US 4385393

I'll run targeted searches for this patent number, including the litigation/docket angle you asked about.

Both searches returned empty. Let me try alternate query formulations.

Let me run one final check specifically on litigation/docket history.

Verification status up front

I ran targeted searches for the exact identifier 4385393 (and its punctuated forms) plus CAFC-2026/docket and litigation-specific queries. Results:

• Patent lookups: hit. Google Patents, USPTO.report, RPX/Insight, Justia, FreePatentsOnline, and the EPO Global Patent Index all returned material matching this exact number.
• "4385393" + CAFC 2026 docket: zero results. "4385393" + litigation/infringement: zero results.
• The only litigation-flavored appearance I found is a citation of US 4,385,393 as prior art in a later Sony reissue (US RE37,864 E), which is a prosecution citation, not a lawsuit.

I did not find any authoritative record of a Court of Appeals for the Federal Circuit 2026 docket involving this patent. Caveat: empty search returns are weak evidence of absence, not proof of it. But it is corroborated a second way — see the expiration note below, which makes 2026 appellate activity on this patent essentially impossible.

Bibliographic summary (US 4,385,393)

Note: one third-party reproduction (the Sony RE37,864 E front page) cites this patent as "Chauve et al., 375/30." I am flagging that literally rather than correcting it; every other source gives the inventor surname as Chaure.

Abstract (as published)

A coding process for adaptive prediction differential PCM. An error signal e_t is formed between a sample of signal y_t to be coded and a prediction signal p_t. The prediction signal p_t is formed from restored signals ȳ_t and ē_t by two linear filtering operations — the first on N successive samples of ȳ_t using coefficients A1_t…AN_t, the second on P successive samples of ē_t using coefficients B1_t…BP_t — with the coefficients sequentially adjusted at each time t so that the mean power of e_t is minimized. Instead of predicting from the restored signals only, the invention also uses the real signal y_t and the real error e_t. A linear filtering of y_t with coefficients A1_t…AN_t over N successive samples, and a linear filtering of e_t with coefficients B1_t…BP_t, are performed, and the resulting quantities are respectively weighted by two coefficients between 0 and 1 that are not simultaneously zero.

Plain-language overview of the independent claims

Claim 1 — the coding process (independent).

A method of ADPCM-type encoding that:

1. Forms an error signal e_t as the difference between an incoming sample y_t and a prediction p_t;
2. Quantizes and codes that error;
3. Reconstructs a "restored" (quantized) error ē_t from the quantized/coded value;
4. Adds ē_t back to p_t to form the restored signal ȳ_t;
5. Builds the prediction p_t from ȳ_t and ē_t with two linear filters — one over N samples of ȳ_t using coefficients A1_t…AN_t, one over P samples of ē_t using coefficients B1_t…BP_t — with the coefficients being sequentially re-adapted at each time t to minimize mean error power;
6. The inventive step: additionally shapes the quantization-noise spectrum so it runs "parallel" to the signal spectrum, by doing at least one of:
   • (a) filtering the real (unquantized) signal y_t with the same A coefficients to get py_t, then blending it with the restored-signal prediction py̅_t: pAR_t = γ_AR·py_t + (1 − γ_AR)·py̅_t, where γ_AR is an adjustable constant in [0, 1]; and/or
   • (b) filtering the real unquantized error e_t with the same B coefficients to get pe_t, then blending: pMA_t = γ_MA·pe_t + (1 − γ_MA)·pe̅_t, with γ_MA in [0, 1];
   • where γ_AR and γ_MA are not both zero (both zero = prior-art ADPCM with flat noise spectrum);
7. Finally adding pAR_t and pMA_t and delaying the sum by one sampling interval to produce p_t.

The claim is written to cover both "raw" samples and "derived samples" — the latter being orthogonalized (decorrelated) versions of the y_t samples.

Claim 3 — the coding apparatus (independent).

The hardware counterpart to claim 1. It recites, in order: a two-input subtractor (y_t and p_t → error e_t); a quantizer plus coder; circuitry to reconstruct a restored error ē_t from the quantized or coded signal; a two-input adder (ē_t + p_t → restored signal ȳ_t); and a predictor fed with ȳ_t and ē_t. The predictor contains the two adaptive linear filters (N multipliers + coefficient-formation circuits + N-input adder producing py̅_t; P multipliers + adder producing pe̅_t) with sequential coefficient adaptation. The shaping means in the predictor comprise (A) at least one of — (a) a y_t filter plus a first "algebraic circuit" that outputs pAR_t = γ_AR·py_t + (1 − γ_AR)·py̅_t, or (b) an unquantized-error e_t filter plus a second algebraic circuit outputting pMA_t = γ_MA·pe_t + (1 − γ_MA)·pe̅_t; (B) an adder combining pAR_t and pMA_t; and (C) a one-sample delay circuit whose output is the prediction signal p_t.

Dependent claims (for context, not independent):

• Claim 2 depends on claim 1 and adds the orthogonalization detail: the N samples of y_t (real and restored alike) are passed through a lattice filter having N−1 recursively adjusted coefficients R1_t…R(N−1)_t, yielding N orthogonal samples ε1_t…εN_t used as the "derivative samples" multiplied by A1_t…AN_t. (The claim text as reproduced is somewhat redundant/garbled in third-party renderings.)
• Claim 4 depends on claim 3 and adds the apparatus counterpart: a first orthogonalizing lattice circuit (reference 28a) ahead of the y_t filter in the predictor, and a second lattice circuit (36) ahead of the y̅_t filter, the second reusing the R coefficients "sampled" from the first.

Docket / litigation check (CAFC 2026)

• No CAFC 2026 docket, appeal, IPR, or district-court case was found referencing 4385393.
• Independently, the patent's anticipated expiration was 2001-04-10 and Google Patents records the status as "Expired – Lifetime." A patent that expired in 2001 cannot support a 2026 infringement suit; it could at most appear in 2026 as cited prior art in someone else's IPR or litigation. I found no such record, but I cannot rule it out from the searches performed.
• The only substantive third-party use of this patent that I could confirm is as a prior-art citation: it is listed on the front page of Sony's reissue patent US RE37,864 E ("Quantizing error reducer for audio signal," Akagiri et al., reissued Oct. 1, 2002) and appears under "Also published as" / citation lists on later ADPCM-family documents.

Discrepancies and uncertainties you should know about

1. Claim text vs. the description's stated invention. The authoritative Google Patents description defines the invention as a blend of real-signal and restored-signal prediction, with orthogonalization presented as an optional refinement (see the "As is known, it may be advantageous… orthogonalization" passage and dependent claims 2/4). The granted independent claims as reproduced by uspto.report and RPX, however, build orthogonalization into the base claim-1 and claim-3 filtering operations ("taking a sequence of N samples from a first sequence… and from a second sequence obtained by orthogonalizing…"). This suggests the granted claims may reflect a Certificate of Correction — USPTO.report notes "Please see images for: (Certificate of Correction)." I could not read the correction document itself, so treat the exact as-issued claim wording as medium confidence.
2. Assignee chain. Sources disagree on labeling: original assignee = Gouvernement de la République Française; the 1981-10-22 assignment is to L'Etat Français / Secrétaire d'Etat aux Postes et Télédiffusion / CNET; Google Patents lists current assignees as France Telecom R&D SA and Gouvernement de la République Française while explicitly warning its assignee list "may be inaccurate." I have not verified the assignment history against USPTO Assignment records directly.
3. Classification conflict. The granted face shows US classes 375/30, 332/11D, 375/26 (per the "Patent History" block), while other databases show 375/249, 341/143, 375/243, 375/245. Both appear in reputable reproductions; I did not resolve which is the authoritative post-reissue/reclassification set.
4. I could not reach the USPTO PatentCenter or any CAFC/CM-ECF docket page directly — the searches surfaced aggregators (uspto.report, Justia, RPX, Google Patents, EPO GPI) rather than primary USPTO or court records. For a filing/assignment/docket question that must be airtight, pull the PatentCenter "Assignment" and "Certificate of Correction" images and run a PACER/CM-ECF docket query, since I cannot confirm those directly.

Primary sources used: https://patents.google.com/patent/US4385393/en · https://uspto.report/patent/grant/4385393 · https://insight.rpxcorp.com/patent/[US4385393A](/patent/US4385393A) · https://patents.google.com/patent/FR2481026B1/en · https://patents.justia.com/patent/[7949650](/patent/7949650) · https://patentimages.storage.googleapis.com/eb/54/f6/02edf8fdee52e6/[USRE37864](/patent/USRE37864).pdf · http://data.epo.org/gpi/EP1186104A4

I'll search for litigation records involving US patent 4385393.

Litigation Search Results for U.S. Patent No. 4,385,393

No litigation involving U.S. Patent No. 4,385,393 could be identified.

I searched multiple sources, including general web queries targeting patent litigation databases (Unified Patents litigation portal, CourtListener, PACER-oriented indices, and Federal Circuit/opinion repositories) using the exact identifiers "4385393" and "4,385,393." No case, complaint, docket, judgment, or settlement referencing this specific patent number was returned.

Important caveats on the search

1. Results returned were for different patents. Every "‑393" hit surfaced by my searches involved unrelated patent numbers, not U.S. 4,385,393. These included:

   • U.S. 3,441,393 (W.R. Grace & Co. v. Pullman, Inc., 446 F. Supp. 771 (W.D. Okla. 1976))
   • U.S. 4,693,399 / 4,936,494 (Weatherchem Corp. v. J.L. Clark, Inc., 163 F.3d 1326 (Fed. Cir. 1998))
   • U.S. 10,407,393 (Eisai R&D Mgmt. v. Shilpa Medicare Ltd., 1:19-cv-19998 (D.N.J.))
   • RE41,393 (Bristol-Myers Squibb v. Dr. Reddy's Labs., No. 3:12-cv-07800 (D.N.J.))

   None of these is the patent at issue, and I am not treating any of them as relevant.

2. No positive confirmation of absence. My searches returned zero results rather than a definitive "no litigation" record from a litigation database. This is not the same as exhaustive confirmation. A complete answer would require direct queries to subscription services (Docket Navigator, Lex Machina, PACER's full docket index) and the Unified Patents litigation portal itself, which I was unable to retrieve results from.

3. Practical context (not a litigation finding). The patent's prosecution/publication record at Google Patents shows it was filed 1981-04-10, granted 1983-05-24, assigned to the French State entity (Centre National d'Études des Télécommunications) and later to France Telecom R&D / Gouvernement de la République Française, with anticipated expiration on 2001-04-10 and current legal status "Expired – Lifetime." Given the 2001 expiration, any infringement suit would necessarily predate that date. This context is consistent with, but does not by itself establish, the absence of litigation.

Bottom line

Based on the searches available to me, there is no known litigation involving U.S. Patent No. 4,385,393. I cannot confirm this with high confidence, because my searches yielded no results at all rather than a verified negative from a litigation database. If a definitive answer is required, the next step would be a direct docket search on PACER/Docket Navigator and a query of the Unified Patents litigation case list for the exact number.

Proceedings overview

Total AIA trial proceedings on US 4,385,393: 0. The canonical source for this analysis — the structured "PTAB proceedings on file" block in this prompt (USPTO Open Data Portal ingest) — returns no IPR, PGR, or CBM proceedings, so the breakdown is: active 0 / claims invalidated 0 / claims sustained 0 / settled 0 / institution denied 0. My own independent web searches (Google Patents, USPTO.report, FreePatentsOnline, Unified Patents portal, CAFC/PTAB decision coverage) likewise surfaced no petition, no institution decision, and no Federal Circuit appeal involving this patent. Bottom line for a defendant: there is no PTAB record to exploit — but you almost certainly do not need one, because the patent expired on 2001-04-10 and is not assertable today. (Note a name-collision trap: search results for "the '438 patent" in BTG Int'l Ltd. v. Amneal Pharm. and for JHU's U.S. 11,591,393 refer to different patents. Do not conflate them with US 4,385,393.)

No proceedings to report

There is no IPR20XX-XXXXX, PGR20XX-XXXXX, or CBM20XX-XXXXX number to list for US 4,385,393. Per the operating rules, I will not invent one. What follows is the reason the null result is credible — and why it is not the usual "absence of IPRs is a negative signal for a defendant" situation.

Strategic summary

Claim status. Because no AIA trial ever reached a Final Written Decision, no claim of US 4,385,393 has been canceled, confirmed, or construed by the PTAB. The claim set as granted on 1983-05-24 — a process claim directed to ADPCM coding with weighting of the filtered real signal y_t and real error e_t by regulatable coefficients γ_AR and γ_MA, plus the corresponding apparatus claim (predictor with the first and second filter/algebraic-circuit groupings, adder, and one-sample delay), as recited in the patent's own claim summary — stands unadjudicated, i.e. UNTESTED. Every claim is simultaneously "never canceled" and "never confirmed." There is no claim-level PTAB record of any kind.

Estoppel landscape. Section 315(e)(2) estoppel is inapplicable in the ordinary sense: there is no petitioner and no FWD, so no party is barred from raising any ground. A defendant today faces no § 315(e)(2) estoppel and would face no § 325(e)(2) estoppel — but this is cold comfort, because the far more fundamental bar is temporal, not procedural. The patent's term ran out on 2001-04-10 (20 years from the 1981-04-10 filing date, which exceeds the alternative 17-years-from-grant date of 2000-05-24; the Google Patents record shows "Anticipated expiration 2001-04-10" and legal status "Expired - Lifetime"). An expired patent cannot be infringed going forward. Any residual past-damages theory for conduct before 2001-04-10 is time-barred by the six-year limitation of 35 U.S.C. § 286 — that window closed in 2007. Practically, every prior-art ground remains "available" precisely because there is nothing left to defend against.

Pattern signals. None of the usual patterns are present. There is no serial petitioner, no defensive aggregator (Unified Patents' portal shows US 4,385,393 only as cited prior art against later patents, e.g. IP Reservoir/Exegy FPGA patents — never as a target). The patent owner (originally the French State / Centre National d'Études des Télécommunications, later appearing as France Télécom R&D SA) never had reason to appeal a PTAB decision, because none issued. The absence of IPRs here is not the tell-tale signal of a patent that "has never been tested" but is live and being asserted; it is the signal of a long-dead 1983 telecommunications patent that no one has any economic reason to challenge.

Recommended next steps

• If you have received a demand citing US 4,385,393: the first response is a term/expiry letter, not an IPR. Ask the asserting party to identify (a) the exact claim numbers asserted, (b) the accused product, and (c) the date of the allegedly infringing acts prior to 2001-04-10, and confirm the § 286 six-year damages window. A demand letter on an expired 1983 patent should be treated as a screening/mass-mail artifact rather than a genuine litigation threat; check whether the sender has any recorded assignment interest (the Google Patents "Current Assignee" field lists France Telecom R&D SA and the Gouvernement de la République Française, and that field carries an express disclaimer that it is neither analyzed nor warranted).
• Do not file an IPR. An IPR against an expired patent is generally permitted only for the limited purpose of challenging claims (no claim amendment is available), and it would burn § 315(b) time and fees to invalidate something that expired 25 years ago and cannot support a live infringement claim. There is no live assertion to defend against, so an IPR would be a solution in search of a problem.
• If a demand is coupled with an assertion of a later, related patent (e.g., an ADPCM/audio-codec patent that cites US 4,385,393 as prior art — the Exegy/IP Reservoir and EP 1 186 104 references above show this patent is heavily cited), then the real dispute is about that patent. US 4,385,393 then becomes prior-art ammunition for you, not a target: a 1980-priority French-origin disclosure of pole-zero/AMAR adaptive prediction with noise shaping is an excellent § 102/§ 103 reference against post-2000 codec claims. Verify its 1980-04-21 priority date and publication facts (FR 2481026, GB 2076240, JP 56-059647, DE 3115859) as the prior-art date anchor.
• Verification trail for the null result: USPTO Patent Center / PTAB E2E for application 06/253,159 and patent 4,385,393; and the Google Patents page at https://patents.google.com/patent/US4385393/en for the "Expired - Lifetime," "Anticipated expiration 2001-04-10," and assignment history entries. If you want a belt-and-suspenders confirmation, run a PTAB E2E party-name search on "Chaure," "Benveniste," "Gilloire," "Centre National d'Etudes des Telecommunications," and "France Telecom" — I found no such proceedings via public search, but E2E is the authoritative docket.

Fabrication guardrail: I found no proceeding number, panel, institution decision, FWD, settlement, or CAFC appeal for US 4,385,393, and I have cited none. The governing fact is that the USPTO ODP shows zero AIA trials on this patent, and the patent's expired status makes an AIA trial both pointless and, in practical terms, unavailable as a defensive tool.

I'll research the assignment record for US 4,385,393.

Assignment & Ownership Analysis — U.S. Patent No. 4,385,393

Data-retrieval caveat up front. My available tools could not render the USPTO Assignment Center (assignmentcenter.uspto.gov) or its indexed mirror (assignment.uspto.gov) for this patent number. Google Patents' legal-events timeline, Espacenet and the patent's own front-page data were retrievable; the Assignment Center's reel/frame numbers and correspondent-of-record fields were not. I therefore report the one assignment event that is documented in the record I can access (Google Patents legal events) and flag every field I could not verify, rather than reconstructing reel/frame values I never saw. No assignments have been fabricated below.

Inventors

Basis for the employer call: the three inventors are the assignors on the 1981-10-22 record conveying "ASSIGNMENT OF ASSIGNORS INTEREST" to L'État Français / CNET (Google Patents legal events). A simultaneous assignment of all co-inventors to a single national research centre is the standard signature of employee-inventor obligations, not of independent inventors. I could not independently verify each inventor's employment contract, so the employer column is an inference, not a documented fact.

Unusual patterns: None detectable, and no evidence of the tell-tale pattern this task asks about. There is no record of any inventor departing, nor any individual (as opposed to corporate) patent transfer, within 12 months of filing. The only conveyance is the inventors → employer assignment. Any "departure precedes fire-sale" narrative is unsupported here — and, as noted below, the assignee is a state entity that never underwent a fire-sale at all.

Original assignee

Named on the issued patent (front page): Gouvernement de la République Française (the French State).
Named in the operative inventor assignment (1981-10-22): L'État Français, représenté par le Secrétaire d'État aux Postes et à la Télédiffusion — Centre National d'Études des Télécommunications (CNET).

These are the same legal person recorded two ways: the French State, acting through the ministry responsible for posts and telecommunications, and specifically through its research centre CNET. This is a correspondence/designation difference, not a chain-of-title break — not a contradiction in the record, but worth flagging because a naïve screenshot comparison of "Original Assignee" vs. "the 1981 reassignment event" looks like two different owners when it is one.

Line of business / product practice: CNET was France's national telecommunications research laboratory, and the patent sits squarely in its ADPCM speech-coding programme (the specification cites CNET/telecom-standard work throughout and cross-references French application EN 80 0883, filed 1980-04-21). The inventions were directed at the then-active standardisation goal of moving telephone transmission from 64 kbit/s down to 32 kbit/s — i.e. the assignee was the entity defining and deploying the standard the claims serve. Whether a specific commercial product embodied these claims is not documented in the record I retrieved, so I cannot state "shipped a product" as a finding.

Current status:

• The Google Patents "Current Assignee" field lists France Telecom R&D SA alongside the Gouvernement de la République Française. This reflects the corporate succession of CNET into France Télécom's R&D arm (and ultimately the Orange group), i.e. a change of name / corporate reorganisation of the same state-derived entity, not a sale to a third party.
• Legal status: Expired – Lifetime, anticipated expiration 2001-04-10 (20 years from the 1981-04-10 filing). The patent has been in the public domain for ~25 years as of today.

Assignment timeline

Recorded assignment events retrievable for US 4,385,393: one (1) — the original inventor-to-employer assignment.

The single recorded assignment

• Executed: not surfaced. Google Patents reports the event at 1981-10-22 (a date six months after filing, consistent with a post-filing recordation of a filing-date assignment). Treat the 1981-10-22 date as the recorded event date, not a verified execution date.
• Recorded: 1981-10-22
• Reel/Frame: NOT RETRIEVED — the Assignment Center pages carrying this field could not be rendered by my tools. I will not invent a number.
• Conveyance: Assignment — USPTO's own extraction reads "ASSIGNMENT OF ASSIGNORS INTEREST."
• Assignor(s): BENVENISTE, Albert; CHAURE, Christian; GILLOIRE, André (all three co-inventors).
• Assignee: L'État Français, représenté par le Secrétaire d'État aux Postes et à la Télédiffusion — Centre National d'Études des Télécommunications (CNET).
• Correspondent of record: NOT RETRIEVED. Because this is the only event in the chain, the "repeat correspondent" signal cannot be evaluated at all (see Signal 3).
• Context: Original employee-inventor assignment to the entity that filed and prosecuted the application — i.e. the founding link of the chain, not a transfer. There is no acquisition, fire-sale, securitization, internal reorg, or transfer-to-asserter in this record.

No post-issuance assignments exist in the retrievable record. There is no conveyance to any LLC, holding company, or licensing vehicle; no security agreement; no merger; no licence record. Consistent with the patent's 2001 expiry and its "Expired – Lifetime" status, the chain of title appears to have rested with the French State / its telecom successor from filing to expiration.

Timeline diagram

timeline
    title Ownership of US 4385393
    1980 : French priority app EN 80 0883 filed
    1981 : US app 06 253 159 filed by French State
         : Inventors assign to French State CNET
    1983 : Patent granted
    2001 : Patent expired

NPE / troll-pattern signals

1. Shell-entity transfer — NOT PRESENT.
No assignment moves this patent to any "IP / Patents / Licensing / Holdings / Ventures" entity. The only conveyance in the record runs to a government research centre (1981-10-22), not away from an operating company to a licensing shell. No registered-agent address, single-member LLC, or product-free assignee appears anywhere in the retrievable chain.

2. Known asserter in the chain — NOT PRESENT.
Neither the original assignee (French State / CNET) nor the listed current assignee (France Telecom R&D SA / Gouvernement de la République Française) matches any of the enumerated asserters (Acacia, Marathon, Intellectual Ventures, IPNav, Wi-LAN, Mosaid/Conversant, Vringo, Pendrell, Innovatio, MPHJ, Lumen View, Round Rock, Document Generation Corp, Spangenberg entities) or any Unified Patents / RPX high-frequency-plaintiff listing surfaced in my searches. Search results for this patent returned unrelated "‑393" patents owned by other parties — none in this chain.

3. Repeat correspondent across the chain — UNCLEAR (unevaluable).
The correspondent-of-record field was not retrievable, and with only one recorded event there is no "chain" over which recurrence could be measured. I am recording this as unclear rather than negative: absence of evidence here is a data gap, not a finding.

4. Cascading transfers — NOT PRESENT.
One recorded assignment in ~20 years of life. There is no sequence of transfers through chained LLCs, in <24 months or otherwise, and no shared correspondent/principal structure to point at.

5. Pre-litigation transfer — NOT PRESENT.
No litigation naming this patent has been identified (see the previously generated litigation section). The only assignment (1981-10-22) predates grant by ~19 months and has no litigation within any relevant window. The patent expired 2001-04-10.

6. Bankruptcy fire-sale — NOT PRESENT.
The assignee is the French State and its telecom successor. There is no Chapter 7/11 proceeding, no Kodak/Nortel/Polaroid-style estate sale, and no auction of this patent in the record.

7. Privateering — NOT PRESENT.
No operating company transferred the patent to an NPE asserting on its behalf. The chain never leaves the state/telecom-successor family.

8. Defensive aggregator (anti-NPE) — NOT PRESENT.
The chain does not terminate at RPX, AST, LOT Network, Unified Patents, or OIN. It terminates at the original government/telecom owner via expiration. This is an inverse-of-privateering outcome by default, but it is not the defensive-aggregator signal the rubric defines, so I mark it not present rather than claiming a neutralisation event that did not occur.

Verdict

Insufficient data (only the original assignment is recorded)

The rubric's "Insufficient data" category is the literal fit: the retrievable record contains exactly one assignment — the 1981-10-22 inventors-to-French-State/CNET conveyance — and no post-issuance transfers, with reel/frame and correspondent fields unretrievable through my tools. That said, the available evidence affirmatively rebuts rather than merely fails to support an NPE reading: all eight signals are not present or, in the single instance of Signal 3, unclear only because the correspondent field could not be pulled. The chain runs from three CNET employee-inventors to the French State's telecommunications research centre (the entity driving the 64→32 kbit/s ADPCM standardisation the claims serve), never leaves the state/telecom-successor family, is never held by a licensing vehicle, is never litigated, and expired 2001-04-10. This is not a patent-troll chain; it is a government research portfolio that ran its full term.

Verification links:

• USPTO Assignment Center (search by 4385393): https://assignmentcenter.uspto.gov/
• Indexed mirror: https://assignment.uspto.gov/patent/index.html
• Source record relied on for the 1981-10-22 event: https://patents.google.com/patent/US4385393/en (Legal Events)

Recommended next step for a definitive reel/frame and correspondent capture: query the Assignment Center's underlying search API directly (or PACER-adjacent patent-assignment data services) for patent number 4385393; the web-indexed pages for this record did not render in my environment.

I'll search the USPTO and related patent databases for the specific patent 4385393 and its cited prior art references.

Let me get the front-page "References Cited" data and foreign citations for this specific patent.

Let me verify the specific cited references and their details.

Prior-Art Analysis — U.S. Patent No. 4,385,393

1. Confirmation of the target patent (USPTO record)

The number 4,385,393 resolves to a single, specific grant — there is no ambiguity and no similar number is being substituted:

Sources consulted: patents.google.com/patent/US4385393, uspto.report/patent/grant/4385393, insight.rpxcorp.com/patent/US4385393A, companyprofiles.justatic.com/patent/4385393, freepatentsonline.com class listing, and the FPO front-matter for US 4,841,571 and US 4,516,241 (which cite '393).

Important caveat on completeness: Google Patents' citation table for this patent was not fully retrievable in my searches (the page returned the specification text rather than a clean "Citations" table). The reference list below therefore rests on the front-matter data reproduced by uspto.report / Justatic-RPX ("Referenced Cited — U.S. Patent Documents"), which returned two U.S. patents. I flag this explicitly: the list is what the sources show, but I cannot certify it is exhaustive. A definitive check would be the printed front page of the grant (or PatentCenter) for US 4,385,393.

2. Patent citations of record on US 4,385,393

Foreign/other documents referred to in the specification (not necessarily "of record" on the front page):

• French Patent Application No. 80 0883 ("EN 80 0883"), filed April 21, 1980, titled (per the description) "Adaptive prediction circuit using a lattice filter and corresponding differential TCM coding or decoding apparatus." Cited in the specification as describing a "higher performance adaptive apparatus." Note that 1980-04-21 is also the listed priority date of '393, so this document sits in the same family/priority window — I could not resolve from the retrieved record whether it is the priority application itself or a same-day sibling. Treat with caution.

Distinction to keep clear: The long list of numbers appearing near this patent — US 4,437,087 (Petr), US 4,411,001, US 4,361,893, US 4,354,273, US 4,317,208 (Araseki), US 4,319,360, US 4,516,241, US 4,841,571, etc. — are forward citations ("Cited By"), i.e., later patents that cite '393, plus class-listing neighbors in FPO's 375/245 page. They are not prior art against '393 and are excluded from this analysis.

3. §102 analysis of each cited patent

3.1 US 4,141,034 — Netravali et al. (Feb. 20, 1979)

• Full citation: U.S. Patent 4,141,034, issued Feb. 20, 1979 (inventor "Netralavi et al." per the printed citation records; the industry-standard spelling is Netravali).

• Date/qualification: Issued Feb. 20, 1979 — more than one year before the 1980-04-21 priority date, so it is a §102(b) statutory-bar reference on any measure (one year before either the U.S. filing date of 1981-04-10 or the priority date of 1980-04-21).

• Brief description — IMPORTANT LIMITATION: In this session I was unable to retrieve the title, abstract, or claims of US 4,141,034 from the available sources. Every search returned the citation identifier only. I therefore will not fabricate its subject matter. What can be said responsibly is: (i) it is a single-examiner citation placed on the front page of an ADPCM noise-shaping patent; (ii) Netravali's body of work at Bell Laboratories at that date centered on predictive/differential coding with quantizer-in-loop feedback; and (iii) the examiner's use of it was evidently to show the generic ADPCM/DPCM encoder architecture (subtracter → quantizer → coder → inverse quantizer → adder → predictor) that appears in Figs. 1–2 of '393. Its actual disclosure must be verified from its own text before any anticipation position is asserted.

• Claims of '393 potentially anticipated: On the record available, none. A §102 anticipation requires a single reference disclosing every element of the claim. Even assuming US 4,141,034 shows a conventional DPCM/ADPCM loop with an adaptive predictor (i.e., the preamble of claim 1 and the corresponding structure in claim 3), it would not disclose:

  • the two-coefficient-set AMAR prediction (A₁…A_N on reconstructed signal samples and B₁…B_P on reconstructed error samples) — claim 1, second filtering operation;
  • operation (a): filtering the real, unquantized signal y_t with the same coefficients A and forming pAR_t = γ_AR·pȳ_t + (1−γ_AR)·py_t, γ_AR ∈ [0,1];
  • operation (b): filtering the real, unquantized error e_t with the same coefficients B and forming pMA_t = γ_MA·pē_t + (1−γ_MA)·pe_t, γ_MA ∈ [0,1], not both zero;
  • claims 3–4 apparatus equivalents and the lattice orthogonalizers (28a, 36) of claims 2 and 4.

  Accordingly, US 4,141,034 is properly characterized as background/§103 material, not §102 anticipation.

• Caveat: This conclusion is provisional because the reference's text was not verified.

3.2 US 4,319,082 — Gilloire et al. (Mar. 9, 1982)

• Full citation: U.S. Patent 4,319,082, Adaptive prediction differential-PCM transmission method and circuit using filtering by sub-bands and spectral analysis, issued Mar. 9, 1982 (Gilloire et al.). Same assignee family as '393, and André Gilloire is a common inventor — the examiner cited the applicants' own earlier U.S. patent.
• Date/qualification: Issued after the 1981-04-10 filing and after the 1980-04-21 priority date. It therefore cannot be §102(a) or §102(b) art by publication. Its only possible prior-art role is pre-AIA §102(e) (a U.S. patent granted on an application filed before the applicant's filing/invention date). Its U.S. filing date was not verified in this session; the citation on the '393 front page is consistent with a pre-April-1981 filing, but that must be confirmed.
• Brief description: An adaptive differential PCM transmission method and circuit in which adaptive quantization and adaptive prediction are combined, and in which the adaptive prediction is carried out in adjacent frequency bands on signals obtained by sub-band filtering and spectral analysis (per the abstract shown in the FPO class-245 listing and the EPO search report for EP 1 186 104, where it is cited as category "XA").
• Claims of '393 potentially anticipated:
  • Claims 1 and 3 (preamble/generic elements): US 4,319,082 appears to disclose an ADPCM encoder loop — error signal formed between input sample and prediction, quantization, coding, reconstructed error, reconstructed signal by addition, and an adaptive predictor whose coefficients are recursively adjusted to minimise prediction-error power. Those are the generic ADPCM elements recited in the preamble of claim 1 and in the corresponding passage of claim 3. If (and only if) it were §102(e) art, it could reach only those elements — and since anticipation requires all elements in one reference, it still cannot anticipate claims 1 or 3.
  • The distinguishing elements are absent: nothing in the retrieved description of US 4,319,082 indicates the γ_AR/γ_MA weighting or the parallel filtering of the unquantized y_t and e_t with the same A/B coefficients. That combination — the entire point of the '393 invention (see the specification's equations (14) and (17) and the EQU7/EQU8 discussion) — is not disclosed. It is also relevant that '393 itself distinguishes the block/sub-band approach: "this shaping of the quantization noise is carried out within the scope of a prediction procedure by linear filter with recursive readjustment and not within the scope of a coding system by sample blocks."
  • Claims 2 and 4: The lattice-filter orthogonalization (coefficients R₁…R_{N−1}; circuits 28a and 36) is not shown in US 4,319,082.
• Conclusion: US 4,319,082 is a §102(e) candidate for the generic ADPCM/adaptive-prediction environment only, and a natural §103 combination reference. It does not anticipate any of claims 1–4.

4. The highest-relevance prior art is actually the non-patent literature of record

Because both cited patents miss the noise-shaping feature, the genuinely most relevant prior art to the '393 claims is the printed-publication art cited in the specification's own background section (all §102(b) publications as of the 1980/1981 critical dates):

* Note the careful reading of '393: the orthogonalization ("derived samples") limitation appears in the granted claim 1 as well as claims 2 and 4 (see the RPX text of claim 1: "taking a sequence of N samples from a first sequence formed by said N samples of y_t and from a second sequence obtained by orthogonalizing said N samples of y_t"). This matters: it makes the lattice-filter art (Makhoul) more directly material to claim 1 than the published specification alone would suggest.

5. Bottom line

1. Only two U.S. patent citations appear on US 4,385,393 in the sources I could retrieve: US 4,141,034 (Netravali et al., Feb. 20, 1979 — §102(b)) and US 4,319,082 (Gilloire et al., Mar. 9, 1982 — at most a §102(e) reference as of its U.S. filing date, which I could not verify).
2. Neither cited patent anticipates any claim of '393 under 35 U.S.C. § 102. Both are relevant only to the generic ADPCM-in-loop elements of the claim 1 / claim 3 preambles. Neither discloses the two-weight (γ_AR, γ_MA) parallel filtering of the unquantized signal y_t and error e_t with the same A/B coefficient sets, which is the sole point of novelty recited in claim 1(a)–(b) and its claim 3 apparatus counterpart.
3. The genuinely most relevant prior art is the printed literature cited in the specification — above all Atal & Schroeder, IEEE Trans. ASSP, June 1979 and Berouti & Makhoul / Makhoul & Berouti (1978–79) for the noise-shaping concept, and Makhoul (1977–78) for the lattice-filter orthogonalization recited in claims 1, 2 and 4. These are §102(b) publications and the realistic §103 combination bases.
4. Explicit uncertainties I am not papering over: (a) I could not retrieve Google Patents' full "Citations" table for '393, so the citation list above may be incomplete; (b) I could not verify the title, abstract, or claims of US 4,141,034, so my §102 assessment of it is provisional; (c) I could not verify US 4,319,082's U.S. filing date, which is dispositive of its §102(e) status; (d) I could not resolve whether FR 80 0883 is the priority application for '393 or a separate sibling application.

Recommended next steps for a definitive answer: pull the printed front page (or PatentCenter "References Cited") of US 4,385,393; pull the full texts of US 4,141,034 and US 4,319,082 directly from USPTO PatentCenter; and obtain the bibliographic record for FR 80 0883 to settle its relationship to the '393 priority claim.

I'll ground the obviousness analysis in the actual prior-art references named on this patent's face and in its specification, so let me verify the two cited U.S. patents and the key noise-shaping literature.

Two useful corrections surfaced. Let me confirm the substance of the two front-page U.S. references and the key noise-shaping literature.

Let me verify two architecture points that materially affect the §103 reasoning: the standard ADPCM predictor structure, and the vintage of "error-feedback" noise shaping.

Obviousness Analysis — U.S. Patent No. 4,385,393 under 35 U.S.C. § 103

Technical/legal analysis, not a legal conclusion. Cross-references the earlier sections of this file rather than repeating them.

1. Governing framework and posture

Which § 103 applies. The application (US 06/253,159) was filed 1981-04-10, with foreign priority to 1980-04-21. That is decades before 2013-03-16, so pre-AIA § 103(a) governs — not the AIA first-inventor-to-file version. This matters because pre-AIA § 102(e), § 103(c) and the "before the invention" framing of the prior-art date are all live here.

Standard. Graham v. John Deere Co., 383 U.S. 1 (1966) (scope/content of art; differences; PHOSITA level; objective indicia) as qualified by KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007) (TSM is not the only route; "a combination of familiar elements according to known methods is likely to be obvious when it does no more than yield predictable results"; a known technique used to improve a similar device in the same way is obvious). Under MPEP 2144.05, varying a known parameter to a disclosed or obvious degree, and selecting from a disclosed range, is an obvious design choice; under MPEP 2144.04, statements in the specification of what was known are usable as applicant admissions.

Practical relevance. As established in the earlier sections, this patent expired 2001-04-10, its PTAB and docket record is empty, and its real-world use today is as § 102/§ 103 prior art against later codec claims. The analysis below therefore serves two purposes: (a) assessing the patent's own validity had it ever been tested, and (b) establishing precisely which claim elements are, and are not, strongly supported when the '393 is used as a validity reference against later patents. Note that the obviousness case assembled here is materially stronger than what a 1983 examiner could have made, because KSR post-dates the grant.

2. The prior-art set on this page

The page's "Prior Art" content has four parts: the keyword field ("signal, circuit, coefficients, samples, output"), the two cited US patents, the 148 citing documents (downstream art — not available against the '393), and the non-patent literature recited in the Description. I built the analysis from the cited patents and the recited NPL.

Two refinements to the earlier "Patent summary" section (flagging, not repeating). (i) The earlier table lists the second cited patent as "US 4,319,082 (Gilloire et al., Mar. 1982)"; the search confirms the correct attribution is Gilloire & Zurcher, with US filing 1979-12-14 and a 1978-12-28 priority — that filing date is what makes it available under pre-AIA § 102(e). It is not a § 102(a)/(b) reference, because it issued 1982-03-09, after the '393 filing. (ii) The lattice-predictor companion case described in the '393 Description as "French Patent Application EN 80 0883 filed on April 21st 1980" corresponds to Le Guyader et al. (US 4,389,726; DE 3115884 A1, filed 1981-04-21) — not to US 4,319,082. I note without auto-correcting that "FR 80 0883" and the priority "FR 80 08884" cited earlier are near-collisions; I have treated them as distinct documents.

3. What the '393 specification itself concedes (admissions)

Three passages do heavy lifting for the obviousness case, and they are applicant admissions under MPEP 2144.04:

1. *"In conventional ADPCM coding, the quantization noise spectrum is in a first approximation roughly flat, which sometimes has the effect of making it clearly perceptible in the frequency ranges where the speech signal to be coded has a low power. Therefore, various devices have been designed … for shaping the quantization noise spectrum so as to eliminate this disadvantage."* → The problem and the goal are admitted prior art.
2. "Such a filter does not make it possible to obtain a very good transmission quality …" / "the adaptation criterion is that the average power of the difference signal … is as low as possible" → the AMAR/pole-zero adaptive predictor is admitted conventional.
3. "As is known, it may be advantageous, in place of directly processing the samples y_t … to carry out beforehand an 'orthogonalization' of said samples … This orthogonalization can be obtained by recursive adjustment of an all-zero lattice filter …" → Claim 2's entire added subject matter is admitted prior art.

The stated point of novelty is therefore narrow and, importantly, is framed as a degree of shaping: γ_AR = γ_MA = 1 gives "noise Δy_t exactly the spectrum of the restored signal"; γ_AR = γ_MA = 0 "gives the noise Δy_t the flat spectrum, which corresponds to the absence of shaping encountered in the prior art"; and "a random intermediate choice is possible leading to a shaping by approximate parallelism."

4. Element-by-element mapping of the independent claims

Claim 1 (process) / Claim 3 (apparatus)

No single reference discloses all elements, so the proper framing is § 103, not § 102 — though Ground I below comes close to anticipation of the operative structure if one reads A&S 1979's feedback-loop implementation as the primary reference.

5. The grounds of rejection

Ground I (primary): Gilloire '082 / Gibson 1978 / AMAR-ADPCM art + Atal & Schroeder 1979 (optionally + Makhoul & Berouti 1979) → claims 1 and 3

The gap the combination fills. Backward-adaptive pole-zero (AMAR) prediction in an ADPCM loop, requiring no transmitted coefficients (Gilloire '082; Gibson 1978; Cohn & Melsa; Cairns et al.), plus noise shaping in which the quantizer-error spectrum is made to follow the signal spectrum by feeding the error through a filter built from the prediction coefficients themselves, with an adjustable factor γ ∈ [0,1] (A&S 1979; Makhoul & Berouti 1979).

Why a PHOSITA would have combined them.

1. Same field, same problem, same constraint. Both address telephone-band speech coding at reduced bit rate (32 kbit/s); Gilloire '082 explicitly targets the 32 kbit/s/64 kbit/s telephone problem; A&S 1979 targets audible quantizer noise in predictive coders for speech.
2. The motivation is in the prior art verbatim. Gilloire '082 states that noise "strictly correlated to the signal is … subjectively only very slightly disturbing." A&S 1979 supplies the mechanism and the psychology (masking): noise is objectionable precisely where the signal is weak — the '393's stated problem, word for word.
3. The γ parameter is the claimed γ. A&S 1979 discloses a shaping factor between 0 and 1 whose endpoints are (γ = 0) flat noise and (γ = 1) noise parallel to the signal spectrum. Claim 1 claims that same range with that same physical meaning, and the '393's own Table equates γ_AR = γ_MA = 0 with the prior art. Choosing an intermediate γ is optimisation of a disclosed parameter (MPEP 2144.05; In re Aller).
4. No new information is transmitted, so there is no reason not to combine. Because the coefficients A and B already exist in the backward-adaptive predictor and the extra filtered terms are computed in the encoder only (the decoder needs only the standard restored-signal predictor), the combination adds no side information — the very objection A&S-type block systems had to answer. The '393's equation (17) confirms this: the decoder remains a plain DPCM decoder and the shaping appears as NTF(z) = 1 + γ_AR·A(z) + γ_MA·B(z). This is the classical error/noise-feedback configuration, and feedback-loop filter synthesis was already published (Proc. IEEE, 1963, cited in A&S 1979).
5. Predictable result, reasonable expectation of success. Placing a known shaping filter in a known DPCM loop yields the known trade-off (shaped noise, slightly reduced segmental SNR). The '393 claims no unexpected result — it claims the spectrum is made "parallel," which is exactly what A&S 1979 set out to do and reported doing.

Claim 3 falls with claim 1: the apparatus is the same structure in block-diagram form (filter + algebraic circuit + adder + one-sample delay), and every block is disclosed in the cited art, including the delay (Netravali '034, element 204).

Ground II: Ground I + Makhoul lattice art → claims 2 and 4

Claim 2 adds only the orthogonalization the specification concedes was known, and Makhoul 1977/1978 and Makhoul & Viswanathan 1978 disclose adaptive lattice/all-zero-lattice orthogonalization with recursively adapted reflection coefficients. The motivation is stated in the lattice literature and repeated in the '393 itself: decorrelating the samples "improve[s] the effectiveness of the least squares estimation recursive algorithms." Claim 4's two-lattice arrangement (lattice 28a on y_t feeding the y_t filter; lattice 36 on ȳ_t driven by the same reflection coefficients) is the standard analysis-filter / replicated synthesis-filter pattern of LPC — the second lattice simply reuses coefficients already computed, i.e. duplication of a known element with a predictable benefit.

Ground III (supporting): Netravali '034 + Gilloire '082 + A&S 1979

Netravali '034 supplies the secondary structural elements the other references state less explicitly: (i) forming the coding error against the true sample, (ii) comparing a filtered version of the true signal against the reconstructed/predicted version as part of adaptation, (iii) the one-sample delay in that comparison, and (iv) applying one filter/predictor configuration to two parallel signals, which is the structural analogue of applying the same A/B coefficients to both ȳ_t/ē_t and y_t/e_t. Because it is § 102(b) art and is a patent, not an article, it also provides the documentary form a rejection would prefer.

6. Where the obviousness case is weakest — fair counterarguments

A rigorous § 103 memo must state these; they are also the points a patentee (or a party relying on the '393 as a validity reference) would press:

1. Encoder/decoder asymmetry. Claim 1 requires the encoder's predictor to use the unquantized y_t and e_t; a PHOSITA in 1980 might reasonably have doubted that a decoder could reproduce a prediction built from signals it does not possess. The answer (and the reason this is not fatal) is that the arrangement is a noise-feedback coder whose decoder is an ordinary DPCM decoder — the '393's own equation (17) and Fig. 2 structure demonstrate it, and the technique is old (feedback-quantizer filter synthesis, Proc. IEEE 1963). Only a reasonable expectation of success is required (Pfizer v. Apotex).
2. Forward vs. backward adaptation. The principal shaping references (A&S 1979; Makhoul & Berouti 1979) are block/forward-adaptive and need side information; the '393 is backward-adaptive. This is the strongest distinction — but Gilloire '082 and Gibson 1978 both criticise exactly that and supply the backward solution, so the combination requires no leap.
3. Two independent γ's, and the MA-side (γ_MA) term in particular. Using the unquantized prediction error e_t as a filtering input is the least conventional element. A&S 1979's two-γ weighting filter, however, already contemplates independent pole-side and zero-side shaping parameters, so mapping them onto the pole and zero sections of an AMAR predictor is routine.
4. Claim drafting. The "blend of real-filtered and restored-filtered outputs with complementary weights" is a formulation of a linear combiner; the underlying physics is a feedback loop, which was not new.
5. Common-ownership caveat on US 4,319,082. André Gilloire is an inventor both on '393 and on '082, and both appear to have been CNET/French-State filings. If the subject matter of '082 and the '393 were commonly owned at the time the '393 invention was made, pre-AIA § 103(c) could disqualify '082 as a § 103 basis entirely. I flag this as an unresolved issue rather than a conclusion (its retrospective application to a 1981 filing is itself moot), and note the easy mitigation: Ground I does not depend on '082 — Gibson 1978 (a § 102(b) printed publication located in the text of '082 itself) supplies the same backward-adaptive-prediction element free of any ownership issue.

7. Objective indicia

• The 148 citing documents and the fact that the '393 is repeatedly cited as background for later ADPCM/audio work are evidence of the field's direction, not of non-obviousness.
• Industry convergence. ITU-T G.721 (1984) adopted a backward-adaptive pole-zero predictor: a 2nd-order pole section a_i(k) driven by the reconstructed signal s_r(k−i) and a 6th-order zero section b_i(k) driven by the quantized difference d_q(k−i), with per-sample gradient updates (https://search.itu.int/history/HistoryDigitalCollectionDocLibrary/4.260.43.es.3014.pdf). That is the AMAR structure of the '393's equations (3) and (7)–(9) with N = 2, P = 6. Important caveat: G.721 is 1984, i.e. not prior art to the '393 — it is evidence that the art regarded this predictor as the settled design choice, and it tends to undercut any "long-felt need" or unexpected-result argument. It does not evidence adoption of the γ noise-shaping blend; I found no standard adopting that specific feedback arrangement, and I make no such claim.
• No contemporaneous evidence of unexpected results appears in the '393: the specification asserts only the expected parallelism of spectra.

8. Verdict

Cross-reference to an open item from the earlier section: the granted claim text may embody a Certificate of Correction, and the USPTO.report/RPX renderings appear to build orthogonalization into base claims 1 and 3. If that rendering is correct, claims 2 and 4 add nothing to claims 1 and 3 respectively, and the lattice art (Makhoul) becomes necessary to the independent claims rather than to the dependents. That does not change the verdicts above — it only moves the Makhoul citation one row up. Treat the exact as-issued wording as medium confidence until the correction document is read.

What to verify before relying on any of this: (i) the USPTO PatentCenter Certificate of Correction for 06/253,159; (ii) the exact US filing date of the Le Guyader companion case (US 4,389,726), to settle whether it is § 102(e) art; (iii) the assignment record for US 4,319,082, to test the pre-AIA § 103(c) common-ownership point; (iv) the printed texts of A&S 1979 and Makhoul & Berouti 1979, since the "γ-scaling inside the predictor loop" reading is the pivot of Ground I. Primary sources used here: https://patents.google.com/patent/[US4385393](/patent/US4385393)/en · https://uspto.report/patent/grant/4385393 · https://patents.google.com/patent/US4319082 · https://uspto.report/patent/grant/4141034 · https://patentimages.storage.googleapis.com/76/16/56/000f07f0e62813/US4319082.pdf · https://search.itu.int/history/HistoryDigitalCollectionDocLibrary/4.260.43.es.3014.pdf