US 6982663

US Patent 6,982,663 — Summary

Note on scope of this answer: I retrieved the authoritative full text (Google Patents, https://patents.google.com/patent/[US6982663](/patent/US6982663)/en). The patent number was searched literally; every figure below comes from either that document or specific search hits cited inline. I flag one date discrepancy explicitly.

1. Bibliographic data

2. Abstract (verbatim)

"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."

3. What the invention actually is (plain language)

Binary arithmetic coders (BACs) only accept binary input, so video symbols (transform-coefficient levels, motion-vector residual magnitudes) must first be binarized. The two prior approaches each have a flaw:

• Unary codes (0, 10, 110, 1110, …) let the coder statistically separate each small symbol, but a large index produces an enormous bin string (potentially tens of thousands of bins).
• exp-Golomb codes cap length but lump small indices together, hurting compression.

The patent's scheme is hybrid: use pure unary below a threshold N, then switch to a code whose prefix is a run of (N−1) ones followed by a truncated unary "length" field, with a γ-bit binary suffix appended. N = 64 is given for quarter-pixel motion-vector residual magnitudes (Table 3 / FIG. 5); N = 16 for coefficient levels (Table 4 / FIG. 6). Construction algorithm (from the specification):

• If v < N: emit unary — v ones terminated by a 0.
• If v ≥ N: (1) initial prefix of (N−1) ones; (2) compute γ = ⌊log₂(v−(N−2))⌋ and emit it unary, appended to the prefix; (3) append the γ least-significant bits of g = v−(N−2)−2^γ as the binary suffix.

The result is a true prefix code that also compresses well for MPEG‑AVC/H.264 (ITU‑T H.264 / ISO/IEC 14496‑10).

4. Independent claims — plain-language overview

Claim 1 — Method, decode side (generating an index value from a codeword):
Look at a codeword's first portion. If that first portion matches a "first pattern" (the all-ones prefix), then: (A) set the index value equal to a threshold; (B) add an offset derived from the second portion of the codeword that follows the first; and (C) add a further value derived from the third portion that follows the second. In essence: the decode-side mirror of the hybrid code — the fixed all-ones prefix means "start counting at N," the second field gives the exp-Golomb length/offset, the third field gives the low-order bits.

Claim 11 — System, decode side:
A system comprising (i) a decoder configured to generate a codeword and (ii) a circuit configured to perform the same three steps as claim 1 (set index to threshold on detecting the first pattern; add offset from the second portion; add value from the third portion). This is the hardware/system counterpart of the claim 1 method, with the decoder recited as a separate element.

Claim 12 — Method, encode side (generating a codeword from an index value):
When the index value is at least the threshold: (A) generate the first pattern in the codeword's first portion; (B) generate a second pattern in the following portion representing an offset of the index value above the threshold; and (C) generate a third pattern in the next portion representing a value of the index value above that offset (the residual low-order bits). This is the encoder-direction counterpart of claim 1.

Claim 21 — System, encode side:
A system comprising (i) a circuit configured to generate the first, second, and third patterns as in claim 12, and (ii) an encoder configured to encode the codeword. The encoder-direction counterpart of claim 11.

Notable dependent-claim limitations (for context, not independent): claims 2/4/5 and 13/15 cover the below-threshold case as a run of zero-or-more first-state bits terminated by an opposite-state bit (classic unary); claim 3/14 require the first pattern to be unique from all possible representations of the below-threshold pattern (preserving the prefix code); claim 7 requires the second pattern to be unary-style; claim 8 requires the third pattern to be a binary number; claims 9 and 20 recite H.264 / ISO‑IEC 14496‑10 compatibility; claims 18–19 state that the second and/or third portions are void when the index is below the threshold.

5. Litigation / 2026 docket check

I found no 2026 USPTO or CAFC docket activity naming US 6,982,663 specifically. The reason is straightforward: the patent's term expired on 2022‑07‑17, so it cannot be asserted in new 2026 infringement actions.

What the record shows for this patent is historical:

• PTAB: IPR2017‑00964 (Petitioner Amazon.com, Inc. / Amazon Web Services, Inc.; Not Instituted – Merits); IPR2017‑01182 (Petitioner Advanced Micro Devices, Inc. / ATI Technologies ULC; terminated by Settlement); IPR2021‑00468 (Petitioner Netflix, Inc. / Netflix Streaming Services, Inc.; Not Instituted – Merits).
• District court / ITC: multiple suits between roughly 2009 and 2020, including Broadcom Corp. et al v. Amazon.com, Inc. et al, C.D. Cal. 8:16‑cv‑01774 (Stanford NPE Litigation Database, https://npe.law.stanford.edu/patent/6982663); an ITC investigation 337‑TA‑837 (LSI v. Realtek et al.); and several N.D. Cal. cases (e.g., 3:20‑cv‑04677, 5:20‑cv‑04677).
• CAFC: the patent's Google Patents litigation record lists older appeals (14‑1410 and 17‑1491), not 2026 proceedings.

2026 Broadcom/video-codec activity that exists but does NOT involve the '663 patent (so it should not be conflated): Netflix, Inc. v. Broadcom Inc., Fed. Cir. No. 26‑1329 (appeal filed Jan. 12, 2026; briefing ongoing through 2026) — this concerns a different set of patents; and Bayerische Motoren Werke AG v. Broadcom Inc. et al, E.D. Tex. No. 2:26‑cv‑00775 (declaratory-judgment action filed 2026, exhibits are US 7,835,389; US 7,936,778; 9,137,672; and others — not 6,982,663). Likewise, TexasLDPC Inc. v. Broadcom Inc., Fed. Cir. No. 25‑1074 (Sept. 14, 2026) concerns LDPC patents, not this one.

Uncertainty note: PACER/CAFC docket listings are not exhaustively indexed by the sources I reached, so I cannot positively exclude a stray 2026 filing referencing an expired '663 patent (e.g., a damages/prior-art dispute or a license-related declaratory action). Based on the patent's expired status and everything I could retrieve, however, there is no active 2026 litigation on US 6,982,663 that I can substantiate.

6. Sources

• Google Patents, US 6,982,663 B2: https://patents.google.com/patent/US6982663/en
• Unified Patents patent page: https://portal.unifiedpatents.com/patents/patent/6982663
• Justia (pre-grant pub. US 2004/0150540): https://patents.justia.com/patent/20040150540
• Stanford NPE Litigation Database: https://npe.law.stanford.edu/patent/6982663
• Fed. Cir. docket 26‑1329 (Netflix v. Broadcom): https://dockets.justia.com/docket/circuit-courts/cafc/26-1329
• E.D. Tex. 2:26‑cv‑00775 (BMW v. Broadcom et al.): https://cdn.pacermonitor.com/public/case/66559993/

I'll research litigation involving US patent 6982663. Let me run several searches.

Let me gather more detail from the key litigation databases.

Litigation Involving U.S. Patent No. 6,982,663 ("Method and System for Symbol Binarization")

Patent identity (confirmed): U.S. Patent No. 6,982,663 B2, application 10/770,213, filed February 2, 2004, granted January 3, 2006; continuation of U.S. 10/191,596 (now U.S. 6,744,387); priority date July 10, 2002; inventor Lowell Winger; original assignee LSI Logic Corp.; current assignee Broadcom Corp. Status: Expired – Lifetime (expired 2022-07-17).

Important caveat before the list. The '663 patent is one member of a two-patent family (with U.S. 6,744,387). Broadcom/LSI historically asserted these H.264-related patents together in the same complaints, so many docket entries below involve the '663 as one of several asserted patents. Google Patents' "Family has litigation" feed for US6982663 lists the cases below; I have independently confirmed the identity of several of them, but I could not verify plaintiff/defendant and outcome for every docket number before hitting my search limit. Where a fact is unverified I say so rather than guess. All alphanumeric identifiers below are reproduced literally as found.

A. District court cases (U.S.)

B. International Trade Commission

C. Federal Circuit appeals

• Appeal No. 14-1410 — listed in the Google Patents litigation feed as a Federal Circuit case for this family. Parties and disposition unverified.
• Appeal No. 17-1491 — listed in the same feed. Parties and disposition unverified.

D. PTAB proceedings (AIA reviews)

Each of these three IPRs identifies Respondent Patent Number 6982663 and application 10/770,213, so they are directly tied to this patent.

E. What I could NOT confirm / negative results

• No litigation is not the correct answer — there is substantial litigation.
• I could not retrieve the Stanford NPE Litigation Database page (npe.law.stanford.edu/patent/6982663) in full; search snippets showed at least two C.D. Cal. entries (8:16-cv-01774 and one more, both naming Broadcom). The full case list there likely covers additional defendants I have not captured.
• I could not confirm plaintiff/defendant or outcome for docket numbers 8:16-cv-01052, 8:20-cv-00529, 2:15-cv-00239, 3:15-cv-04525, 3:15-cv-05427, 5:15-cv-04307, 3:16-cv-00451, 2:12-cv-02047, 3:11-cv-02709, or the two Federal Circuit appeals (14-1410, 17-1491).
• Not attributed to this patent: the Dutch proceeding Broadcom v. Netflix (Rechtbank Den Haag, ECLI:NL:RBDHA:2025:24060), which concerns a "single binary tree" / HEVC prediction- and partition-mode encoding patent — a different patent from the '663. I am flagging it only to avoid conflating it with the U.S. '663 litigation.

Key sources

• Google Patents, US6982663B2 (litigation, family, PTAB, assignment data): https://patents.google.com/patent/US6982663/en
• Unified Patents patent page: https://portal.unifiedpatents.com/patents/patent/6982663
• Unified Patents litigation/case list: https://portal.unifiedpatents.com/litigation/caselist
• PTAB case details (GreyB/IPVerse): IPR2017-00964, IPR2017-01182, IPR2021-00468 — https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2017-00964 ; .../IPR2017-01182 ; .../IPR2021-00468
• Stanford NPE Litigation Database: https://npe.law.stanford.edu/patent/6982663
• SanDisk v. LSI, N.D. Cal. C 09-02737: https://docs.justia.com/cases/federal/district-courts/california/candce/3:2009cv02737/[218036](/patent/218036)/55
• Broadcom v. Netflix, N.D. Cal. 3:20-cv-04677 (Joint Claim Construction Statement; Netflix trial brief re '663, '387, '283): https://storage.courtlistener.com/recap/gov.uscourts.cand.[362366](/patent/362366)/
• Avago/Broadcom v. Amazon, C.D. Cal. 8:16-cv-01774 docket: https://portal.unifiedpatents.com/litigation/California%20Central%20District%20Court/case/8:16-cv-01774
• TIPO (Taiwan IP Office) research report on ITC Inv. No. 337-TA-837, listing US 6,982,663 B2 as asserted patent [No. 02]: https://tiponet.tipo.gov.tw/downloads/module030/communication_103_3_2.pdf and https://tiponet.tipo.gov.tw/downloads/module030/communication_103_5.pdf

Bottom line: U.S. Patent No. 6,982,663 has been litigated extensively — most prominently in the ITC Section 337 investigation 337-TA-837 (LSI v. Realtek), the SanDisk v. LSI N.D. Cal. action (C 09-02737), the Avago/Broadcom v. Amazon C.D. Cal. actions (8:16-cv-01774 and 8:16-cv-01052), and Broadcom v. Netflix (3:20-cv-04677, N.D. Cal.), plus three PTAB IPRs (all either denied institution or settled) and at least two Federal Circuit appeals. Several docket numbers from the Google Patents feed remain unverified as to parties and outcome, and should be checked in PACER/Unified Patents before reliance.

I'll start by reconciling the structured data (which reports no ODP trials) against the litigation data embedded in the patent record, then verify each proceeding via web search.

Three IPRs surface despite the ODP block. Let me verify each and check for appeals and the Netflix IPR.

Let me pin down the denial reasoning, the AMD settlement, and whether any of these were appealed.

PTAB / AIA Trial Proceedings — US 6,982,663 ("Method and system for symbol binarization")

Sourcing note / discrepancy flag. The structured "PTAB proceedings on file" block reports no AIA trial proceedings from the USPTO ODP ingest. That appears to be an ODP indexing gap, not the true state of the world. The patent record itself (Google Patents litigation tabs) and several independent sources (PTAB E2E petition files, Patexia, GreyB/IPVerse) surface three IPR petitions naming this patent. I report those three below, flagged by source, and I do not invent any proceeding numbers. Where a source is silent (e.g., panel composition for two of the three), I say so.

Proceedings overview

Three (3) IPR petitions have been filed against US 6,982,663 — none produced a Final Written Decision: two were denied institution (IPR2017-00964, IPR2021-00468) and one was terminated by settlement before institution (IPR2017-01182), leaving 0 claims canceled, 0 claims sustained on the merits, and all 21 claims nominally intact. The bottom-line defensive posture is therefore mixed but net-favorable on paper and net-unfavorable in practice for the patent owner: the patent has never been invalidated at the PTAB, but the absence of any FWD means no statutory estoppel was ever created against the three challengers, the claims have come out of district court claim construction materially narrowed as to claim 12's step order (N.D. Cal., 2022-05-23), and — decisively for a defendant today — the patent expired on 2022-07-17, so it is a damages-only artifact with a fixed (and rapidly closing) accrual window.

IPR2017-00964 — 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 the status verbatim as "Not Instituted - Merits"; Patexia/GreyB record "Institution Denied")
• Judge panel: James A. Worth, James B. Arpin, Melissa A. Haapala (per Patexia); APJ Arpin authored the 2017-08-18 decision on petitioners' motion to withdraw backup counsel (Paper 14)
• Petition grounds: Challenged all of claims 1–21. The petition asserted § 102(a)/(b) and § 103 grounds over the Wiegand "Draft ISO/IEC 14496-10" contribution (publicly available May 2002) and Cheung, "Configurable Variable Length Code for Video Coding" (publicly available 2001-05-11), among other exhibits, including Teuhola, "A Compression Method for Clustered Bit-Vectors," the '663 and '387 prosecution histories, and an email from inventor Lowell Winger to the JVT experts reflector (Exs. 1001–1017). Petitioners also concurrently filed on the parent patent (see pattern signals below).
• Institution decision: Denied — 2017-09-13. The denial on the merits is corroborated procedurally by petitioners' Request for Refund of Fees (2017-10-26) and the Notice of Refund (2017-11-02). I could not retrieve the Board's substantive reasoning verbatim in this session; based on the record structure and the Patent Owner's Preliminary Response (Paper 11, 2017-06-15) relying on Winger/JVT documentation and the Itoh reference, the fight was over whether the 2002-era JVT/Wiegand materials qualified as § 102(a)/(b) printed publications and whether the combination rendered the hybrid unary/exp-Golomb swapping obvious. I am not asserting a specific holding quote I have not verified.
• Final Written Decision: None issued. No claim-level verdict exists for this proceeding.
• Settlement / termination: Not settled; denial terminated the case with claim 1–21 untouched.
• Appeal: No FWD to appeal. (The patent record separately lists Federal Circuit case numbers 14-1410 and 17-1491 in the family litigation history, but I could not confirm either is an appeal of an IPR on the '663 patent — the IPRs here never reached a final written decision, so there is nothing appealable from them.)
• Defensive value: The Wiegand and Cheung references were presented and considered by the Board and rejected at the threshold — they are exhausted as a third-party petitioning strategy and carry substantially weakened value in a later § 103 case unless you can materially improve the motivation-to-combine showing the Amazon petition apparently failed to make.

Sources: Patexia docket summary, GreyB/IPVerse, PTAB E2E petition filing

IPR2021-00468 — Netflix, Inc. / Netflix Streaming Services, Inc. v. Broadcom Corporation

• Type: Inter Partes Review
• Filed: 2021-02-03
• Status: Institution Denied (Google Patents records "Not Instituted - Merits"; GreyB/IPVerse records "Institution Denied")
• Judge panel: Not confirmed in available public sources — I will not guess APJ names.
• Petition grounds: Challenged a subset of the claims rather than all 21. A Netflix petition on the '663 patent in this era requested cancellation of claims 12–16, 18, and 19, with grounds under §§ 102/103 and a substantial § 112 indefiniteness theory on claim 13 ("a fourth pattern in said first portion … in response to said index value being below said threshold") and on the "wherein said [second/third] portion is void" limitations. Caveat: I could not conclusively tie that specific petition excerpt to IPR2021-00468 rather than to a companion Netflix filing; treat the claim list as provisional.
• Institution decision: Denied — 2021-08-09. Google Patents classifies the denial as merits-based ("Not Instituted - Merits"), not procedural. The parallel N.D. Cal. docket reflects Broadcom/Avago filing a "NOTICE of Denial of Institution of IPR as to '663 Patent" — i.e., the patent owner promptly used the denial to carve the '663 out from any stay. I have not verified the Board's verbatim reasoning (e.g., whether it turned on the printed-publication status of the JVT materials or on the merits of the § 112/motivation arguments).
• Final Written Decision: None issued.
• Settlement / termination: Not settled.
• Appeal: None (no FWD). Notably, however, the district court litigation continued: the Court granted a stay only as to patents actually under PTAB review and denied the stay as to patents not under review (Judge Donato, 2021-09-15), so the '663 patent stayed live in N.D. Cal. after the denial.
• Defensive value: Netflix — a sophisticated, well-funded defendant with every incentive to kill this patent — could not get past the threshold on the '663 patent. That is a meaningful signal that the hybrid-binarization claims are harder to invalidate on paper than the "unary + exp-Golomb were both known" narrative suggests. But because no FWD issued, Netflix is not statutorily estopped and its art remains available to others.

Sources: GreyB/IPVerse, CourtListener docket 3:20-cv-04677

IPR2017-01182 — Advanced Micro Devices, Inc. / ATI Technologies ULC v. Avago Technologies General IP (Singapore) Pte. Ltd.

• Type: Inter Partes Review
• Filed: 2017-03-28
• Status: Terminated (Google Patents records the status verbatim as "Settlement"; GreyB/IPVerse records "Terminated")
• Judge panel: Not confirmed in the sources retrieved.
• Petition grounds: Challenged the '663 patent (application 10/770,213, Tech Center 2800). Specific claims and art not confirmed in the retrieved record; AMD's counsel was Brian Oaks and Avago's counsel was Daniel Young. This petition was filed as a companion to the Amazon filings (the Board's 2017-08-18 order groups IPR2017-01181, -01182 and IPR2017-00963/-00964 as related).
• Institution decision: No institution decision was reached — the proceeding ended before the § 314 deadline.
• Final Written Decision: None.
• Settlement / termination: Terminated 2017-09-06, before institution, on settlement. This is consistent with the global pattern of 2017 Avago/AMD-LSI licensing resolution; the terms are confidential and I have not verified any specific license or payment terms.
• Appeal: None.
• Defensive value: Settled pre-institution IPRs are near-worthless as invalidity ammunition — there are no Board findings, no estoppel, and no claim-level rulings. The only takeaway is structural: the '663 patent was being challenged in coordinated multi-front campaigns by both device makers (Amazon) and chip designers (AMD/ATI) at the same time it was in the ITC (337-TA-837) and multiple district courts.

Sources: GreyB/IPVerse, PTAB related-proceedings order, 2017-08-18 (IPR2017-00964 Paper 14)

Strategic summary

Claim status: nothing is canceled, nothing is sustained-on-the-merits, everything is untested at the PTAB. All 21 claims survive. The two "losses" for the patent owner are non-institution denials, which carry zero res judicata and zero statutory estoppel effect — a denial is not an adjudication of validity, and under § 315(e)(2) estoppel only attaches to a petitioner whose IPR has been instituted and reached a final written decision. There are no FWDs on the '663 patent, so Amazon, AMD/ATI, Netflix and their privies are not estopped, and neither are you by anything done here. The practical narrowing of the patent instead came from the district court, not the Board: on 2022-05-23 the N.D. Cal. (Judge Donato) construed claim 12 of the '663 patent to require its three codeword-generation steps to be performed in the recited order — "the plain language of the claim and the specification establishes that the steps in claim 12 must be performed in the recited order" — and rejected Netflix's indefiniteness attack on claim 13. Broadcom asserted claims 12, 13, and 20 of the '663 patent against Netflix; Netflix's non-infringement theory is that x264 generates the "third pattern" (code line 1448) before the "second pattern" (line 1450) and therefore cannot infringe an ordered-step claim.

Estoppel landscape. Because no FWD issued in any of the three proceedings, § 315(e)(2) is not triggered against any petitioner. For a defendant being asserted against today, that means the prior art raised by Amazon (Wiegand "Draft ISO/IEC 14496-10," Cheung's configurable VLC, Teuhola), by AMD/ATI, and by Netflix is not statutorily foreclosed in district court — but note two practical limits: (i) Netflix itself is pursuing invalidity under §§ 101/102/103 and § 112 ¶ 4 in N.D. Cal. using Kerofsky (2001) and the admissions of inventor Dr. Winger, so the art set is broader than what the Board saw; and (ii) any new IPR against the '663 is now both time-barred under § 315(b) for anyone served more than a year ago and, more fundamentally, moot because the patent expired 2022-07-17 (Google Patents: "Expired - Lifetime … Adjusted expiration 2022-07-17"). An IPR can technically be sought on an expired patent, but it buys a defendant nothing beyond what a district-court § 101/§ 112 motion already buys.

Pattern signals. This is a classic serial-assertion pattern with a coordinated-challenge response: Amazon filed two petitions on the same day (IPR2017-00963 on the parent US 6,744,387 and IPR2017-00964 on the '663), and AMD/ATI filed a parallel pair (IPR2017-01181 and IPR2017-01182). Both fronts failed at the threshold or settled. Netflix later filed IPR2021-00468 on the '663 as part of a broader multi-patent campaign stemming from Broadcom Corp. v. Netflix, Inc., 3:20-cv-04677 (N.D. Cal.) and the PTAB denied institution again. I see no defensive aggregator (e.g., Unified Patents) as petitioner — Unified Patents appears only as a litigation-data source on the patent record, not as a party, despite the record suggesting otherwise (the "Unified Patents" and "IPR2021-00468 / IPR2017-00964 / IPR2017-01182" entries on the Google Patents page are data-source attributions). The '663 patent was also enforced in ITC Inv. No. 337-TA-837 (LSI v. Realtek) and in the Avago v. ASUSTeK cases, so the assertion footprint is broad and old.

Recommended next steps

1. Do not build a § 315(e)(2) estoppel argument. There are no FWDs, so there is no estoppel. Anyone who tells you "Amazon is estopped on the Wiegand art" is wrong.
2. Lead with expiry and the district-court claim construction, not the PTAB. The '663 expired 2022-07-17. Damages are limited to infringement accruing within the § 286 six-year lookback from the complaint, and there is no injunction available. The binding construction that claim 12's steps (A)→(B)→(C) must be performed in order (N.D. Cal. 2022-05-23, Dkt. 218) is the single most useful invalidity/non-infringement lever — Broadcom's own expert conceded the accused x264 code emits the third pattern before the second.
3. If you need the merits records, pull them from PTAB E2E directly — the two denials (IPR2017-00964 Paper 15, 2017-09-13; IPR2021-00468, 2021-08-09) and the IPR2017-01182 termination (2017-09-06) each contain the Board's/Petitioner's framing of the art, even though they are non-precedential and non-estopping. I have not verified the verbatim reasoning of either denial, so do not quote them until you retrieve the papers.
4. Assert § 101 and § 112 ¶ 4 in parallel with any § 102/§ 103 case. Netflix's operative position — that the asserted claims are directed to combining two known mathematical algorithms (unary and exp-Golomb) with no inventive concept, and that claim 13 fails to further limit claim 12 under § 112 ¶ 4 — was never tested by the Board because institution was denied. It is the most thoroughly developed invalidity theory on this patent and it is still fully available.
5. Check the sibling patents separately. US 6,744,387 (parent) and the other Avago/Broadcom "encoding" patents in the same campaign (the '283/'183/'976/'138 family) have their own dockets — including IPR2021-00542 (Netflix v. Avago) — and should not be conflated with the '663. An IPR on the '387 does not estop or bind anything on the '663.

I'll ground the assignment record in the authoritative patent text provided, and search for the correspondent-of-record details that the text doesn't include.

US 6,982,663 B2 — "Method and system for symbol binarization" — Assignment & Ownership Reconstruction

Scope note up front (important): The authoritative patent record provided (Google Patents) exposes the reel/frame numbers and execution/recording dates for every post-issuance assignment, but it does not expose the correspondent-of-record field. I attempted to retrieve correspondents via live search; the only attorney name surfaced is the prosecution attorney of record (Christopher P. Maiorana, per Justia's patent history), which is not the same as the assignment-recording correspondent. I therefore cannot confirm correspondents for any reel/frame below and have marked that signal unclear rather than guess. Everything else below is grounded in the reel/frame entries and dates in the provided record.

Inventors

• Lowell Winger — sole named inventor. The application was filed by and assigned to LSI Logic Corporation (Milpitas, CA); Dr. Winger is listed with a Waterloo, Ontario residence on the face of the patent (Justia patent history). His employer at filing was therefore LSI Logic (or its Canadian design operation) — this is inferred from the assignee naming, not from a separately recorded employment agreement. Note the Netflix briefing in Broadcom v. Netflix refers to "the deposition testimony of the inventor himself, Dr. Winger," confirming he is the inventor of record and a witness in later litigation.
• Unusual patterns: None. There is a single inventor, and the provided record contains no evidence of inventor departure within 12 months of filing (no employment/assignment re-recordings, no inventor-to-third-party conveyances). The only "inventor-side" documentation gap is that the pre-issuance inventor→LSI Logic assignment is not visible in the Google Patents legal-events list (either never separately recorded in a way Google indexes, or recorded under the parent/pipeline). This is common for large-filer pre-AIA cases and is not a fire-sale tell.
• The patent is a continuation of US Ser. No. 10/191,596 (issued as US 6,744,387 B2), same sole inventor, same assignee, priority 2002-07-10. The two siblings travel together through the entire chain below.

Original assignee

• LSI Logic Corporation (Milpitas, California) — a fabless semiconductor company and one of the largest ASIC/SoC vendors of its era. Primary line of business: custom ASICs, storage/network controllers, and consumer media/video codec silicon (MPEG-2, then H.264). It also maintained an active patent-licensing program.
• Products embodying the claims: The claimed subject matter (hybrid unary/exp-Golomb binarization for a binary arithmetic coder) is part of the H.264 / MPEG-4 AVC CABAC entropy-coding path. LSI Logic / LSI Corporation built H.264 video codec IP and consumer media silicon, and the patent was later asserted against H.264/H.265 encoder pipelines — and Broadcom concedes a RAND commitment covering the '663 and '387 patents (Netflix opposition brief, N.D. Cal. 3:20-cv-04677). That RAND commitment is strong evidence the patent is treated as standards-essential, hence practiced by the assignee's own codec products as well as the industry. Direct product-level proof of "shipping a product embodying the claims" was not independently verified from the provided record; the H.264 SEP status is the strongest available proxy.
• Current status: No longer an independent entity. LSI Logic Corporation → LSI Corporation (name change, effective 2007-04-04, reel 027836/0054) → assets acquired by Avago Technologies Limited on 2014-05-06 (confirmed in the N.D. Cal. briefing, which states Avago acquired LSI's assets on May 6, 2014, with KPMG valuation dated 2014-09-18). LSI as a brand was retired; Avago later renamed itself Broadcom Inc. after acquiring Broadcom Corporation in Feb 2016. So the original assignee is acquired/dissolved, not operating and not in bankruptcy.

Assignment timeline

Chronological. All reel/frame numbers and dates are from the Google Patents Legal Events record for US 6,982,663. Correspondents are not exposed by that source (see scope note).

• 2004-02-02 (executed) / recorded 2004-02-02 — no reel/frame (application filing, not an assignment)

  • Conveyance: Application filed by LSI Logic Corp (continuation of 10/191,596)
  • Assignor: n/a · Assignee: LSI Logic Corporation
  • Correspondent: Christopher P. Maiorana (prosecution attorney of record per Justia) — flagged only as the prosecution correspondent; not evidence of an assignment-chain repeat player.
  • Context: Original filing by the operating company.

• 2007-04-04 (executed) / recorded 2012-03-09 — Reel 027836/0054

  • Conveyance: Change of Name
  • Assignor: LSI Logic Corporation · Assignee: LSI Corporation
  • Correspondent: not retrievable from sources consulted.
  • Context: Internal reorg / corporate name change only — no change in beneficial ownership.

• 2014-05-06 (executed) / recorded 2014-05-08 — Reel 032856/0031

  • Conveyance: Patent Security Agreement (security interest, not a title transfer)
  • Assignor: LSI Corporation; Agere Systems LLC · Assignee: Deutsche Bank AG New York Branch, as Collateral Agent
  • Correspondent: not retrievable.
  • Context: Securitization/financing lien recorded against the patent rights.

• 2014-08-04 (executed) / recorded 2015-02-17 — Reel 034974/0873

  • Conveyance: Assignment of Interest
  • Assignor: LSI Corporation · Assignee: Avago Technologies General IP (Singapore) Pte. Ltd.
  • Correspondent: not retrievable.
  • Context: M&A — Avago acquired LSI's assets (closed 2014-05-06); title moved into Avago's Singapore IP-holding subsidiary.

• 2016-02-01 (executed) / recorded 2016-02-02 — Reel 037684/0039

  • Conveyance: Termination and Release of Security Interest (releases RF 032856-0031)
  • Assignor: Deutsche Bank AG New York Branch, as Collateral Agent · Assignee/beneficiary: LSI Corporation; Agere Systems LLC
  • Correspondent: not retrievable.
  • Context: Lien release — the 2014 collateral interest is cleared.

• 2016-02-01 (executed) / 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 retrievable.
  • Context: Replacement financing lien (contemporaneous with Avago's Feb 2016 acquisition of Broadcom Corporation).

• 2017-01-19 (executed) / recorded 2017-02-03 — Reel 041710/0001

  • Conveyance: Termination and Release of Security Interest
  • Assignor: Bank of America, N.A., as Collateral Agent · Assignee/beneficiary: Avago Technologies General IP (Singapore) Pte. Ltd.
  • Correspondent: not retrievable.
  • Context: Lien release — financing collateral cleared.

• 2018-05-09 (executed) / recorded 2018-10-04 — Reel 047196/0097

  • Conveyance: Merger
  • Assignor: Avago Technologies General IP (Singapore) Pte. Ltd. · Assignee: Avago Technologies International Sales Pte. Limited
  • Correspondent: not retrievable.
  • Context: Internal corporate merger of the Singapore IP entities.

• 2018-09-05 (executed) / recorded 2019-03-06 — Reel 048555/0510

  • Conveyance: Corrective Assignment (corrects the execution date recorded at 047196/0097)
  • Assignor: Avago Technologies General IP (Singapore) Pte. Ltd. · Assignee: Avago Technologies International Sales Pte. Limited
  • Correspondent: not retrievable.
  • Context: Correction only.

• 2020-03-10 (executed) / recorded 2020-03-10 — Reel 052065/0838

  • Conveyance: Assignment of Interest
  • Assignor: Avago Technologies International Sales Pte. Limited · Assignee: Broadcom Corporation
  • Correspondent: not retrievable.
  • Context: Intra-group transfer-to-asserting-entity — executed the same day it was recorded and immediately before the March 2020 Broadcom v. Netflix complaints (C.D. Cal. 8:20-cv-00529; later N.D. Cal. 5:20-cv-04677). Same corporate family, not a third-party sale.

Timeline diagram

timeline
    title Ownership of US 6982663
    2002 : Priority date from parent case
    2004 : Filed by LSI Logic Corp
    2006 : Patent issued
    2007 : LSI Logic renamed LSI Corporation
    2012 : Change of name recorded
         : ITC action against Realtek
    2014 : Avago acquires LSI assets
         : Deutsche Bank security interest
         : Assigned to Avago General IP Singapore
    2015 : Avago assignment recorded
    2016 : Deutsche Bank lien released
         : Bank of America security interest
         : Suit filed against Amazon
    2017 : Bank of America lien released
    2018 : Merger into Avago International Sales
    2019 : Corrective merger recording
    2020 : Assigned to Broadcom Corporation
         : Netflix suit filed

NPE / troll-pattern signals

1. Shell-entity transfer — not present. The only "IP-holding" names in the chain are Avago Technologies General IP (Singapore) Pte. Ltd. (reel 034974/0873) and Avago Technologies International Sales Pte. Limited (reels 047196/0097, 048555/0510, 052065/0838). These are wholly owned IP subsidiaries of an operating semiconductor company (Avago/Broadcom), used for intra-group title-holding and financing — not licensing-only shells with registered-agent addresses or single-member LLC structures. No Delaware/Texas single-purpose NPE LLC appears anywhere in the record.

2. Known asserter in the chain — not present. No assignee matches any public NPE roster (Acacia, Marathon, IV, IPNav, Wi-LAN, Mosaid/Conversant, Vringo, Pendrell, Innovatio, MPHJ, Lumen View, Round Rock, etc.). The current assignee is Broadcom Corporation, and Stanford's NPE Litigation Database itself classifies the asserter in these cases as "IP subsidiary of product company" and "Product company" — i.e., not an NPE.

3. Repeat correspondent across the chain — unclear / not determinable. The sources available to me do not expose the correspondent-of-record field for any of the nine reel/frame entries. I will not infer a repeat player from the prosecution attorney (Christopher P. Maiorana), who appears only as the attorney of record on the patent face (Justia), not as a recording correspondent. Action item for verification: pull each reel/frame in the USPTO Assignment Center and compare the recorded correspondent names — this is the one signal in this chain that cannot be resolved from the provided record.

4. Cascading transfers — partially present, but benign. There are nine recorded events between 2012 and 2020, several within 24 months of each other (2014 lien → 2015 Avago assignment; 2016 release → 2016 new lien → 2017 release). However, the pattern is corporate M&A plus secured-financing housekeeping (an acquisition, a merger, two security agreements and their releases, a change of name, a correction) — not chained LLC-to-LLC churn with shared principals. Marking the NPE version of this signal not present; noting the high recording count as a factual matter.

5. Pre-litigation transfer — present. Reel 052065/0838, executed 2020-03-10 (recorded the same day), moved title from Avago Technologies International Sales Pte. Limited to Broadcom Corporation, days before the March 2020 Netflix complaints (C.D. Cal. 8:20-cv-00529; N.D. Cal. 5:20-cv-04677 filed 2020-07-14). This is a standing/venue cleanup within the same corporate family to place title in the U.S. operating entity that sued — it supports assertion but is not a third-party transfer to an NPE.

6. Bankruptcy fire-sale — not present. LSI was acquired by Avago in a going-concern M&A transaction (closed 2014-05-06); there is no Chapter 7/11, no §363 sale, and no distressed disposition anywhere in the record.

7. Privateering — not present. The classic pattern (operating company hands patents to an independent NPE that sues on its behalf) is absent: the operating company itself sues. The plaintiffs in the enforcement actions are LSI / Avago / Broadcom, and the Stanford categorization confirms product-company/its-IP-subsidiary status. (Mild caveat: title does sit in Singapore IP subsidiaries that are co-plaintiffs — e.g., "Avago Technologies International Sales PTE Limited" alongside Broadcom Corp. in 3:20-cv-04677 — but these are controlled affiliates, not rent-a-plaintiff trolls.)

8. Defensive aggregator — not present. The chain does not terminate at RPX, AST, LOT, Unified Patents, or OIN. If anything, Unified Patents-adjacent challengers moved against the patent: IPR2017-00964 (Amazon, filed 2017-02-22, instituted-denied 2017-09-13), IPR2017-01182 (AMD/ATI, filed 2017-03-28, terminated 2017-09-06 — settlement), and IPR2021-00468 (Netflix, filed 2021-02-03, institution denied 2021-08-09). Those challenges confirm the patent is being asserted, not neutralized.

Verdict

Operating-company assertion.

Justification: The chain terminates at Broadcom Corporation by assignment executed 2020-03-10 (reel 052065/0838), a large semiconductor/software product company whose predecessors (LSI Logic → LSI Corporation → Avago) designed H.264 codec silicon and were the named plaintiffs throughout. The patent was asserted by the owner itself against actual competitors — Realtek at the ITC (337-TA-837), Amazon (8:16-cv-01774 / IPR2017-00964), AMD (IPR2017-01182), and Netflix (8:20-cv-00529, 5:20-cv-04677 / IPR2021-00468) — with Stanford's database categorizing the asserter as a product company / its IP subsidiary, not an NPE. The only "NPE-ish" features are the Singapore IP-holding subsidiaries and the 2020-03-10 intra-group transfer immediately preceding the Netflix suit; both are normal operating-company mechanics, not a shell/IP-holding NPE chain.

Verify the chain at: https://assignmentcenter.uspto.gov/ (Patent Assignment Search → enter patent number 6982663), mirrored at https://assignment.uspto.gov/patent/index.html#/patent/search?q=6982663 — with particular attention to the correspondent-of-record field on reels 027836/0054, 032856/0031, 034974/0873, 037684/0039, 037808/0001, 041710/0001, 047196/0097, 048555/0510, and 052065/0838, which is the single unresolved item in this reconstruction.

I'll research US patent 6982663 and its cited prior art. Let me run several searches.

Let me research the remaining cited references.

Let me research the remaining two references.

US Patent 6,982,663 — Prior Art Analysis

1. Patent identified (literal reading of the number)

A search of the USPTO/Google Patents record for 6982663 returns the same reference set as the authoritative full text supplied in your message — this analysis is grounded on that text plus verification searches on the individual references.

Claim structure (21 claims):

• Claims 1–10 — decoding: generating an index value from a codeword (threshold + offset + value per first/second/third patterns).
• Claim 11 — decoding system.
• Claims 12–20 — encoding: generating a codeword from an index value (literal-hybrid unary/exp‑Golomb).
• Claim 21 — encoding system.

The critical date for § 102 is the 2002‑07‑10 priority date (the 2004 continuation carries the parent's priority). Anything published/effective before that date is § 102 prior art; US patents/applications filed before that date but publishing later can qualify only under § 102(e).

2. Patent citations (6) — as listed on the face of US 6,982,663

(a) US 5,471,207 A — Zandi, Stork & Allen; Ricoh Company Ltd.

• Title: Compression of palettized images and binarization for bitwise coding of M‑ary alphabets therefor
• Filed / published: US appl. 08/200,233 filed 1994‑02‑23; granted 1995‑11‑28
• Description: Binarizes M‑ary input symbols via a reindexing table mapping each symbol to a number of binary values, with the binarization chosen to minimize bitwise entropy and fed to a binary encoder (context modeler → probability estimator → bit generator). Also covers parallel encoding/decoding with bit reordering.
• § 102 relevance: Statutory § 102(b) art (publication >1 year before priority). It discloses binarization for a binary arithmetic-style coder in general, which is the subject-matter field of claims 1–21, and its "map each symbol to a sequence of binary values" teaching is the closest prior art for the broad binarization concept. However, it does not disclose the claimed threshold-switched concatenation of a unary prefix and an exp‑Golomb suffix, the "set index to a threshold + add offset + add value" decode of claim 1, or the corresponding encode of claim 12.
• Assessment: Potential § 102 challenge is weak/not cleanly anticipatory. Best characterized as § 103 background art against the preamble/subject-matter of claims 1, 11, 12, 21. Confidence: medium. (I could not confirm a passage teaching a threshold-dependent code switch.)

(b) US 6,236,960 B1 — Peng, Cruz Zeno & Ashley; Motorola, Inc.

• Title: Factorial packing method and apparatus for information coding
• Filed / granted: 1999‑08‑06 / 2001‑05‑22
• Description: "Factorial pulse coding" for speech/audio/image/video — generates a codeword as a sum of offsets/indices representing pulse number, position and magnitude (combinatorial expression F(n,d)).
• § 102 relevance: § 102(b) art. Relevant to the general notion of building a codeword from index/offset arithmetic (compare claim 1's "add an offset … add a value"), but it is a multi-stage offset summation enumerated coding scheme, not a unary/exp‑Golomb hybrid with a switching threshold.
• Assessment: Not anticipatory of any claim as written. Possible § 103 art on the "add offset / add value" concepts of claims 1 and 12 only. Confidence: medium-high that it does not anticipate.

(c) US 6,636,222 B1 — Valmiki et al.; Broadcom Corporation

• Title: Video and graphics system with an MPEG video decoder for concurrent multi‑row decoding
• Filed / granted: 1999‑11‑09 / 2003‑10‑21
• Description: Hardware video/graphics decoding system — row decoding engines that concurrently decode multiple SLICEs of MPEG‑2 video, memory/bus architecture, video compositor.
• § 102 relevance: Filed before the 2002 priority but granted after → at most § 102(e) art. It is directed to video decoder hardware architecture, and does not describe entropy binarization at all.
• Assessment: Not anticipatory of any claim. Background art only (and notably the assignee family that later acquired the patent). Confidence: high.

(d) US 6,850,568 B1 — Williams, Luna & Wang; Sony Corporation

• Title: Method and apparatus for decoding MPEG video signals
• Filed / granted: 2000‑01‑12 / 2005‑02‑01
• Description: MPEG decoder using a VLIW core processor + co‑processor with a variable length decoder (VLD); macroblock/slice handling, motion compensation, fake slice start codes.
• § 102 relevance: § 102(e) art (filed 2000‑01‑12). It discloses MPEG variable-length decoding, i.e., the inverse of codeword construction, and an "apparatus for decoding" (cf. claim 11's "decoder … and a circuit"). But its VLD is not the claimed threshold hybrid codeword decode.
• Assessment: Not anticipatory. Only conceivable relevance is to the general environment of claims 9/20 (MPEG‑AVC/H.264 compatibility) and the decoder-recited claim 11 — and even there the specific limitation is absent. Confidence: medium-high.

(e) US 2004/0114683 A1 — Heiko Schwarz (Fraunhofer)

• Title: Method and arrangement for coding transform coefficients in picture and/or video coders and decoders …
• Priority / published: 2002‑05‑02 / 2004‑06‑17
• Description: Coding of transform-coefficient levels in image/video coders using code tables — directly parallel to the "coefficient level binarization" (Table 4, N=16) described in 6,982,663.
• § 102 relevance: Priority 2002‑05‑02 predates the 2002‑07‑10 priority of 6,982,663, so if its US filing is before the invention date it could be § 102(e) art. Its subject matter (coding coefficient levels / transform coefficients) overlaps the coefficient-level application of the invention.
• Assessment: Not a clean anticipation. It is the strongest of the "same field" citations for the coefficient-level use case (claims 12–16/Table 4), and a plausible § 103 combination reference, but I have no evidence it discloses the specific unary→exp‑Golomb threshold switch. Confidence: medium. (The targeted verification search on this reference returned a step-limit message, so I could not fully confirm its disclosure — treat with caution.)

(f) US 6,744,387 B2 — Winger; LSI Logic Corporation

• Title: Method and system for symbol binarization
• Filed / granted: 2002‑07‑10 / 2004‑06‑01
• Description: This is the parent application (US 10/191,596) of the patent under review; same inventor, same disclosure, same priority date.
• § 102 relevance: It is not prior art — it is the priority/parent document of 6,982,663 and shares the identical specification. Listing it among the "citations" reflects the family relationship, not a third‑party anticipation.
• Assessment: Cannot anticipate (same invention/own work). Confidence: high.

3. Non‑patent literature cited (4)

4. Bottom line

• Which claims could be reached under § 102: On the record available, no single cited reference discloses all limitations of any of claims 1–21. The references fall into three buckets:
  1. General binarization/bin coding — US 5,471,207 (claims 1, 11, 12, 21 subject matter; weak § 102, primarily § 103).
  2. Adjacent code-construction / coefficient coding — US 6,236,960 and US 2004/0114683 (offset/value arithmetic of claims 1 & 12; coefficient-level use of claims 12–16; § 103, not clean § 102).
  3. Video-decoder architecture / standards background — US 6,636,222, US 6,850,568 (claims 9/11/20 environment; background only).
• US 6,744,387 is the patent's own parent (same invention) and is not prior art.
• The four NPL items are foundational entropy-coding/coding-theory references and are background, not anticipatory.

Important caveats:

• A definitive anticipation conclusion requires element-by-element comparison against the full text and claims of each reference; I was able to verify bibliographic data and abstracts/claims for most, but a full § 102(e) disclosure review of US 2004/0114683 was not completed (search limit reached), and the identity of the 1966 IEEE paper is unspecified in the citation list.
• I did not auto-correct or substitute any patent number; all identifiers are reported literally as they appear on US 6,982,663.
• These citations were identified from the patent's own cited-art list; the patent is expired (2022‑07‑17), but its cited art remains the relevant prior-art set for validity/§ 102–103 analysis.

Sources: Google Patents US6982663B2 (authoritative text supplied); Google Patents/USPTO and free full-text records for US5471207, US6236960, US6636222, US6850568, US20040114683, US6744387; Unified Patents portal for US‑6982663‑B2 (examiner: Young, Brian).

I'll start by verifying the content of the cited prior art references so my obviousness analysis is grounded in what they actually disclose.

Good grounding. Let me verify the remaining references and check the actual IPR petitions filed against this family (which will indicate how the art was actually combined).

Obviousness Analysis — U.S. Patent 6,982,663 ("Method and System for Symbol Binarization")

Analyst note on sources and confidence. I grounded this analysis in (a) the full text of US 6,982,663 (https://patents.google.com/patent/US6982663/en), (b) the references listed on that page under the citation sections (the "Prior Art" set: 6 Patent Citations and 4 Non-Patent Citations), and (c) verification searches for the key references. Where I relied only on an abstract/partial text and did not read a reference end-to-end, I say so. I did not auto-correct any patent number or identifier; all are reproduced as listed on the page.

1. Governing framework

• Statute: Pre-AIA 35 U.S.C. § 103(a). The '663 patent claims priority to 10 July 2002 (continuation of Ser. No. 10/191,596, filed 2002-07-10, now US 6,744,387). Its effective filing date is therefore no later than 10 July 2002, and the invention date could be earlier.
• Standard: KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007) — a combination of known elements is obvious where the improvement is a predictable use of prior-art elements according to their established functions, and where there was a known problem for which the combination supplies an apparent solution. "Obvious to try" is a valid rationale where the prior art identifies a finite number of identified, predictable solutions.
• Critical caveats:
  1. US 6,744,387 B2 appears in the page's citation list but is the parent of the same family, same inventor (Lowell Winger), same specification. It is not prior art against the '663 claims (common ownership/inventive entity; § 103(c)/double-patenting concerns). I exclude it.
  2. The page's "Cited By / Families Citing this family" entries (e.g., US 8,730,070; US 2014/0247165; KR 101391601; EP 2141815; EP 2164176) are later-dated and are not prior art. Several are, however, relevant as evidence of the state of the art / obviousness-adjacent art, and I note them where useful.
  3. US 2004/0114683 A1 (Schwarz, Fraunhofer) is listed with the examiner's asterisk, i.e. it was cited by the examiner against '663. Its § 102(e) availability turns on whether its 2 May 2002 German priority can be carried back under post-AIPA § 102(e). I flag this in § 8.

2. What the claims actually require

The independent claims come in two directions:

• Decoding direction — claim 1 (and system claim 11): from a codeword, (A) set the index value to a threshold when a first portion has a "first pattern"; (B) add an offset based on a second, following pattern; (C) add a value based on a third, following pattern.
• Encoding direction — claim 12 (and system claim 21): (A) generate a first pattern when index ≥ threshold; (B) generate a second pattern representing an offset above the threshold; (C) generate a third pattern representing the value above the offset.

The dependents cabin this into the disclosed hybrid: unary below the threshold (claims 2–7, 13–15), a fixed run of ones as the "first pattern" (claims 6, 10, 14), the first pattern being prefix-unique from the fourth/unary pattern (claims 3, 10, 14), a binary "third pattern" (claim 8), differing representations for offset and value (claim 16), void second/third portions below threshold (claims 18–19), and H.264/ISO-IEC 14496-10 compatibility (claims 9, 20).

Two drafting wrinkles worth flagging (they matter for mapping):

• Claim 17 (and 10(iv)) requires the "second pattern" to have the same number of bits as the "third pattern." In the disclosed algorithm (spec: prefix of N−1 ones; then a unary representation of γ+1; then γ LSBs), the two fields have γ+1 vs. γ bits — i.e., the literal limitation is not exhibited by the patent's own embodiment. This claim reads on a narrower/different variant and is the hardest claim to invalidate.
• The flowchart description (step 106: "an initial prefix of N 1's") conflicts with the algorithm text ("prefix of (N−1) 1's"). Mapping should follow the algorithm + Figs. 5–6.

3. The prior art of record on this page

Most probative reference: Schwarz US 2004/0114683 A1. Its own disclosure (§§ 0071–0073) states:

"For coefficient magnitudes in the interval [1; 14], this binarization corresponds to a unary binarization. The binarization for coefficient magnitudes larger than 14 consists of a prefix part consisting of 14 ones and a suffix part representing a 0th order exp-golomb code for the symbol **(ABS−15)… The binarization formed of a prefix part and a suffix part consisting of a 0th order exp-golomb code… has the advantage that a special non-adaptive context with symbol probabilities 0.5 can be used without sacrificing in coding efficiency… whereby the amount of calculating for encoding and decoding can be reduced."

That is the '663 invention — a threshold-switched hybrid unary/exp-Golomb binarization with a fixed one-prefix, in H.264/AVC CABAC, for the same purpose (bounded codeword length + distinct small values + lower BAC computation), published in the same standard context.

4. Primary ground of rejection: Schwarz in view of Teuhola

4.1 Element-by-element, claim 12 (encoding) — Schwarz alone

Result: Claim 12 (and 13–16, 18–20) is prima facie obvious over Schwarz alone, with the only arguable distinction being that Schwarz frames the "index value" as a coefficient magnitude (ABS) rather than a generic "code symbol index." The '663 specification forecloses that distinction: it states the same technique is applied to "transformed-quantized picture differences and motion vector residuals," and that "any system utilizing BAC may make use of the present invention." Substituting a general symbol index for a coefficient magnitude is the epitome of a predictable variation. The residual difference in threshold (14 vs. 16 in the patent's Fig. 6 table) is a non-claimed, optimization-level design choice.

4.2 Claim 1 / 11 (decoding)

Claim 1 is the inverse mapping: set index to threshold, add the offset field, add the value field. This is exactly the inverse of Schwarz's construction and the inverse of the '663 algorithm (v = (N−2) + 2^γ + g). Two responses to a "Schwarz is encoder-only" objection:

• H.264/AVC is a decoder standard; a compliant bitstream is by definition decodable, and the inverse transform of a prefix code is inherent.
• If explicit decoding hardware/firmware disclosure is required, US 6,850,568 (Sony) and US 6,636,222 (Broadcom) supply a conventional video decoder with entropy/VLD circuitry, and US 5,471,207 (Zandi) supplies the encoder/decoder binarization pair explicitly ("An encoder converts input symbols into codewords, while a decoder converts the codewords into the input symbols"). Combining an entropy decoder with the hybrid binarization front-end is routine.

4.3 Why Teuhola is added

Teuhola supplies the exp-Golomb code itself and, critically, its stated engineering rationale: it "encodes both short and long runs efficiently," has a "worst-case behaviour… far better than that of the Golomb code," and "could e.g. be applied to… transmission or storage of digital images." That is the exact problem statement of the '663 specification and the exact functional justification for putting an exp-Golomb suffix above threshold while keeping unary below it. Teuhola closes any gap on the "exp-Golomb" limitation and supplies the motivation.

5. Motivation to combine (the KSR prongs)

1. Same field, same problem, same subsystem. Both unary and exp-Golomb binarizations feed a binary arithmetic coder; the '663 specification admits this: "Among the commonly used binarization methods are the following: unary, binary, Golomb, and exp-Golomb."
2. Complementary, non-overlapping functions — a classical predictable combination. The specification itself states the problem: unary gives per-codeword distinctness but "in the worst case, [as many] bins as the largest symbol index (which may range into the tens of thousands)," while exp-Golomb "does not permit codewords with a small symbol index (other than index 0) to be uniquely distinguished." Using each technique in the range where it functions well is the stated objective — "retains the most valuable properties of the unary and exp-Golomb binarizations." Where the patent's own background articulates the problem and the solution as the combination of two known techniques, the § 103 case is strong (the specification is an admission of the prior art and of the motivation).
3. The combination is mechanical / predictable. Concatenating a fixed one-prefix with an exp-Golomb suffix adds no new principle of operation; it preserves the prefix-free property (the run of ones cannot appear in a unary code) and requires only routine selection of a threshold and a routine code-space offset so that the first exp-Golomb codeword does not collide with the last unary codeword.
4. Reasonable expectation of success — demonstrated. Schwarz built and published the combination for H.264/AVC CABAC, and stated that it reduces "the amount of calculating for encoding and decoding" — the very benefit '663 asserts.
5. Finite, predictable design space. Once one decides to switch models at a threshold, the only free parameters are (i) the threshold N and (ii) the boundary offset; both are identified, bounded choices — KSR "obvious to try."
6. Teaches away? No. Nothing in Teuhola, Zandi, Schwarz, Huffman or O'Neil criticizes the hybrid; Teuhola expressly praises efficiency across both short and long runs.

6. Alternative and secondary grounds

7. What actually happened in the real world (important for calibration)

The '663 patent page records the following challenges to this family:

• IPR2017-01182 (Petitioner: Advanced Micro Devices / ATI) — Settlement.
• IPR2017-00964 (Petitioner: Amazon.com / AWS) — Not instituted (merits).
• IPR2021-00468 (Petitioner: Netflix) — Not instituted (merits).

A petition in the family (USPTO PTAB exhibit, https://ptacts.uspto.gov/ptacts/public-informations/petitions/[1483354](/patent/1483354)/download-documents?artifactId=cZ7f4HudOSHIETgddV6RPXWpKSS1XAUqwpnVIdS_BzHWANB0x3fsILE) argued that Teuhola is prior art and "is directed to a method of binarization," quoting Teuhola's statement that the method "encodes both short and long runs efficiently" and mapping its exp-Golomb tables. The petition targeted the '387 parent, whose claims differ.

Implication: the obviousness theory above is not novel speculation — it is substantially the theory actually advanced. But the PTAB declined to institute on at least two merits petitions. That is a meaningful counter-signal: either (a) the art was presented against different (narrower) claims, (b) the petitions failed on the § 102(e)/prior-art-status or evidentiary prongs (e.g. the '663 examiner listed Huffman with "no date given"), or (c) the Board credited a nexus/secondary-consideration or claim-construction argument. Any competent analyser must weigh this against a myopic "Schwarz alone" conclusion.

8. Weaknesses, counterarguments, and open questions

1. § 102(e) status of Schwarz. US 2004/0114683 A1 published 17 June 2004, after the '663 priority date (10 July 2002). It is available only under pre-AIA § 102(e), which requires a US filing before the applicant's invention. Its US filing appears to be December 2003; its § 102(e) date depends on carry-back of the 2 May 2002 DE priority under post-AIPA § 102(e). The examiner cited it (asterisk), which suggests the Office treated it as prior art — but I could not independently verify the exact US filing date or the priority chain from the page text. If § 102(e) does not reach back to 2002-05-02, Schwarz drops out as prior art and the case rests on Teuhola + Zandi (+ admitted prior art). This is the single biggest vulnerability — and also the biggest lever — in the obviousness case, and it should be independently verified.
2. Claim 17 / 10(iv) — "same number of bits." As noted, this limitation is inconsistent with the patent's own embodiment (γ+1 vs. γ). It is therefore not met by Schwarz's 0th-order exp-Golomb suffix either. Claims 17 (and the 10(iv) variant) are materially harder to invalidate; a ground directed at them would need a reference teaching equal-width offset and value fields (e.g., a symmetric partitioned code), or would fail. Recommend not attacking claim 17 on Grounds A–B without additional art.
3. Claim 1's "set said index value to a threshold." Literal mapping requires the decode offset to resolve to (N−2) (per the algorithm) rather than N. Confirm against the issued claim construction; a court could read "threshold" more loosely, which helps the challenger.
4. Non-Patent Citation evidentiary gaps. The page lists Huffman with "no date given" and O'Neil/IEEE with bibliographic shorthand. Prior-art status of these NPL items should be established with authenticated copies before relying on them.
5. Secondary considerations (not addressed by the page). H.264/AVC CABAC's widespread commercial adoption of unary+exp-Golomb hybrid binarization could be asserted as industry praise/adoption. The nexus is contestable, however, because the Fraunhofer/Schwarz CABAC proposal (May 2002) predates and independently embodies the approach — under Orion/WBIP, a nexus requires the success to be attributable to the claimed invention rather than to the standard.
6. Scope honesty. I verified the Schwarz and Teuhola disclosures from source text and abstracts; I did not line-by-line review the full claim text of US 5,471,207, US 6,236,960, US 6,636,222, or US 6,850,568. Grounds C and E are therefore preliminary and would need full-text verification before being asserted.

9. Bottom line

• Claims 12–16 and 18–20 (encoding) — strongly obvious over US 2004/0114683 A1 (Schwarz) alone, or Schwarz + Teuhola; the hybrid unary/exp-Golomb, one-prefix, threshold-switched binarization for H.264/AVC CABAC is expressly disclosed, including the recited "offset" and "value" portions and the reason to do it (bounded length, lower BAC computation).
• Claims 1–9 and 11 (decoding / system) — obvious over Zandi (US 5,471,207) + Teuhola, or over Schwarz + one of Sony/Broadcom (for the explicit decoder/hardware recitation). The decode direction is the routine inverse of an admitted, standard, prefix-free code.
• Claims 21 and the system aspects — obvious; the recited "circuit + encoder/decoder" is a conventional implementation vehicle (specification concedes hardware or software).
• Claim 17 (and 10(iv)) — the strongest validity position for the patent; the "same number of bits" limitation is not met by the disclosed embodiment or by Schwarz's 0th-order exp-Golomb, and needs separate art or a broader construction.
• Net posture: a prima facie § 103 case exists and mirrors the theory actually litigated (Teuhola). It is materially undercut by the § 102(e) date question for Schwarz and by the PTAB's two merits-based refusals to institute. My confidence is high that Schwarz's hybrid binarization is the closest art; moderate that it qualifies as § 102(e) prior art; moderate that a well-pleaded Schwarz + Teuhola ground would survive institution if the claim-construction and priority issues are resolved correctly.