In my previous article, I introduced my goal of building a DIY Active Noise Cancellation (ANC) system using an ESP32. The long-term objective is to explore whether similar techniques could eventually be applied to reduce low-frequency noise inside a vehicle cabin.
Since then, I have moved beyond collecting hardware and started implementing an actual FxLMS-based adaptive controller.
The good news is that the system is running.
The bad news is that ANC in the real world is much harder than it looks on paper.
From Theory to Reality
The basic idea of ANC seems straightforward:
- Measure incoming noise with a reference microphone.
- Generate an anti-noise signal.
- Play the anti-noise through a speaker.
- Minimize the remaining sound measured by an error microphone.
In simulation, this process often looks clean and elegant.
In practice, every component introduces imperfections:
- Microphone noise
- Speaker delays
- Processing latency
- Acoustic reflections
- Wiring issues
All of these affect the system’s ability to cancel sound effectively.
Current Hardware Setup
The current prototype uses:
- ESP32-WROOM-32
- INMP441 I2S microphone (Error Microphone)
- MAX4466 analog microphone (Reference Microphone)
- MAX98357A I2S amplifier
- Small speaker
The reference microphone listens to the incoming noise, while the error microphone measures the residual sound after cancellation.
The controller runs an FxLMS algorithm in real time on the ESP32.
The Unexpected Challenges
1. Software Issues Took Longer Than Expected
Getting the adaptive algorithm running was not the first challenge.
Before I could even evaluate ANC performance, I had to solve several ESP32-related problems:
- I2S configuration conflicts
- ADC driver issues
- Core dump crashes
- Memory management problems
At multiple points, the hardware appeared functional while the software silently failed.
A significant amount of development time was spent simply getting stable audio input and output streams.
2. Wiring Matters More Than I Expected
One of the biggest surprises was how much wiring quality affected the results.
Initially, I focused on tuning the algorithm:
- Step size (μ)
- Filter length
- Delay compensation
However, some of the most noticeable improvements came from basic hardware practices.
For example, twisting microphone wires reduced noise pickup and improved signal quality more than several rounds of parameter tuning.
As engineers, we often want to optimize algorithms first. This project reminded me that signal quality is just as important.
3. Delay Is Everything
ANC is extremely sensitive to timing.
The anti-noise signal must arrive at nearly the same time as the original noise. Even small delays can dramatically reduce cancellation performance.
This is easy to understand in theory, but seeing it happen in a real system was eye-opening.
A few milliseconds may seem insignificant in many embedded applications, but for ANC they can make the difference between successful cancellation and no noticeable effect at all.
Is It Actually Working?
This is probably the question everyone wants answered.
The honest answer is: partially.
The adaptive filter coefficients are changing as expected, indicating that the FxLMS algorithm is learning.
I can also observe changes in the measured error signal under controlled conditions.
However, the reduction in audible noise is still limited. The system is not yet producing the dramatic “silence” effect that many people associate with commercial ANC headphones.
At this stage, I would describe the project as “technically functioning, but not yet optimized.”
What I Learned So Far
The biggest lesson from this project is that ANC is not primarily an algorithm problem.
The FxLMS algorithm itself is well established and relatively straightforward to implement.
The real challenges are:
- Signal quality
- Acoustic paths
- Latency
- Sensor placement
- Hardware integration
In other words, the difficult part is making the physical system behave the way the mathematics assumes it should.
Next Steps
There is still plenty of work to do.
My next experiments will focus on:
- Improving secondary path estimation
- Optimizing filter length
- Adjusting adaptation parameters
- Testing different microphone placements
- Measuring system latency more accurately
I also plan to perform more structured experiments using controlled noise sources to better quantify the cancellation effect.
Looking Toward Vehicle ANC
My long-term goal remains unchanged.
This desktop prototype is ultimately a learning platform for understanding the practical challenges of active noise cancellation.
Modern vehicles increasingly use ANC to reduce low-frequency cabin noise, and I am curious how far a low-cost DIY implementation can be pushed.
The project is still far from that goal, but each experiment provides a better understanding of what is required to make ANC work outside of simulations.
I’ll continue sharing updates as the system improves.