FPGA implementation of a general purpose HMM processor based on token passing algorithm

In this paper, the FPGA implementation of a general purpose hidden Markov model (HMM) processing unit is illustrated. The HMM is based on the token passing algorithm, resulting in a simplified hardware topology, since a typical HMM based system can be seen as a number of HMM processors working in parallel. Each HMM processor has been designed as a parametric unit, reusable in a number of applications (speech recognition, protein structure prediction, control systems, etc). A VHDL RTL model has been developed and results obtained for the implementation on a Xilinx VirtexII FPGA are discussed.