Fig. 2: Normalized power consumption for different video resolutions with mp4 and mpeg formats this phenomenon is not specific to a particular video type. Fig. 3: Overall work-flow and mode of operations. L is the battery-level and AL is change in battery level a full buffer. To ensure that we do not underflow or overflow the buffer the buffer needs to be filled as soon as it is about to get empty, with the filling of the buffer occurring concurrently with its draining. This is necessary to provide amenable and seamless viewing experience to the user. The lower bound for P should also be greater than the total switching delay (ny, + 71H) as we do not want the device to perform some redundant switching without transitioning to LPS which may incur large latency and power overheads. We observe that this condition gets satisfied automatically when we test using real values. The energy consumption for our chunking model can be expressed as: Pe 5 tint We can maximize the power savings by using the maximum available bandwidth as proved here. This can be explained from the knowledge that a larger value of B means a corresponding increase in h and decrease in w. This decreases the overall data reception duration which in turn enables the radio to remain in LPS for a longer interval. So the chunking model aids the transcoding procedure for further minimizing the power consumption. While we considered a battery aware protocol that focuses on improving battery life-time and used the available bandwidth to estimate the chunk-size, other decision variables such as signal strength or bit-error rate of wireless channel can also be used. Fig. 7: Normalized” power consumption for different player buffer sizes and video resolutions using WiFi (a) and LTE (b) Fig. 6: Energy consumption for different bandwidth utilization using 4G LTE (a), 3G (b), and WiFi (c) (simulated).
