A Convolution-Neural-Network Feedforward Active-Noise-Cancellation System on FPGA for In-Ear Headphone

A real-time streaming feedforward active-noise-cancellation (ANC) system for an in-ear headphone was demonstrated in a real application scenario, by implementing 10-layer dilated convolutional-neural-network (CNN) on field programmable gate array (FPGA). 16 × systolic used the FPGA, to speed up model computation. The latency 170.6 μs, at clock frequency of 120 MHz. CNN 3232 parameters. Due large input receptive field, 327 ms, this work achieved total power reduction, 14.8 dB and 14.3 noise incident direction 0∘ 90∘, respectively, attenuation bandwidth 2000 Hz both angles; all results were superior those conventional FxLMS algorithm.