Singular value decomposition FPGA implementation for tactile data processing

Embedded electronic systems for tactile data processing capture the attention of recent researchers because of its importance in many domains. Machine learning based on tensorial kernel approach has proven its effectiveness in processing tactile information. Computing tensorial kernel corresponds to computing the singular value decomposition. This paper presents an FPGA implementation of singular value decomposition for tensorial kernel computation. The design is implemented for an arbitrary m×n matrix with fixed point arithmetic. The results figure out a tradeoff between the accuracy of computation and the input data resolution. The experimental results demonstrate the efficiency of our design by increasing the accuracy of computation and by providing comparable results in terms of time latency.