I am doing an implementation of feed-forward Filtered X-LMS for active noise cancellation in lab set up with an engine accelerometer data of low frequency (<500Hz) and Fs = 2000Hz. I do not have a directional speakers and high efficient mic. I have a Creative SBS 2.1 Speaker and a Table Desk Microphone to capture the data in real time. The algorithm is modeled in simulink. Will I be able to carry out the experiment efficiently (with 10dB noise reduction).
I tried out offline secondary path estimation, but the #LMS filter weights did not converge with very low error. I doubt it is because the acoustic signal captured by the mic is not accurate. Should I do any additional processing such that mic capture an accurate acoustic output so that LMS works fine for secondary path identification.
I believe your task of discerning your primary signal from noise and room reflections requires multiple microphones arranged in a beam forming array. Consider using several MEMS microphones such as those from InvenSense (formerly made by Analog Devices). Even then, some fancy DSP will be required. Applications such as this require a mic with low self noise. The vibrations from a desk mounting will be prohibitive. I realize that soldering those MEMS microphones presents a problem (non-standard footprint) and that the sample boards are a bit pricey. As an alternative, try the electret Microphones from Adafruit, Product ID 1713. The preamps are built in and they are less than $8. Remember that there is a 2.5V dc offset on electrets. You might also start with a tiny speaker as a point source to simplify your task. This will minimize room reflections from a multi-speaker system. Start with little steps and get that to work.
Also, consider a cup-like shape behind your mics. Let the shapes of animal's outer ears guide you.
Let me ping a few members in case they can help you.
@Tim Wescott, @mnentwig, @CarpetOfStars, @spetcavich
But no one replied !!!