US 6424635

I'll verify the current status of US 6,424,635 through targeted searches, then summarize.

US 6,424,635 — Verification note first

I searched for this exact number (6424635 / 6,424,635) and found no USPTO litigation record, no CAFC 2026 docket entry, and no 2026 appellate activity. Targeted searches for the number combined with "CAFC," "2026," "litigation," and "CourtListener" returned either empty result sets or unrelated patents (the well-known Genband v. Metaswitch case, 2:14-cv-33-JRG, asserted Nos. 6,791,971 / 6,885,658 / 6,934,279 / 7,047,561 / 7,184,427 / 7,990,984 / 7,995,589 — not 6,424,635). I cannot confirm any adjudication of the '635 patent, so treat "no litigation" as absence of evidence in my searches, not as a definitive negative.

Status per Google Patents: Expired – Lifetime; anticipated expiration 2018-11-10 (20 years from filing). No live patent rights remain.

Bibliographic summary

Abstract (as printed): An echo canceler including a processor to adjust a dynamic clipping amplitude threshold for use in determining whether or not to clip outgoing audio. The dynamic clipping amplitude threshold is increased when the ratio of residue echo over echo is decreased over a prescribed period of time. The residue echo is taken as a difference between a true echo and an estimated echo initially emulated by the echo canceler. Likewise, when this ratio is increased over the prescribed period of time, the dynamic clipping amplitude threshold is decreased.

Plain-language overview of the independent claims

A note on one recurring limitation. Nearly every independent claim recites the same unusual language: the threshold is based on ERLE, "which based on a new average signal value of a residue echo, the new average signal value of the residue echo being the ones complement of a number between zero and one times an average signal value of the residue echo plus a current absolute value of residue echo times the number." Read literally, this is the exponential-average recursion of specification equation (4): RES_ECHO_ave(new) = (1−β)·RES_ECHO_ave(old) + β·|RES_ECHO_current|, with 0 < β < 1.

Claim 1 (method). A method of making an echo canceler less likely to clip outgoing audio wrongly. Produce a clipping amplitude threshold that is dynamically adjusted based on ERLE, where ERLE is derived from the new average residue-echo value computed by the recursion above. Increase the threshold when the ERLE value goes up over a period of time.

Claim 7 (method). Same core as claim 1, but the threshold is produced by three explicit steps: (a) determine a minimum amplitude level for clipping outgoing audio, (b) determine the ERLE value, and (c) add ERLE to that minimum level.

Claim 11 (communication device). An apparatus: a substrate, a processor on it, and a memory holding a nonlinear processor (NLP) module. When executed, the NLP module produces an amplitude threshold that reduces erroneous clipping of digitized audio as the ERLE value varies over time, with the threshold dynamically adjusted per the same residue-echo recursion.

Claim 18 (communication device). Reads as a standalone restatement of claim 11 with the negative-feedback behavior folded in: the NLP module controls the processor to decrease the clipping threshold when ERLE decreases over a prescribed period. (As printed, claims 11/12 and 18 overlap substantially — I flag this as an apparent claim-drafting redundancy rather than asserting a legal conclusion about it.)

Claim 19 (machine-readable medium). A storage medium carrying a program with a first routine that determines an ERLE signal value and a second routine that applies ERLE to a clipping amplitude threshold used to remove residue echo from digital signals output by the machine — the threshold again adjusted per the residue-echo recursion.

Claim 24 (machine-readable medium). Same as claim 19, plus the limitation that the machine includes a processor operating as an echo canceler using the NLP module.

Claim 25 (echo canceler). A three-block apparatus: (1) an adaptive filter producing an average amplitude level of an estimated echo from the incoming digital signals; (2) a summing circuit that subtracts that estimated-echo average from the average true echo of the outgoing digital signals; and (3) an NLP that clips the outgoing signals using a clipping threshold dynamically modified by adjusting the ERLE value, per the residue-echo recursion. (Note: the printed claim uses the apparent typo "ERLB"; claim 26's parallel text says "ERLE." I read this literally and do not auto-correct it.)

Claim 26 (echo canceler). Same three blocks as claim 25, plus additional circuitry to determine the average amplitude level of the true echo on the outgoing signals before they reach the summing circuit.

Claim 28 (communication device, means-plus-function). A device with (a) means for storing software that produces an adjustable clipping threshold for reducing erroneous clipping of digitized audio, the software including an NLP module, and (b) means for processing that software to alter the threshold based on ERLE — increasing it when ERLE rises and decreasing it when ERLE falls — with the new average residue-echo value computed as the old average residue echo (the complement of a 0–1 factor) plus the current absolute residue echo multiplied by a factor between 0 and 1.

Dependent claims in brief: 2 sets a ceiling (threshold capped at a predetermined maximum, ~−37 dBm0 per spec); 3 determines the average residue-echo value; 4–5 compare the average residue echo to the threshold and clip / refrain from clipping accordingly; 6 decreases the threshold as ERLE falls; 8–10 detail ERLE derivation (estimate echo, find true echo, subtract, clip; decibel-power levels; ratio + logarithm + multiplier); 12–17 add device-level variants and memory placement (on-chip vs. off-chip non-volatile); 20–23 add increasing/decreasing routines and echo-subtraction subroutines; 27 details the NLP's residue-echo/ERLE/clipping circuitry.

Context worth noting

• Prior art cited on the face (14 references), notably US 5,274,705 (Tellabs, "Nonlinear processor for an echo canceller and method"), US 4,679,230 (TRT, center clipper control), US 5,687,229 (Qualcomm), US 5,535,194 (Motorola, double-talk immunity), and US 5,668,865 (Lucent, E-side speech detector).
• Forward citations are extensive and indicate the family's technical footprint, including US 7,050,576 (Texas Instruments, "Double talk, NLP and comfort noise"), US 8,750,491 (Microsoft, adaptive NLP), US 9,042,535 (Cisco, echo control optimization), US 7,558,729 (Mindspeed, music detection for echo cancellation), US 7,693,276 / EP 1 691 490 (Mitel, residue-echo threshold), and a large recent Shure Acquisition Holdings cluster on acoustic echo cancellation and NLPs.
• Uncertainty I want to flag: I could not independently verify the 2004 certificate of correction's substance (it may have altered claim text or the "ERLB" typo), and the Google Patents "Current Assignee" label ("Genband US LLC") is inconsistent with the later-recorded Ribbon assignments. Anyone relying on ownership or exact claim scope should pull the USPTO PatentCenter documents and the certificate of correction directly.

I'll research litigation involving US patent 6424635 using the sources you indicated.

Initial searches returned nothing. Let me try the specific litigation databases and other phrasings.

Litigation Search Report — U.S. Patent No. 6,424,635

Bottom line

I found no litigation specifically asserting U.S. Patent No. 6,424,635. None of the sources searched (Unified Patents litigation portal, Law.com Radar, Docket Alarm, UniCourt, Justia-style dockets, IPWatchdog, Law360, SEC litigation disclosures, and general web search) returned a case in which this patent number appears as an asserted patent — in a complaint, counterclaim, IPR, or otherwise.

Per your instruction, I did not substitute results for similarly numbered patents. No case number, plaintiff, defendant, or outcome can be reported for 6,424,635 because no such record was retrieved.

What I could confirm about the patent itself (from the patent record)

These are ownership/status facts, not litigation:

The ownership trail includes a chain of security interests and assignments (One Equity Partners III, Comerica Bank, Silicon Valley Bank, Citizens Bank, N.A.) that are financing/reassignment records, not infringement suits. A certificate of correction was recorded 2004-06-15.

Related enforcement activity by the patent's owners (not this patent)

The patent's owners (Genband/Ribbon/Sonus) did run a well-documented patent campaign in this technology space, but the sources I retrieved do not identify 6,424,635 among the asserted patents:

• GENBAND US LLC v. Metaswitch Networks Corp. — E.D. Tex., Marshall Division, filed January 21, 2014. Jury verdict 2016-01-15 ($8.2M past royalties); final judgment 2018-03-22 ($8.9M plus costs). Federal Circuit appeal (Fed. Cir. 2017) vacated/remanded the permanent-injunction denial; injunction again denied on remand. Sources describe the asserted patents as relating to telephony gateways, call routing, softswitch overload, network management interfaces, and traffic classification/grouping — not echo cancellation.
• Sonus Networks, Inc. v. Metaswitch Networks Ltd. et al. — E.D. Tex., No. 2:18-cv-00058, filed 2018-03-08 (ten additional Sonus patents); dismissed with prejudice 2019-07-02 after a $63M global settlement of the patent and antitrust claims.
• Metaswitch Networks Ltd. v. Sonus Networks, Inc. (d/b/a Ribbon Communications Operating Company) — PTAB IPR2019-00051, filed 2018-10-17, terminated (settled) 2019-06-18 — this IPR is directed to a different patent (inventors Kaczmarczyk et al.), not 6,424,635.
• Charter Communications Operating, LLC v. Sonus Networks, Inc. and Ribbon Communications Operating Company, Inc. — N.Y. Sup. Ct. No. 653435/2022 (and a parallel Delaware action). This is Ribbon defending an indemnity claim tied to the earlier Sprint patent suit against Charter, not an assertion of 6,424,635.
• Morris Routing Technologies, LLC v. Ribbon Communications, Inc. — E.D. Tex. 4:25-cv-00852, filed 2025-08-07 (SR-MPLS/SRv6 routing patents) — unrelated subject matter.
• Far North Patents, LLC v. Ribbon Communications Inc. et al. — E.D. Tex. 4:19-cv-00945, filed 2019-12-26 — unrelated subject matter.

Caveats on this negative result

1. Tooling limits. I was not able to query PACER directly or run a full-text litigation database search keyed to the number 6,424,635 (e.g., Docket Navigator, Lex Machina, RPX, Unified Patents' per-patent page, or the CAFC docket search). My conclusion rests on general web and secondary-source searching plus the patent record page itself, which contains no litigation section.
2. Expiration constrains the window. The patent expired 2018-11-10 (20 years from filing), well before the June 2018 TC Heartland convenience notwithstanding. Any infringement action would have had to be filed before expiration, and damages would be limited by the 35 U.S.C. § 286 six-year lookback. This makes an unlisted, late-filed suit unlikely but not impossible.
3. A patent can be asserted without prominent coverage. A 2000s–2010s assertion of an echo-cancellation NLP patent could exist in a settlement-only case with little public reporting. I could not verify one either way.

Recommendation

To get a definitive answer, run the exact string "6,424,635" (and "6424635") in:

• Docket Navigator or Lex Machina (patent-number field, all district courts + PTAB + ITC),
• PACER Case Locator / CAFC docket search,
• Unified Patents per-patent litigation page and RPX / RPX Insight, and
• USPTO Patent Center "Litigation" tab and the PTAB API for IPR/PGR/CBM filings.

If you can share access to any of those, I can search and report the specific case metadata (plaintiff, defendant, jurisdiction, case number, filing date, status) directly.

I'll start by verifying the structured data against live sources.

Let me run a few more targeted checks before concluding.

A couple final checks on litigation/assertion history that might have spawned proceedings.

PTAB Proceedings — US 6,424,635 B1

"Adaptive nonlinear processor for echo cancellation" · Inventor: Wei-jei Song · Filed/Priority: 1998-11-10 · Granted: 2002-07-23 · Assignee chain: Northern Telecom → Nortel Networks → Genband US LLC → Ribbon Communications Operating Company, Inc. · Status: Expired – Lifetime (anticipated expiration 2018-11-10)

Proceedings overview

Total AIA trial proceedings on file with the USPTO Open Data Portal: zero (0). The structured "PTAB proceedings on file" block returns no AIA trials — so the breakdown is 0 active / 0 claims invalidated / 0 claims sustained / 0 settled / 0 institution denials, and the bottom-line defensive posture is the opposite of a "hardened" patent: no petitioner has ever tested these claims at the Board, so there is no IPR estoppel, no cancellation order, and every claim 1–28 remains presumptively valid — but the patent has also been expired since 2018-11-10, which means the practical fight is over past damages, not injunctive relief.

My independent web checks — queries on the patent number paired with "IPR," "PTAB," "inter partes review," petitioner names, and Unified Patents — surfaced no IPR, PGR, or CBM proceeding involving this patent. Consistent with the operating rule, I treat the ODP result as canonical and report no PTAB activity on file. I did not find proceedings on the USPTO PTAB Decisions page, PTAB E2E, or in CourtListener-indexed Federal Circuit activity. I cannot rule out an unfiled/very-recently filed petition that ODP has not yet ingested; that caveat is the only residual uncertainty.

Because there are no proceedings to enumerate, the per-proceeding template below is intentionally empty. Sections marked "N/A — no proceeding" are not omissions; they reflect the actual docket.

Proceedings overview (detail)

{No proceeding number} — N/A (no petitioner of record)

Not applicable. No petitioner has filed against US 6,424,635. There is no institution decision date, no panel, no grounds, no FWD, no settlement, and no appeal to report. Reporting any of these would require fabrication, which I will not do.

Family / related-proceeding note (not a proceeding on this patent)

• Family: US-only family (single application US 09/190,377; no foreign counterparts listed). Family ID 22701087.
• Forward citations of interest (these are patents citing the '435, not PTAB proceedings): US 7,050,576 B2 (Texas Instruments, "Double talk, NLP and comfort noise"), US 8,750,491 B2 (Microsoft, "Mitigation of echo in voice communication using echo detection and adaptive non-linear processor"), US 9,042,535 B2 (Cisco, "Echo control optimization"). None of these embodies an AIA challenge to the '435.
• No defensive aggregator in the chain. No Unified Patents, RPX, or similar entity appears in the assignment history. Ownership passed through ordinary corporate M&A (Nortel → Genband → Ribbon), i.e., an operating company, not a PAE.

Strategic summary

Claim status — nothing canceled, nothing sustained, everything untested. Because no IPR, PGR, or CBM was ever instituted, all claims 1–28 of US 6,424,635 stand unadjudicated and carry the statutory presumption of validity under 35 U.S.C. § 282. There are no sustained claims in the IPR sense (a claim is only "sustained" if a panel upheld it after trial) and no canceled claims. Independents 1, 7, 11, 18, 19, 24, 25, 26, and 28, plus dependents 2–6, 8–10, 12–17, 20–23, and 27, are all UNTESTED. For a defendant, that cuts both ways: you cannot point to a cancellation order, but you also have a clean slate on estoppel.

Estoppel landscape — there is none, and that is the single most important defensive fact. Section 315(e)(2) estoppel binds only a petitioner (and its privies/real parties in interest) that has been through an IPR that reached an FWD. With zero proceedings, no party is estopped from anything. Every invalidity ground — § 102 anticipation, § 103 obviousness, and (in litigation) § 112 written-description/enablement/definiteness — remains fully available in district court or at the ITC. There is also no § 315(b) one-year bar running against any defendant, because that clock starts only on service of a complaint alleging infringement and there is no indication of a live assertion. Practically: if you are sued, you retain the complete invalidity playbook, and you can also choose to file an IPR yourself without worrying about prior petitioners' estoppel shadow. One timing wrinkle: this is a pre-AIA patent (effective filing 1998-11-10), so Board and court challenges proceed under pre-AIA §§ 102/103, and PGR is unavailable (PGR applies only to first-inventor-to-file patents). CBM is also unavailable — an echo-canceler/nonlinear-processor patent is not a "covered business method" patent under AIA § 18(d)(1), which is confined to financial products or services.

Pattern signals — the notable pattern is silence. No repeat petitioner, no serial IPR practice on this patent, and no evidence the patent owner has ever pursued a PTAB appeal. That is unsurprising and largely explained by the calendar: the patent issued 2002-07-23 and expired 2018-11-10 at the 20-year statutory term. The modern AIA-trial era (IPR available from 2012-09-16) overlapped this patent's life by only about six years, and during that window there is no public record of a litigation campaign that would have triggered petitions. The most consequential fact for a defendant today is not PTAB-related at all: under 35 U.S.C. § 286, damages are recoverable only for infringement occurring within six years before the complaint is filed. The patent could only be infringed on or before 2018-11-10. As of today (2026-09-10), more than six years have elapsed since expiration, so a complaint filed now would appear to have no recoverable damages period. Have counsel confirm there is no tolling, fraudulent-concealment, or continuing-infringement theory (e.g., method claims argued to run to a later date) — but this is a far stronger defensive posture than any IPR outcome would give.

Two claim-drafting irregularities worth pricing into a defense. First, claim 4 and claim 6 state that clipping occurs "if the average signal value of the residue echo exceeds the dynamic clipping amplitude threshold," which is the inverse of the specification's stated operation at FIG. 6 block 630 ("If the RES_ECHO_ave does not exceed a clipping amplitude threshold ... the outgoing digital signal is clipped"). Second, the independent claims recite the residue-echo average as "the ones complement of a number between zero and one times an average signal value of the residue echo plus a current absolute value of residue echo times the number" — language that does not cleanly map onto specification equation (4), RES_ECHO_ave = RES_ECHO_ave + β × |RES_ECHO_current|. These read as drafting/transcription artifacts and may create claim-construction and § 112 exposure in a pending case; I flag them as observations to investigate against the certified copy and the 2004-06-15 Certificate of Correction, not as adjudicated holdings.

Recommended next steps

1. Treat "no PTAB activity" as the finding. State it plainly in any invalidity/risk memo: there is no FWD to cite, no cancellation to quote, and no institution decision to distinguish. Do not let a demand letter imply otherwise.
2. Run the primary-source confirmation yourself before relying on this. Verify at PTAB E2E (search by patent number 6,424,635), the USPTO PTAB Decisions page, and the USPTO Open Data Portal API for any petition filed in the last few weeks that ODP has not yet ingested. Cross-check litigation and any CAFC activity via CourtListener and PACER. The patent's authoritative full text and assignment history are at Google Patents US6424635B1.
3. Lead with the § 286 damages-window argument, not validity. If a demand letter cites claims 1–28 today, the threshold question is whether any actionable damages period remains given expiration on 2018-11-10. That argument is cheap, dispositive if correct, and independent of the merits.
4. Preserve the full invalidity record — nothing is estopped. Since no IPR exists, assemble § 102/§ 103 grounds under pre-AIA law with no estoppel constraint, and note that PGR and CBM are unavailable for this patent (pre-AIA; non-financial subject matter) — so IPR is the only available AIA trial vehicle if you choose to file one.
5. If you do file an IPR, note the expired-patent dynamics. The Board will institute on an expired patent, but claims cannot be meaningfully amended, so the proceeding becomes an all-or-nothing validity ruling useful mainly for collateral/defensive effect against past-damages theories. Weigh that against the cost given the § 286 point above.
6. If litigation is already pending, attack the claim/specification mismatch. Depose on the claim 4/claim 6 "exceeds" language versus FIG. 6 block 630, and on the "ones complement" residue-echo averaging recitation versus equation (4). Build indefiniteness and written-description positions in parallel with prior art; these defects were never vetted by an examiner post-issuance (the only post-grant event is the 2004-06-15 Certificate of Correction).
7. No trial-stage milestones to track. With zero active proceedings, there is no institution-decision deadline, no oral hearing date, and no statutory one-year FWD due date to calendar.

Confidence statement: I have high confidence that no AIA trial proceeding is on file for US 6,424,635 as of the most recent ODP ingest, corroborated by independent web searches that returned nothing on point. I have no information about any petition that may be pending ingestion, and I did not locate any Federal Circuit appeal. Where the record is silent, I have said so rather than inferring events.

I'll work from the authoritative patent text provided, then verify key facts with live sources.

Let me verify the Genband→Sonus/Ribbon chain and check for any litigation history.

Let me attempt to surface the assignment correspondents (attorney of record) for the key reels.

US 6424635 — Assignment & Ownership Reconstruction

Patent: US 6,424,635 B1 — “Adaptive nonlinear processor for echo cancellation”
Application: 09/190,377 · Filed: 1998‑11‑10 · Issued: 2002‑07‑23
Status (per Google Patents legal events): Expired – Lifetime; anticipated expiration 2018‑11‑10 (full 20‑year term, all maintenance fees paid)
Current assignee of record: Ribbon Communications Operating Company, Inc. (f/k/a GENBAND US LLC and Sonus Networks, Inc.)

Sourcing note on correspondents: The assignment records available to me (Google Patents legal‑events feed, which mirrors the USPTO assignment metadata) expose owner, conveyance text, effective date and reel/frame, but not the correspondent attorney of record. The USPTO Assignment Center (https://assignmentcenter.uspto.gov/) is an interactive query system and I could not retrieve the correspondent field for these reels from any accessible source. I therefore do not name correspondents anywhere below — doing so would be fabrication. This directly limits signal #3 in the NPE analysis. If you have Assignment Center access, pull each reel/frame in the timeline and read the "Correspondent" block to complete that column.

Inventors

Unusual patterns: None detected. This is a single‑inventor patent with a standard, promptly‑recorded inventor→employer assignment (executed ~6 weeks after filing, recorded 1998‑12‑30). There is no evidence in the record of the inventor departing the original assignee within 12 months, and no evidence of a portfolio fire‑sale tied to inventor departure.

Specification-based inference (flagged as inference, not record evidence): the only commercial platform named in the patent is the PASSPORT™ 4400 from “MICOM Communications Corporation of Simi Valley, Calif., a Northern Telecom Company,” and the invention sits squarely in the carrier/data echo‑cancellation space. That suggests the inventor was in the Northern Telecom / MICOM enterprise data‑networking group rather than the carrier CVAS group whose assets were later sold to GENBAND — but note the patent nonetheless transferred with the CVAS asset package in 2010 (see below). I cannot confirm the specific internal business unit from the record.

Original assignee

Entity on the issued patent: Nortel Networks Limited (Canada) — original assignee of record was Northern Telecom Limited, renamed through the 1999/2000 change‑of‑name chain described below.

• Primary line of business: Nortel was, until 2009, one of the world’s largest telecommunications equipment manufacturers — carrier VoIP/softswitch, optical, wireless (CDMA/LTE), and enterprise networking. In the 1997–2000 window relevant to this patent, it was a direct, high-volume competitor of Lucent, Alcatel, Siemens, Ericsson and Cisco.
• Did they ship a product embodying the claims? Yes — this is an operating‑company patent, not a paper asset. Nortel’s PASSPORT/data‑networking and later carrier gateways and softswitches used adaptive echo cancellation with a nonlinear processor; the specification expressly describes deployment in that product line. The claims (e.g., claim 1’s ERLE‑driven dynamic clipping threshold; claims 25–27’s “adaptive filter + summing circuit + nonlinear processor” architecture) map to signal‑processing firmware/DSP implementations in those platforms.
• Current status: Dissolved / wound down. Nortel filed Chapter 11 on 2009‑01‑14 in the U.S. Bankruptcy Court for the District of Delaware, Case No. 09‑10138 (jointly administered). It never reorganized; it auctioned its business units and wound down. The Carrier VoIP and Application Solutions (CVAS) business — which included this patent — was sold to GENBAND (stalking‑horse bid December 2009; combined U.S./Canadian bankruptcy court approval 2010‑03‑03; transaction closed 2010‑05‑28 at ~$182M net). Nortel’s separate, much larger patent portfolio went to the Rockstar Consortium in 2011; this patent did not go to Rockstar — it went with CVAS to GENBAND. Sources: Bankr. D. Del. 09‑10138, D.I. 5292 (CVAS sale order enforcement); GENBAND CVAS acquisition completion, 2010‑05‑31.

Assignment timeline

Chronological, all recorded entries. Note that the official record interleaves title transfers, change‑of‑name records, security interests (liens), lien releases, and a certificate of correction — only the title/name records move ownership.

• 1998‑12‑20 (executed) / recorded 1998‑12‑30 — Reel 009673/0820

  • Conveyance: Assignment of Assignors’ Interest
  • Assignor: Song, Wei‑Jei (individual inventor)
  • Assignee: Northern Telecom (Northern Telecom Limited, Canada)
  • Correspondent: not retrievable — see sourcing note above.
  • Context: Standard inventor‑to‑employer assignment of the application at filing.

• 1999‑04‑29 (executed) / recorded 1999‑12‑20 — Reel 010504/0903

  • Conveyance: Change of Name
  • Assignor: Northern Telecom Limited
  • Assignee: Nortel Networks Corporation
  • Correspondent: not retrievable.
  • Context: Internal corporate reorg / rebranding (Northern Telecom → Nortel Networks).

• 1999‑04‑29 (executed) / recorded 1999‑12‑23 — Reel 010567/0001

  • Conveyance: Change of Name
  • Assignor: Northern Telecom Limited
  • Assignee: Nortel Networks Corporation
  • Correspondent: not retrievable.
  • Context: Duplicate of the same‑day, same‑parties name change above, filed three days later under a second reel — a recording redundancy, not a second transfer.

• 2000‑08‑30 (executed) / recorded 2000‑08‑30 — Reel 011195/0706

  • Conveyance: Change of Name
  • Assignor: Nortel Networks Corporation
  • Assignee: Nortel Networks Limited
  • Correspondent: not retrievable.
  • Context: Internal reorg — the Canadian operating entity is renamed to the name that appears as assignee on the issued patent.

• 2004‑06‑15 — Certificate of Correction (event code CC; not an assignment, no reel/frame cited). No change in ownership.

• 2005‑12‑28 and 2009‑12‑22 — Maintenance fee payments (4th and 8th year). Not assignments; relevant only as evidence the patentee was maintaining the asset at full value through the 2009 bankruptcy.

• 2010‑05‑27 (executed) / recorded 2010‑06‑02 — Reel 024468/0507

  • Conveyance: Change of Name
  • Assignor: GENBAND Inc.
  • Assignee: GENBAND US LLC (Texas)
  • Correspondent: not retrievable.
  • Context: Acquirer’s own entity conversion — GENBAND Inc. to GENBAND US LLC — executed the day before the CVAS closing.

• 2010‑05‑28 (executed) / recorded 2010‑06‑18 — Reel 024555/0809

  • Conveyance: Patent Security Agreement (lien, not a title transfer)
  • Assignor: GENBAND US LLC
  • Assignee: One Equity Partners III, L.P., as Collateral Agent
  • Correspondent: not retrievable.
  • Context: Securitization — acquisition financing for the Nortel CVAS purchase (One Equity Partners was GENBAND’s private‑equity sponsor and co‑investor in the CVAS deal).

• 2010‑05‑27 (executed) / recorded 2010‑08‑25 — Reel 024879/0519

  • Conveyance: Assignment of Assignors’ Interest
  • Assignor: Nortel Networks Corporation
  • Assignee: GENBAND US LLC
  • Correspondent: not retrievable.
  • Context: Bankruptcy fire‑sale asset transfer — this is the operative conveyance of this patent as part of the Nortel CVAS sale (closing 2010‑05‑28; court approval 2010‑03‑03).

• 2010‑10‑28 (executed) / recorded 2010‑11‑09 — Reel 025333/0054

  • Conveyance: Security Agreement (lien)
  • Assignor: GENBAND US LLC
  • Assignee: Comerica Bank (Michigan)
  • Correspondent: not retrievable.
  • Context: Securitization — working‑capital/general credit facility collateral.

• 2010‑05‑27 (executed) / recorded 2012‑03‑30 — Reel 027992/0443

  • Conveyance: Corrective Assignment (to reel 024879/0519)
  • Assignor: Nortel Networks Limited and Nortel Networks Corporation
  • Assignee: GENBAND US LLC
  • Correspondent: not retrievable.
  • Context: Chain‑of‑title cleanup — the original 2010 assignment named only Nortel Networks Corporation; the corrective filing adds Nortel Networks Limited (the entity actually named on the face of the patent) as an assignor. This closes the gap between the patent’s named assignee and the seller of record. (Note: the 2012 record in the Google feed lists the assignee owner name as “CENBAND US LLC,” an OCR/typographic corruption of GENBAND — flagged per the “interpret literally, do not auto‑correct” rule, but it is the same entity.)

• 2012‑12‑19 (executed) / recorded 2014‑01‑10 — Reel 031968/0955

  • Conveyance: Release by Secured Party
  • Assignor: One Equity Partners III, L.P., as Collateral Agent
  • Assignee: GENBAND US LLC
  • Correspondent: not retrievable.
  • Context: Lien release following refinancing — closes out the 2010 One Equity security interest.

• 2014‑01‑23 — Maintenance fee payment (12th year). Not an assignment.

• 2016‑07‑01 (executed) / recorded 2016‑07‑06 — Reel 039269/0234

  • Conveyance: Patent Security Agreement (lien)
  • Assignor: GENBAND US LLC
  • Assignee: Silicon Valley Bank, as Administrative Agent
  • Correspondent: not retrievable.
  • Context: Securitization / refinancing — SVB replaces Comerica as secured lender.

• 2016‑07‑01 (executed) / recorded 2016‑07‑07 — Reel 039280/0467

  • Conveyance: Release and Reassignment of Patents
  • Assignor: Comerica Bank, as Agent
  • Assignee: GENBAND US LLC
  • Correspondent: not retrievable.
  • Context: Lien release on the same closing date as the new SVB facility.

• 2016‑07‑01 (executed) / recorded 2017‑01‑03 — Reel 041422/0080

  • Conveyance: Corrective Assignment (to reel 039269/0234; removes an erroneously listed patent number — US 6,381,239 — from the SVB collateral schedule)
  • Assignor: GENBAND US LLC
  • Assignee: Silicon Valley Bank, as Administrative Agent
  • Correspondent: not retrievable.
  • Context: Administrative correction to the security schedule; no ownership effect.

• 2017‑12‑21 (executed) / recorded 2017‑12‑29 — Reel 044986/0303

  • Conveyance: Termination and Release of Patent Security Agreement
  • Assignor: Silicon Valley Bank, as Administrative Agent
  • Assignee: GENBAND US LLC
  • Correspondent: not retrievable.
  • Context: Lien release in connection with the Sonus/GENBAND merger financing.

• 2017‑12‑29 (executed) / recorded 2018‑01‑02 — Reel 044978/0801

  • Conveyance: Security Interest (lien)
  • Assignor: GENBAND US LLC and Sonus Networks, Inc.
  • Assignee: Silicon Valley Bank, as Administrative Agent
  • Correspondent: not retrievable.
  • Context: Securitization — post‑merger collateral package covering the combined Ribbon group. Establishes that by end‑2017 the patent sits inside the merged Sonus/GENBAND → Ribbon Communications entity (merger completed 2017‑10‑27).

• 2020‑03‑03 (executed) / recorded 2020‑03‑03 — Reel 052076/0905

  • Conveyance: Security Interest
  • Assignor: Ribbon Communications Operating Company, Inc.
  • Assignee: Citizens Bank, N.A., as Administrative Agent
  • Correspondent: not retrievable.
  • Context: Securitization / refinancing — Citizens Bank replaces SVB as agent. The assignor name confirms GENBAND US LLC and Sonus Networks, Inc. had by then been consolidated/renamed into Ribbon Communications Operating Company, Inc. (designated “f/k/a GENBAND US LLC AND SONUS NETWORKS, INC.” in the recorded release at reel 058949/0497).

• 2020‑03‑03 (effective) / recorded 2021‑12‑06 — Reel 058949/0497

  • Conveyance: Termination and Release of Patent Security Agreement (re reel 044978/0801)
  • Assignor: Silicon Valley Bank, as Administrative Agent
  • Assignee: Ribbon Communications Operating Company, Inc. (f/k/a GENBAND US LLC and Sonus Networks, Inc.)
  • Correspondent: not retrievable.
  • Context: Lien release on the SVB facility.

• 2024‑06‑20 (executed) / recorded 2024‑06‑24 — Reel 067822/0433

  • Conveyance: Release by Secured Party
  • Assignor: Citizens Bank, N.A.
  • Assignee: Ribbon Communications Operating Company, Inc. (f/k/a GENBAND US LLC and Sonus Networks, Inc.)
  • Correspondent: not retrievable.
  • Context: Final lien release; the patent is now unencumbered and expired.

There is an assignment record for this patent — a long and well‑documented one. No gap requiring a “no records” finding.

Timeline diagram

timeline
    title Ownership of US 6424635
    1998 : Filed by Nortel
         : Inventor assigns to Northern Telecom
    1999 : Corporate name change to Nortel Networks
    2000 : Renamed Nortel Networks Limited
    2002 : Patent issued
    2009 : Nortel files Chapter 11
    2010 : CVAS unit sold to GENBAND
         : GENBAND Inc renamed GENBAND US LLC
         : Liens recorded to One Equity and Comerica
    2012 : Corrective assignment adds Nortel Networks Limited
    2017 : Sonus and GENBAND merge into Ribbon
    2018 : Patent reaches full term
    2020 : Lien recorded to Citizens Bank
    2024 : Citizens Bank lien released

NPE / troll‑pattern signals

1. Shell‑entity transfer — NOT PRESENT. Every assignee in the chain is an operating entity or a bank acting as collateral agent: Northern Telecom → Nortel Networks (two name changes, reels 010504/0903, 010567/0001, 011195/0706) → GENBAND US LLC (reel 024879/0519 plus corrective 027992/0443) → Ribbon Communications Operating Company, Inc. No assignee carries an “IP / Patents / Licensing / Holdings / Ventures” suffix; no registered‑agent‑service address appears; GENBAND/Texas and Ribbon/Massachusetts are corporate operating addresses. The 2010 and 2016–2020 “One Equity / Comerica / SVB / Citizens” entries are liens (reels 024555/0809, 025333/0054, 039269/0234, 044978/0801, 052076/0905), not divestitures to licensing vehicles.

2. Known asserter in the chain — NOT PRESENT. Neither the current assignee (Ribbon Communications Operating Company, Inc.) nor either of its predecessors (Sonus Networks, Inc.; GENBAND) appears on the standard public NPE rosters (Acacia, Marathon, IV, IPNav, Wi‑LAN, Mosaid/Conversant, Vringo, Pendrell, Innovatio, MPHJ, Lumen View, Round Rock, Spangenberg entities). Both GENBAND and Sonus were Nasdaq‑listed communications‑equipment vendors and are now consolidated into Ribbon Communications Inc. (Nasdaq: RBBN). I found no RPX/Unified Patents listing for any entity in this chain as a high‑frequency plaintiff.

3. Repeat correspondent across the chain — UNCLEAR / NOT ASSESSABLE. This is the one signal I cannot responsibly score: the correspondent field is not exposed by any source I could reach (see sourcing note). Practically, the pattern one would expect here is a small number of repeat law firms — one handling the 1999–2000 Nortel name changes, one handling the 2010 GENBAND acquisition and financing filings, and one handling the 2016–2020 SVB/Citizens lien filings — because that is how corporate IP‑recording counsel normally works on a 25‑year chain. But that is an expectation, not a finding, and I am not asserting it. To close this signal, read the correspondent block on reels 024879/0519, 027992/0443, 039269/0234 and 052076/0905 in Assignment Center. A single recurrence across the 2010 GENBAND cluster would still be ordinary deal counsel, not an NPE tell.

4. Cascading transfers through chained LLCs in <24 months — NOT PRESENT (for NPE purposes). There are dense clusters (four name/grant records 1998–2000; four GENBAND records within ~5 months in 2010) but no sequence of transfers through successively‑created holding LLCs, no shared‑correspondent‑address pattern observable, and no common‑principal LLC chaining. The 2010 cluster is an acquisition closing (assignment + entity conversion + acquisition financing lien recorded in a single month) — a normal M&A/security recording burst, not an assertion‑engineering cascade.

5. Pre‑litigation transfer — NOT PRESENT. I found no infringement litigation naming US 6,424,635. There is consequently no assignment dated within six months of a first suit, and no arrangement of the chain to set venue or standing.

6. Bankruptcy fire‑sale — PRESENT (but benign as to NPE risk). The operative transfer, reel 024879/0519 (executed 2010‑05‑27, recorded 2010‑08‑25), conveyed the patent out of Nortel Networks’ Chapter 11 (Bankr. D. Del., Case No. 09‑10138) as part of the CVAS business sale to GENBAND. Court approval 2010‑03‑03; close 2010‑05‑28. This is a genuine distressed sale — but the buyer was a strategic operating purchaser acquiring a business unit (products, customer contracts, and ~2,000 employees), not a patent‑only aggregator. That distinction is what keeps this from being a troll signal.

7. Privateering — NOT PRESENT. No evidence that Nortel, GENBAND, Sonus or Ribbon transferred the patent to an NPE to assert on its behalf. The only outbound transfers are to lenders as collateral, all subsequently released (reels 031968/0955, 039280/0467, 044986/0303, 058949/0497, 067822/0433).

8. Defensive aggregator — NOT PRESENT. The chain does not terminate at RPX, AST, LOT, Unified Patents, or OIN. It terminates at an operating vendor.

Additional neutralizing fact: the patent reached full term and expired 2018‑11‑10 (status “Expired – Lifetime”; 4th, 8th and 12th‑year maintenance fees all paid). Even if a litigation‑funding entity had wanted this asset, it has been unenforceable for roughly eight years as of today, which is consistent with the absence of any assertion record.

Verdict

Operating-company assertion — with the “assertion” prong explicitly unverified.

This is, on the record, an operating‑company‑owned patent, not an NPE asset. The chain is continuous and fully documented from the inventor’s 1998 assignment (reel 009673/0820) through Nortel’s 1999–2000 name changes (reels 010504/0903, 010567/0001, 011195/0706), the 2010 Chapter 11 asset sale to GENBAND (reel 024879/0519, corrected at reel 027992/0443), and the 2017 Sonus/GENBAND merger into today’s assignee, Ribbon Communications Operating Company, Inc. Every intermediate assignee is an operating communications‑equipment vendor or a bank acting as collateral agent, and every lien was released — there is no shell entity, no known NPE, and no cascading LLC chain. The only genuine “distress” signal, the Nortel bankruptcy fire‑sale (reel 024879/0519), ran to a strategic buyer of a whole business unit, so it does not read as privateering or aggregation; and with the patent expired 2018‑11‑10, no assertion is possible regardless.

Caveat on the verdict label: the verdict options require “is suing actual competitors” for the operating‑company‑assertion category, and I found no litigation naming this patent — so the accurate reading is operating‑company ownership, non‑asserting. I selected the operating‑company verdict over “defensive/non‑asserting” because the chain does not terminate at a defensive aggregator (RPX/AST/LOT/Unified/OIN), which is that category’s defining criterion.

Verify: USPTO Assignment Center · Google Patents legal events for US 6,424,635 · FreePatentsOnline record

I'll verify the patent record and each cited reference against live sources before building the anticipation mapping.

Let me verify the remaining face references.

I hit the search-step cap before completing live verification of every one of the 14 face references, so I'll flag below exactly which entries rest on the authoritative patent text + confirmed searches and which rest on the face record alone.

Prior-Art Review for Claim-Anticipation Purposes — US 6,424,635 B1

0. Verification note (what I searched, what it returned)

• The number 6424635 resolves, per the USPTO/Google Patents record, to US 6,424,635 B1, "Adaptive nonlinear processor for echo cancellation," Wei-jei Song, app. 09/190,377, filed 1998-11-10, granted 2002-07-23, Nortel Networks Limited. Live searches on the exact string returned the patent itself (Google Patents, FreePatentsOnline, Justia) — no substitute patent was returned and I have not auto-corrected the number. (US6424635B1; FPO; Justia)
• The face of the patent lists 14 cited references, all of which I reproduce below. I confirmed the disclosure content of US 4,064,379, US 4,679,230, US 4,751,730, US 4,998,241 (via its apparent EP 0 371 567 family), US 5,274,705, and US 5,351,291 (via its apparent EP 0 530 423 family) through live searches. The remaining entries (US 4,912,758, US 4,965,822, US 4,991,166, US 5,535,194, US 5,668,865, US 5,687,229, US 5,809,085, US 6,061,651) rest on the face citation data and the descriptions carried in the patent record, confirmed only partially.
• Contradiction check against the prior sections: none. My earlier "Obviousness" section and "Patent summary" section treated the same 14 references and the same supplementary Tellabs reference (US 6,198,819). One thing to state plainly, because it changes the framing: those sections analyzed § 103 combination grounds. This task is § 102 anticipation, which is a narrower question, and the answer is materially more negative (see § 3).

1. The § 102 standard, and the structural point that governs this whole review

For a reference to anticipate a claim under 35 U.S.C. § 102 (pre-AIA, which governs this 1998-filed patent), it must disclose every element of the claim, arranged as recited, in a single reference. Net MoneyIN, Inc. v. VeriSign, Inc., 545 F.3d 1359 (Fed. Cir. 2008); In re Paulsen, 30 F.3d 1475 (Fed. Cir. 1994). Missing even one element defeats anticipation. Two combinable references are a § 103 question, not a § 102 question. And because every dependent claim incorporates its parent, a reference cannot anticipate any of claims 2–6, 8–10, 12–17, 20–23, or 27 unless it first anticipates the independent claim they depend from.

Every independent claim of the '635 (1, 7, 11, 18, 19, 24, 25, 26, 28) requires both:

1. a clipping amplitude threshold that is dynamically adjusted based on an echo return loss enhancement (ERLE) signal value; and
2. the ERLE is "based on a new average signal value of a residue echo, the new average … being the ones complement of a number between zero and one times an average signal value of the residue echo plus a current absolute value of residue echo times the number" — i.e., the spec's Eq. (4) recursion RES_ECHO_ave(new) = (1−β)·RES_ECHO_ave(old) + β·|RES_ECHO_current|, 0<β<1.

(Interpreting literally and without auto-correction: claim 25 prints "ERLB," claims 26/1/7/28 print "ERLE"; claim 28's recursion is worded slightly differently — "an average signal value of the residue echo and a current absolute residue echo multiplying a number between 0 and 1.")

No reference cited on the '635 face discloses an ERLE-driven clipping threshold built on that recursive average of |residue echo|. Consequently, under § 102 taken reference-by-reference, none of the 14 face references anticipates any of claims 1–28. What follows maps each reference to the elements it does disclose, identifies the closest candidates, and flags the one non-face document that could support a genuine single-reference § 102(e) challenge.

2. Reference-by-reference: citation, dates, disclosure, and § 102 reach

2.1 US 4,064,379 — Horna / Communications Satellite Corporation

• Full citation: U.S. Patent No. 4,064,379, "Logarithmic echo canceller," granted Dec. 20, 1977 (filed June 11, 1976). (link)
• Description (confirmed): Digital echo canceller using logarithmic (A-law) coding; computes a pseudo-RMS / average value of speech samples in the X-register; adaptive center clipper with a clipping level "automatically adjust[ed] to the optimum value in accordance with the echo producing signal level," driven by a cross-correlation/RMS-derived bias. Analog subtraction of the simulated echo in the return path.
• § 102 reach: Discloses a center clipper with a variable bias and averaging of magnitude samples → touches the genus of claims 1/4/5 (clip based on a variable threshold; compare-and-clip). Does not disclose ERLE as the control variable, nor the "ones complement" recursion. Anticipates: none of claims 1–28. Best characterized as background art for the variable-center-clipper concept and as a § 103 secondary reference (already treated in the Obviousness section).

2.2 US 4,679,230 — Lassaux et al. / T.R.T.

• Full citation: U.S. Patent No. 4,679,230, "Echo canceller and center clipper control arrangement," granted July 7, 1987 (filed May 15, 1984). (link; EP family EP 0 164 159)
• Description (confirmed): Center clipper in the send path whose clipping threshold varies as an increasing function of the level N_AV; the clipper is activated by comparing the ratio N_AP/N_AV (post-subtractor level over pre-subtractor level) against a threshold; multiple non-zero clipping thresholds assigned by level range.
• § 102 reach: The N_AP/N_AV ratio is a cancellation-quality measure functionally akin to ERLE, and the "increasing function of level" is a variable-threshold teaching. But the '635 requires the threshold be adjusted based on an ERLE signal value derived from a recursive average of |residue echo|, which '230 does not describe. Anticipates: none of claims 1–28 as printed. This is the single most relevant face reference for the dynamic-threshold concept, and the lead reference in the § 103 grounds already set out.

2.3 US 4,751,730 — Galand et al. / IBM

• Full citation: U.S. Patent No. 4,751,730, "Process and system for improving echo cancellation within a transmission network," granted June 14, 1988 (filed Apr. 30, 1985). (link)
• Description (confirmed): Shortened echo-canceler filter G preceded by flat delay D; residual-echo clipping using an "accurate determination of a decision threshold" that is computed from the level of the residual echo (H1+H2 portion of the hybrid impulse response), to remove residual echo without clipping local speech.
• § 102 reach: Discloses the residue-echo level → clipping decision chain, i.e., the subordinate steps recited in claims 8, 15, 23, and 27 ("determining a true echo … subtracting the estimated echo … to produce the residue echo … clipping the residue echo"). But those are dependent limitations; the independent claims still require the ERLE recursion. Anticipates: none of claims 1–28. Strong § 103 partner for the residue-echo-derivation limitations.

2.4 US 4,912,758 — Arbel / IBM

• Full citation: U.S. Patent No. 4,912,758, "Full-duplex digital speakerphone," granted Mar. 27, 1990 (filed Oct. 26, 1988). (link)
• Description: Processor-based full-duplex speakerphone platform (acoustic echo path, DSP control).
• § 102 reach: Platform art only. Relevant to the substrate/processor/memory recitals of claims 11, 18, and the means-plus-function structure of claim 28, but discloses nothing of ERLE or the recursion. Anticipates: none.

2.5 US 4,965,822 — Williams / Videotelecom Corp.

• Full citation: U.S. Patent No. 4,965,822, "Full duplex speakerphone," granted Oct. 23, 1990 (filed Apr. 10, 1989). (link)
• Description: Full-duplex speakerphone hardware/firmware platform.
• § 102 reach: Same category as '758 — conventional platform art. Anticipates: none.

2.6 US 4,991,166 — Shure Brothers Incorporated

• Full citation: U.S. Patent No. 4,991,166, "Echo reduction circuit," granted Feb. 5, 1991 (filed Oct. 28, 1988). (link)
• Description: Echo-reduction circuit for audio/speakerphone applications.
• § 102 reach: Background art for echo reduction generally; discloses no ERLE-controlled threshold. Anticipates: none.

2.7 US 4,998,241 — Brox et al. / U.S. Philips Corporation ⚠️ closest face reference to claim 7

• Full citation: U.S. Patent No. 4,998,241, "Echo canceller," granted Mar. 5, 1991 (filed Dec. 1, 1988), U.S. Philips Corporation. (link; apparent EP family EP 0 371 567)
• Description (confirmed via the EP 0 371 567 family): A control unit computes, from mean values of the signals, a threshold that "bears a proportion to the mean value of the echo-cancelled signal e(k)," and controls both the adaptation rate and a center clipper 39, "in which the magnitude of the signal amplitudes to be suppressed is … controlled in dependence on the ratio of the threshold to the mean value of the echo-cancelled signal." Its claim 2 recites a threshold consisting of "the sum of a time-constant part and … the ratio … to an estimate of the total echo return loss," and claim 4 states the total echo return loss is the product of the echo return loss of the echo path and the echo return loss of the adaptive filter — i.e., explicitly ERL × ERLE.
• § 102 reach — the strongest but still incomplete single-reference case. '241 discloses (i) a threshold derived from the average/mean of the echo-cancelled (residue) signal, touching the "average signal value of the residue echo" element, and (ii) a time-constant part + a return-loss correction term, which maps structurally onto claim 7's "minimum amplitude level + ERLE." It is the only face reference that puts an echo-return-loss-enhancement-related quantity into the threshold arithmetic. However it does not disclose: the threshold being increased as ERLE increases over a period of time in the claimed closed-loop sense, nor the "ones complement … plus … current absolute value" recursion of Eq. (4). Anticipates: none of claims 1–28 as printed — but claim 3 and claim 9 (determining/using the average signal value in decibel terms) are the dependent limitations most exposed, and claim 7's additive threshold formula is the independent limitation most exposed. ⚠️ Caveat: I matched the U.S. text to the EP family because the U.S. specification body was not retrieved; confirm the identical disclosure in the US 4,998,241 image file before relying on this mapping.

2.8 US 5,274,705 — Younce et al. / Tellabs ⚠️ closest face reference to the NLP architecture

• Full citation: U.S. Patent No. 5,274,705, "Nonlinear processor for an echo canceller and method," granted Dec. 28, 1993 (filed Sept. 24, 1991), Tellabs, Inc. (link; PDF)
• Description (confirmed): Center-clipper NLP receiving e[n] = send-in minus echo estimate; adaptive suppression threshold that is a function of the echo estimate signal ŷ[n] generated by the adaptive filter; dual adaptive thresholds; an adaptive noise-transparency threshold; and noise-matching (pseudo-random noise injection) when the NLP activates.
• § 102 reach: Discloses the '635's Figure 2 center-clip transfer function ("Prior Art" in the '635 itself), the adaptive-filter-plus-subtractor-plus-NLP architecture later recited in claims 25–26, and compare-and-clip behavior (claims 4–5). But the threshold is keyed to the echo estimate, not to measured ERLE, and there is no recursive |residue| average. Anticipates: none of claims 1–28. This is the primary § 103 reference against the NLP-architecture apparatus claims.

2.9 US 5,351,291 — Menez et al. / IBM

• Full citation: U.S. Patent No. 5,351,291, "Adaptive echo cancellation method and device for implementing said method," granted Sept. 27, 1994 (filed Sept. 6, 1991), IBM. (link; EP family EP 0 530 423)
• Description (confirmed via the EP 0 530 423 family): Computes a threshold from averaged coefficient magnitudes: Thrshld(n+1) = α · [Σ|h_i(n)| / (Last−First)], 0<α≤1, and redefines coefficient boundaries by comparing |h_i| against the threshold.
• § 102 reach: Supplies the "scaled average of magnitudes" arithmetic that the '635 expresses as its recursion — but as a filter-coefficient threshold, not a residue-echo clipping threshold, and with no ERLE. Anticipates: none. Relevant only as a § 103 teaching of the averaging primitive.

2.10 US 5,535,194 — Ashley et al. / Motorola

• Full citation: U.S. Patent No. 5,535,194, "Method and apparatus for echo canceling with double-talk immunity," granted July 9, 1996 (filed July 14, 1994), Motorola, Inc. (link)
• Description: Echo canceling with double-talk immunity — suppressing false clipping during simultaneous near/far speech.
• § 102 reach: Addresses the same failure mode the '635's Background identifies (erroneous clipping during mixed amplitude/double-talk conditions), but discloses no ERLE-controlled threshold and no recursion. Anticipates: none.

2.11 US 5,668,865 — Duttweiler et al. / Lucent Technologies

• Full citation: U.S. Patent No. 5,668,865, "Echo canceler E-side speech detector," granted Sept. 16, 1997 (filed Feb. 26, 1996), Lucent Technologies Inc. (link)
• Description: E-side speech detector operating on echo-return-loss / residual-signal level comparisons.
• § 102 reach: Discloses echo-return-loss measurement and level-ratio comparison logic — the raw material for the ERLE-derivation limitations of claims 8–10 — but not an ERLE-driven clipping threshold or the recursion. Anticipates: none.

2.12 US 5,687,229 — Sih / Qualcomm

• Full citation: U.S. Patent No. 5,687,229, "Method for controlling echo canceling in an echo canceller," granted Nov. 11, 1997 (filed Sept. 25, 1992), Qualcomm Incorporated. (link)
• Description: State machine that measures far-end/near-end/residual energy, forms ratios, and determines echo return loss produced by the filters; synthesizes background (comfort) noise in the far-end-only state.
• § 102 reach: Discloses energy-ratio-based echo-return-loss determination and comfort-noise generation (the '635's substituted comfort noise). Still no ERLE-adjusted clipping threshold and no Eq. (4) recursion. Anticipates: none.

2.13 US 5,809,085 — Goodson et al. / Motorola

• Full citation: U.S. Patent No. 5,809,085, "Apparatus and method for detecting and discriminating various signal types in the presence of impulse distortions," granted Sept. 15, 1998 (filed June 28, 1995), Motorola Inc. (link)
• Description: Signal-type detection/discrimination amid impulse distortion.
• § 102 reach: Peripherally related signal-classification art; discloses nothing of the claimed ERLE/NLP subject matter. Anticipates: none.

2.14 US 6,061,651 — Nguyen / Speechworks International ⚠️ § 102(e)-eligible, but not anticipatory

• Full citation: U.S. Patent No. 6,061,651, "Apparatus that detects voice energy during prompting by a voice recognition system," granted May 9, 2000 (filed May 21, 1996), Speechworks International, Inc. (link)
• Description: Voice-energy detection used during prompts in an interactive voice-recognition system.
• § 102 reach: Granted after the '635's 1998-11-10 filing, so it could only be prior art under pre-AIA § 102(e) (U.S. patent granted on an application filed before applicant's invention). Its subject matter is unrelated to ERLE-controlled center clipping. Anticipates: none.

3. Synthesis — the § 102 answer

Bottom line: Measured under § 102 alone, the 14 face references collectively anticipate no claim of US 6,424,635. Every independent claim requires the two-element core in § 1 above; no face reference discloses both elements in a single document's four corners, and because dependents incorporate their parents, no dependent claim falls alone either.

Ranking of face references by § 102 proximity (for prioritization, not as a finding of anticipation):

The one genuine single-reference § 102 candidate is not on the face. As flagged in the earlier Obviousness section and confirmed by search, US 6,198,819 (Tellabs Operations, Farrell & Liu) discloses a "non-linear processor … [with] non-linear thresholds at values of ±T_NLP" whose "value of T_NLP is dynamically dependent, at least in part, on echo return loss measurements," and whose specification computes the threshold from ERL and ERLE in dB. (US6198819 PDF) Its application was filed 1997-11-14, before the '635's 1998-11-10 filing, making it potential pre-AIA § 102(e)(2) prior art that was apparently not before the examiner. Even this reference, however, does not appear to disclose the '635's signature recursive |residue-echo| average in the "ones complement" form, so the cleanest characterization is: a very strong § 102/§ 103 single-reference challenge to the ERLE-driven-threshold limitation, but not a complete anticipation of any independent claim as printed. Verify the '819 text and its pre-1998-11-10 filing/priority against the USPTO image file before relying on it.

4. Caveats a validity opinion must carry forward

1. The 2004-06-15 Certificate of Correction (from the patent record) may have altered claim text — including possibly the claim 25 "ERLB"/"ERLE" discrepancy and the claim 4/6 "exceeds" language that inverts the specification's FIG. 6 block 630. Pull the certified copy before relying on any claim text above.
2. US 4,998,241 and US 5,351,291 disclosure details were confirmed through their apparent EP family members (EP 0 371 567 and EP 0 530 423) rather than the U.S. specification bodies; the U.S. texts should be pulled directly.
3. "No anticipation" is a § 102 statement only. It is fully consistent with — and does not undercut — the earlier section's conclusion that the independent claims are obvious under § 103(KSR) over combinations such as Lassaux '230 + Brox '241 + Sih '229.
4. Confidence: High that the 14 face references do not singly anticipate any claim, because the ERLE-plus-recursion core is absent from all of them. Moderate on the exact disclosure content of the four entries I could not live-verify (US 4,912,758, US 4,965,822, US 4,991,166, US 5,535,194, US 5,809,085, US 6,061,651).

Primary sources: US 6,424,635 · US 4,064,379 · US 4,679,230 · US 4,751,730 · US 4,998,241 · US 5,274,705 · US 5,351,291 · US 5,668,865 · US 5,687,229 · US 6,198,819.

I'll research the key prior-art references cited on this patent's face to ground the obviousness analysis.

Let me pull details on the remaining key references to complete the combination analysis.

Obviousness Analysis — U.S. Pat. No. 6,424,635 B1 under 35 U.S.C. § 103

Scope note. This is an analyst-grade obviousness study built on the 14 references cited on the '635 face (the "Prior Art" section of the patent page), plus one Tellabs reference surfaced during research that is not listed on the face but sits in the same field. I treat the printed claim text literally — including the apparent "ERLB" typo in claim 25 and the "ones complement" recursion language — and I do not auto-correct any identifier. Consistent with the earlier sections of this analysis, no court has construed these claims (no litigation was found for 6,424,635), so all claim scope below is my own reading, not a judicial construction.

I. The governing framework

Under Graham v. John Deere Co., 383 U.S. 1 (1966), obviousness turns on (1) the scope and content of the prior art, (2) the differences between the prior art and the claims, (3) the level of ordinary skill, and (4) secondary considerations. Under KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007), a combination is obvious where the improvement is the "predictable use of prior-art elements according to their established functions," where there is a "design incentive" or "market demand" to solve the problem, or where the disclosed approach is "one of a finite number of identified, predictable solutions." A teaching, suggestion, or motivation to combine need not be explicit in any single reference. In re Kahn, 441 F.3d 977 (Fed. Cir. 2006).

Key structural implication for this patent: every independent claim is a combination claim. No single face reference discloses all elements, so this is a § 103 combination case, not a § 102 anticipation case (with one possible § 102(e) caveat discussed in § VII).

II. Level of ordinary skill in the art (POSITA)

Based on the field and the '635 disclosure, a POSITA at the Nov. 10, 1998 filing date would be:

• a person with a B.S. in electrical engineering (or equivalent) and roughly 2–4 years of experience in digital telephony / line-echo cancellation, or an M.S. plus 1–2 years;
• familiar with the standard echo-canceler architecture of FIG. 1 of the '635 (hybrid → adaptive filter → summing junction → center-clipper NLP) and with ITU-T G.165/G.168 terminology;
• conversant in adaptive filtering (LMS/NLMS), center-clipper NLPs, comfort-noise injection, dBm0 level conventions, and echo return loss (ERL) and echo return loss enhancement (ERLE) as standard performance measures; and
• practiced in ordinary DSP building blocks — in particular first-order (exponential / leaky-integrator) recursive averaging of signal magnitudes.

This is an unusually mature art. The '635 specification itself concedes that "center clipping" NLPs, comfort noise, and dynamic threshold control were all conventional — its stated contribution is which parameter drives the threshold.

III. Element-by-element scope of the independent claims

Dependent claims add well-known incidentals: a maximum-threshold cap (claim 2; spec: ~−37 dBm0), averaging the residue echo (claim 3), compare-and-clip / refrain (claims 4–5), decrease on ERLE fall (claim 6), ERLE derivation detail (claims 8–10), on-chip vs. off-chip memory (claims 16–17), and routine-level restatements (claims 20–23, 27).

Observation: the only elements that are not plainly conventional in this art are (i) using ERLE as the control variable for the NLP threshold and (ii) the specific exponential-average recursion. Everything else — dynamic clipping thresholds, center clipping, comfort noise, the adaptive-filter/summer/NLP architecture, and "minimum level + a correction term" threshold formulas — is old and admitted.

IV. The face references, and what each contributes

Cross-reference flag / contradiction check: The earlier "Patent summary" section identified US 6,198,819 only indirectly (via the forward Tellabs family) and did not treat it as prior art. It is not on the '635 face. I introduce it here only as a supplementary, same-field secondary reference (see §VII) and flag that its relevance was not captured in the earlier sections — not a contradiction, but an addition.

V. Primary grounds of rejection

Ground 1 — Lassaux ('230) in view of Brox ('241), further in view of Sih ('229)

This is the strongest ground because all three references address the same problem the '635 identifies: setting the center-clipper threshold so residue is suppressed without erroneously clipping near-end speech.

Ground 2 — Younce ('705) in view of Brox ('241), further in view of Duttweiler ('865) or Sih ('229)

• '705 teaches the exact center-clipper NLP of FIG. 2/'635 and a threshold that is a function of the echo estimate — i.e., a threshold that tracks the echo-canceler's own performance rather than being a fixed preset.
• '241 teaches that the threshold may instead be computed from the mean of the residual (echo-cancelled) signal.
• '865 / '229 supply ERL/ERLE measurement in dB.
• Motivation: '705's threshold responds to the echo estimate, not to how well the canceller is actually converging; a POSITA seeking to avoid "erroneous clipping" (the '635's stated object) would naturally key the threshold to the measured ERLE, which is the industry-standard convergence metric (the same metric Tellabs used in its dual-H family).

Ground 3 — '705 + '241 + '291 (for the averaging-recursion limitations specifically)

'291's THRSHLD = Σ|h| / (LAST−FIRST) followed by THRSHLD = ALPHA·THRSHLD shows a POSITA deriving a working threshold from a scaled average of absolute signal-derived values. Mapped onto '241's "mean value of the echo-cancelled signal," this is the '635's Eq. (4) recursion. In re Stempel / In re Boesch: expressing a known averaging operation in an alternative (complement-form) mathematical notation does not confer patentability where the steps and their function are the same.

Ground 4 — Apparatus / medium claims (11–18, 19, 24, 25–28)

Once the method above is obvious, the apparatus and media claims fall with it:

• The substrate + processor + memory of claims 11/18, the on-chip vs. off-chip memory of claims 16–17, and the machine-readable medium of claims 19/24 reflect conventional DSP platform design — the '635 spec itself calls the platform "a digital platform such as … PASSPORT™ 4400" and states echo cancellation "may include dedicated circuitry," and it explicitly contemplates storing modules in off-chip memory. '758 and '522 show processor-based full-duplex telephony platforms.
• The adaptive filter + summing circuit + NLP of claims 25–26 is the admitted FIG. 1/FIG. 5 architecture, and is shown in '230, '705, '241, '229, and '865.
• For claims 19/24 ("machine readable medium … computer program"), In re Beauregard, 53 F.3d 1583 (Fed. Cir. 1995), and KSR confirm that implementing a known process in software on a known medium is obvious absent a new and unobvious functional relationship.

VI. Motivation to combine (the Kahn / KSR rationale)

A POSITA would have been motivated to make these combinations for reasons rooted in the references themselves, independent of hindsight:

1. Same field, same problem, same architecture. '230, '241, '705, '229 and '865 all address line-echo cancellation with a center-clipper/NLP, and all are classified with the '635 (US Cl. 370/286–289; 379/406–410). '230 even cites the Horna COMSAT article ('379's lineage), showing a shared literature.
2. The problem the '635 admits. The '635's own background states that a threshold computed from far-end incoming level "may be accidentally clipped" when the far-end is ~10 dB louder than near-end. '230 addresses exactly this failure mode (its background criticizes activation logic keyed on N_REC/N_AV), and '194 (Motorola) is directed to double-talk immunity against false clipping. A POSITA would look to a cancellation-quality metric — ERLE — precisely to avoid tying clipping to raw far-end level.
3. Finite, predictable solutions. KSR: where the art teaches a small set of candidate control variables for a threshold (echo level ('230), echo estimate ('705), mean residual ('241), ERL/ERLE ('229/'865/'819)), selecting ERLE is a predictable design choice with an expected benefit (track actual convergence).
4. Established function, predictable result. Averaging a magnitude to smooth a control signal is the established function of exponential averaging; using it to stabilize the threshold yields nothing more than its expected benefit (less jitter, fewer clip errors) — "predictable use of prior-art elements according to their established functions."
5. Threshold formulas were already additive/multiplicative in a correction term. '230 uses "an increasing function of level"; '291 scales by ALPHA; the supplementary '819 computes T_NLP = σx2 − ERL − ERLE. '635 claim 7's "minimum level + ERLE" is a routine algebraic rearrangement of such formulas (cf. In re Stempel).
6. Simple substitution of a known, equivalent metric. Swapping a residual/echo level ratio for its dB-domain ERLE (10·log or 20·log of the same quotient) is a change in units/mathematical form with no change in function — a paradigm case of obviousness.

VII. Anticipation (§ 102) edge and the § 102(e) caveat

• No face reference anticipates. '705 controls the threshold on the echo estimate, not on ERLE. '230 controls on level N_AV and the N_AP/N_AV ratio, not on an averaged residue echo. '241 uses residual-signal means but not ERLE. So no single reference contains every element of claim 1, 7, 11, 18, 19, 24, 25, 26, or 28.
• § 102(e) caveat — US 6,198,819 / US 7,020,278 (Tellabs, Farrell & Liu). If this reference is treated as prior art, it is dangerously close: its claim 1 recites a center-clipping NLP whose "value of T_NLP is dynamically dependent, at least in part, on echo return loss measurements," and the specification computes σ̃e′2 (dB) = σx2 (dB) − ERL (dB) − ERLE (dB) and sets T_NLP to overbound it. That is structurally the '635's Eq. (1) (Default + ERLE) with a different arrangement of terms. It is not on the '635 face, so it was presumably not before the examiner — but it was filed Nov. 14, 1997 (four days after the '635's own filing window opens in earnest) and published/granted after the '635 filed, which makes it classic § 102(e)/§ 102(a)(2) prior art on the application's filing date. I flag this as the single most potent single-reference § 102 or § 103 challenge I identified. (Caveat: I have not verified the exact '819 text from the USPTO record; the quote above is from the secondary sources cited in §IV.)

VIII. Dependent claims — obviousness is even easier

IX. Weaknesses in the § 103 case (diligence on the other side)

To keep this balanced, a patentee (or an examiner resisting rejection) would press these points:

1. Direction-of-change tension. The '635 adds ERLE to a default (threshold rises as ERLE rises). The supplementary Tellabs '819 computes T_NLP = σx2 − ERL − ERLE — which would make T_NLP fall as ERLE rises. A patentee may argue the prior art taught away from the claimed direction. Rebuttal: '230 sets the clipping threshold as an increasing function of a level, and '635's "default + ERLE" is just an additive form; the '635 gives no demonstrated criticality for the sign, and KSR allows a known technique's obvious variant. This is the patentee's best argument but is unlikely to carry the independent claims alone.
2. The recursion's exact wording. The "ones complement … times an average … plus a current absolute value … times the number" phrasing is unusual. A patentee could argue no reference discloses it. Rebuttal: it is the standard exponential-moving-average update; '241 (short-time averages) and '291 (ALPHA·-scaled averaging) supply the concept, and In re Stempel forecloses patentability from mere reformulation (though the claims do tie the average to a concrete clipping decision, so § 101 is not the issue here).
3. Secondary considerations. No evidence of record — the patent expired 2018-11-10 and, per the earlier sections, no litigation was located. There is therefore no demonstrated commercial success, licensing, or copying nexus, and long-felt-need arguments would be weak given the crowded art.
4. My negative-results caveat carries over: I could not confirm the substance of the 2004-06-15 certificate of correction (flagged earlier), which may have altered claim text; the exact '819 disclosure is from secondary sources, not the USPTO image file; and "no litigation" is absence of evidence. Anyone relying on this analysis for a validity opinion should pull the PatentCenter documents, the certificate of correction, and the '819/'278 image files directly.

X. Bottom line

Every independent claim of US 6,424,635 is vulnerable to a § 103 rejection. The strongest ground is Lassaux '230 (variable center-clipper threshold, controlled by a post-/pre-subtractor level ratio ≈ ERLE) in view of Brox '241 (threshold derived from the mean of the echo-cancelled signal) and Sih '229 (echo-return-loss measurement in an energy-ratio state machine), with the exponential-average recursion supplied by '241/'291 and ordinary DSP knowledge. A parallel ground runs Younce '705 (adaptive NLP threshold as a function of the echo estimate) + '241 + '865/'229.

The only genuinely novel-seeming limitation — using ERLE as the NLP threshold control variable, expressed as a first-order recursive average of |residual echo| — is, at the filing date, a predictable substitution of a well-known convergence metric combined with a textbook averaging primitive, satisfying KSR's "finite number of identified, predictable solutions" standard. The closest thing to a single-reference kill shot is the not-listed Tellabs reference US 6,198,819 / US 7,020,278, which teaches a center-clipping NLP threshold dynamically dependent on echo return loss and computes it from ERL/ERLE in dB — I recommend verifying that reference against the USPTO record before finalizing any validity opinion.