Embedded Reconfigurable Solution for OFDM Detection Over Fast Fading Radio Channels

Abstract

OFDM demodulation under fast fading radio channels is very computationally demanding, making the implementation of Software Defined Radio (SDR) solutions problematic. A sub-optimal demodulation algorithm based on QR decomposition of blocks of the channel transfer matrix offers near optimal performance at lower computational cost, but hardware support is still needed. We first propose a COordinate Rotation DIgital Computer (CORDIC) rotator in reconfigurable hardware to expose and then exploit at software level the intra-block paralellism of the QR decomposition. In particular, we show that although the rotator is deeply pipelined, the scale factor inherent to CORDIC algorithm can still be distributedly compensated throughout the pipeline at no additional cycle time penalty. Then, for a Nios II processor augmented with a Reconfigurable Functional Unit (RFU) that incorporates the proposed CORDIC rotator, we also propose a computing scenario that keeps all the data to be processed inside the RFU, to minimize overhead of the data trafic between the Register File and the CORDIC rotator. Overall, we show that OFDM demodulation under fast-fading can be performed in fixed-point arithmetic and in real-time on a Nios II reconfigurable embedded system, proving that an SDR solution for OFDM demodulation under fast fading is possible.