US 4317208

I'll verify the patent record and check for any litigation/docket information before summarizing.

US Patent 4,317,208 — Verification Summary

Important framing note: I retrieved the authoritative full text of this patent (Google Patents, fetched 2026-09-30) and ran targeted searches. I did not find any CAFC 2026 docket entry or district-court litigation involving U.S. 4,317,208. I also did not find it in any litigation database indexed by these searches. One search for the bare number "4317208" returned an unrelated Argentine trademark record (Acta 4317208, Class 25, apparel) — that is a different identifier entirely and is not related. Treat "no litigation found" as a negative search result, not proof that none exists.

Bibliographic data

Inventorship/assignment: Assignment executed 1979-09-20, recorded 1981-09-10 (Reel 003909/0075), transferring Araseki and Ochiai's rights to Nippon Electric Co., Ltd.

Family/parallel filings (same priority JP 12325578): AU 524633B2, CA 1148661A, FR 2438384A1/B1, GB 2033702A/B, JP JPS5550738A (later JPH0519331B2). A separate US patent, US 4,354,273, appears in Google's list of "Similar Documents" with the same title — that listing is a similarity citation, not a confirmed family relationship, so I will not assert they are one family without further evidence.

Abstract (as published)

"An adaptive differential pulse code modulation (ADPCM) system includes a predictor which predicts a sample value based on past prediction errors and coefficients which are adaptively corrected to lessen the difference, i.e. the prediction error, between the predicted values and the actual values. The predictor is duplicated in the receiver, has no feedback loop and thus instability due to transmission errors is eliminated. The system can also include a second predictor whose output is combined with that of the first predictor to obtain the predicted value. The second predictor output is based on past sums of the prediction error and the predicted value and coefficients which are adaptively corrected. The second predictor is in a feedback loop but instability is prevented by choosing the coefficients used."

Plain-language overview of the independent claims

Claim 1 — ADPCM system (transmitter + receiver).
A transmitter has three parts: a subtractor that forms the difference E_j between input sample X_j and a predicted value x̂_j; a quantizer that quantizes that difference to produce E_j; and a transmit prediction means with variable coefficients and no feedback loop that takes the quantizer output and generates the predicted values. A receiver mirrors this: a receive prediction means, also with variable coefficients and no feedback loop, generates predicted values from the received quantizer outputs, and an adder sums the received quantizer outputs with those predicted values to produce the reproduction signal. The core point: predictions are driven by the quantized error signal, not by the reconstructed signal, so there is no closed loop anywhere and a channel error cannot drive the decoder unstable.

Claim 2 — Transmitter for an ADPCM system.
Same transmitter as in claim 1, recast as a standalone apparatus: subtractor → quantizer → transmit decoder containing a no-feedback predictor that derives predicted values from the quantizer output.

Claim 3 — Receiver for an ADPCM system.
Same receiver as in claim 1, standalone: a no-feedback receive prediction means generating predicted values from received quantizer output signals, plus an adder that combines them into the reproduction signal X_j.

Claim 6 — ADPCM system with two predictors (the "second embodiment," FIG. 3).
The predicted value is split into two parts, Y_j + Z_j. The transmitter has: a first no-feedback predictor (50) producing Y_j from the quantizer output; a first adder that adds Y_j and Z_j to produce the overall predicted value fed to the subtractor; a second adder producing a reconstructed signal; and a second predictor producing Z_j from that reconstructed signal (this second predictor is in a feedback loop). The receiver mirrors it: adder 160 combines received predicted values with quantizer outputs to yield the reproduction signal; no-feedback predictor produces Y_j; adder 170 combines Y_j and Z_j into the receive predicted value; and a second receive predictor produces Z_j from the reproduction signal. The zero-based (no-feedback) predictor provides stability; the pole-based (feedback) predictor provides extra gain, with stability ensured by constraining its coefficients.

Claim 7 — Transmitter version of claim 6. The transmit-side structure only: first no-feedback predictor → adder with Z_j → subtractor; second adder forms the reconstructed signal, which feeds the second predictor that generates Z_j.

Claim 8 — Receiver version of claim 6. The receive-side structure only: adder 160 forms the reproduction signal; no-feedback predictor generates Y_j; adder 170 combines Y_j + Z_j into the receive predicted value; second receive predictor generates Z_j from the reproduction signal.

Dependent claims (brief):

• Claim 4 (depends on any of 1–3): defines the predicted value as the sum over i of B_i^j · E_{j−i}, with the coefficient update B_i^{j+1} = (1−δ)·B_i^j + g·E_{j−i}·E_j, where δ ≪ 1 and g is a positive constant.
• Claim 5 (depends on 4): numeric ranges M ≈ 7 and δ ≈ 2^−6.
• Claim 9 (depends on 6 or 7): defines Y_j by the same B_i^j recursion, and Z_j = A_l^j · X_{j−l} with A_l^{j+1} = (1−δ)·A_l^j + g′·E_j·X_{j−l}.
• Claim 10 (depends on 9): 0 ≤ A_l^j ≤ 0.9 (the stability constraint).
• Claim 11 (depends on 9): M ≈ 3 and A_l^j ≈ 0.9.
• Claim 12 (depends on 6 or 8): receive-side definitions of Y_j and Z_j, mirroring claim 9.

Technical background the specification emphasizes: conventional ADPCM (FIG. 1A) uses a feedback loop predictor (30/130) fed by the reconstructed signal, which can oscillate after a transmission error; the usual mitigation is a "leakage" term δ in the coefficient update, which trades error-recovery speed against prediction gain. The invention's contribution is to make the primary predictor zero-based (driven by the quantized error) so instability is structurally eliminated, and optionally to add a small pole-based predictor whose coefficient count is deliberately limited to one so stability is trivially guaranteed (|A_l^j| < 1).

Prior art and citations of record

• U.S. patents cited: US 4,071,842 (Bell Telephone Laboratories); US 4,093,962 (Nippon Electric); US 4,144,543 (Nippon Electric).
• Non-patent literature cited: IBM Technical Disclosure Bulletin, vol. 15, no. 11, April 1973, pp. 3338–3341, "Adaptive Rate Delta Modulator."
• Non-patent literature discussed but not formally cited: Cohn et al., "The Residual Encoder — An Improved ADPCM System for Speech Digitization," IEEE Trans. Commun., COM-23(9), Sept. 1975, pp. 935–941; Cummiskey et al. (adaptive quantization), BSTJ 52(7), Sept. 1973, pp. 1105–1118; Duttweiler, "A Twelve-Channel Digital Echo Canceler," IEEE Trans. Commun., COM-26(5), May 1978, pp. 647–653 (source of the FIG. 4 predictor structure).

Later-art relevance: The patent is repeatedly characterized in subsequent literature as an early zero-based / residual-driven predictor design. For example, US 4,593,398 (Northern Telecom, "residual-driven adaptation of feedback predictor") and US 2001/0040927 both distinguish their approaches from "the Araseki system," where coefficients derive in part from a reconstructed signal — a useful signal of how the art treated this patent. Google lists 21 examiner/third-party forward citations (including US 4,470,146, US 4,593,398, US 4,654,863, and a long Broadcom "noise feedback coding" lineage).

Uncertainty / caveats

• The full US 4,317,208 text above came from Google Patents; I did not independently pull the USPTO PatentCenter image file or the CAFC docket system directly. The provided text is authoritative for the claim language.
• I found no evidence of any 2026 CAFC or district-court activity on this patent. Given that it expired for failure to pay maintenance fees with an anticipated expiration of 1999-09-28, litigation exposure is functionally nil, but I state this as a search-based inference, not a verified docket fact.
• US 4,354,273 (same title, NEC) appears in Google's "Similar Documents" list. I have not confirmed whether it is a continuation/division of 4,317,208 or an independent filing — do not treat them as the same family without checking.

I'll search for litigation involving US patent 4317208.

Let me try more targeted litigation sources.

Let me run a few additional checks against litigation-specific databases.

Answer: No litigation is known involving U.S. Patent No. 4,317,208

I searched litigation-focused sources (Unified Patents litigation portal, RPX Empower/litigation.rpxcorp.com, Stanford NPE Litigation Database, CourtListener/RECAP, and general web/Google Patents) for the exact patent number 4,317,208 ("ADPCM System for speech or like signals," Nippon Electric Co., Ltd. / NEC Corp.; inventors Araseki and Ochiai; filed 1979‑09‑28; granted 1982‑02‑23). I found no records of any district court action, ITC investigation, CAFC appeal, or PTAB proceeding asserting or challenging this patent. The searches returned zero case hits for this patent number.

What I checked and what I found

Important caution — a similar-number decoy

A search for this patent surfaced a case involving a different patent with a superficially similar number. I am flagging it so it is not mistakenly attributed to 4,317,208:

• Siemens‑Elema AB v. Puritan‑Bennett Corp., 925 F.2d 1480 (Fed. Cir. 1991) (S.D. Cal. No. 86‑1728) — this concerns U.S. Patent No. 3,741,208, a medical ventilator patent. Per your instruction to interpret identifiers literally and not return results for similar numbers, this case is not related to U.S. 4,317,208 and should be excluded.

Contextual note on why no litigation is likely

The Google Patents record for US 4,317,208 lists its anticipated expiration as 1999‑09‑28 (20 years from the 1979‑09‑28 filing), with legal status "Expired – Lifetime." U.S. patent litigation against this patent would therefore have had to occur between 1982 and its 1999 expiration. The absence of any litigation record is consistent with an expired, relatively old telecommunications coding patent. Its significance is primarily as prior art/technology lineage (it is cited by later ADPCM and predictive‑coding patents, e.g., U.S. 4,593,398 to Northern Telecom, U.S. 5,291,284 to British Telecommunications), rather than as a case‑law subject.

Confidence and limitations

• High confidence that no litigation for 4,317,208 appears in the public litigation databases and aggregators searched.
• Caveat: Some historical district court records from the 1980s–1990s are incomplete in free databases; PACER coverage for older closed cases and pre‑2004 filings can be sparse. I cannot rule out an obscure or long‑closed suit that never generated a searchable record. If you need a definitive answer, a paid comprehensive check (e.g., Docket Navigator or a full PACER/derivative‑index search) is advisable.
• I did not fabricate any case; where no data existed, I report none.

Bottom line: Based on the sources searched, there is no known litigation involving U.S. Patent No. 4,317,208.

I'll verify the structured data with targeted searches before writing the report.

One more verification pass, including reexamination and any AIA challenge filed pre-2020.

Proceedings overview

Total AIA trial proceedings on file for U.S. Patent No. 4,317,208: ZERO. The structured "PTAB proceedings on file" block included in this prompt — the canonical list sourced from the USPTO Open Data Portal — returns no IPR, PGR, or CBM proceedings, and my independent web/PTAB E2E sweeps found no proceedings the ODP has failed to index. Breakdown by status is therefore trivially: active 0 / claims invalidated 0 / claims sustained 0 / settled 0 / institution denied 0.

Bottom-line defensive posture: This is not "a hardened patent" and it is not "a patent with dead claims." It is a patent that AIA trials never touched, because it was already dead before the AIA trial regime existed. US 4,317,208 expired 1999-09-28 (20 years from its 1979-09-28 filing date, per the Google Patents legal-events record: "1999-09-28 — Anticipated expiration"), roughly thirteen years before the first AIA petition could be filed on 2012-09-16. There is no PTAB estoppel to leverage, no FWD to quote, and — more importantly — no infringement exposure to defend against, since an expired patent cannot be asserted for prospective infringement. If someone is waving this patent number at you in a demand letter, the PTAB calendar is irrelevant; the expiration date is the kill shot.

Individual proceedings

None to itemize. There is no {PROCEEDING_NUMBER} to report, no petitioner, no panel, no institution decision, no Final Written Decision, and no Federal Circuit appeal for U.S. 4,317,208. I will not manufacture one.

What I can report is the near-miss set — patents whose numbers superficially resemble "4,317,208" or "…208," which repeatedly contaminated the searches. Per the operating instruction to interpret identifiers literally and not auto-correct them, none of these is U.S. 4,317,208, and none should be attributed to it:

I am stating the CPC/Apple matters as decoys only because they dominated the search results for "208" + PTAB. I did not independently verify every procedural detail of those proceedings, and they are irrelevant to your patent regardless.

Strategic summary

Claim status: all 12 claims of US 4,317,208 are UNTESTED at the PTAB and all 12 stands as issued. Claims 1, 2, and 3 are the independent claims (system, transmitter, receiver). Claims 4–5 depend from "any one of claims 1–3" and recite the prediction/adaptation equations (M≈7, δ≈2⁻⁶). Claims 6–8 are the second-embodiment independent claims (dual-predictor system, transmitter, receiver); claims 9–11 depend from claims 6 or 7 (0≦A_l^j≦0.9; M≈3, A_l^j≈0.9); claim 12 depends from claims 6 or 8. Because no AIA trial occurred, there is no statutory disclaimer, no certificate-canceled claim, and no substitute claim in the file. Nothing has been narrowed, and nothing has been fortified.

Estoppel landscape: empty, and structurally irrelevant. 35 U.S.C. § 315(e)(2) estoppel attaches only to a petitioner in an IPR that "results in a final written decision under section 318(a)." There is no such FWD here, so no party is estopped by anything on this patent. The corollary is what matters to you: no prior-art ground has been burned. Every § 102/§ 103 theory against claims 1–12 remains theoretically available in any forum — but see the timing problem below, which is the real constraint, not estoppel.

Pattern signals: none. No repeat petitioner (no petitioner at all), no patent-owner PTAB appeals, no defensive aggregator (Unified Patents, RPX, etc.) in the chain. The only post-grant history in the record is prosecution and assignment: filed 1979-09-28 by Nippon Electric Co., Ltd.; inventorship assignment recorded 1979-09-20 (effective), REEL/FRAME 003909/0075; granted 1982-02-23; expired 1999-09-28. Family members — JP 53‑123255 (JPS5550738A), CA1148661A, FR2438384A1, GB2033702B, AU524633B2 — are all similarly long-expired. The patent's only live significance is as prior art and technology lineage: Google Patents lists it among "Cited By" of later work (e.g., US 4,593,398 to Northern Telecom, US 5,291,284 to British Telecommunications) and among "Similar Documents" alongside US 4,354,273, US 4,554,670/CA1,156,369 (NEC's own follow-on ADPCM filings, US4354273A and CA1156369A/ADPCM system for speech or like signals). Note the cross-reference: the previously generated litigation summary reached the same conclusion (no litigation), so the two sections are consistent, not contradictory.

Avenues foreclosed by statute: CBM review is unavailable — US 4,317,208 is a speech-coding patent, not a "covered business method" patent under AIA § 18(d)(1), and the CBM program sunset on 2020-09-16 in any event. PGR is unavailable — the patent's effective filing date (1978-10-05) is decades before the 2013-03-16 PGR cutoff, so it is not a "first-inventor-to-file" patent. IPR is the only AIA vehicle whose statutory window was theoretically open to this patent, and it was never used.

Recommended next steps

• If you are a defendant or a demand-letter recipient: the dispositive fact is the 1999-09-28 expiration (Google Patents legal events, "Anticipated expiration"; status "Expired – Lifetime" — https://patents.google.com/patent/[US4317208A](/patent/US4317208A)/en). An expired patent cannot support prospective infringement liability for post-expiration conduct, and there is no PTAB FWD to cite because none exists. Your response is an expiration/freedom-to-operate argument, not an IPR strategy. There is no claims-invalidated FWD to link to, because no claim of this patent was ever adjudicated — do not let anyone hand you the CPC/Apple IPR2022-00601 FWD as if it applied here.
• If you are evaluating this patent as prior art: its pre-AIA status, 1979-09-28 filing, 1978-10-05 JP priority, and 1982-02-23 grant make it solidly available as § 102(a)/§ 102(b)/§ 102(e) art (as applicable to the target's own priority scheme) against later ADPCM and linear-prediction filings. Use the published family (JPS5550738A, CA1148661A, FR2438384A1, GB2033702B) for corroboration and the JP priority document for the earliest date.
• If you need a belt-and-suspenders check: AIA trial data is reliably covered by PTAB E2E/ODP, but pre-2012 administrative activity is not well covered by IPR/PGR/CBM databases by definition, and ex parte reexamination certificate histories for pre-2000 patents can be thin in free sources. My searches surfaced no reexamination of US 4,317,208, but if reexam history is outcome-determinative for you, pull the paper file wrapper directly via USPTO Patent Center for application no. 06/079,724 rather than relying on aggregators. I did not find a reexamination record and I am not asserting one is absent with the same confidence I have about the (structurally guaranteed) absence of AIA trials.
• Milestone note: with zero active proceedings, there are no institution deadlines, no oral-hearing dates, and no § 316(a)(11) one-year FWD due dates to calendar. If a counterparty tells you a PTAB proceeding on this patent is "pending," that representation is false on the record I can see — ask them for the proceeding number and check it at https://ptacts.uspto.gov/ptabweb/ before taking it seriously.

Confidence: High that no AIA trial proceeding exists for US 4,317,208 (ODP canonical list is empty; independent web and E2E sweeps found nothing for the literal number). High confidence in the expiration date and the resulting defensive posture. Lower confidence on the absence of ex parte reexamination, as noted above. I did not fabricate any proceeding number, petitioner, panel, or outcome.

I'll reconstruct the assignment chain using the authoritative patent text plus targeted searches of USPTO assignment records and related corporate-history sources.

Let me query the USPTO assignment and PEDS endpoints directly, plus check for any post-issuance transfers.

I was unable to get a live, parsed response from the USPTO Assignment Center's JS-driven search app (the queries returned unrelated hits — a Norwegian electrical catalogue number and an Argentine trademark, both of which are different identifiers and must not be conflated). What follows is therefore reconstructed from the authoritative patent text you supplied (its "Legal Events" block reproduces the recorded assignment) plus corroborating corporate-history sources. I flag every place where I am inferring rather than reading a record.

Inventors

Pattern check: The assignment was executed 1979-09-20, i.e. 8 days before the US filing date (1979-09-28), and roughly 11.5 months after the JP priority (JP 12325578, 1978-10-05). This is the ordinary "sign the assignment on the way out the door to the PCT/US filing" pattern for a corporate research lab — not a departure signal. No evidence was found of either inventor leaving NEC within 12 months of filing, and neither inventor name reappears in the forward-citation record as assignor or in any later-recorded conveyance. There is no "all inventors departed" fire-sale precursor here.

Original assignee

Nippon Electric Co., Ltd. (renamed NEC Corporation in 1983 for its English-language name; Japanese legal name remains 日本電気株式会社).

• Named on the issued patent: yes — the face of US 4,317,208 is Nippon Electric Co., Ltd.; Google Patents' "Current Assignee" field normalizes this to "NEC Corp."
• Product embodying the claims: NEC was a full-line telecommunications and electronics manufacturer at the time and shipped the enabling hardware (it introduced the µPD7710 digital signal processor in 1980 and built out digital switching/transmission product lines through the 1980s). The patent is an algorithm/architecture patent for a speech-quality ADPCM codec — the kind of thing NEC implemented in its own transmission equipment rather than as a standalone part. This is a finding of operating-company ownership, not a licensing vehicle.
• Primary line of business: IT/network solutions and electronics (computers, semiconductors, telecom switching, mobile). Member of the Sumitomo Group.
• Current status: Operating. NEC remains a publicly traded company (TSE: 6701), headquartered at 7-1, Shiba 5-chome, Minato-ku, Tokyo — the modern address of the "33-1, Shiba Gochome, Minato-ku, Tokyo" recorded on the 1981 assignment.

Note the address continuity: the recorded assignee address (33-1 Shiba Gochome) and NEC's current registered address (7-1, Shiba 5-chome) are the same Shiba district block — consistent with a single continuous corporate entity, not a shell.

Assignment timeline

One recorded assignment exists. The chain below is exhaustive as to what appears in the record:

• 1979-09-20 (executed) / recorded 1981-09-10 — Reel 003909/0075
  • Conveyance: Assignment (Assignment of Assignors' Interest)
  • Assignor: Takashi Araseki; Kazuo Ochiai (jointly)
  • Assignee: Nippon Electric Co., Ltd., 33-1, Shiba Gochome, Minato-ku, Tokyo, Japan
  • Correspondent: Not stated in the surfaced record. The reel/frame format (003909/0075) is an old microfilm-reel series consistent with the 1981 recording date, and the Google Patents legal-event extract does not reproduce a correspondent/attorney field for this entry. I cannot name the recording correspondent without pulling the original recorded cover sheet from the Assignment Center. Flagging this as an evidentiary gap rather than guessing.
  • Context: Initial inventor-to-employer assignment — standard corporate prosecution practice, executed immediately before filing.

No post-issuance assignments are recorded. There is no evidence of any transfer away from NEC/Nippon Electric at any time in the patent's life. The patent's legal status is "Expired – Lifetime," anticipated expiration 1999-09-28 (20 years from the 1979-09-28 US filing).

One caveat on the 1983 rename: NEC changed its English company name to "NEC Corporation" in 1983. A pure name change is not an assignment and would not necessarily generate a recorded Reel/Frame entry — and none appears in this patent's record. The 1981 assignment (which used the "Nippon Electric Co., Ltd." name) simply remains the operative ownership record, with the change of name carried by the corporate entity itself. Google's mapping of the current assignee to "NEC Corp" reflects that rename, not a new recorded transfer.

Timeline diagram

timeline
    title Ownership of US 4317208
    1978 : JP priority application filed
    1979 : Inventors assign rights to Nippon Electric
         : US application filed
    1981 : Assignment recorded Reel 003909/0075
    1982 : US patent 4317208 issued
    1983 : Nippon Electric renamed NEC Corporation
    1999 : Patent expires

NPE / troll-pattern signals

1. Shell-entity transfer — NOT PRESENT. The only recorded assignee is Nippon Electric Co., Ltd. / NEC Corporation, an operating manufacturer, per Reel 003909/0075 (recorded 1981-09-10). There is no downstream LLC, no "IP/Holdings/Licensing" suffix, and no registered-agent address anywhere in the chain.

2. Known asserter in the chain — NOT PRESENT. Neither assignee appears on any NPE directory (Acacia, Marathon, IV, Wi-LAN/Conversant, Round Rock, Pendrell, etc.). NEC itself has appeared in litigation databases but only as a party on unrelated patents — never on the '208 patent, and never as an NPE.

3. Repeat correspondent across the chain — NOT PRESENT / UNCLEAR. With only a single recorded link and no correspondent field surfaced, there is no recurrence to test. This signal requires ≥2 appearances by the same recording attorney, and the chain is only one link long, so the signal cannot arise.

4. Cascading transfers — NOT PRESENT. Zero consecutive downstream assignments exist, let alone multiple LLC hops within 24 months. The chain is a single link (inventors → NEC).

5. Pre-litigation transfer — NOT PRESENT. No infringement suit naming US 4,317,208 was located (see the earlier litigation section). With no suit and no transfer, the predicate does not exist.

6. Bankruptcy fire-sale — NOT PRESENT. No bankruptcy sale, no Chapter 7/11 proceeding, and no patent-sale estate touching this patent. NEC was never in bankruptcy; the patent simply expired at end of term.

7. Privateering — NOT PRESENT. No operating company → NPE transfer of this patent exists in the record, so there is nothing to assert against competitors on NEC's behalf.

8. Defensive aggregator (anti-NPE) — NOT PRESENT. The chain does not terminate at RPX, AST, LOT, Unified Patents, or OIN. It terminates at the original operating assignee, where it stayed until expiration.

Verdict

Insufficient data (only the original assignment).

Per your rule, "no post-issuance assignments — that is itself a finding," the only recorded conveyance is Reel 003909/0075 (executed 1979-09-20, recorded 1981-09-10), a straight inventor-to-employer assignment from Araseki and Ochiai to Nippon Electric Co., Ltd. No later transfer, no shell entity, no asserter, and no defensive aggregator appears anywhere in the record; the patent stayed with NEC until it expired for end of term on 1999-09-28. Ownership never left the operating company, so none of the eight NPE signals can even attach to a supporting reel/frame. This is the cleanest possible ownership profile — a corporate captive patent that was never traded.

Verification link: USPTO Assignment Center — search by patent number 4317208 (enter application 06/079,724 or patent 4,317,208; alternately assignment.uspto.gov legacy search).

Confidence / limitations:

• High confidence that Reel 003909/0075 is the only recorded assignment — it is the sole assignment event in the authoritative Legal Events block, and no source surfaced any other.
• Lower confidence on the correspondent field: I could not retrieve the recording attorney's name from this patent's cover sheet because the Assignment Center's search app did not return parsed results to my queries. If you need the correspondent, the cover sheet image for Reel 003909/0075 must be pulled directly from the Assignment Center.

I'll verify the patent record and each cited reference before analyzing anticipation.

Now let me verify the remaining cited references.

Let me verify the remaining references and the family/foreign citations.

Prior-Art Analysis — U.S. Patent No. 4,317,208

Scope note (identifier discipline): Every ID below is taken literally. "4317208" = US 4,317,208 only. I discarded all search hits for look-alike numbers (e.g., Dacom's Dacom 450 fax code, the Argentine trademark Acta 4317208, the El-number/EFO "4317208" proximity-sensor listing, and EPO provisional application "US 4317208 P" used as a priority document in unrelated cases). None of those is the patent at issue.

Method/source caveat: I was able to retrieve and verify the patent and its citation record through indexed public records of the USPTO data (Google Patents, USPTO.report, FreePatentsOnline/Justia mirrors, and the EPO/Google full-text of the cited documents). I could not open PatentCenter's image file directly in this session, and I reached the tool-call limit before retrieving the full text of the single non-patent reference (the IBM TDB article). Where content was not retrieved, I say so.

1. The patent under analysis (verified)

Critical date for prior art: the effective filing/priority date is 1978-10-05. Any reference must predate that date (or qualify under pre-AIA §102(e) as of its U.S. filing date) and disclose every element of a claim to anticipate it.

The single novel limitation running through all independent claims: a prediction means having no feedback loop that is driven by the quantizer output E_j (the zero-based / residual-driven predictor), with the receiver duplicating it. Everything below turns on whether a cited reference shows that.

2. Citations of record (the "patent citations for 4317208")

Google Patents/FreePatentsOnline list three U.S. patents cited by the examiner plus one non-patent citation, and a separate list of family citations (references cited in the foreign counterparts, not necessarily against the U.S. claims).

There are no citations of Araseki/Ochiai's own earlier work in the record. Note the unusual fact that two of the three examiner citations (R2, R3) are the applicant's own employer's patents (NEC) — the same corporate family later characterized this patent as an advance over feedback-loop prediction, so R2/R3 are best understood as the "state of the art at NEC" backdrop.

3. Reference-by-reference §102 analysis

R1 — US 4,071,842 (Tewksbury, Bell Telephone Laboratories)

• Citation: US 4,071,842, "Apparatus for analog to digital conversion," inventor Stuart Keene Tewksbury, assignee Bell Telephone Laboratories, Inc.; filed 1975-08-28, granted 1978-01-31.
• Predates priority? Yes (granted ~9 months before the 1978-10-05 priority). Qualifies as §102(a)/(b) art.
• Description: An oversampled predictive coder / error-feedback coder with a signal-independent feedback filter. In the predictive coder (FIG. 1/4), quantizer 101 receives the difference e_n = s_n − f_n; adder 104 forms X_n = E_n + f_n; and feedback filter 102 is fed from X_n so that f_n is the prediction of the next sample. The disclosure says explicitly that the feedback network is "maintained independent of the input signal," and the summary states the filter output f_n "is a good prediction of the next input signal sample s_n." Alternative integrator-chain implementations are given, plus separate error-feedback embodiments (FIGS. 7–8).
• Anticipation assessment: Does not anticipate any claim of 4,317,208. R1 does disclose a subtractor (105) + quantizer (101) + predictor (102), i.e., elements (a) and (b) of claim 1 and much of claim 2. But R1's predictor is in the feedback loop and is driven by the reconstructed signal X_n, not by the quantizer output alone — the express opposite of claim 1's "prediction means … having … no feedback loop" that "receiv[es] quantized output signals … and generat[es] therefrom said predicted values." Because anticipation under §102 requires every element in a single reference, R1 cannot anticipate claims 1, 2, or 3. It is, at most, a §103 reference for the transmitter front end and a secondary reference for the Z_j (feedback) predictor of claims 6–12.
• Claims it is "relevant to" (not anticipatory): 1, 2 (subtractor/quantizer only), and the Z_j-predictor sub-element of claims 6, 7, 9, 12.

R2 — US 4,093,962 (Nippon Electric Co., Ltd.)

• Citation: US 4,093,962, "Adaptive predictive encoder," assignee Nippon Electric Co., Ltd.; priority 1976-12-01, granted 1978-06-06.
• Predates priority? Yes.
• Description: A predictive encoder for variable-length coding with buffer-occupancy control. Subtractor 12 subtracts the local demodulator output from the input to form a predictive error signal; quantizer unit 14 (with a controllable factor k) quantizes it; and demodulator 13 comprises adder 31 summing the quantized error with the demodulator output, feeding predictive filter 32 that produces the local prediction. It also cross-references the earlier NEC application of Tatsuo Ishiguro (Ser. No. 657,799, now US 4,027,100).
• Anticipation assessment: Does not anticipate any claim of 4,317,208. R2 discloses a subtractor + quantizer + local decoder/predictor, but the predictor (32) is again inside a feedback loop driven by the reconstructed adder output, and the invention's thrust (variable-length coding / buffer-overflow control / adaptive quantization) is orthogonal to claim 1. The "no feedback loop" prediction means is absent, so no independent claim is anticipated. Relevant as §103 background showing the conventional closed-loop ADPCM encoder the patent set out to improve.
• Claims it is "relevant to": 2 (encoder skeleton), and the Z_j (feedback) portion of claims 6, 7, 9.

R3 — US 4,144,543 (Nippon Electric Co., Ltd.)

• Citation: US 4,144,543, "Predictive codec capable of selecting one of at least three prediction signals in two steps," assignee Nippon Electric Co., Ltd.; priority 1976-12-16, granted 1979-03-13.
• Predates priority? Granted after the 1978-10-05 priority, so it is not §102(a)/(b) art by grant date. It can only be prior art under §102(e) if its U.S. filing date (≈1977, on a 1976-12-16 JP priority) precedes 1978-10-05 — which it appears to. I flag this as the one reference whose status as art depends on the §102(e) filing-date rule; verify against the U.S. filing date in the file wrapper.
• Description: A predictive codec that produces at least three prediction signals from the decoder local signal, selects the best two by an encoder comparator over a block, then switches between them at each sample. The prediction error is produced from the autocorrelated/input signal and a switched prediction signal, and the local signal is formed from the error plus the switched signal. Decoder mirrors with decoder prediction means driven by the decoder local signal.
• Anticipation assessment: Does not anticipate any claim of 4,317,208. R3 is fundamentally a multi-predictor selection / switching scheme whose predictors are fed by the decoder local signal (feedback), not by the quantizer output, and it has no "no-feedback, E_j-driven" prediction means. It does not disclose the adaptive coefficient recursions of claims 4/5 (B_i^j) or 9–12 (Y_j/Z_j split). No full-element read on any claim.
• Claims it is "relevant to": conceptually adjacent to the two-predictor concept of claims 6–8 (the idea of combining more than one prediction source), but it discloses selection/switching rather than the claimed Y_j + Z_j sum, and its predictors lack the required zero-based/no-feedback character — so it is §103 material, not §102.

R4 — IBM Technical Disclosure Bulletin, "Adaptive Rate Delta Modulator" (Apr. 1973)

• Citation: IBM Technical Disclosure Bulletin, vol. 15, no. 11, April 1973, pp. 3338–3341, "Adaptive Rate Delta Modulator." (Cited by the examiner as the sole NPL of record.)
• Predates priority? Yes (April 1973 printed publication ⇒ §102(b)).
• Description: As the title indicates, a delta modulator with adaptive-rate step control. Delta modulators are, by construction, feedback (integrator-loop) encoders: the transmitter integrates the transmitted sign stream and compares it with the input.
• Anticipation assessment: Does not anticipate any claim of 4,317,208. A feedback delta modulator lacks the ADPCM subtractor/quantizer structure and the no-feedback, quantizer-output-driven prediction means. It cannot read on claim 1/2/3/6/7/8. It is cited as general background on adaptive step-size control.
• Caveat: I was unable to retrieve the article's full text within this session; the assessment above rests on its title, its classification as an adaptive-rate delta modulator, and the well-established topology of delta modulators. Treat this as medium confidence on the specific content, high confidence that it is non-anticipatory regardless.

R5 — US 4,107,610 (Dacom, Inc.) — family citation, not an examiner citation

• Citation: US 4,107,610, "Data transmission system using a sequential approximation encoding and decoding technique," assignee Dacom, Inc.; priority 1973-03-19, granted 1978-08-15.
• Description: A sequential-approximation ("SAED") encoder/decoder for analog data (video/fax): comparator means develop an error signal from the input and a feedback signal; a detector develops the encoded signal; an adaptive RC filter in the decoder reconstructs the analog signal. The disclosure touts insensitivity to transmission errors and mentions adaptive filtering.
• Anticipation assessment: Does not anticipate any claim of 4,317,208. This is a feedback-comparator, sequential-approximation coder with an analog adaptive RC filter decoder — not an ADPCM system with a quantizer-output-driven adaptive predictor and a coefficient recursion. Note the provenance: R5 appears in the "Families Citing this family" list (i.e., cited during prosecution of a counterpart), not in the U.S. examiner's citation list, so it is of lower evidentiary weight against the U.S. claims. It is relevant only as generic §103 background on error-robust adaptive coding.

R6 — JP S53-46209 A (Matsushita) — family citation, not an examiner citation

• Citation: Japanese Unexamined Application JP S53-46209 A, "Sectional DPCM system," Matsushita Electric Industrial Co., Ltd.; filed 1976-10-08, published 1978-04-25.
• Anticipation assessment: Does not anticipate any claim of 4,317,208. A "sectional DPCM" scheme is a band/segment-partitioned DPCM, not a no-feedback residual-driven adaptive predictor. It is at most §103 background and, being a foreign publication cited only in a family member, carries reduced weight. (I did not retrieve the full Japanese text; treat the specific-content characterization as medium confidence, the non-anticipatory conclusion as high confidence.)

Non-patent literature discussed in the specification (not citations of record)

These are named in the 4,317,208 description but were not listed as references cited against the claims. I include them because they are the strongest background art and matter for a complete §102/§103 picture:

• D. L. Cohn et al., "The Residual Encoder — An Improved ADPCM System for Speech Digitization," IEEE Trans. Commun., COM-23(9), Sept. 1975, pp. 935–941. This is the closest art conceptually: the specification itself says the conventional predictor "may be composed of the type shown in FIG. 1 on page 936" of Cohn. A "residual encoder" is a residual/error-driven ADPCM design. It is the reference most likely to be argued against claims 1/2/3, and I recommend obtaining the paper to check precisely which signal drives Cohn's prediction and whether it is loop-free. I could not verify its exact predictor topology in this session.
• P. Cummiskey et al., "Adaptive Quantization in Differential PCM Coding of Speech," BSTJ 52(7), Sept. 1973, pp. 1105–1118. Cited in the specification as the source for implementing quantizer 20. Anticipates nothing (teaches only the quantizer).
• D. L. Duttweiler, "A Twelve-Channel Digital Echo Canceler," IEEE Trans. Commun., COM-26(5), May 1978, pp. 647–653. Cited in the specification as the source of the FIG. 4 adaptive-predictor structure (delay taps, convolution multipliers, coefficient-correcting means, summation accumulator) used for predictors 50/150. Relevant only to the implementation of the predictor, not to the no-feedback limitation.

4. Element-mapping summary

The decisive gap in every reference is identical: none discloses a feedback-loop-free predictor that derives the predicted value from the quantized error signal E_j. That is the whole point of the Araseki/Ochiai claims (see the specification's statement that predictions come "from the output signal E_j of the quantizer, not from X̂_j," in a structure that "avoids the adoption of a closed circuit in both the transmitter and receiver").

5. Bottom line

1. No reference of record — R1 through R6 — anticipates any of claims 1–12 under 35 U.S.C. §102. Each U.S. patent citation (R1, R2, R3) discloses at most a conventional closed-loop ADPCM/DPCM encoder (subtractor + quantizer + feedback predictor), which is exactly the architecture the patent distinguishes. Anticipation requires all elements in one reference; the "prediction means … having … no feedback loop … receiving quantized output signals" element is missing from all of them.
2. The strongest non-anticipatory (i.e., §103) exposure is the Z_j/feedback predictor sub-element of claims 6–12, which R1 (feedback filter 102), R2 (predictive filter 32) and R3 (switching predictors) all show in general form. The Y_j/zero-based predictor and the specific coefficient recursions (claims 4, 5, 9, 11, 12) are not shown by any of them.
3. The citation that most needs a closer look is the Cohn et al. "Residual Encoder" paper (1975) — discussed but not cited of record — because a residual/error-driven predictor is the nearest concept to the claimed no-feedback predictor. That, not R1–R4, would be the reference to pull in full before concluding on §102 for claims 1–3.
4. Tooling caveats, stated plainly: (a) I could not open the USPTO PatentCenter image/ifw for 4,317,208 directly in this session, so the citation list is taken from indexed USPTO-derived records; (b) I could not retrieve the IBM TDB article's text (R4) or the Japanese text of R6, so those two descriptions are inferential; (c) R3's status as §102(e) art should be confirmed against its U.S. filing date in the file wrapper. I did not fabricate any reference content, and where the record was silent or unreachable I say so rather than filling the gap.

If you want, I can next attempt the specific claim-by-claim §103 combination (e.g., R2 in view of the Cohn 1975 paper) or try again to pull the Cohn paper's predictor diagram to nail down the §102 question on claims 1–3.

I'll ground this analysis in the actual prior-art references of record before reasoning about motivation to combine.

Let me dig into the specific disclosures of the key references and the non-patent literature the patent itself discusses.

Obviousness Analysis — U.S. Patent No. 4,317,208 (ADPCM System for Speech or Like Signals)

Note on scope. I am building on the claim summary and prior-art inventory already generated (claims 1–3 = zero-based/open-loop predictor; claims 6–8 = dual-predictor "second embodiment"; claims 4–5, 9–12 = adaptation recursions and parameter ranges). Below I do not repeat those descriptions; I use the reference set of record on the Google Patents page as my universe of prior art and analyze what combinations a POSITA would have made. Where I could not retrieve a reference's full text, I say so rather than inventing a disclosure.

1. Legal framework and assumptions

• Governing statute: The application was filed 1979-09-28 with a 1978-10-05 JP priority. It is therefore governed by pre-AIA 35 U.S.C. § 103(a) (Graham v. John Deere; KSR Int'l v. Teleflex, 550 U.S. 398 (2007)).
• Critical date: 1978-10-05 (foreign priority). References must pre-date it under § 102(a)/(b), or qualify under § 102(e) (U.S. patents/applications by another filed before the invention date).
• PHOSITA: A person with a B.S. in electrical engineering and 2–3 years in digital speech/data transmission, or an M.S. with equivalent graduate exposure — familiar with DPCM/ADPCM (H03M 3/04, H03M 3/042), adaptive quantization (Cummiskey-style), backward-adaptive prediction, LMS/stochastic-gradient coefficient updating, first- and second-order predictors, and predictor stability criteria (poles inside the unit circle).
• Claim-construction caveat (affects everything below). Claim 1 recites a prediction means "having … no feedback loop" that receives the quantizer output. There is still a physical path predictor → subtractor → quantizer → predictor. I read "no feedback loop" in the specification's intended sense — the predictor is not fed by the reconstructed signal X̂ (there is no adder inside the prediction loop that would let an unrecovered channel error circulate). If "no feedback loop" were read literally/broadly, the claim would arguably not cover its own FIG. 1A/FIG. 2 embodiment. I analyze under the intended (narrower, meaningful) reading, because that is the only reading in which the limitation has prior-art significance.

2. What each reference of record actually teaches (as far as I could verify)

Key takeaway: Every patent reference of record teaches the reconstruction-driven feedback predictor — the very structure the patent's FIG. 1A admits. None of the cited references teaches the claimed feature of driving a no-feedback predictor from the quantized prediction error E_j. That is the entire § 103 battleground.

3. The gap that any § 103 ground must bridge

The invention's independent claims 1–3 reduce to a single departure from the admitted FIG. 1A art:

• Admitted art (FIG. 1A; Cohn; US 4,093,962; US 4,144,543): X̂j = Σ A_i^j · **X̄{j−i}** (past reconstructed samples) — closed loop containing the adder.
• Claimed: X̂j = Σ B_i^j · **E{j−i}** (past quantized prediction errors) — no reconstruction feedback.

Consequences a POSITA would recognize: the receiver becomes X̄_j = (1 + Σ B_i z^{−i}) E_j, i.e., a pure feed-forward filter of the codec output. Channel errors can no longer circulate; instability from out-of-range coefficients is structurally eliminated (but quantizer noise is now inside the prediction recursion, and prediction gain may suffer).

So the obviousness question is narrow and specific: was it obvious, as of 1978-10-05, to replace the reconstruction-signal predictor input with the quantized-error input?

4. Proposed § 103 grounds

Ground A — Claims 1, 2, 3: Cohn in view of Duttweiler

Combination: Cohn (ADPCM with a backward-adaptive predictor, adaptive quantizer, engineered for channel-error robustness) + Duttweiler's adaptive transversal filter (delay taps, convolution multipliers with per-tap coefficient correctors, summation accumulator) as the predictor implementation — plus the admitted FIG. 1A architecture.

Motivation: (i) The problem of decoder instability/oscillation after transmission errors in feedback-predictor ADPCM is squarely at issue in this field and is the stated object of the patent; Cohn already frames the design goal as channel-error robustness. (ii) A POSITA facing that problem is led to the only alternative signal available at both ends — the quantized error E_j already present at the predictor's doorstep — and to an FIR (taps + multipliers + accumulator) realization, which Duttweiler supplies ready-made. Reasonable expectation of success: linear prediction from past samples is standard; the patent's own spec works the transfer function out algebraically (B(z)-domain), which is routine design math.

Weakness of this ground (important): Duttweiler is an echo canceler, and nothing in Cohn directs the POSITA away from the reconstruction signal or toward the residual. The move is more "a known design alternative" than "a teaching." Expect a strong non-obviousness rebuttal (see §6). This ground is arguable but not clean.

Ground B — Claims 1, 2, 3: US 4,093,962 (or Cohn) in view of US 4,071,842

Combination: US 4,093,962 (subtractor→quantizer→local demodulator with feedback predictor; buffer-adaptive quantization) + US 4,071,842's error-feedback coder teaching that a coder may use a feedback network "maintained independent of the input signal" and driven from the quantizer output.

Motivation: US 4,071,842 establishes, in the coding arts, the design pattern of taking a quantizer-output-derived signal into a filter path that is decoupled from the input signal. A POSITA seeking to decouple the ADPCM predictor from the reconstructed signal would find direct support there, and would combine it with the speech-ADPCM frame of US 4,093,962.

Weakness: US 4,071,842's "error feedback" filters the quantization error q_n, not the prediction error E_j, and its purpose (noise shaping) differs from prediction. Bridging that difference requires an argument that conflating the two error signals was routine; that argument is soft. This ground is weaker than Ground A.

Ground C — Claims 4 and 5: leaky-LMS adaptation and numeric ranges

Claim 4 recites the adaptation B_i^{j+1} = (1−δ)B_i^j + g·E_{j−i}·E_j. This is a stochastic-gradient (LMS/sign-LMS) update with a leakage/de-emphasis factor (1−δ). The patent's own background concedes that the (1−δ) form was the known remedy for the transmission-error problem:

"…the following equation to correct the prediction coefficients is used: A_i^{j+1} = A_i^j·(1−δ) + g·F₁(X_{j−i})·F₂(E_j)… δ is a positive value much smaller than 1…"

Cohn's ADPCM and Duttweiler's coefficient correctors supply the "update from the product of a delayed sample and the current error" teaching. Under KSR, adapting a known technique (leaky LMS) to a new but analogous predictor input is within the level of ordinary skill, and the change is a predictable use of a known algorithm.

Claim 5 (M ≈ 7; δ ≈ 2⁻⁶) is a numerical range. Under In re Aller / KSR, ranges discovered through routine experimentation and not shown to be critical are obvious. The specification itself calls δ ≈ 2⁻⁶ "a practical value" and M ≈ 7 a selected number — language that reads as optimization, not criticality. Claims 4–5 are the most vulnerable claims in the patent.

Ground D — Claims 6, 7, 8, 9, 10, 11, 12: dual-predictor embodiment

Combination: The Ground-A zero-based predictor + a conventional reconstruction-driven predictor (Cohn; US 4,093,962; US 4,144,543), combined by simple summation (Y_j + Z_j).

Motivation: US 4,144,543 shows the field already used multiple prediction signals and selected/summed among them to improve prediction. Adding a pole-based contribution (to recover the prediction gain the open-loop path gives up) to a zero-based contribution (to guarantee stability) is the kind of predictable combination of known elements KSR endorses, especially where the patent's own text calls the result "comparable to" the first embodiment and merely "favorable… since the number of prediction coefficients is reduced."

Claims 10 and 11 (0 ≤ A ≤ 0.9; M ≈ 3, A ≈ 0.9) are stability constraints. |A| < 1 for a first-order recursive predictor is textbook pole-location analysis; clamping to 0.9 with a stated allowance for no performance loss is routine margin-setting. Highly vulnerable to a KSR/Aller rejection.

Critical statutory caveat on Ground D: US 4,144,543 issued 1979-03-13, after the 1978-10-05 priority date. It can therefore only be § 102(e) art (pre-AIA), and — because it is commonly owned by NEC (the same assignee as the applicant here) — pre-AIA § 103(c) disqualifies § 102(e)/(f)/(g)-only art from use in an obviousness rejection. So Ground D cannot lawfully rest on US 4,144,543 unless common ownership did not exist at the time the invention was made. (By contrast, US 4,407,842 [4,071,842] and US 4,093,962 issued before the priority date and are § 102(a)/(b) art, so § 103(c) does not shield them.) I flag this because it materially weakens the strongest-looking "multiple prediction signals" combination. I have not independently verified US 4,144,543's exact U.S. filing date; if it post-dates 1978-10-05, it would not even be § 102(e) art.

5. Consolidated motivation-to-combine narrative

For a POSITA at 1978-10-05, the record supports these motivations:

1. A recognized, unsolved problem. ADPCM with a backward-adaptive feedback predictor is unstable after channel errors; the accepted mitigation (leakage δ) trades error recovery against prediction gain and is inadequate. The patent's own background — and, for the problem's persistence, the later NEC patent US 4,554,670, which cites US 4,317,208 and states "…there arises the problem of predictor stability… the oscillation … prevents precise prediction" (US4554670 PDF) — confirm the problem was real and known. (US 4,554,670 is post-date and cannot be prior art; I cite it only as evidence of the problem's persistence.)
2. Eliminating the loop is the natural remedial direction. If leakage is a partial fix, removing reconstruction feedback altogether is the next step, and the only signal available at both transmitter and receiver to drive the predictor without a reconstruction path is the quantized error E_j.
3. The implementation is off-the-shelf. Duttweiler's transversal filter (adopted by the patent itself) supplies taps, multipliers, coefficient correctors, and an accumulator with an update that correlates a tap signal with the current error — structurally identical to the claimed B_i machinery.
4. The adaptation law is known. Leaky LMS/stochastic-gradient updating with an update ∝ (delayed sample)·(current error) was conventional.

6. Non-obviousness rebuttals a challenger would have to overcome

These are substantial and, in my view, make claims 1–3 and 6–8 defensible:

1. Uniform contrary teaching. Every reference of record uses the reconstructed/local signal — the patent's FIG. 1A, Cohn, US 4,093,962, US 4,144,543, and US 4,071,842's predictive coder. The art was not drifting toward a residual-driven predictor; it was converging on feedback predictors with leakage. A "design alternative that nobody in the art took" is thinner obviousness evidence than KSR's "market demand / design incentive" scenarios.
2. Predictable performance penalty supports a teaching-away argument. Feeding quantizer noise into the prediction recursion is precisely the classic reason the art feeds back the reconstruction — a POSITA would expect degraded prediction gain and would not have a reasonable expectation that a no-feedback predictor could hit the claimed ≥10 dB/14 dB improvements. The specification's own concession that δ "degrad[es] the prediction performance" shows the field viewed these as competing trade-offs, not a free replacement.
3. The prior art cited by the examiner was directed elsewhere. The three U.S. patents of record map onto the transmitter side (quantization, buffer control, multi-signal selection), not onto the residual-driven-predictor insight.
4. Objective indicia (limited but real). The '208 patent (and its NEC siblings) are repeatedly treated in later literature as a distinct "zero-based"/residual-driven predictor lineage; later patents (US 4,593,398 to Northern Telecom; US 5,291,284 to British Telecom; US 4,554,670 to NEC) had to address predictor instability, indicating the art did not treat the open-loop solution as trivially available. Examiner allowance itself is weak but nonzero evidence.

Net: Claims 4–5 are weak on their own merits. Claims 1–3 and 6–8 turn on whether "use the quantized error as the predictor input instead of the reconstruction" is characterized as a mere substitution of a known element (KSR) or as a departure from a uniform, contrary art. I would expect a well-argued § 103 rejection to be close but likely unsuccessful against claims 1–3 as a combination; the strongest rejection would target claims 4–5 (and 9–12 as to the recursions) as obvious-over-optimization.

7. Bottom line

Overall: No single reference of record anticipates any claim. The most plausible § 103 attack is Cohn (channel-robust ADPCM) in view of Duttweiler's adaptive transversal filter for claims 1–3, and the same base in view of the admitted leaky-coefficient-update art for claims 4–5 and the recursions of 9/12. Claims 4, 5, 10, 11 (and the formula-only portions of 9/12) are the most likely to fall to an obviousness challenge; the structural "no feedback loop" limitations of claims 1–3 and 6–8 are the defensible core.

8. Confidence and limitations

• High confidence in the claim mapping and in the observation that all three cited U.S. patents use reconstruction-driven, closed-loop predictors (verified from FPO/US4071842, uspto.report/US4093962, and Google/US4144543).
• Lower confidence / unverified: (a) the internal structure of Cohn's predictor — I could not confirm whether it is residual-driven (I assume it is not, per the patent's own admission); (b) the contents of the IBM TDB "Adaptive Rate Delta Modulator" and of US 4,107,610 / JPS5346209A (not retrieved — do not attribute specific disclosures to them); (c) the exact U.S. filing date of US 4,144,543, which controls the § 102(e)/§ 103(c) analysis in Ground D.
• Flagged inconsistency with the earlier section: the prior summary notes Google's "Similar Documents" list contains US 4,354,273 (same title, NEC) and correctly declines to assume a family relationship. My searches additionally surfaced US 4,554,670 (Aiko et al., NEC, filed 1983), which expressly cites US 4,317,208 — that is a citation lineage, not evidence of common priority/family, and I have not treated 4,354,273 or 4,554,670 as family members.
• Not legal advice. This is a technical § 103 assessment against the references of record; a full validity opinion would require the complete texts of the uncited NPL and foreign references and verification of the § 103(c) common-ownership facts.