Skip to content
Back to search results

FxLMS Active Noise Control

Introduced by
Bernard Widrow, Samuel Stearns
Year
1975
Type
Adaptive noise cancellation algorithm
Family
Signal processing, Adaptive filtering
Font size

What does it do?

The Filtered-x Least Mean Squares (FxLMS) algorithm is an adaptive filter used in active noise control (ANC) systems to reduce unwanted sound by generating anti-noise.

Pioneered by Widrow and Stearns in 1975 and refined by Eriksson and colleagues, FxLMS is the most widely deployed algorithm in commercial noise-canceling headphones, hearing aids, automotive cabins, and industrial noise barriers. It works by continuously learning the acoustical path and dynamically adjusting a canceling signal in real time.

The intuition behind the method

Imagine if a speaker could instantly emit an anti-sound that perfectly cancels incoming noise—destructive interference. FxLMS realizes this idea: it monitors the noise, predicts what anti-noise is needed, and generates it before the noise reaches your ear. The challenge is that the system must account for the acoustic path (how the anti-noise travels to the error microphone) and adapt continuously as conditions change. FxLMS uses feedback from an error microphone to adjust its filter coefficients, learning the optimal anti-noise pattern in real time.

History

Widrow and Stearns developed the LMS algorithm and its application to active noise control in the 1970s. Eriksson extended LMS to the FxLMS form (filtered-x variant) in the late 1980s, addressing the secondary path problem critical for practical ANC. The first commercial ANC headphones emerged in the 1990s; by 2000s, ANC became standard in premium headphones, hearing aids, and automotive cabins. Modern variants include multirate FxLMS, leaky FxLMS, frequency-domain implementations, and nonlinear extensions.

When to use it

Use FxLMS when active noise control is needed in low-frequency-dominated environments (aircraft cabins, road noise, HVAC hum) where passive absorption is inefficient. FxLMS is ideal for narrowband or tonal noise (sine, multi-tone). For wideband noise or high frequencies, passive approaches are often more cost-effective. FxLMS requires low-latency hardware (fast DSP or microcontroller) and suitable reference and error microphone placement.

Limitations and considerations

  • Causality constraint: FxLMS must predict and generate anti-noise before the noise arrives; long acoustic paths (>1 meter) introduce delays that limit effectiveness.
  • Algorithm convergence depends on accurate secondary path estimation; poor secondary path modeling leads to instability or negative gain (noise amplification).
  • Less effective for wideband noise; noise reduction is typically narrowband. Multichannel systems are required for broadband control, increasing complexity.
  • Requires multiple microphones (reference and error); sensor placement and spacing critically affect performance and stability.
  • Nonlinear acoustic behavior (e.g., saturation at high SPL) and non-causal noise (noise components not in the reference signal) are not handled well by linear FxLMS.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Highly effective for low-frequency tonal and periodic noise where passive absorption is inadequate; can achieve >10–20 dB attenuation below 500 Hz.
  • Adaptive algorithm continuously learns changing acoustic conditions; performance remains high even as room characteristics or noise patterns vary.
  • Low computational cost compared to other adaptive methods; efficient enough for real-time implementation on embedded DSP hardware.
  • Maturity and widespread deployment; decades of research and commercial implementations ensure robustness and reliability.
  • Enables compact, lightweight active noise control solutions (headphones, small cabins) where large passive absorbers would be impractical.

Applications

  • Noise-canceling headphones and earbuds: actively cancel ambient noise in commuting, air travel, and office environments; improve hearing aid comfort.
  • Aircraft cabin noise control: reduce low-frequency engine and wind noise in aircraft cabins and cockpits for passenger comfort and crew safety.
  • Automotive interior acoustics: reduce road and engine noise in car cabins; particularly effective for low-frequency rumble below 500 Hz.
  • Industrial noise control: attenuate noise in factories, compressor rooms, and HVAC systems where passive absorption is space-constrained.
  • Hearing aid feedback cancellation: prevent acoustic feedback (whistling) by generating anti-feedback signal; improves gain before feedback and subjective comfort.
  • Underwater communication and sonar: suppress ocean ambient noise in underwater acoustic systems; enhance signal-to-noise ratio for long-range communication.

How it works

How it works

1

Mount a reference microphone near the noise source (to capture the noise before it travels to the listener) and an error microphone near the listener's ear or noise target. Connect a loudspeaker that will emit the anti-noise signal.

2

Calibrate microphones and speaker; measure the acoustic transfer path from the loudspeaker to the error microphone (often called the secondary path). This characterization is critical for algorithm stability and performance.

3

Monitor the reference noise signal (input to the FxLMS filter) and the error signal (output of the error microphone). Compute the filter output as a weighted sum of past reference samples; update filter weights based on the error gradient.

4

Apply the FxLMS update rule: compute the gradient of error power with respect to filter coefficients, accounting for the secondary acoustic path. The path is filtered through an estimated secondary path model to ensure convergence and stability (the 'x' in FxLMS stands for this filtering).

5

Monitor noise reduction (attenuation in dB) achieved as a function of frequency. Measure convergence speed (how quickly the algorithm learns), residual error power, and computational cost. Validate against performance metrics (signal-to-noise ratio improvement, speech intelligibility in noise).

Common pitfalls

  • Inadequate reference microphone placement; the reference must capture noise energy before it reaches the error microphone. Poor placement reduces correlation and algorithm effectiveness.
  • Neglecting secondary path estimation; an inaccurate secondary path model can cause instability. Always measure or estimate the secondary path carefully.
  • Choosing step size too large (fast convergence) or too small (slow learning); step size tuning is critical. Adaptive step size algorithms improve robustness.
  • Assuming FxLMS works in all conditions; some acoustic environments (multipath, diffuse fields) resist active control. Test in representative conditions before deployment.
  • Ignoring latency and computational constraints; real-time FxLMS requires low-latency audio processing. High-latency systems may diverge or become unstable.

Frequently asked questions

Why is the secondary path so important in FxLMS?

The secondary path is the acoustic transfer function from the loudspeaker to the error microphone—it determines how the anti-noise signal reaches the error measurement point. Without accounting for this path, the algorithm cannot correctly predict what filter output is needed. An inaccurate secondary path estimate causes the algorithm to diverge or provide negative gain (amplification instead of cancellation).

What frequency range can FxLMS effectively control?

FxLMS is most effective below 1–2 kHz, particularly for tonal or low-frequency noise (<500 Hz). At higher frequencies, the wavelength becomes shorter and causality delays become problematic. Multifrequency and multichannel systems extend the range, but passive absorption is often more practical for high-frequency noise.

How long does FxLMS take to converge and adapt to changing noise?

Convergence speed depends on step size, filter length, and signal statistics. Typical convergence is 0.5–2 seconds for stationary noise. Adaptation to slowly changing noise (speech, music, time-varying machinery) occurs within similar timescales. Fast transient noise (impacts, sudden changes) may not be canceled because the algorithm cannot react instantaneously.

Can FxLMS cancel broadband noise like rain or wind?

FxLMS is less effective for broadband noise because a single reference signal cannot predict all frequency components of wideband noise. Multichannel systems (multiple reference microphones and adaptive filters) can improve broadband control, but computational cost increases. For broadband noise, passive absorption or hybrid passive-active systems are often more practical.

What happens if the reference and error microphones are too far apart?

If the distance is large (>1 meter), acoustic propagation delay becomes significant. The reference signal cannot predict the noise at the error microphone due to causality constraints (the algorithm cannot look into the future). This limits the control bandwidth and effectiveness. Practical ANC systems keep microphone spacing small (<10 cm for headphones, <0.5 m for cabin systems).

Related methods

Sources

  1. Widrow, B., & Stearns, S. D. (1975). Adaptive signal processing for active vibration and noise control. IEEE Transactions on Acoustics, Speech, and Signal Processing, 23(5), 440–453.

    Open DOI reference
    Copied.
  2. Eriksson, L. J., Allie, M. C., & Greiner, R. A. (1988). The selection and application of an IIR adaptive filter for use in active sound attenuation. IEEE Transactions on Acoustics, Speech, and Signal Processing, 36(11), 1879–1891.

    Open DOI reference
    Copied.
  3. Kuo, S. M., & Morgan, D. R. (2002). Active Noise Control Systems: Algorithms and DSP Implementations. John Wiley & Sons.

    ISBN: 978-0-471-49663-5
    Copied.