High-Speed Turbo Equalization for GPP-Based Software Defined Radios

Abstract

High data rate waveforms for software defined radios (SDR) have to cope with frequency selective fading due to the mobile use in different harsh transmission environments. The received signal needs to be equalized in order to restore the transmitted information. Turbo equalization is a promising approach to deal with the inter-symbol interference occurring at the receiver. The iterative exchange of soft information between the equalizer and the decoder improves the decision reliability and hence reduces the bit error probability compared to conventional receivers. However, the necessary resource-demanding soft-input soft-output algorithms require a high processing performance to ensure real-time capability. In this paper, we will present a high-speed implementation of a turbo equalizer for SDRs with digital signal processing being performed on general purpose processors (GPP). The implementation will utilize linear MMSE filtering and suboptimal algorithms like overlapping sub-trellis MAX-Log-MAP decoding, approximations of mathematical operations, parallelization methods such as threading-based pipelining, and processor specific optimizations like single instruction multiple data (SIMD) commands. We will present the processing gains for each optimization level, highlight the performance loss for the suboptimal modifications and analyze the latency introduced by the pipelined processing. So far, transmissions with data rates up to 5.4 Mbit/s can be decoded in real-time with negligible performance loss and tolerable delay.