US 4354273

I'll verify the patent's current status and check for any litigation or CAFC docket activity involving this specific number.

US 4,354,273 — Verified Summary

Important caveat on scope of my checks: The searchable public sources I can reach (Google Patents, Justia, FreePatentsOnline, uspto.report) do not expose a queryable USPTO PatentCenter docket or a CAFC 2026 argument/decision index. My docket check therefore consists of targeted web searches rather than a direct docket lookup. No CAFC or district-court activity involving the literal number 4354273 surfaced. I cannot affirmatively certify a negative, so treat "no litigation found" as an absence of evidence, not proof.

Bibliographic data (as listed on the authoritative patent record)

The assignment recorded 1982-03-19 (effective 1980-05-27) names both inventors as assignors to Nippon Electric Company, Ltd.

Abstract (verbatim)

"In an adaptive differential pulse code modulation (ADPCM) system for frequency band compression of speech or like signals, the coefficient of the synthesis filter in both the transmitter and receiver is varied in accordance with the normalized error e_j /Δ rather than the error itself, thus providing greater frequency band compression and preventing transmission errors from rendering the synthesis filter unstable."

Independent claim overview

The patent carries 9 claims, of which only claim 1 is independent. Claims 2–3 depend from claim 2/1; claim 4 depends from claim 2; claims 5 and 6 depend from claim 4; claim 7 depends from claim 4; claim 8 depends from "any one of claims 1–5"; claim 9 depends from "any one of claims 4, 6 or 7." So there is no second independent claim to summarize — I'm flagging this because a request for "each independent claim" could otherwise be read as expecting more than one.

Claim 1 — An end-to-end ADPCM system for speech or like signals, made of two halves:

• Transmitter half: (a) a subtractor that takes input signal X_j and a predicted value and emits the difference e_j; (b) an adaptive quantizer that divides e_j by an adaptively varying normalization coefficient Δ and outputs the quotient (e_j/Δ) in coded form; (c) an adaptive inverse quantizer that decodes that coded output and multiplies it back by Δ to produce e_j; and (d) an adaptive predictor containing a digital filter whose coefficients are corrected by a control signal dependent upon the decoded output (i.e., the normalized-domain quantity), not upon the raw error.
• Receiver half: a receive adaptive inverse quantizer that decodes the transmitted output and multiplies by the adaptively varying Δ; and a receive adaptive predictor with a digital filter producing a predicted value from the inverse-quantizer output, its coefficients likewise corrected by a control signal dependent upon the decoded output.

Plain language: the inventive core of claim 1 is what the predictor's coefficient-update signal is derived from — the normalized error e_j/Δ, rather than the un-normalized error. That single change is what the specification argues (a) cuts the arithmetic needed for adaptation and (b) keeps the transmit and receive synthesis filters from diverging when transmission errors occur.

Dependent claims in brief (for context, since only one claim is independent)

• 2 – Specifies the transmit adaptive quantizer as a normalizing divider (÷Δ) plus coding means.
• 3 – The coding means quantizes the divider output into a plurality of discrete levels.
• 4 – Specifies the transmit adaptive inverse quantizer: decoder producing e_j/Δ, a control circuit generating Δ from the coder output, and a multiplier combining the two.
• 5 – Recursive-filter form of the transmit predictor: adder forming X_j = X̂_j + e_j, filter control terminal coupled to the decoder output.
• 6 – Non-recursive-filter form of the transmit predictor: filter control terminal coupled to decoder output, filter input is ê_j.
• 7 – Adds noise-shaping: a second subtractor forming quantization noise N_j = e_j − ê_j, a second filter with the same coefficients as the first, and a third subtractor feeding X̂_j − N̂_j to the first subtractor (the FIG. 8 embodiment).
• 8 – Receiver structure corresponding to the recursive (claims 1–5) configuration.
• 9 – Receiver structure corresponding to the non-recursive (claims 4, 6, 7) configuration.

Technical notes worth flagging

• The mathematical justification in the specification is that e_j/Δ ≈ e_j/e, and since Δ ≈ average |e_j|, the algorithm A_i = A_i + g′(e_j/Δ)·sign(X_{j−i}) is approximately equivalent to the reference Cohn algorithm when g′ ≈ g/3n. The claim, however, is drafted on the structural/signal-flow level (control signal "dependent upon said decoded output"), not on the equivalence.
• A copending U.S. application Ser. No. 79,724, filed 1979-09-28 (Araseki et al.) is referenced for the non-recursive predictor details. A related earlier patent in the same family area is US 4,317,208, also titled "ADPCM System for speech or like signals."
• Discrepancy I am not auto-correcting: the body text of the patent describes the Cohn et al. reference as appearing in the September 1975 issue of IEEE Trans. on Communications, while the front-page non-patent-citation list renders it "Sep. 1970." Both appear on the same public record; I am reporting them literally rather than picking one.

Litigation / CAFC status

I found no evidence of any CAFC 2026 appeal or district-court case involving US 4,354,273. Given the patent's expired status (anticipated expiration 2000-06-09, over 25 years ago), any live 2026 appeal concerning this patent would be unusual and would most likely concern a reissue, a later-filed continuation, or an interference-style collateral matter rather than the '273 claims themselves. I have no authoritative confirmation either way and recommend a direct PatentCenter and CAFC docket query if a formal negative is required.

I'll search for litigation involving US patent 4354273.

Let me check specialized litigation databases and the patent's litigation history directly.

Let me try a couple more targeted searches with different formatting.

Findings: Litigation involving U.S. Patent No. 4,354,273

No litigation is known. I was unable to identify any judicial proceeding — district court, ITC, Federal Circuit, or foreign — in which U.S. Patent No. 4,354,273 ("ADPCM System for speech or like signals," Araseki et al., granted Oct. 12, 1982, assigned to Nippon Electric Co., Ltd. / NEC Corp.) was asserted, challenged, or otherwise at issue.

What I searched

Important screening notes (false positives I discarded)

The string "4354273" appears frequently in unrelated records, and I excluded all of these rather than report them as litigation:

• FR trademark no. 4354273 ("ORDIM," VG Investissement, France) — French trademark register, not a patent (data.inpi.fr).
• Argentine trademark Act 4354273 (Class 25, "MARCIAL Y DEPORTIVA," Lopez Leandro) — Argentine trademark bulletin, not a patent.
• Samsung v. Wistron, N.D. Cal. — This case involves U.S. Patent No. 5,333,273 (the "'273 patent"), a computer keyboard/interrupt-signal patent. That is a different patent from U.S. 4,354,273, sharing only the digits "273." It is not the patent you asked about.
• Par Pharm. v. Amneal (D.N.J.) — Refers to a "'435 patent" in an ANDA/vasopressin case. The "435" there is a different patent (and the reference is to a patent number ending in 435, not 4,354,273). Not related.
• Incidental numeric matches in academic papers (e.g., a regression coefficient "0.4354273"; a Russian library cross-reference "MR 4354273") — not litigation.

Context supporting the "no litigation" conclusion

• Status (per Google Patents): Legal status is "Expired – Lifetime," with an anticipated expiration of 2000-06-09 — i.e., the patent term ran out roughly 26 years ago based on its June 9, 1980 U.S. filing date. An expired patent cannot support a live infringement action.
• Assignment history is administrative only. The only recorded USPTO assignment is the March 19, 1982 assignment from inventors Takashi Araseki and Kazunori Ozawa to Nippon Electric Company, Ltd. (reel/frame 003954/0850, effective May 27, 1980). There is no record of assignment to a litigation-focused entity (e.g., an NPE) that typically precedes assertion suits. The same tooling also notes a Certificate of Correction dated April 19, 1983 — again, prosecution-side, not litigation.
• The record shows no litigation events on the patent's Google Patents "Legal Events" and "Status" timelines (only assignment, grant/publication, and expiration entries).

Caveats on confidence

1. I could not run a direct structured query against PACER, Docket Navigator, or the Unified Patents litigation portal in this session; my searches were web-index based. A docket-level database (PACER, Docket Navigator, Lex Machina, or Darts-IP) is the authoritative way to confirm a negative, and I recommend running the patent number there before relying on this as a definitive clearance-type conclusion.
2. My searches covered U.S. federal courts and general litigation reporting. It is remotely possible that a very old, unreported, or foreign proceeding (e.g., a 1980s-era dispute involving NEC and a licensee) exists without a web-visible record; I found no evidence of one, and I am not asserting one exists.
3. Per the operating instruction not to auto-correct identifiers: I treated the number literally as 4,354,273 throughout, and did not fold in results for look‑alike numbers such as 5,333,273 or the various "'435" patents.

Bottom line

If a compilation of "known litigation involving US 4,354,273" is required, the correct entry is: None identified. The patent appears to have lived out its full term (1980 filing → ~2000 expiration) without any recorded assertion or validity challenge in the sources searched.

Proceedings overview

Total AIA trial proceedings on file for US 4,354,273: 0. Breakdown: 0 active, 0 claims invalidated, 0 claims sustained, 0 settled, 0 institution denied — the USPTO Open Data Portal "PTAB proceedings on file" block for this patent is empty, and independent web searches (PTAB E2E/PTABTS document hits, Google Patents litigation/proceedings tabs, CourtListener discussion) surfaced no IPR, PGR, or CBM naming 4,354,273 as the challenged patent.

The bottom line for a defendant is therefore not "the patent survived IPRs and is hardened" — it is the opposite: this patent is dead by expiration, and the absence of PTAB activity is fully explained by its age. Per the patent's own record, the term ran out on 2000-06-09 ("Anticipated expiration," status "Expired - Lifetime"), roughly twelve years before the first AIA petition could even be filed (AIA trials became available 2012-09-16 under 35 U.S.C. §§ 311–319, 321–329). Nobody IPR'd a patent that no longer existed. If a demand letter cites 4,354,273 today, the correct answer is a date-of-expiration letter, not a PTAB petition — a petition would be an expensive way to prove a point that the front page of the patent already proves.

No proceedings to report

Because the structured ODP list is empty, there is no proceeding for me to break out under the ### {PROCEEDING_NUMBER} — {Petitioner} v. {Patent Owner} heading. I will not invent one. For the record, here is what I did verify, so the negative finding is auditable:

Caveat on completeness: Google Patents' "Legal Events" tab for this patent shows only the assignment, the grant, and the certificate of correction. Google's coverage of pre-2000 ex parte reexamination certificates is historically incomplete, so I cannot affirmatively rule out a 1980s–1990s reexam from that tab alone. That said, a reexam would not change the analysis below in any material way, and it is not an AIA trial proceeding.

Strategic summary

Claim status: all nine claims (1–9) expired untested at the PTAB. No claim of 4,354,273 has ever been canceled, confirmed, or construed by the Board, because no AIA petition was ever filed. Claim 1 is the sole independent claim — a system claim spanning both transmitter (subtractor, adaptive quantizer with normalizing divider, adaptive inverse quantizer, adaptive predictor with digital filter whose coefficients are corrected by a signal "dependent upon said decoded output") and receiver (receive adaptive inverse quantizer plus receive adaptive predictor). Claims 2–5 depend from the quantizer/inverse-quantizer/predictor structure of claim 1; claim 6 recites the nonrecursive-filter variant of the transmitter predictor; claim 7 adds the second subtractor plus noise-shaping filter 70 of the third embodiment; claims 8–9 are multiple-dependent claims covering the corresponding receiver architectures. Cancelled: none. Sustained: none. Untested: claims 1–9, all of them — not because the claims are strong, but because the patent's term expired before the mechanism existed.

Estoppel landscape: irrelevant. 35 U.S.C. § 315(e)(2) estoppel only attaches to a petitioner that filed an IPR that reached a final written decision. With zero proceedings, there is no estopped party and no prior-art ground that has been "used up." That is a purely academic observation here, because the more dispositive bars are temporal: the patent expired 2000-06-09, and § 286's six-year damages lookback means any infringement complaint filed after roughly 2006-06-09 captures no compensable infringement period at all (70-year-plus prior art aside, "you cannot infringe an expired patent" does the work that an IPR would). An IPR against an expired patent is legally possible in narrow circumstances (to defeat a claim for back damages), but where every possible act of infringement predates 2000 and the six-year window closed two decades ago, the proceeding would serve no purpose.

Pattern signals: none — no petitioner, no aggregator, no appeal history. The patent was owned by NEC (Nippon Electric) for its entire life, and there is no record of a defensive aggregator such as Unified Patents, RPX, or a joinder-heavy serial petitioner ever attacking it. Nor did NEC ever need to defend it at the PTAB. For context on the owner rather than on this patent, NEC is an active patent owner at the Board generally — e.g., Peloton Interactive, Inc. v. NEC Corporation, IPR2023-01241 (Patent No. 9,769,427), in which NEC has filed a Patent Owner Response — but that is an unrelated later patent and does not involve 4,354,273. The closest thing to a "paper trail" on 4,354,273 is backward-looking: later ADPCM patents cite it as background art (e.g., US 5,550,837, which describes Araseki as varying "the coefficient of the synthesis filter ... in accordance with the normalized error rather than with the error itself"), and it sits in the same NEC/Araseki family as US 4,317,208 ("ADPCM System for speech or like signals") and the copending application Ser. No. 79,724 filed 1979-09-28 referenced in the specification. None of those forward citations is a validity challenge.

Recommended next steps

• If you are a defendant or recipient of a demand letter citing 4,354,273, do not file an IPR. State the date and ground the letter on the face of the patent: the patent issued 1982-10-12 and its anticipated expiration was 2000-06-09 (https://patents.google.com/patent/US4354273/en). With the patent expired for over 26 years as of today (2026-09-27) and the § 286 six-year damages lookback having closed in 2006, no actionable infringement theory survives. Ask the sender to identify any act of infringement occurring before the expiration date and any complaint filed within six years of it.
• If you nonetheless need validity ammunition, it is free and public, not PTAB-sourced. The examiner cited US 3,502,986 (Bell Telephone, adaptive prediction for redundancy removal), US 3,973,199 (Philips, prediction DPCM with adaptive compounding), and US 3,931,596 (Bell Telephone, adaptive quantizer with training mode), plus the Cohn Residual Encoder paper (IEEE Trans. Comm., COM-23(9), Sept. 1975, pp. 935–941) and Jayant, "Adaptive Quantization with a One-Word Memory," BSTJ Sept. 1973, pp. 1119–1144. Those are district-court / ITC invalidity materials now, not IPR grounds.
• If you are looking for a PTAB hook because a co-asserted, unexpired patent is in the same family, look elsewhere. Nothing in the 4,354,273 record helps or hurts you at the Board; check whether the assertion actually rests on a live NEC continuation (the 1979-09-28 application Ser. No. 79,724 and US 4,317,208 are the candidates) and evaluate those patents for AIA-trial availability and § 315(b) timing.
• Deadline math for any active proceeding: none applies here, but for the record — an instituted IPR runs on a statutory one-year clock from the institution decision to the Final Written Decision (35 U.S.C. § 316(a)(11)), with the § 314(b) institution decision due within six months of the petition's filing date. No such clock is or has ever been running on this patent.

Bottom line: no PTAB proceedings exist on US 4,354,273, and the reason is not that the patent is strong — it is that the patent term ended on 2000-06-09, a decade before AIA trials began. The defensive posture this gives a defendant is the strongest available: the patent cannot be asserted against anyone.

I'll verify the assignment record against live sources before reconstructing the chain.

Inventors

Pattern assessment — no "inventor departure" red flag. Both inventors appear to have remained at NEC well past the 12-month window: Araseki and Ozawa are again listed together as inventors on CA 1,252,568 (application filed 1985-12-23; granted 1989-04-11), applicant NEC Corporation — i.e., five-plus years after the '273 filing and after the 1983 English-name change from "Nippon Electric Company" to "NEC Corporation." A "Ozawa" also appears as inventor on much later cited NEC-lineage patents (e.g., US 9,984,712; US 10,008,230; US 10,170,144 in the 2018–2019 window), but I cannot confirm that is the same Kazunori Ozawa and do not treat it as a finding. The one documented pattern here is the opposite of a fire-sale precursor: continuous same-employer inventorship.

Original assignee

Nippon Electric Company, Ltd. (rendered on the Google Patents record as "Nippon Electric Co Ltd"; current-assignee field renders NEC Corp).

• Primary line of business: Japan-headquartered multinational telecommunications, computers, and semiconductor manufacturer — an operating company, not a holding vehicle.
• Status: Operating. Nippon Electric Company, Ltd. changed its English corporate name to NEC Corporation in 1983 (the Japanese name 日本電気株式会社 was unchanged); NEC Corporation continues to trade today on the TSE. No bankruptcy, dissolution, or portfolio-wide Chapter 7/11 event is associated with NEC in any source I reached.
• Did they ship a product embodying the claims? Not verified. Claims 1–9 are directed to an ADPCM codec architecture (adaptive quantizer/inverse-quantizer plus a coefficient-adapting predictor driven by the normalized error e_j/Δ). NEC was a first-tier manufacturer of speech-coding DSP and codec silicon in exactly this era, and the '273 sits in the same inventive line as NEC's subsequent low-bit-rate codec work (CA 1,252,568, 1985/1989). But I have no product-level evidence tying a specific NEC part number to claim 1, and I will not manufacture one. Treat "NEC shipped an embodying product" as probable on industry context, unproven on the record.
• Post-issuance ownership: The record continues to show NEC-lineage ownership. No assignment away from NEC/Nippon Electric appears anywhere on this patent's file. Note carefully: the name change from "Nippon Electric Company, Ltd." to "NEC Corporation" is not reflected as a recorded Change of Name assignment on this patent — Google Patents' legal-events tab stops at the 1983 certificate of correction. So the current-assignee field reading "NEC Corp" reflects NEC's own corporate renaming/its portfolio database, not a recorded USPTO conveyance.

Assignment timeline

Coverage caveat, stated plainly: I was unable to complete a direct, live query of USPTO Assignment Center (https://assignmentcenter.uspto.gov/, legacy https://assignment.uspto.gov/patent/index.html) — my tooling returned a step limit before I could pull the Assignment Center record itself. What follows is reconstructed from the patent's own legal-events/assignment rendering on the authoritative Google Patents record, which reproduces the USPTO assignment data (reel/frame, conveyance, assignor, assignee). I have no independent Assignment Center confirmation of the reel/frame, and the correspondent-of-record field is not exposed in the source I could reach. That field must be pulled directly from Assignment Center to complete signal #3.

The complete recorded chain is one link:

• 1980-05-27 (executed) / recorded 1982-03-19 — Reel 003954/0850
  • Conveyance: ASSIGNMENT OF ASSIGNORS INTEREST
  • Assignor: ARASEKI, TAKASHI; OZAWA, KAZUNORI (both, jointly)
  • Assignee: NIPPON ELECTRIC COMPANY, LTD., 33-1 Shiba Gochome, Minato-ku, Tokyo, Japan
  • Correspondent: Not available from the sources I reached. No attorney/agent/firm is rendered in the assignment rendering on the patent record; the only filing-side identifier given is the reel/frame. This is the single most important gap in this analysis — see Signal 3 below. (Because it is a one-link chain, the "repeat correspondent" test cannot fire regardless of who the correspondent was.)
  • Context: Employee-invention assignment to employer — the standard inventor→corporate-assignee conveyance, executed ~2 weeks before the 1980-06-09 US filing and recorded ~22 months later, shortly before the 1982-10-12 grant. Not an acquisition, not a fire-sale, not a securitization.

Adjacent, non-assignment events on the same record (not ownership transfers, listed for completeness):

• 1982-10-12 — Patent granted (US 4,354,273 A).
• 1983-04-19 — Certificate of Correction (post-grant correction; does not alter ownership).

Termination: Anticipated expiration 2000-06-09 (20 years from the 1980-06-09 filing); Google Patents status "Expired – Lifetime." The enforceable term ended over 25 years ago, so no assignment after 2000 could create assertion exposure on the '273 claims.

Finding: There are no recorded post-issuance transfers, no security interests, no releases, no licenses, and no change-of-name records on this patent. The original assignee's successor (NEC Corporation) is the effective terminal owner of record.

Timeline diagram

timeline
    title Ownership of US 4354273
    1979 : JP priority applications filed
    1980 : Inventors assign to Nippon Electric
         : US application filed
    1982 : Assignment recorded reel 003954 frame 0850
         : Patent issued
    1983 : Certificate of correction
    2000 : Patent expired
    2026 : No further assignments of record

NPE / troll-pattern signals

Number-collision warning first, because it matters for verification. A bare search on "4354273" is badly polluted by trademark registrations carrying the identical number: US Trademark Reg. No. 4,354,273 (DRYBAR, Drybar Holdings LLC, Dallas TX; registered 2013-06-18), French mark FR 4354273 (ORDIM, VG INVESTISSEMENT SARL), and unrelated municipal-procurement docket numbers. Any docket or assignment query on this number must be filtered to US utility patent 4,354,273 / application 06/156,990 or it will return false positives.

Verdict

Insufficient data — but with an unusually strong negative NPE finding attached.

The entire recorded ownership history of US 4,354,273 is one conveyance: reel 003954/0850, executed 1980-05-27 and recorded 1982-03-19, moving the application from inventors Araseki and Ozawa to their employer Nippon Electric Company, Ltd. That is the classic employee-invention assignment, and no subsequent assignment, security interest, license, release, or change-of-name has ever been recorded against the patent — which remained with NEC-lineage ownership until it expired 2000-06-09. Per the rubric I must select "Insufficient data" because only the original assignment exists, but that category should not be read as ambivalence here: there is no shell entity, no known asserter, no cascading LLC transfer, no pre-suit transfer, no bankruptcy sale, and no privateering anywhere in this chain, and none of the evidence is merely an absence of a hit on a name — it is the absence of any record at all after link one.

Two truthful limits on this verdict. (1) I could not complete a live Assignment Center query, so the reel/frame, conveyance wording, and the all-important correspondent of record for reel 003954/0850 are taken from the patent record's rendering of the assignment data rather than from Assignment Center itself; the correspondent is unretrieved, not merely unremarkable. (2) Given expiry in 2000, an NPE chain on these claims would be commercially pointless — the absence of one is consistent with the incentives, not just with luck.

Verify at: USPTO Assignment Center (search patent number 4354273, or application 06/156,990) — and cross-check the legacy interface at https://assignment.uspto.gov/patent/index.html. Filter out US Trademark Reg. No. 4,354,273 (DRYBAR) and French mark FR 4354273 (ORDIM), which collide on the bare number.

I'll verify each cited reference against live sources before mapping them to the §102 analysis.

Prior-Art Analysis — U.S. Patent No. 4,354,273

0. Identification and scope (read this first)

Target patent, taken literally: U.S. 4,354,273 — ADPCM System for speech or like signals, application 06/156,990, filed 1980-06-09, granted 1982-10-12, inventors Takashi Araseki / Kazunori Ozawa, assignee Nippon Electric Co., Ltd. (NEC). Verified on the patent record at https://patents.google.com/patent/[US4354273](/patent/US4354273)/en and https://uspto.report/patent/grant/4354273.

Number-collision filter applied (per the operating rule against auto-correction). Do not conflate 4,354,273 with:

• U.S. 5,333,273 — the "'273 patent" in Samsung v. Wistron, a computer keyboard/interrupt patent. Different patent.
• U.S. Trademark Reg. 4,354,273 (DRYBAR) and FR 4,354,273 (ORDIM) — trademark registers.
• Any bare "4354273" string hits in municipal dockets or academic regressions.

Verification limitation stated plainly: I do not have a queryable USPTO PatentCenter or Assignment Center session in this task. My verification of the target patent and of each cited reference runs through the Google Patents rendering of the USPTO record, the patentimages PDFs of the cited documents themselves, and uspto.report. Those are reproductions of the official data; they are not a live PatentCenter query. Where I could not read a document's full text, I say so rather than paraphrasing as if verified.

1. Legal framework for this §102 pass

The application was filed 1980-06-09 (priority 1979-06-12). Pre-AIA 35 U.S.C. §102 governs. Anticipation requires that a single reference disclose every element of a claim, arranged as in the claim (all-elements rule). Because claim 1 is the only independent claim, every one of claims 2–9 carries all claim-1 limitations; a reference that fails claim 1's elements cannot anticipate claims 2–9 either.

Priority-date arithmetic for the cited art (all comfortably §102(b) art):

Critical point for the anticipation question: none of the cited references is commonly owned with NEC, and none is by Araseki/Ozawa, so there is no "same inventive entity" or common-ownership obstacle to their use under §102. (That is not true of U.S. 4,317,208 and the copending Ser. No. 79,724 — see §5.)

2. The citation set of record (US 4,354,273)

The patent's own front page carries three patent citations and three non-patent citations. (The 23/42-entry "Cited By" lists are forward citations — later art citing the '273 — and are not prior art against it. Several earlier generated sections treat them as citation traffic only; that reading is correct.)

Two record corrections I am carrying forward, not silently fixing.

1. The Google Patents front-page non-patent citation renders Cohn as "Sep. 1970." The specification's body text and every independent source give September 1975 (vol. COM-23, no. 9, pp. 935–941). Per the instruction to prefer search results over stale renderings, 1975 is used here. Immaterial to §102 either way (both predate 1979-06-12).
2. The body text twice renders a result quantity as "the normalized signal from multiplier 23 is represented by e_j" and elsewhere "ê_j." I treat the multiplier-23 output as the reconstructed residual (the '273's own equations require it: X̂_j + ê_j). I flag this because claim construction of "decoded output" in claim 1 turns on it — see §4.

3. Requirement-by-requirement map of claim 1 (the anticipation template)

I decompose the sole independent claim into six elements, keyed to the claim text:

The two elements that decide every single-reference §102 question below are E2/E3 (the Δ-normalizing quantizer pair) and E4/E6 (a predictor whose coefficients are adapted from the decoded output). The cited art characteristically supplies one and not the other.

4. Reference-by-reference §102 analysis

PA-1 — US 3,502,986 (Lucky)

Full citation: U.S. Patent 3,502,986, Adaptive prediction for redundancy removal in data transmission systems, Robert W. Lucky, assignee Bell Telephone Laboratories, Inc.
Filed: 1967-12-14. Issued: 1970-03-24. §102(b) status: barred art (issued ~9.5 years before the 1979-06-12 priority). Non-patent counterpart: Lucky, "Adaptive Redundancy Removal in Data Transmission," BSTJ 47(4):549–573 (1968).
Brief description: A tapped-delay-line (transversal) predictor for data compression. A subtractor (24/44) forms the error e between the present digit a and the predicted digit. The error is correlated in multipliers 48 with the delayed prior digits on leads 52; the correlated products are integrated and averaged in low-pass filters 47 to form control signals that set the coefficients of attenuators 43 — i.e., the predictor taps are adapted from the error. The receiver uses a "bootstrap" predictor of "exact counterparts" (delay line 32, attenuators 33 with coefficients identical to the transmitter's), and the specification expressly discusses cumulative error propagation when the receiver predictor diverges.
Source: https://patents.google.com/patent/US3502986 ; https://patentimages.storage.googleapis.com/90/75/c2/5b4020ead68b0f/US3502986.pdf

(a) Planes-of-attack against claim 1

(b) §102 verdict

• Anticipates no claim. Lucky is missing E2 and E3 entirely (no adaptive quantizer / no normalization coefficient Δ whatsoever) and therefore also E5. Because claims 2–9 all depend from claim 1 (directly or through the claim-2/claim-4 chain), a claim-1 failure is dispositive for all of them.
• Closest claims: claim 1 (via E4/E6 — error-driven tap adaptation with a matched receiver predictor) and, derivatively, claims 5 and 8 (recursive predictor + matched receiver). But claims 5 and 8 require the claim-4/claim-1 quantizer structure, which Lucky lacks.
• Do not overread Lucky as a §102 reference. It is powerful §103 art on the "adapt the predictor from the residual" concept (the earlier obviousness section already uses it that way) and it is the best-art evidentiary support for the error-propagation problem the '273 claims to solve — but error propagation is background, not a claim limitation.

PA-2 — US 3,973,199 (Widmer / Philips)

Full citation: U.S. Patent 3,973,199, Prediction differential pulse code modulation system with adaptive compounding, assignee U.S. Philips Corporation. (I did not independently verify the named inventor's full name in this session; the record identifies the assignee as U.S. Philips.)
Filed: 1973-09-03. Issued: 1976-08-03. §102(b) status: barred art.
Brief description: A DPCM speech system with dynamic-range (companding) control of the quantizer. The transmitter (FIG. 1) has band-pass filter 2, sampler 3, non-uniform quantizing circuit 6, PCM coder 7, local receiver with predictor 8, adder 9 and expander 10, and a difference producer 11. The heart of the invention: "a control generator fed by the input signal from the predictor, said generator being provided with a storage network and an averaging network for obtaining a control signal which corresponds to the average of the absolute values of the predictor input signal over a limited number of sampling periods, said control signal being applied to a control input of the dynamic control means." FIG. 2 gives the matching receiver (regenerator 12, PCM decoder 13, expander 10′, adder 9′, predictor 8′). The disclosure also emphasizes the system's self-correcting behaviour after overload errors (error propagation promptly extinguished).
Source: https://patents.google.com/patent/[US3973199A](/patent/US3973199A)/en ; https://patentimages.storage.googleapis.com/be/b3/33/1a31fcc018f62c/US3973199.pdf

(a) Planes-of-attack against claim 1

(b) §102 verdict

• Anticipates no claim. Widmer fails E4 and E6 outright — it has no coefficient-adaptive predictor at all, which is the single inventive element of claim 1 as drafted ("the coefficients … being corrected in accordance with a filter control signal dependent upon said decoded output"). It also fails E2/E3 on the strict "divide/multiply by Δ" reading.
• Closest claims: none as an anticipation reference; its value is §103 and claim-construction support, because the "control signal … corresponding to the average of the absolute values" is precisely the Δ ≈ mean|e| premise the '273 specification later relies on to justify its substitution (see the earlier obviousness section, rationale (B)). That is a motivation-to-combine point, not an anticipation point.
• Do not treat Widmer's "self-correcting after overload" as the '273's anti-divergence feature. They are different mechanisms (range re-adaptation vs. non-recursive filter stability), and neither is a claim-1 limitation.

PA-3 — US 3,931,596 (Gersho & Goodman)

Full citation: U.S. Patent 3,931,596, Adaptive quantizer apparatus using training mode, Allen Gersho and David Joel Goodman, assignee Bell Telephone Laboratories, Incorporated; application 05/510,412. Verified at https://uspto.report/patent/grant/3931596.
Filed: 1974-09-30. Issued: 1976-01-06. §102(b) status: barred art.
Brief description: The specification's FIG. 1 is expressly "a prior art adaptive differential pulse code modulation (ADPCM) system" — differential amplifier 11, adaptive quantizer 12, encoder 10, summing amplifier 14 with first-order prediction network 15, and adaption logic network 16 that "monitors the output of encoder 10 … and provides for adaption of the quantizer step size on the basis of the most recently quantized output." FIG. 2 shows the 2N-level quantizer characteristic; FIG. 3 shows the adaption logic implemented with ROM 17 producing multipliers m_i, multiplier 29, and accumulator 21 (Δ ← Δ·m_i), and FIG. 4 shows the same multiplication performed by a shift register 18 — "a simple shift right operation."
Source: https://patents.google.com/patent/US3931596 ; https://patentimages.storage.googleapis.com/eb/b9/a9/bfee829dc43426/US3931596.pdf

Note the near-identity to the '273's own FIG. 1/FIG. 3B: the '273 depicts its control circuit 22 as "a register 22-1 which stores the normalization coefficient Δ and ROM 22-2 which … gives a new Δ (=Δ×m)." That is Gersho's FIG. 3 apparatus. The '273's Table 1 (m = 0.8/0.8/1.2/2.0) is Jayant's.

(a) Planes-of-attack against claim 1

(b) §102 verdict

• Anticipates no claim. Gersho is the closest single reference on E1–E3 and E5, but it fails E4 and E6 — the coefficient-adapting digital filter of claim 1 and its receiver counterpart. That failure is dispositive for claim 1 and therefore for all of claims 2–9.
• Closest claims: claims 2, 3 and 4 (normalizing divider + coder; plural discrete quantization levels — Gersho's FIG. 2 is literally an "adaptive quantizer with 2N output levels"; and the decoder/Δ-control-circuit/multiplier triple of claim 4, which maps onto ROM 17/multiplier 29/accumulator 21). Claims 2–4 nevertheless depend from claim 1 and cannot be anticipated on this reference alone.
• Two further points a challenger should know. (i) Gersho's own FIG. 1 is labelled prior art, i.e., the reference itself is evidence that the ADPCM transmitter/receiver architecture of the '273 was old by 1976 — useful for the §102(e)/§102(b)-style "the whole system was known" narrative even though it does not complete claim 1. (ii) The shift-right multiplier teaching ("the multiplication of factor m_i by a_k may be accomplished by a simple shift right operation") is the direct antecedent of the '273's own stated advantage that "multiplication by g′ … requires no actual operation." That is §103 material, already developed in the earlier obviousness section.

NPA-1 — Cohn & Melsa, "The Residual Encoder"

Full citation: D. L. Cohn et al., "The Residual Encoder — An Improved ADPCM System for Speech Digitization," IEEE Transactions on Communications, Vol. COM-23, No. 9, pp. 935–941, September 1975 (front page renders "Sep. 1970" — see §2, item 1).
Published: 1975-09. §102(b) printed publication: yes (predates 1979-06-12 priority by ~3.8 years).
Brief description: Adaptive differential PCM for speech with (i) a difference/subtractor stage, (ii) an adaptive quantizer, (iii) a local-decoder inverse quantizer, (iv) a pole predictor whose coefficients are adaptively corrected from the residual, and (v) a matching receiver predictor. The '273 specification expressly maps its own prior-art FIGS. 1–2 onto Cohn's FIG. 1(a)/(b) and reproduces Cohn's coefficient-correction algorithm as its own equation (2).

(a) §102 exposure — this is the single most dangerous reference in the set

Cohn is the only cited reference that supplies E4 and E6 (a coefficient-adaptive predictor fed from the residual) in combination with E1–E3 and E5 (the full normalized ADPCM quantizer/inverse-quantizer pair). If "dependent upon said decoded output" in claim 1 is construed broadly — to cover a control signal derived from Cohn's reconstructed/residual signal rather than from the specific quantity e_j/Δ — then Cohn potentially anticipates claim 1 outright under §102(b), and derivatively claims 2–5 and 8.

Why that construction is genuinely available: the '273 drafted the limitation as a structural/signal-flow recitation ("dependent upon said decoded output"), not as the mathematical quantity e_j/Δ. The specification's own distinction from Cohn is that Cohn's algorithm (eq. 2) uses the error where the '273 uses the normalized error (eq. 3) — and the patent then argues the two are "approximately identical" with g′ = g/3n. A patentee who argues equivalence to the prior art on the merits is poorly positioned to claim the distinction is a claim limitation.

Evidentiary limitation I am obliged to state: my reading of Cohn is drawn from the abstract, the reference list, and the '273 patent's own characterization of Cohn's FIG. 1(a)/(b). I have not read a full-text copy of the Cohn paper in this session. A §102 rejection or invalidity contention built on Cohn must be run against the paper's actual figures and text.

(b) §102 verdict: potentially anticipates claim 1 (and, derivatively, claims 2–5 and 8) under a broad construction of "decoded output"; does not anticipate claim 6 or 9 (Cohn is a recursive-pole predictor, not the non-recursive residual-driven FIR of the second embodiment); does not anticipate claim 7 (no noise-shaping filter 70).

NPA-2 — Jayant, "Adaptive Quantization with a One-Word Memory"

Full citation: N. S. Jayant, "Adaptive Quantization with a One-Word Memory," Bell System Technical Journal, Vol. 52, No. 7, pp. 1119–1144, September 1973.
Published: 1973-09. §102(b): yes.
Brief description: The one-word-memory step-size rule Δ_new = Δ·m(M_i) with the multiplier table the '273 reproduces verbatim as its Table 1 (M₁/M₋₁ = 0.8; M₂/M₋₂ = 0.8; M₃/M₋₃ = 1.2; M₄/M₋₄ = 2.0). Covers the quantizer characteristics (decision levels / representative levels) and the choice of the Δ-update multipliers.
Sources: cited on the '273 face; corroborated at https://www.scilit.net/publications/76183f662fb275dda9234da16a8f10af and https://patents.google.com/patent/US3931596

(a) §102 exposure: Jayant discloses E2's adaptation rule and quantization-level structure and the control-circuit data (claim 4's control circuit). It does not disclose E1, E4, E6 (or, standing alone, E3/E5 as claim-1 structures).

(b) §102 verdict: anticipates no claim. It is a support reference — it makes claims 3 and 4 (plural discrete levels; Δ generated from the coder output) trivially easy and supplies the exact Table 1 the '273 uses, but it lacks the predictor elements entirely.

NPA-3 — Mick, Digital Signal Processing Handbook (AMD, 1976, p. 47)

Full citation: Mick, Digital Signal Processing Handbook, p. 47, Advanced Micro Devices, Inc., Sunnyvale, CA (1976).
Published: 1976. §102(b): yes.
Brief description: Circuit-level implementation of a multiplier/divider — the '273 itself cites this page as the known way to build the multiplier 23 of its inverse quantizer and the divider 11 of its quantizer ("can respectively be composed of a multiplier/divider of the type illustrated in FIG. 6 in p. 47 of … [Reference 3]").

(a) §102 exposure: Mick discloses only implementation detail for E2/E3 arithmetic elements. It is an admission-against-interest reference: the '273 concedes these blocks were conventional.

(b) §102 verdict: anticipates no claim. It is a §103/secondary-reference exhibit (and an evidentiary admission of conventionality).

5. Consolidated §102 claim matrix

✓? = anticipates only under a broad construction of "control signal dependent upon said decoded output" (i.e., treating Cohn's residual-derived predictor control as the recited "decoded output"). On the narrow construction (the control signal must be the normalized quantity e_j/Δ), no cited reference anticipates any claim, and the invalidity case is §103, as developed in the earlier obviousness section.

Missing-element summary — what each reference lacks versus claim 1:

• Lucky: lacks E2/E3/E5 (no quantizer, no normalization coefficient Δ, no decoded output) → cannot reach any claim.
• Philips '199: lacks E4/E6 (no coefficient-adaptive predictor at all) → cannot reach any claim.
• Gersho: lacks E4/E6 (fixed first-order predictor; adaptation only to quantizer step size) → cannot reach any claim.
• Jayant / Mick: lack E1, E4, E6 → cannot reach any claim.

6. Two references the record does not support as §102 art (important non-findings)

The task asks specifically for cited references, but a defensible analysis must also say what cannot be used, so the §102 posture is not overstated:

• US 4,317,208 (ADPCM System for speech or like signals, Araseki, issued 1982-02-23) appears only in the "Similar Documents" list of the '273 record — it is not a citation of the '273. It postdates the '273's 1980-06-09 filing, and it names the same inventive entity (Araseki). It is therefore not §102(b) and not §102(e) "by another" art against the '273. Its utility is obviousness-type double patenting, as the earlier obviousness section notes — not anticipation.
• Copending U.S. Ser. No. 79,724, filed 1979-09-28 (Araseki et al.), cited in the '273 specification for the non-recursive predictor details, postdates the 1979-06-12 priority and, again, is the applicants' own work. Not §102 art against the '273; it is the source of the claim-6/claim-9 subject matter and is best handled as a same-family / ODP / later-application issue.

7. Bottom line

1. USPTO record for the literal number 4354273 identified and confirmed: the '273 is ADPCM System for speech or like signals, app. 06/156,990, filed 1980-06-09, granted 1982-10-12, NEC, expired 2000-06-09. No similar-number substitution was made, and the Samsung/Wistron 5,333,273 and trademark 4,354,273 collisions are excluded.

2. The examiner-cited reference set is three patents and three publications — and on a strict all-elements pre-AIA §102 reading, none of the three cited patents anticipates any of claims 1–9. Lucky is missing the entire adaptive quantizer/normalization-coefficient architecture (E2/E3/E5); Philips '199 and Gersho are each missing the coefficient-adaptive predictor (E4/E6); Jayant and Mick supply only quantizer-adaptation and arithmetic-block details.

3. The genuinely dangerous §102 reference is not a patent at all — it is Cohn & Melsa (1975), the very publication the '273 maps onto its own prior-art figures. Cohn may anticipate claim 1 (and, derivatively, claims 2–5 and 8) if "dependent upon said decoded output" is construed broadly; on the narrow, normalized-quantity construction it does not, and the case becomes one of §103 (Cohn + Jayant/Gersho + Philips '199 + Lucky), as already developed.

4. Claims 6, 7 and 9 are not anticipated by anything in the citation set, and cannot be reached by any single cited reference: claim 6/9 need the non-recursive residual-driven FIR (the applicants' own Ser. No. 79,724 lineage), and claim 7 needs the noise-shaping/error-feedback concept (Atal & Schroeder 1978 / Kimme & Kuo 1963 / Cutler) that is not among the cited art.

5. Practical weight (unchanged from the earlier sections, and it governs everything above): the patent expired 2000-06-09, a decade before AIA trials existed and roughly two decades before the §286 six-year damages window closed. This §102 mapping is therefore an historical validity-quality assessment and a defensive template for live members of the NEC/Araseki ADPCM family, not a ground for a current PTAB petition or an assertion against the '273.

6. Open verification items, stated rather than papered over. (a) I could not run a live PatentCenter/Assignment Center query, so bibliographic and assignment data are taken from the Google Patents rendering of the USPTO record and uspto.report. (b) I did not retrieve the '273 file wrapper; the fact that the patent granted over these references does not tell me the examiner's exact grounds, and I do not assert them. (c) I read Cohn, Jayant and Mick only via abstracts, citation lists and the '273's own characterization — the Lucky, Philips '199 and Gersho quotations above are from their published specifications (PDFs linked inline). Any §102 contention on Cohn must be re-run against the paper's full text. (d) The "Sep. 1970" vs. "Sept. 1975" Cohn date discrepancy is reported as found, not corrected on the record.

I'll ground the analysis in the actual prior-art references listed on the patent record before building the §103 case.

Let me verify the sibling patent and the noise-shaping art for claims 6–9.

Obviousness Analysis — US 4,354,273 under pre-AIA 35 U.S.C. §103

0. Framework and two flags before the merits

Flag 1 — date inconsistency. Your task header says "Current Date: April 26, 2026," while the session date I was given is 2026-09-27. Nothing in this §103 analysis turns on which is correct (the patent expired 2000-06-09), but I am not silently harmonizing them.

Flag 2 — a citation error on the record. The Google Patents front page renders the Cohn non-patent citation as "IEEE Trans. on Comm., vol. COM-23, No. 9, pp. 935-941, Sep. 1970." Every independent source I reached gives September 1975 (vol. 23, no. 9, pp. 935–941, DOI 10.1109/TCOM.1975.1092919; https://www.scilit.net/publications/76183f662fb275dda9234da16a8f10af ; https://www.semanticscholar.org/paper/8f1c4513a63d10c19b50a8a819708930ce3d7512). The specification body says 1975. Per the operating rule to prefer search results over stale renderings, I use 1975. It is immaterial to §103 (both dates predate the 1979-06-12 priority), but a citation to "Cohn (1970)" in any brief would be wrong.

Legal frame. Application US 06/156,990 was filed 1980-06-09, so pre-AIA §103(a) governs; a reference qualifies if it is prior art under pre-AIA §102 (here, §102(b) printed publications/patents more than one year before filing — Cohn 1975, Jayant 1973, Gersho 1976, Widmer 1976, Lucky 1970 all qualify, and all also predate the 1979-06-12 JP priority). The Graham v. John Deere factors apply, as glossed by KSR Int'l v. Teleflex (obvious to try; predictable variation; known technique improving a similar device in the same way; simple substitution of a known equivalent).

1. Person of ordinary skill in the art (POSITA)

A designer of digital speech codecs with a B.S./M.S. in EE and 2–4 years' experience in waveform coding, familiar with: adaptive quantization with backward-estimated step size (the Jayant one-word-memory scheme), gradient/sign-based adaptation of predictor tap coefficients, and the arithmetic cost of digital filter implementation (fixed-point multiply, ROM look-up, shift-and-add). This is the level at which the "fewer operations" motivation operates, and it is the level at which the case must be judged.

2. The prior-art set of record (and what each actually discloses)

Non-record art relevant only to claim 7 (noise shaping): Cutler's differential-quantization patents; Kimme & Kuo, "Synthesis of optimal filters for a feedback quantization system," IEEE Trans. Circuit Theory (Sept. 1963); and Atal & Schroeder, "Predictive coding of speech signals and subjective error criteria" (ICASSP Apr. 1978; IEEE ASSP 1979). These are what a modern challenger would add for the FIG. 8 embodiment (see §6).

3. The gap between claim 1 and the art

Claim 1 is a two-half system claim. Stripped to its elements, the art supplies:

• subtractor (X_j − X̂_j) → Cohn, Widmer, Gersho;
• adaptive quantizer = normalizing divider ÷ Δ + coder → Gersho (Fig. 1 + adaption logic 16) and Jayant (Δ rule);
• adaptive inverse quantizer = decoder + multiplier × Δ + Δ control circuit → Gersho (ROM 17 → mᵢ → accumulator), Jayant (Table 1), '273's own FIG. 3B;
• adaptive predictor = digital filter with coefficients corrected by a control signal derived from a system signal → Cohn (predictor), Lucky (A_i ← A_i + g·sign(e)·sign(a_{j−i})-style correlation of the error with past data, integrated to form tap control), Widmer (control generator averaging |·| of the predictor input);
• matching receiver inverse quantizer + predictor → Cohn FIG. 1(b); Lucky's "exact counterparts" bootstrap receiver.

The single delta is this: the coefficient-correcting control signal is taken from the decoder output, i.e. the normalized error e_j/Δ, instead of from the un-normalized error. The '273 specification admits as much when it says only "the parts related to adaptive prediction" differ from FIGS. 1–2, and when it argues equation (3) — A_i = A_i + g′(e_j/Δ)·sign(X_{j−i}) — is "approximately equal to" the prior Cohn algorithm (equation (2)) with g′ ≈ g/3n.

4. Motivation to combine (the KSR rationales, grounded in the references)

(A) The normalized quantity is already computed, free of charge — "use of a known technique to improve a similar device in the same way." In Cohn/Gersho-type architecture the decoder output is e_j/Δ; it exists as an intermediate node before multiplier 23. Tapping that node for the predictor's adaptation input adds no hardware. An engineer optimizing Cohn (whose complexity is expressly "on the same order as other ADPCM systems," i.e. not free) would see the normalized residual as the obvious, costless adaptation variable.

(B) The normalization removes signal-level dependence from the adaptation gain, and Widmer already teaches that Δ ≈ mean|e|. Widmer (US 3,973,199) discloses a control signal "correspond[ing] to the average of the absolute values of the predictor input signal over a limited number of sampling periods" and applies it to the difference-signal path. That is precisely the mathematical premise the '273 patent uses to justify its substitution ("Δ is close to the rms value of the prediction error… approximately equal to the average of the absolute value of e_j"). Where a reference supplies the reason the substituted quantity is equivalent to the old one, the substitution is an evident design choice.

(C) The substitution delivers a concrete, recognized engineering payoff: multiplication by a power-of-two constant becomes free. The '273 patent's own stated advantage — with g′ a power of 2, "multiplication by g′ can be achieved merely by appropriate wiring of the circuit… no actual operation is needed" — is the classic design incentive rationale. And the technique for cheap power-of-two scaling was itself known: Gersho performs mᵢ·a_k by "a simple shift right operation" and by ROM look-up. Combining Gersho's shift-multiplier teaching with Cohn's adaptive predictor to eliminate a per-sample multiply is a textbook KSR motivation.

(D) Same field, same problem, same solution type. All references are speech/voiceband waveform coders attacking band compression; Lucky and Widmer both discuss error propagation / self-correction in the receiver predictor, which is the '273 patent's asserted advantage. Corporate- and journal-common provenance (Bell Labs: Cohn sub-reference, Lucky, Gersho, Jayant) is additional KSR evidence of combinability.

(E) The result is mathematically predictable, not unexpected. The patent's own derivation reduces the new algorithm to the old one within a constant (g′ = g/3n, X/e ≈ 3–4). An invention whose benefit is demonstrated by showing it equals the prior algorithm is the paradigm case of a predictable variation; there is no unexpected-result story to tell because the specification never asserts one with data.

Net: Cohn (the complete ADPCM transmitter/receiver) + Jayant/Gersho (the Δ adaptation and normalizing quantizer, including a ROM/shift implementation) + Widmer (control signal = average |difference signal|) + Lucky (error-driven tap adaptation and a matching receiver predictor) renders claim 1 obvious. Cohn alone comes very close; Widmer supplies the missing reason to normalize.

5. Claim-by-claim map

6. Where the §103 case is genuinely contestable

1. Claim scope ambiguity cuts both ways. "Coefficients… corrected in accordance with a filter control signal dependent upon said decoded output" is broad. The specification's own FIG. 1 prior-art circuit feeds the predictor from the un-normalized ê_j and the decoder output e_j/Δ separately; if "decoded output" is read to cover either, Cohn anticipates claim 1 outright (§102), which is a stronger and simpler attack than §103. If it is read narrowly to require the normalized value, the §103 case in §4 above carries the burden. Patentees in this posture usually cannot have it both ways — the §112 indefiniteness/§102 squeeze is the practical leverage.

2. "Teaching away" is not available. No reference disparages normalizing the update signal; the worst that can be said is that Lucky and Cohn adapt on the raw error. KSR is explicit that the absence of a suggestion is not teaching away.

3. No secondary considerations are evidenced. The specification asserts simplified circuitry and greater compression but supplies no comparative test data, no unexpected-results showing, and no nexus evidence. Cohn and the '273 patent report essentially the same SNR class. Commercial success (NEC's ADPCM family) might be argued, but the nexus would have to run to the normalized-update limitation specifically, not to ADPCM generally.

4. Claim 7 and claims 6/9 are the only real footholds, for the reasons in the table — and claim 7's foothold disappears the moment Atal–Schroeder (1978) or Kimme–Kuo (1963) is put in the record.

7. Bottom line

• Claims 1–5 and 8 are, on this record, highly vulnerable to §103 over Cohn (1975) in view of Jayant (1973) and/or US 3,931,596 (Gersho), further in view of US 3,973,199 (Widmer), and US 3,502,986 (Lucky). The motivation is not merely "these are all ADPCM references"; it is concrete and triple-sourced: (i) the normalized residual is a free, already-computed node (Cohn/Gersho architecture); (ii) Widmer expressly teaches that the normalization control signal equals the average absolute value of the very signal being coded — the equivalence the '273 patent relies on; and (iii) Gersho expressly teaches implementing the adaptive multiplier by ROM look-up and shift, which is exactly the "no actual operation is needed" advantage the '273 patent claims.
• Claim 3 and claim 4 are the easiest (explicit level tables and explicit ROM/accumulator/multiplier structure in Jayant, Gersho, Cohn).
• Claim 7 (and its receiver analogue) is the hard one unless Atal & Schroeder (ICASSP 1978 / IEEE ASSP 1979) or Kimme & Kuo (1963) is added for the noise-shaping/error-feedback concept.
• Claims 6 and 9 are better attacked as obviousness-type double patenting over US 4,317,208 (Araseki, issued 1982-02-23) than as §103 over the cited art, because the residual-domain FIR predictor appears to be the same family's own contribution (copending Ser. No. 79,724, filed 1979-09-28) rather than another's prior art.

Practical weight: as the earlier sections of this analysis establish, no claim of 4,354,273 was ever tested — no PTAB proceeding, no identified litigation — because the term expired 2000-06-09, before AIA trials existed and more than six years before the §286 damages window closed. This §103 analysis therefore matters (a) as a historical/validity-quality assessment, (b) as a defensive showing if any party ever asserted the expired patent for pre-2000 conduct, and (c) as a template for the still-live members of the NEC/Araseki ADPCM family, where §103(c)/AIA §102(b)(2)(C) common-ownership carve-outs and the sibling US 4,317,208 should be checked before assuming the Cohn/Jayant/Widmer/Lucky combination above is still available as a ground.

Caveat on completeness: my reading of Cohn is drawn from the abstract, reference list, tables, and the '273 patent's own characterization of it, not from a full-text copy of the paper; the Lucky and Widmer quotations above are from their published specifications. Where I have not read full text, I say so rather than paraphrase as if verified.