Real-time synthetic aperture beamforming: practical issues for hardware implementation

We have implemented a synthetic aperture beamformer for real-time 3D imaging using a network of digital signal processors. This system is capable of beamforming 6.6 million points per second. Using simulated acoustical inputs, the point response of the beamformer has been evaluated using the actual hardware. The output of the beamformer using pre-calculated inputs is in agreement with simulations. The dynamic range of the system as a function of noise at the inputs has been determined. The experimentally measured root-mean-square noise level of the A/Ds is less than 1 mV, which provides over 65 dB of noise-free dynamic range in the image. Calibrations to remove DC offsets, to correct for gain differences, and to correct for relative delays between channels have been developed. The relative delays can be calibrated to within less than 1/1000^th of a cycle. A real-time DC correction has been implemented, which corrects for the DC offsets introduced by the truncation error associated with the 16-bit fixed-point summing. The preliminary performance of the beamformer has been characterized using a limited number of narrowband inputs for beamforming. Images made at 650 frames per second using sine wave inputs are in good agreement with simulations