US 4998241

I'll verify against live sources before summarizing.

Verification of identifiers

I searched for the exact string 4998241 only (no truncation or normalization). Findings:

• US 4,998,241 A — confirmed as "Echo canceller," Google Patents (https://patents.google.com/patent/US4998241/en) and USPTO-mirror https://uspto.report/patent/grant/4998241. Consistent with the authoritative full text supplied.
• Number collision to flag: the same digit string 4998241 is also a Japanese patent number (JP 4998241, a wire-saw workpiece cutting method, app. 2007‑320200) — https://jglobal.jst.go.jp/en/detail?JGLOBAL_ID=201303059386980481 — and appears coincidentally as a locker number in an unrelated NC AG register PDF. Neither relates to the US patent.
• CAFC / litigation: I found no CAFC 2026 docket entry, and no district-court or PTAB proceeding, tied to US 4998241. My search tooling gave no docket hits in either attempt. I cannot treat that as proof of "no litigation" — only that I could not verify any via the sources available. Also note the patent lapsed in 2003 (see below), so a 2026 CAFC appeal concerning it would be unusual.
• Caveat on sources: I did not query USPTO PatentCenter/PAIR or the CAFC docket system directly (my tooling is general web search). Metadata below comes from Google Patents and mirror sites, not a live PAIR fetch.

Bibliographic summary — US 4,998,241 A

Abstract (as published)

An echo canceller with an adaptive filter (15) whose adaptation rate is controlled by a control unit (29) responsive to its input signals (x(k), u(k), e(k)). Adaptation should occur only when the far-end subscriber is talking and the near-end subscriber is silent; because an erroneous automatic decision can noticeably degrade call quality, the adaptation rate is controlled gradually by the control unit in response to suitably chosen mean values of its three input signals when the far-end subscriber is talking.

Plain-language overview of the independent claims

Claim 1 — control-unit centric

An echo canceller with two cooperating parts:

1. An adaptive filter that has an input, a control input, and an adaptation rate that can be varied by a control signal on that control input.
2. A control unit that generates that control signal, containing:
   • several inputs for incoming signals that each have a signal power;
   • circuitry that computes a mean value for each signal based on its power — including the signal fed to the filter's input (the far-end/remote signal) and the echo-cancelled signal;
   • a decision circuit that tests whether the mean value of the filter's input signal is above or below a first threshold, and that forces the adaptation rate to zero when that mean value is below the threshold;
3. Threshold-setting circuitry that supplies a second threshold value computed as a function of the mean value of the echo-cancelled signal, where that second threshold governs the adaptation rate (per that function) whenever the decision circuit has not already zeroed the rate.

In short: a hard on/off gate (far-end speech present?) plus a soft, continuously variable rate control derived from the residual-echo level.

Claim 2 (dependent on 1)

Specifies the signal set: the control unit's inputs include the local microphone signal, and the signal at the adaptive filter's first input is the remote transmitter signal.

Claims 3–6 (dependent on 1)

• 3: The second threshold = a time-constant part plus the ratio of (mean of the filter-input signal) to (an estimate of total echo return loss).
• 4: In intervals where the rate has not been zeroed and the ratio of threshold to echo-cancelled mean value exceeds a limit, the total-ERL estimate is updated.
• 5: That total ERL estimate is the product of two partial echo return losses — the echo path's ERL and the adaptive filter's ERL.
• 6: The echo-cancelled signal passes through a centre clipper, whose suppressed-amplitude magnitude is also controlled in dependence on the ratio of the second threshold to the echo-cancelled mean value.

Claim 7 — system/architecture centric (independent)

A full echo-canceller arrangement:

• First input coupled to a local microphone signal u(k);
• Second input for the remote signal x(k);
• A subtractor that subtracts the estimated echo y(k) from the local signal to yield the echo-cancelled signal e(k);
• An adaptive filter receiving the remote signal and the echo-cancelled signal, with a control input for the adaptation-rate control signal, and an output providing y(k);
• A control unit receiving the local, echo-cancelled, and remote signals, and producing the control signal, including:
  • circuitry producing short-time averages E_x(k), E_u(k), E_e(k) of the remote, local, and echo-cancelled signals;
  • first means that set the adaptation rate to zero when the remote signal's short-time average falls below a first threshold;
  • second means that, when the rate is not zeroed, set the rate according to a second threshold that is a function of the short-time average of the echo-cancelled signal.

Claims 8–10 (dependent on 7)

• 8: The short-time averages are computed per E_w(k) = (1−β)·E_w(k−1) + β·w(k), where k is the discrete sample index, w stands for any of the local/echo-cancelled/remote signals, and β is a smoothing parameter (specification prefers β ≈ 0.05 at 8 kHz sampling).
• 9: The second means computes μ(k) = f(S(k)/E_e(k)), where f is a monotonically increasing function bounded between 0 and 1, and S(k) = S_o + E_x(k)/ERL with S_o a fixed startup constant and ERL an estimate of the long-time average total echo return loss.
• 10: ERL is computed as the product of two ratios: (mean remote / mean local) × (mean local / mean echo-cancelled).

Context, citations, and residual uncertainty

• Prior art cited on the face: US 4,360,712; US 4,479,036; EP 0 053 202 B1; US 4,535,206; US 4,648,111 (examiner-cited); DE 37 33 132 A1; plus non-patent literature: Yamamoto et al., "An Adaptive Echo Canceller with Variable Step Gain Method" (1/82) and "Echo Canceller Systems" (8/84). The specification identifies EP 0 053 202 B1 as the closest prior art (binary zero/maximum adaptation rate), which the invention replaces with gradual control.
• The patent's stated core insight: rather than a yes/no double-talk decision, the step-size μ is treated as a probability that the near-end subscriber is not talking, so degradation from erroneous decisions is reduced (see Abstract and the passage describing μ as that probability).
• Downstream citation evidence of relevance: later art cites US 4,998,241, e.g., EP 0 733 838 A3 categorizes it as "A" against claims 1–8; EP 0 731 593 A3 cites it as "X" against claim 1 (column 4 line 48 – column 5 line 6, Fig. 2); EP 0 721 274 A3 cites it as "Y" against claims 1–19; and US 5,390,250 and US 9,502,020 list "4998241, March 5, 1991, Brox" among references.
• Uncertainty I cannot resolve: current assignee of record after Philips' corporate changes; any unrecorded reassignments; and whether any post-2003 litigation or IPR ever named this patent. No CAFC 2026 docket match for 4998241 was returned by my searches could be located — treat that as "not found," not "confirmed absent."
• Interpretation rule observed: all identifiers were read literally; I did not "correct" u(f) appearing in the published abstract text (the description uses u(k)) nor any other alphanumeric string.