquisition — the startup’s rights in this patent move to the acquiring semiconductor company (an asset assignment, not a merger; the reel records an assignment of interest).

3. 2003-07-09 (executed) / recorded 2003-12-17 — Reel 014897/0605

• Conveyance: Assignment (confirmatory / “dirty-chain” cleanup)
• Assignors: Video Locus Inc. and LSI Logic Corporation
• Assignee: LSI LOGIC CORPORATION
• Correspondent: not exposed in the sources reviewed
• Context: internal reorg / curative — re-recording the same 2003-07-09 transaction with both grantors named to make the LSI chain of title clean before issuance.

4. effective 2007-04-04 / recorded 2012-03-09 — Reel 027836/0054

• Conveyance: Change of Name
• Assignor: LSI Logic Corporation
• Assignee: LSI Corporation
• Correspondent: not exposed in the sources reviewed
• Context: change of name only — no change in beneficial ownership. Note the 4.9-year gap between legal effectiveness (2007) and recording (2012), typical of a bulk re-recording of a corporate rename.

5. effective 2014-05-06 / recorded 2014-05-08 — Reel 032856/0031

• Conveyance: Patent Security Agreement (collateral pledge)
• Assignor: LSI Corporation and Agere Systems LLC
• Assignee: Deutsche Bank AG New York Branch, as Collateral Agent
• Correspondent: not exposed in the sources reviewed
• Context: securitization — a lien on the portfolio securing corporate debt, not a transfer of title. Effective date corresponds to the LSI/Avago deal closing window.

6. effective 2014-08-04 / recorded 2014-11-06 — Reel 034178/0284

• Conveyance: Assignment
• Assignor: LSI Corporation
• Assignee: Avago Technologies General IP (Singapore) Pte. Ltd.
• Correspondent: not exposed in the sources reviewed
• Context: acquisition — post-closing transfer of the acquired portfolio into the acquirer’s IP-holding subsidiary.

7. effective 2014-08-04 / recorded 2015-02-19 — Reel 035051/0694

• Conveyance: Corrective Assignment (corrects the underlying document of Reel 034178/0284)
• Assignor: LSI Corporation
• Assignee: Avago Technologies General IP (Singapore) Pte. Ltd.
• Correspondent: not exposed in the sources reviewed
• Context: corrective filing — same transaction as #6, re-papered.

8. effective 2014-08-14 / recorded 2015-04-03 — Reel 035390/0388

• Conveyance: Assignment
• Assignor: LSI Corporation
• Assignee: Avago Technologies General IP (Singapore) Pte. Ltd.
• Correspondent: not exposed in the sources reviewed
• Context: third recording of the same 2014 LSI→Avago transfer. Three reels (#6, #7, #8) all describe one economic event — useful caution for anyone counting “transfers” in automated databases.

9. effective 2016-02-01 / recorded 2016-02-02 — Reel 037684/0039

• Conveyance: Termination and Release of Security Interest (releases Reel 032856/0031)
• Assignor: Deutsche Bank AG New York Branch, as Collateral Agent
• Assignee/Releasee: LSI Corporation and Agere Systems LLC
• Correspondent: not exposed in the sources reviewed
• Context: securitization unwind — not a title transfer.

10. effective 2016-02-01 / recorded 2016-02-11 — Reel 037808/0001

• Conveyance: Patent Security Agreement
• Assignor: Avago Technologies General IP (Singapore) Pte. Ltd.
• Assignee: Bank of America, N.A., as Collateral Agent
• Correspondent: not exposed in the sources reviewed
• Context: securitization — new lien replacing #9; effective the same day the Avago/Broadcom Corporation merger closed.

11. effective 2017-01-19 / recorded 2017-02-03 — Reel 041710/0001

• Conveyance: Termination and Release of Security Interest (releases Reel 037808/0001)
• Assignor: Bank of America, N.A., as Collateral Agent
• Assignee/Releasee: Avago Technologies General IP (Singapore) Pte. Ltd.
• Correspondent: not exposed in the sources reviewed
• Context: securitization unwind.

12. effective 2018-05-09 / recorded 2018-10-04 — Reel 047195/0026

• Conveyance: Merger
• Assignor: Avago Technologies General IP (Singapore) Pte. Ltd.
• Assignee: Avago Technologies International Sales Pte. Limited
• Correspondent: not exposed in the sources reviewed
• Context: internal reorg — intra-group merger of the IP-holding entities.

13. effective 2018-09-05 / recorded 2018-11-05 — Reel 047477/0423

• Conveyance: Corrective Merger (corrects the effective date of Reel 047195/0026)
• Assignor: Avago Technologies General IP (Singapore) Pte. Ltd.
• Assignee: Avago Technologies International Sales Pte. Limited
• Correspondent: not exposed in the sources reviewed
• Context: corrective filing.

14. effective 2020-03-10 / recorded 2020-03-10 — Reel 052065/0838 (current title holder)

• Conveyance: Assignment
• Assignor: Avago Technologies International Sales Pte. Limited
• Assignee: Broadcom Corporation (California)
• Correspondent: not exposed in the sources reviewed
• Context: internal reorg — title consolidated into the U.S. operating entity. Same-day execution and recording, and ~4 months before the N.D. Cal. action in which Broadcom Corporation appears as a plaintiff.

Two related legal events that are not assignments: Certificates of Correction dated 2012-05-08 and 2015-03-10 (Examiner’s amendments), and the 2006-12-07 change of entity status from small to large (LSI stopped claiming small-entity status). The patent is recorded Expired – Lifetime, adjusted expiration 2022-07-17 — it is now expired.

Data-quality warning for the analyst using secondary sources: the KIPO/JETRO-derived reporting on this family lists only two transfer records (2012-03-09 LSI Logic → LSI Corporation; 2014-11-06 LSI Corporation → Avago Technologies General IP (Singapore)). That source filters out the security agreements, the releases, the corrective recordings, and the 2020 transfer to Broadcom Corporation. Do not reconcile this chain against third-party “assignment history” counts; go to the reel/frame records.

Timeline diagram

timeline
    title Ownership of US 6744387
    2002 : Filed by VideoLocus Inc
         : Winger assigns to VideoLocus
    2003 : Assigned to LSI Logic Corp
    2004 : Patent issued
    2007 : LSI Logic renamed LSI Corporation
    2014 : Lien granted to Deutsche Bank
         : Assigned to Avago General IP Singapore
    2015 : Corrective re-recordings of 2014 transfer
    2016 : Deutsche Bank lien released
         : Lien granted to Bank of America
    2017 : Bank of America lien released
         : Amazon and AMD file IPRs
    2018 : Merged into Avago International Sales
    2020 : Assigned to Broadcom Corporation
         : Broadcom sues Netflix
    2025 : Netflix claims dismissed with prejudice

NPE / troll-pattern signals

1. Shell-entity transfer — not present.
No assignee in this chain is a single-purpose licensing LLC. Every transferee is an operating company or the wholly-owned IP-holding subsidiary of one: LSI Logic/LSI Corporation (semiconductor supplier), Avago Technologies General IP (Singapore) Pte. Ltd. and Avago Technologies International Sales Pte. Limited (Singapore IP-holding vehicles of an operating semiconductor group, per Reels 034178/0284, 035390/0388, 047195/0026), and finally Broadcom Corporation (Reel 052065/0838). No Delaware or Texas single-member LLC appears; no registered-agent service address appears in the recorded party data. The Singapore “General IP” entities are the closest thing to a holding vehicle, but they sit inside a product company and were created by merger, not for assertion — and the chain ends at a U.S. operating entity, which is the reverse of the shell-entity pattern.

2. Known asserter in the chain — not present.
No assignee matches any published NPE/PAE list (Acacia, Marathon, IV, Wi-LAN, Mosaid/Conversant, Vringo, Pendrell, Innovatio, Round Rock, MPHJ, or Spangenberg entities). The Stanford NPE Litigation Database entry for patent 6744387 classifies the asserting parties in the identified case(s) as “Product company” and “IP subsidiary of product company” — i.e., precisely not the PAE categories (npe.law.stanford.edu/patent/6744387). Broadcom is a large operating semiconductor company with an active IP & Licensing Division; aggressive licensing by an operating company is not the same finding as an NPE. (Whether the specific claims are practised by the owner’s products is contested — Netflix argues Broadcom practises neither the ’387 nor the ’663 patent, Dkt. 625 — but that is a damages apportionment argument, not evidence of a shell structure.)

3. Repeat correspondent across the chain — unclear / not assessable.
I could not obtain correspondent-of-record data for any of the fourteen entries, so I cannot test the recurrence of a single recording attorney. Hypothesis, not finding: given that twelve of the fourteen events are intra-group recordings by one corporate family (LSI → Avago → Broadcom), the recordings were almost certainly made by that family’s in-house IP counsel or a consistent outside firm, and one would expect a repeating correspondent. Verification requires pulling the reel images in the USPTO Assignment Search and reading the correspondent box on Reels 013100/0205, 014256/0099, 014897/0605, 027836/0054, 032856/0031, 034178/0284, 035051/0694, 035390/0388, 037684/0039, 037808/0001, 041710/0001, 047195/0026, 047477/0423 and 052065/0838. I am not naming anyone absent the record. The only attorney name I can source is Brian K. Young, listed as the correspondence address of record for the continuation prosecution (US 10/770,213), at LSI Logic, 1621 Barber Lane, Milpitas — a prosecution correspondent, not an assignment correspondent.

4. Cascading transfers — not present as an NPE indicium.
The chain does show many recordings (three of them, Reels 034178/0284, 035051/0694, 035390/0388, all documenting the single 2014-08-04/2014-08-14 LSI→Avago transfer), but they are corrective duplicates within one corporate family, not successive transfers through unrelated chained LLCs. The substantive title events are separated by years: 2003 → 2014 → 2018 → 2020. No two consecutive transferees are unrelated parties, and no <24-month chain of distinct assignees appears.

5. Pre-litigation transfer — present as timing; not probative.
Title moved to Broadcom Corporation on 2020-03-10, executed and recorded the same day (Reel 052065/0838), months before the N.D. Cal. action (No. 3:20-cv-04677) in which Broadcom Corporation is a named plaintiff. That is the classic “clean standing record before suit” timing. It is not evidence of venue- or assertion-manufacturing here, for two reasons: (a) it is an intra-group transfer between affiliated Broadcom/Avago entities, and (b) this patent had already been asserted by the same corporate family years earlier, while title sat with the Avago entities — Broadcom Corp. v. Amazon.com, 8:16-cv-01774 (C.D. Cal.), per the Stanford NPE Litigation Database. The transfer consolidated title; it did not create the assertion.

6. Bankruptcy fire-sale — not present.
No assignor in the chain filed Chapter 7/11. The LSI→Avago and Avago→Broadcom transitions were negotiated M&A transactions (the security agreements effective 2014-05-06 and 2016-02-01 track those deal closings), not court-supervised asset sales. No reorganisation reels appear.

7. Privateering — not present.
The patent is asserted in the name of the operating company and its affiliated Singapore IP entity (plaintiffs in 3:20-cv-04677 are Broadcom Corporation and Avago Technologies International Sales Pte. Ltd.). There is no separate NPE asserting on Broadcom’s behalf against Broadcom’s competitors, and no third-party NPE appears anywhere in the chain.

8. Defensive aggregator (anti-NPE) — not present.
The chain terminates at Broadcom Corporation (Reel 052065/0838), a large operating company with an active assertion programme — not at RPX, AST, LOT, Unified Patents, or OIN. The inverse signal (patent neutralised) does not apply. Note, however, that third parties have repeatedly attacked the patent defensively from outside the chain: IPR2017-00963 (Amazon, filed 2017-02-22, institution denied) and IPR2017-01181 (Advanced Micro Devices/ATI, filed 2017-03-28, terminated by settlement), plus IPR2021-00542 (Netflix v. Avago) referenced in the N.D. Cal. claim-construction record — all filed by competitors being accused or threatened, the mirror image of NPE litigation.

Verdict

Operating-company assertion.

The chain runs VideoLocus → LSI Logic (Reel 014256/0099, 2003) → LSI Corporation by name change (Reel 027836/0054) → Avago Technologies General IP (Singapore) (Reels 034178/0284 and 035390/0388, 2014) → Avago Technologies International Sales (Reel 047195/0026, 2018) → Broadcom Corporation (Reel 052065/0838, 2020). Every link is either an acquisition or an internal corporate step inside one semiconductor corporate family; the only non-title recordings are two collateral security agreements and their releases (Reels 032856/0031, 037808/0001, 037684/0039, 041710/0001), which is classic large-cap corporate securitisation, not NPE behaviour. The patent was asserted by the operating family itself against actual competing products — Broadcom Corp. v. Amazon.com, 8:16-cv-01774 (C.D. Cal.), classified by the Stanford NPE Litigation Database as “Product company / IP subsidiary of product company,” and Broadcom Corp. v. Netflix, 3:20-cv-04677 (N.D. Cal.), where the claims were ultimately dismissed with prejudice per the court’s order of 2025-04-07 (CourtListener docket 17347564, Dkt. 730).

Two qualifiers for the record: (i) the patent expired 2022-07-17 (adjusted expiration), so no further assertion value remains; and (ii) the current owner runs a deliberate video-codec licensing programme and asserts standard-essential H.264/H.265 claims against implementers, with the defendants countering on RAND and patent-misuse theories — an aggressive operating-company monetisation posture, which is a business-model observation, not an NPE classification. The one unverified item that could sharpen this analysis is the correspondent of record on the fourteen reels above; if that box shows a single outside firm running the Avago/Broadcom recordings rather than corporate in-house counsel, it would be worth a second look — but on the evidence actually recorded, all eight NPE signals are absent or non-probative.

I'll begin by verifying the patent via search, then work through each cited reference.

Let me check the IPR proceedings that challenged this patent, since those often surface the most probative prior art.

Prior-Art Analysis — U.S. Patent No. 6,744,387 B2

Important framing note up front: U.S. 6,744,387 carries exactly one (1) U.S. patent citation on its face. The remainder of its citation list is non-patent literature (4 items). I have kept the scope strictly to 6,744,387 and have not substituted results for the similarly-numbered US 6,982,663 (its own continuation) or US 6,744,382 (which appears in the same examiner search log).

1. Reference Identification (the patent under analysis)

Note: a Certificate of Correction issued 2012-05-08. Per the PTAB, corrected claim 1 reads "determining a code symbol index value; if said code symbol index value is less than a threshold value, constructing a codeword using a unary binarization; and if said code symbol index value is not less than said threshold value, constructing a codeword using a exp-Golomb binarization."

2. The Complete Patent Citation List of US 6,744,387

That is the entire patent-citation set. Google Patents separately lists a "Family Cites Families" group (US 6,636,222 B1; US 6,236,960 B1; US 6,850,568 B1; ES 2270246 T3), but these are family-level citations (associated with the '663 continuation), not citations of record against 6,744,387 itself. They are flagged here for completeness only and should not be attributed to the '387 patent.

3. Reference-by-Reference Analysis

3.1 US 5,471,207 A — Zandi, Stork & Allen (Ricoh)

Full citation: U.S. Patent No. 5,471,207; inventors Ahmad Zandi (Cupertino, CA), David G. Stork (Sanford), James Allen (Mountain View); assignees Ricoh Company Ltd. (Tokyo, JP) and Ricoh Corporation (Menlo Park, CA); Appl. No. 200,233; primary examiner Howard L. Williams.
Dates: U.S. priority/filing 1994-02-23; granted / published 1995-11-28.
Foreign family: FR 2 716 587 A1 (pub. 1995-08-25), CN 1117222 A, DE 695 158 32 T2 (EP counterpart family).
Sources: https://patents.google.com/patent/US5471207 ; https://patentimages.storage.googleapis.com/fa/ea/26/af741ef68aaafc/US5471207.pdf ; https://patents.justia.com/patent/5471207

Brief description. The reference addresses compression of palettized images, generalized as compressing a sequence of M-ary alphabet input symbols. Its core contribution is a reindexing table that maps each input symbol to a binary sequence; the mapping is chosen to minimize bitwise entropy and is transmitted with the compressed data as overhead (or fixed in advance). Binarization is context-model driven (bit-position, bit-plane, bitwise-dependent, or mixed context models). The compressed bitstream then passes through a context modeler, a probability-estimation module (PEM), and a bit generator — i.e., a binary arithmetic-coding architecture. A second aspect concerns a parallel compression/decompression pipeline with bit buffering and reordering so that context bits are decoded before the bits that depend on them. Representative claims include claim 1 (compression method with context model), claim 14 (compression engine with reindex table generator), claim 16 (binary sequences of fixed length K ≥ log₂M), and claim 17 (compressor comprising context modeler, probability estimation module, bit generator).

Statutory footing relative to 6,744,387 (filed 2002-07-10). Granted 1995-11-28, i.e. more than one year before the '387 filing date. It therefore qualifies as prior art under pre-AIA § 102(a), § 102(b) (statutory bar), and § 102(e) (U.S. patent granted on an application filed before the applicant's invention date).

Anticipation analysis — § 102.

Conclusion — anticipation: No claim of US 6,744,387 is anticipated by US 5,471,207 under § 102(a), (b), or (e). The reference's binarization is table-driven and entropy-optimized, not threshold-switched between a unary region and an exp-Golomb region. It contains no disclosure of a threshold value N (=16 or =64 in the '387 Tables 3–4), no unary code of v 1's terminated with a 0, and no exp-Golomb suffix of the form claimed. The one point of genuine overlap — both convert M-ary symbols to binary for a binary arithmetic coder — is a field-of-use overlap, not an element-by-element identity, and therefore cannot support anticipation.

Proper role of this reference: It is best characterized as a § 103(a) background/obviousness reference — evidence that (i) binarizing M-ary symbol indices ahead of a BAC was known, and (ii) selecting a binarization to suit the encoder's statistics was known. It supplies the general framework but contributes nothing to the novel threshold-switching mechanism. Note that neither the Examiner nor either IPR petitioner ultimately relied on it as the primary anticipatory reference.

3.2 Non-Patent Citations (4) — as listed on the face of 6,744,387

These are the remaining citations. They are listed literally as they appear on the patent; two are incompletely identified in the source.

§ 102 assessment for the NPL citations: None of the four anticipates any of claims 1–6. Huffman (1952) and the 1966 IEEE paper are general entropy-coding/prefix-code references; O'Neil concerns predictive quantizing, not binarization; Teuhola (1978) is the only one approaching the claimed subject matter, and it discloses clustered bit-vector compression, not the specific threshold-switched unary/exp-Golomb construction recited in claims 1–2. All four are properly § 103(a) background art.

4. Most Relevant Prior Art Overall — and the Reason It Is Not the Examiner's Citation

The most probative prior art against 6,744,387 was never cited on the patent's face. It surfaces in IPR2017-00963 (Amazon.com, Inc. et al. v. Avago Technologies General IP (Singapore) Pte. Ltd.), filed 2017-02-22, decided by the Board on 2017-09-14 (Decision Denying Institution, Paper 10; APJs Arpin, Worth, Haapala).
Source: https://ptacts.uspto.gov/ptacts/public-informations/petitions/[1483354](/patent/1483354)/download-documents?artifactId=erfaGUCMkxkRnFPg9tXZlJZqtOCqHndLovHzulAmdu-3hYeHFaq-uw4

Grounds asserted against claims 1–6:

• Fenwick — Peter Fenwick, "Experiments with a Block Sorting Text Compression Algorithm," Technical Report 111, ISSN 1173-3500, Dept. of Computer Science, University of Auckland, May 17, 1995.
• Talluri — U.S. Patent No. 6,304,607 B1, published Oct. 16, 2001 (cited at 4:43–60, 5:08–6:14, Table 1 and Table 2).
• Teuhola — as in § 3.2, item 1 above.
• Wien — Matthias Wien, "ABT Coding for Higher Resolution Video," JVT of ISO/IEC MPEG & ITU-T VCEG.

Outcome: The Board denied institution on all four grounds — the petitioner failed to show a reasonable likelihood of prevailing on any of claims 1–6. The IPR2017-01181 petition by Advanced Micro Devices, Inc. / ATI Technologies (filed 2017-03-28, effective date 2017-03-28) was terminated by settlement. https://natlawreview.com/article/pto-litigation-report-march-29-2017

Why this matters for your question: Any competent § 102 analysis of 6,744,387 must lead with the fact that the examiner-cited art (US 5,471,207) does not anticipate, while the litigation-tested art (Wiegand, Cheung) was asserted as § 102(a)/(b) anticipation art and failed at the institution stage. Wiegand (the May 2002 H.264 Joint Committee Draft) is notable because it post-dates the '387 priority date (2002-07-10) by only two months and was therefore asserted under § 102(a) — a tight but potentially fatal date problem had the Board accepted the petition's claim mapping.

5. Bottom-Line Table

6. Explicit Caveats

1. There is no second patent citation. If you expected a longer citation list under 6,744,387, it does not exist; US 5,471,207 is the sole patent reference of record. Do not conflate it with the family citation set belonging to continuation US 6,982,663 (which lists US 6,236,960 B1, US 6,636,222 B1, US 6,850,568 B1, and US 2004/0114683 A1).
2. I have not verified the 1966 IEEE article's authorship or exact title. The patent's face omits both. I am declining to supply an identification I cannot ground.
3. Claim 3 and claim 6 contain apparent drafting errors on the printed face — claim 3's second "means for" recites the exp-Golomb branch as operating "if said code symbol index value is less than a threshold value" (it should read "is not less than"), and claim 6 recites "equal in number to said threshold value" (omitting "minus one") plus the literal string "v−(N=2)". I have interpreted these literally and not auto-corrected them, consistent with your instruction. This matters for any § 112 ¶ 6 analysis of claims 3–4, which the Board addressed in its claim-construction section.
4. IPR art is not § 102 "prior art of record." The Wiegand/Cheung/Fenwick/Talluri/Teuhola/Wien references were never cited by the Examiner against 6,744,387 and appear nowhere on its face; they enter the analysis only via the 2017 PTAB petitions.

I'll research the prior art references identified on this patent's page before building the § 103 analysis.

Key finding: the "Article — A Compression Method for Clustered Bit-Vectors" is Teuhola 1978 — the origin of the Exp-Golomb code. Let me dig into the IPR/litigation prior art and the remaining references.

This is highly informative. Let me pull the actual IPR petition grounds and the specific references invoked against this patent.

Obviousness Analysis — US 6744387 B2 ("Method and system for symbol binarization")

0. Scope, caveats, and the evidentiary record I'm working from

Governing law. US 6744387 was filed 2002‑07‑10 and granted 2004‑06‑01, so it is a pre‑AIA patent. The applicable test is pre‑AIA 35 U.S.C. § 103(a) as applied through Graham v. John Deere (scope/content of prior art; differences; PHOSITA level; secondary considerations) and KSR Int'l v. Teleflex (predictable combinations of known elements, "finite number of identified, predictable solutions," design incentives).

Important scope note. The prior-art of record on the face of the patent is thin — one patent citation and four non-patent citations. All of it is old, general-purpose coding literature. The real § 103 attack on this patent has come from litigation/IPR art (Wiegand, Cheung, Fenwick, Talluri, Teuhola, Wien), which appears on the page only through the litigation links. I analyze both sets, and I flag which is which, because that distinction drove the actual PTAB outcome.

Claim-text caveat (I am not auto-correcting). As fetched, several claims contain apparent typographical/scriveners' errors: claim 3's exp‑Golomb means-element reads "less than a threshold value" (surely meant "not less than"); claim 4 recites "v−(N=2)−2**γ"; claim 6 recites an initial prefix "equal in number to said threshold value" (no "minus one"), "v−(N=2)", and "exp‑Golomnb." Two certificates of correction were issued (2012‑05‑08 and 2015‑03‑10) per the page's Legal Events, but the corrected text is not in the record supplied to me. I analyze the claims as reproduced and flag where the errors matter. (Confidence: the certificates exist; their content is unverified.)

PHOSITA. From the parallel-family record, a POSITA is someone with a bachelor's degree in EE/computer engineering/mathematics plus ~2 years of data-compression experience (or a master's with a compression specialization), who understands information theory, data compression, and video standards (JPEG, MPEG‑1/2, H.261, H.263), and who can "combine, customize, and/or optimize coding techniques based on the characteristics of the data to be compressed." (Netflix petition excerpt, https://ptacts.uspto.gov/ptacts/public-informations/petitions/[1546464](/patent/1546464)/)

1. What the claims actually require

The only thing separating claim 1 from the admitted background is the threshold switch. Unary and exp‑Golomb binarization are both expressly described in the patent's own Background as known, with known trade-offs ("small codewords should be distinguishable as with a unary binarization, while large codewords should have their binarization limited to a reasonable length"). That passage is a binding admission of the prior art and of the problem to be solved — a powerful starting point for a § 103 case.

2. The prior art available on the page

(a) Of record on the patent:

• US 5,471,207 A (Zandi et al., Ricoh, 1995‑11‑28) — "Compression of palettized images and binarization for bitwise coding of M‑ary alphabets therefor." (https://patents.google.com/patent/US5471207) Discloses binarizing M‑ary symbols, with the binarization selected to give good compression by a binary coder, via a reindexing table that maps each symbol to a number of binary values and is a local minimum of bitwise entropy.
• Teuhola, "A Compression Method for Clustered Bit-Vectors," Inf. Process. Lett. 7(6):308‑311 (Oct. 1978) — the origin of the exp‑Golomb code ("Teuhola code"); the literature universally cites it for exp‑Golomb. (https://scite.ai/reports/a-compression-method-for-clustered-1maX0n; https://researchr.org/publication/Teuhola78)
• Golomb, "Run-length encodings," IEEE Trans. Info. Theory, pp. 399‑401 (1966) — the Golomb/Rice family: unary prefix + binary postfix with parameter k.
• Huffman, "A Method for the Construction of Minimum-Redundancy Codes" — variable-length/prefix coding fundamentals.
• O'Neil, "Predictive Quantizing Systems" (ms. rec'd 1965‑12‑27) — predictive quantization; residuals whose magnitudes are concentrated at small values.

(b) Cited by family members: US 6,636,222 B1 (Broadcom), US 6,236,960 B1 (Motorola — factorial packing), US 6,850,568 B1 (Sony — MPEG video decoding), ES 2 270 246 T3 (Fraunhofer — transform-coefficient coding; underlying priority 2002‑05‑02).

(c) Litigation/IPR art (from the page's litigation links): Wiegand (Draft ISO/IEC 14496‑10 AVC, May 2002), Cheung ("Configurable Variable Length Code for Video Coding," May 2001), Fenwick ("Experiments with a Block Sorting Text Compression Algorithm," 1995), US 6,304,607 (Talluri), Teuhola, and Wien ("ABT Coding for Higher Resolution Video," Jan. 2002).

3. The combinations

Ground A — Teuhola 1978 + Golomb 1966 (+ Huffman; + O'Neil)

• Unary binarization: Golomb 1966 and Huffman 1952 supply unary-prefix coding of small indices; the patent concedes unary binarization is known.
• Exp‑Golomb binarization: Teuhola 1978 discloses exactly the exp‑Golomb code — alphabet divided into groups of exponentially increasing size, with a unary-length prefix and a binary suffix.
• Threshold switch: the gap. Neither Teuhola nor Golomb describes switching from a pure unary code to exp‑Golomb at an index threshold.
• Motivation: the patent's own Background frames the invention as the obvious compromise between the two known codes' known drawbacks (unbounded length on one side; loss of per-bin distinguishability on the other). Under KSR, combining two known coding schemes, both designed for the same purpose, to obtain the predictable benefit of each, is the paradigm of obviousness. O'Neil supplies the statistical predicate — quantized residuals cluster at small magnitudes, so a code that is short near zero and length-bounded at large values is exactly what the data calls for.

Assessment: strong as to claims 1, 3, 5 if a threshold-switching teaching is supplied from somewhere in the art; without such a teaching this combination is incomplete, because claim 1's only point of novelty is the threshold.

Ground B — US 5,471,207 (Zandi) + Teuhola + Golomb

Zandi supplies the missing design methodology: binarization is a design choice, selected so the mapped binary string compresses well in the following binary coder, with the mapping optimized as a function of the symbol distribution. That is precisely the rationale a POSITA would use to select unary for the low-index region (where it uniquely distinguishes each symbol and permits separate context statistics per bin — the very advantage the patent recites) and exp‑Golomb for the tail (where bounded length matters more than distinguishability). Under KSR's "design incentive" and "predictable results" rationales, this is a textbook combination.

Assessment: good for claims 1/3/5; Zandi also helps on claims 2/4/6 by framing the codeword mapping as a tunable design parameter.

Ground C — Fenwick (or Wien) + Talluri + Teuhola — the combination actually petitioned in IPR2017‑00963

This is the strongest § 103 theory on the record, and it maps element-for-element:

• Fenwick ("Experiments with a Block Sorting Text Compression Algorithm," 1995) discloses two coding regimes separated by a threshold: unary-style coding for low values, and "for higher values (where the distributions flatten out), a more conventional arithmetic coding is appropriate," with a described prefix of six 1's. That is the threshold-based switch.
• Talluri (US 6,304,607) supplies exp‑Golomb coding in a video/image context ("very well matched to the run-length coded data that occur in quantized image transforms").
• Teuhola independently supplies exp‑Golomb.
• Petitioner's stated motivation: "substituting exp‑Golomb coding for Fenwick's coding is simply a design choice from a finite number of predictable solutions with a reasonable expectation of success." (Amazon petition, Ground 3; quoted in Avago's Preliminary Response, https://ptacts.uspto.gov/ptacts/public-informations/petitions/[1483354](/patent/1483354)/)

Ground D — Wien + Talluri + Teuhola substitutes Wien ("ABT Coding for Higher Resolution Video," Jan. 2002) for the threshold teaching, coupling it to exp‑Golomb from Talluri/Teuhola.

Assessment: Ground C/D are the analytically complete attacks. Their weakness is articulated motivation, not element coverage.

4. Claims 2, 4, 6 (the specific codeword construction)

The recited construction — (N−1) ones, then unary γ where γ = ⌊log₂(v−(N−2))⌋, then the γ LSBs of v−(N−2)−2^γ — is the standard exp‑Golomb structure with a constant offset (N−2) chosen so the code remains a true prefix code continuous across the threshold. Teuhola already discloses the length-prefix + γ-LSB architecture; the offset is the mathematically compelled adjustment when you splice a bounded-length unary region onto an exp‑Golomb tail (otherwise codewords collide or gaps appear). A POSITA implementing the claim‑1 hybrid has essentially one degree of freedom at the junction, and KSR treats such an optimization as "a design choice from a finite number of identified, predictable solutions."

Counterpoint (real, and material): the offset algebra is specific, and a challenger must still show why a POSITA would land on exactly N−2. In IPR2017‑00963 the patent owner argued claims 2/4/6 were not shown to be met at all by the proposed combination. Also, the claim-text errors in claims 4 and 6 ("N=2," missing "minus one") would need to be resolved against the certificate of correction before any final validity opinion.

5. Why a POSITA would combine — consolidated motivation

1. Same field, same problem — all references address entropy coding of symbol sequences fed to a binary coder.
2. Known, documented trade-off — the patent itself states the unary vs. exp‑Golomb trade-off; selecting the better code by index magnitude is a predictable solution.
3. Express teaching of the threshold mechanism — Fenwick (and Wien) teach unary for low values and a different coder for higher values precisely because the distribution flattens out — the identical rationale.
4. Express teaching of exp‑Golomb in video — Talluri and Teuhola.
5. Design-incentive rationale — arithmetic-coder complexity scales with the number of bins, so truncating unary growth at a threshold is an efficiency-driven, expected result.
6. Zandi's optimization methodology — binarization is selected to optimize compression in the following binary coder.

6. Counterevidence the analysis must confront

• The claim-1 threshold did real work at the PTAB. Institution was denied in IPR2017‑00963 (Amazon) on 2017‑09‑13; the Board found Amazon had failed to establish most references as prior art, and considered only Cheung on the merits — rejecting it for failing to disclose the "threshold." (https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2017-00963; https://services.patexia.com/lawsuits/Amazoncom-inc-et-al-v-Avago-Technologies-General-IP--Singapore--Pte-Ltd-id-74249) The identical reasoning appears in the parallel '663 decision, IPR2017‑00964 Paper 15.
• IPR2017‑01181 (AMD/ATI) settled before institution; the Board never considered that art.
• Wiegand is not § 102(a) art: the ITC determined the relevant sections of the H.264/JVT draft were solely attributable to Winger, the named inventor (his 2002‑05‑02 email to jvt-experts@mail.imtc.org is Ex. 1013). An inventor's own work is not prior art under § 102(a) and, being within the grace period, is not a § 102(b) bar either.
• Patent owner's attacks on motivation: Fenwick is text compression with a 256‑symbol arithmetic coder for <100 ranked symbols, while video symbol indices "may range into the tens of thousands" — i.e., a POSITA would not start from Fenwick. That is a genuine teaching-away/field-of-invention argument and it succeeded at the institution stage.
• No established secondary considerations appear in this record (no nexus, licensing, copying, or industry-praise evidence identified). Conversely, the accused-infringer set (Amazon, AMD/ATI, ASUS, Sony) itself evidences commercial significance, which cuts both ways.

References that are NOT prior art (so they cannot support a § 103 ground): the "Cited By" and "Families Citing this family" entries all post-date 2002‑07‑10 — e.g., EP 2141815 A1 (Thomson, "Method for a hybrid Golomb-Elias gamma coding," 2008), KR 101391601 B1 (Samsung, exp‑Golomb with optimal thresholds, 2007), US 8,854,233 B2 (MediaTek, de‑binarization, 2013), and the Fraunhofer US 2005/0117652 A1 (published 2005). They are useful only as hindsight confirmation that the field converged on hybrid unary/exp‑Golomb codes — which is exactly why they should not be used in the analysis.

7. Bottom line

• Claims 1, 3, 5 are, in my assessment, vulnerable but not clearly invalid on the art listed: the combination of Teuhola 1978 + Golomb 1966 + (Zandi US 5,471,207 or Fenwick or Wien) supplies every element, and the KSR motivation is well supported by the patent's own admission of the unary/exp‑Golomb trade-off plus Zandi's "select-the-binarization" methodology. The single hardest limitation to prove is the threshold comparison, and the PTAB's refusal to credit Cheung on that point shows it is a real, non-trivial narrowing limitation rather than a formality.
• Claims 2, 4, 6 are more likely obvious on the same art, because the recited construction is the standard exp‑Golomb code with the offset arithmetically required to preserve prefix-freeness at the junction — a predictable design choice. Confidence here is lower until the certificate-of-correction text is obtained, given the "N=2" and "threshold value" (vs. "threshold value minus one") errors in the printed claims.
• The best-supported single ground is Fenwick (threshold switch) + Talluri (exp‑Golomb, video) + Teuhola (exp‑Golomb), i.e., Amazon's Ground 3; Wien + Talluri + Teuhola (Ground 4) is a close second, with the caveat that Wien's own public availability (Jan. 2002) must be proven.
• Practical caveat: the historical record shows this exact combination was petitioned and denied institution on the merits of admissibility and the Cheung ground — not on a full merits adjudication of Fenwick/Talluri/Teuhola, which the Board never reached because the parties settled or the references were not established as prior art. A new challenge that properly authenticated Fenwick's and Wien's public accessibility and supplied a rigorous, non-hindsight motivation to combine (with emphasis on the bin-count/complexity rationale in the specification) would be materially stronger than the 2017 petitions.

Confidence: high on the identity, dates, and content of Teuhola 1978 (it is the exp‑Golomb source) and on the IPR grounds/outcomes; moderate on the precise teachings and public-availability dates of Fenwick and Wien, which I have only through the petition and patent-owner characterizations; and explicitly unknown as to the corrected text of claims 2, 4 and 6 after the 2012‑05‑08 and 2015‑03‑10 certificates of correction.