Novel Direct Digital Frequency Synthesis With Direct Analog Output Architecture Based On Artificial Neural Networks

Direct digital frequency synthesizers (DDSs) are of interest in modern digital communication systems and high-precision function generation. Wide frequency synthesis range, agile and continuous phase frequency switching, fine frequency resolution, and the ability to synthesize arbitrary waveforms as well as sinewaves are some of the important advantages of this family of frequency synthesizers. A conventional DDS usually consists of a phase accumulator, a sine–cosine generator, a digital-to-analog converter (DAC), and a low -pass filter (LPF), as shown in Fig. 1. The sine generator is a look-up table in such a ROM-based DDS. The synthesized sinewave digital signals can be either directly used in digital systems, or converted to analog through a digital-to-analog converter (DAC) and a low-pass reconstruction filter (LPF). In most applications, the parameter which properly characterizes the DDS outputs spectral purity is the spurious free dynamic range (SFDR), defined as the ratio (in decibels) of the amplitude of the desired frequency component to that of the largest undesired frequency component. Decides the accumulating step in the phase accumulator is the operation frequency of the DDS. Many implementations simply use a ROM lookup table to implement the generator block. Unfortunately, in order to achieve an acceptable SFDR at the DDS outputs, large ROMs are needed. The power dissipation of these large ROMs is a significant limitation in portable applications. Furthermore, large ROMs provide slow access time which degrades DDFS circuits timing performances. There are many approaches to convert phase samples at the output of the phase accumulator into the associated sine amplitude samples [1-8].