VLSI Computational Architectures for the Arithmetic Cosine Transform
This work addresses hardware efficiency for signal processing applications like image and biomedical sensors, though it is incremental as it builds on the recently proposed ACT algorithm.
The paper tackled the problem of computing the discrete cosine transform (DCT) efficiently in hardware by proposing the arithmetic cosine transform (ACT), which enables exact computation for null mean sequences using only additions and integer constant multiplications, resulting in low area complexity and power consumption, with physical implementations showing feasibility on FPGA and 45 nm technology.
The discrete cosine transform (DCT) is a widely-used and important signal processing tool employed in a plethora of applications. Typical fast algorithms for nearly-exact computation of DCT require floating point arithmetic, are multiplier intensive, and accumulate round-off errors. Recently proposed fast algorithm arithmetic cosine transform (ACT) calculates the DCT exactly using only additions and integer constant multiplications, with very low area complexity, for null mean input sequences. The ACT can also be computed non-exactly for any input sequence, with low area complexity and low power consumption, utilizing the novel architecture described. However, as a trade-off, the ACT algorithm requires 10 non-uniformly sampled data points to calculate the 8-point DCT. This requirement can easily be satisfied for applications dealing with spatial signals such as image sensors and biomedical sensor arrays, by placing sensor elements in a non-uniform grid. In this work, a hardware architecture for the computation of the null mean ACT is proposed, followed by a novel architectures that extend the ACT for non-null mean signals. All circuits are physically implemented and tested using the Xilinx XC6VLX240T FPGA device and synthesized for 45 nm TSMC standard-cell library for performance assessment.