Low-complexity audio CODEC for resource-scarce embedded system with fixed-point arithmetic

Abstract

This work presents a low-complexity audio coder-decoder (CODEC) based on fixed-point arithmetic to save the usage of system resources in a low-end embedded system. To reduce time complexity and memory usage, a simplified discrete wavelet transform (DWT) with only integer operations is developed. The simplified DWT reduces the computation that is required for the trigonometric operations in the fixed-point system. Moreover, separating the even and odd orders of wavelet coefficients halves the computation time of the original DWT. Additionally, a tri-mode zeroes recording algorithm (TZRA) is proposed. The proposed TZRA utilizes different encoding modes with their corresponding bit data structures to record the locations of zero wavelet coefficients, and determines the optimal encoding mode that uses the fewest bits. To further improve the compression performance, a dynamic wavelet level selecting algorithm is developed to decide the wavelet level for compressing each of frame. This algorithm can dynamically select the optimal wavelet level with the fewest bits by using the determined mode from the TZRA. The experimental results herein reveal that the encoder in the proposed CODEC can multiply computation by factors of 164 and 112 and reduce the size of the execution file by approximately 95.61 and 98.42% relative to baseline audio compression CODECs. Such results indicate that the efficiency when used in real consumer products with fixed-point arithmetic.