Improving performance of GPU-based software beamforming using transform-based channel data compression

Abstract

The massive data transfer of ultrasound data becomes a critical issue of real time transmission for a GPU-based beamformer. Our previous study proposed a real time lossless compression/decompression algorithm to truncate unnecessary bits of ultrasound baseband data and provided compression ratio around 1.7. The compression ability of the previous method depends on the largest signal within a batch, if the location of the largest signal within a batch can be identified and suppressed; higher compression ratio can be achieved. Therefore, we propose the use of the fast Walsh transform (FWT) associated with the lossless compression method to suppress amplitude data to enhance the compression ability of the lossless compression approach. From the simulation results, the compression ratios and PSNRs reached nearly 2.1-2.7 and 42-48 dB, respectively. The extra processing times for FWT performed in hardware front end required several clock cycles. In addition, the inverse FWT of a frame data implemented in a GPU required several milliseconds only. These results show that FWT accompany with the lossless compression method can effectively compress data with reasonable latency. The proposed compression method is also embedded in a 64 channel ultrasound imaging system to verify the feasibility for a GPU-based beamforming system.