FM synthesis has been widely used in almost thirty years because not only is its algorithm simple but it can also generate complex harmonic spectrum with small size of memory for parameters. However, the realistic sound cannot just produce by the basic structure of FM synthesis. In this thesis, the architecture of FM synthesizer which generates more realistic sound than conventional structure is proposed.
The proposed architecture uses six operators to produce one instrumental sound with a modified accumulation unit which has more flexibility that it can implement various connections of operators. The envelope curve which is divided into 5-region is proposed instead of conventional ADSR curve.
As a result, the proposed FM synthesizer is designed in 5-stage architecture that is able to produce 16 sounds simultaneously. The sampling rate of output is 50kHz and the operating frequency is 20MHz. The proposed architecture uses 3.4kB of RAM for internal register files and 16.1kB of ROM for storing the parameters of 128 instruments. To implement hardware with fixed-point arithmetic, the optimized bit width of parameters and registers is determined by simulation. Parameters of an instrument are estimated By Genetic Algorithm. The proposed architecture is verified with simulations to show improvements from conventional architectures.