Selecting FFT Word Length in OFDM systems with Undersampled Sparse Input-IEEE MOCAST2020

Subtitles available in Greek A critical module in telecommunication applications that are based on Orthogonal Frequency Division Multiplexing (OFDM) is the Fast Fourier Transform (FFT). The rounding error caused by the limited word length for the various FFT parameters has been extensively studied in the literature. The minimum acceptable FFT word length is investigated for a specific OFDM architecture where undersampling is applied at the receiver when sparse data are exchanged. A configurable new FFT architecture has been developed in VHDL in order to test various FFT sizes, word lengths and Quadrature Amplitude Modulations (QAM). The simulation results show that the error floor caused by the undersampling in the specific OFDM system makes unnecessary the use of more than 8 bits as word length when 256-pt or 1024-pt FFT and 16QAM or 4QAM is used. The Normalized Mean Square Error (NMSE) and the Symbol Error Rate (SER) are measured for 8 and 6-bit fixed point word lengths. The simulation results show that the use of a 6-bit word length degrades significantly NMSE and SER while their degradation is small if 8 bits are used since the error caused by the employed undersampling procedure is dominant in this case. These effects are studied for several levels of input sparsity.