A reconfigurable beamformer for audio applications

Beamforming is a signal processing technique that improves the signal strength received from a specific location. It is already used for many decades in telecommunications, while over the last years, it has been adopted by the audio research society, mostly to enhance speech recognition. In this paper, we propose a scalable organization for a hardware time-invariant beamformer that can be used in small handheld devices and complete 3D-audio systems. Our design can be configured according to the number of input channels. Furthermore, all critical internal modules, such as decimators, FIR filters and interpolators can be adjusted to support various input sampling rates. We developed a hardware prototype in VHDL targeting the Xilinx ML410 board incorporating Virtex4 FX60 FPGA. Following a constrained approach regarding FPGA resource utilization, our hardware prototype occupies 21% of the aforementioned FPGA when instantiating 16 beamforming modules, and consumes approximately 2 Watts of power. Furthermore, our design achieves a speedup of 28 compared to an OMP-annotated software implementation running on a Pentium D at 3.4 GHz. We also compared our design against prior related work. Results suggest that it can extract an audio source up to 11 times faster compared to a reconfigurable adaptive beamformer, and up to 19 times faster compared to DSP implementations.