Power efficient SDS motion estimation architecture using dynamic iteration control and hierarchical adder compressors for real time HDTV video coding

Abstract

This paper presents a high performance, power efficient and low hardware cost architecture for motion estimation (ME) targeting portable consumer applications. This hardware uses the Sub-sampled Diamond Search algorithm (SDS) with a Dynamic Iteration Control (DIC). The SDS–DIC algorithm can significantly reduce the number of SAD (Sum of Absolute Difference) calculations for block matching, thus enabling the development of an efficient hardware design for the ME. The DIC technique allows for the required throughput to be achieved with a restriction in the number of iterations, which contributes to the reduction in the overall number of clock cycles needed for the motion vector calculation. The processing units (PU) of the ME were developed by using efficient hierarchical adder-compressors, where simultaneous additions of more than two operands can be performed. The results we present show that, by using both the adder compressors in the PU and the DIC technique, it is possible to obtain an efficient ME architecture with higher performance and reduced power consumption. The architecture that implements this algorithm and the PUs was described in VHDL. Hardware synthesis results are presented for a 0.18 μm CMOS standard cell library. The architecture can reach real time for HDTV 1080p with less than 40 mW of power consumption.