Memory and computation efficient hardware design for a 3 spatial and temporal layers SVC encoder

Spatial and temporal scalability in Scalable Video Coding (SVC) compression enables a video encoder to generate bit streams efficiently for various resolutions and frame rates. However, doing this requires more complex computations and greater memory bandwidth than H.264/AVC compression. In this paper, the performance and memory bandwidth for a SVC hardware encoder with three spatial and temporal layers are analyzed. Based on the analysis, a novel method is proposed for the source and interlayer data load. Experimental results show that the memory bandwidth is reduced by 77%. Furthermore, the memory access latency of the source data for the base layer is reduced by creating a data load for the base layer overlap with the execution of the enhancement layer. To satisfy the latency requirement, a mode pre-decision algorithm for a hardware SVC encoder is proposed. It reduces the computation of the fractional motion estimation (FME) and the inter-layer residual prediction by 80%. Simulation results show that the proposed methods decrease the BD-PSNR by 0.05 dB and increase the BD-BR by 1.64%, an amount that can be considered negligible in terms of degradation, while an encoding speed of 30 fps for Full HD (1920×1080) videos is achieved at an operating clock frequency of less than 200 MHz¹.