On the performance of real-time DSP on Android devices

With the increasing availability of mobile devices comes the possibility of using (relatively) cheap, wireless hard-ware embedded with plenty of sensors to perform real-time Digital Signal Processing on live artistic performances. The Android Operating System represents a milestone for mo-bile devices due to its lightweight Java Virtual Machine and API that makes it easier to develop applications that run on any (supported) device. With an appropriate DSP model implementation, it is possible to use the values of sensors as input for algorithms that modify streams of au-dio to generate rich output signals. Because of memory, CPU and battery limitations, it is interesting to study the performance of each device under real-time DSP condi-tions, and also to provide feedback about resources con-sumption as basis for (user or automated) decision making related to the devices ’ use. This work presents an object oriented model for performing DSP on Android devices, and focuses on measuring the time taken to perform com-mon real-time DSP tasks (such as input/output, FIR/IIR filtering and computing FFTs of varying sizes) and esti-mating the maximum order of FIR filters that can be run on different combinations of software and hardware. We obtain and discuss statistics of some specific combinations of device model and operating system version, and our ap-proach can be used on any Android device to provide the user with important information that can aid aesthetic and algorithmic decisions regarding real-time DSP feasibility. 1