Estimating Software and Hardware Video Decoder Energy Using Software Decoder Profiling

In this paper, we develop models for the energy demand of hardware video decoders based on highly accurate energy models for software video decoding. While the standardization of video coding specifications has improved compression efficiency by permitting coding tools that are by the encoder, these tools also result in increased complexity. However, during the development of a video codec, the area and energy demand of a hardware implementation are unknown. To address this issue, we propose new models for estimating the hardware decoder energy demand. Our approach involves the profiling of a software decoder, which is available during standardization, to estimate the hardware energy demand using linear regression modeling. Thereby, the complexity of coding tools for future hardware implementations can be estimated already during the standardization process. We evaluate and discuss three types of models in terms of estimation accuracy. Our results show that pure software decoder modeling has an average estimation error of 1.63%. This concept is then applied with a software decoder that is profiled to estimate the complexity of a hardware decoder. Thereby, it is possible to estimate the energy demand of the hardware decoder with an average error of 13.14%.