A Practical Scheme for Implementing Dynamic Spectral Precoding in OFDM

MARIANO LIZÁRRAGA; GRACIELA CORRAL BRIONES

Cancún

Conferencia; ReConFigurable Computing and FPGAs (ReConFig), 2014 International Conference on; 2014

IEEE

In wireless dynamic spectrum access (DSA) networks, orthogonal frequency-division multiplexing (OFDM) is an attractive modulation technique to enable coexistence of primary (legacy) and secondary users. In these systems, the use of cognitive radio (CR) promises to achieve high levels of efficiency in spectrum management. Focused on the operation of a secondary transmitter, emphasis is placed on the necessity of adapting the spectral shaping dynamically. To this end, we propose an architecture suitable for updating the spectral precoder according to changing wireless link conditions. In particular, in this paper we consider precoding techniques based on orthogonal projection of the information symbols. This specific precoding requires the generation of a constraint matrix that involves nonlinear operations, and the inversion of matrices with relatively large dimensions. Thus the overall processing remains challenging for FPGA implementations. A novel implementation scheme is proposed for fixed-point arithmetic which is appropriate when the constraint matrix satisfies certain conditions that ensures that involved variables can be adequately represented in this numerical system. Moreover, we identify the critical operations that firstly affect the stability of the updating processing and propose an hybrid architecture with fixed-point calculations and floating-point counterparts. The perspective in this stage is to deal with spectral shaping settings that may incur in an increased demand of stability. Finally, we analyze numerical conditions that require full floating-point calculations in order to achieve a numerically stable solution. A comparison between a floating-point computational model and the hybrid architecture implemented indicates high numerical robustness for the proposed approach.