BER enhancements for practical interference alignment in the frequency domain

The K user interference alignment scheme with symbol extensions proposed by Cadambe and Jafar achieves K/2 degree of freedom in theory for high signal-to-noise ratios (SNRs). However, lower SNR ranges appear in many practical scenarios. Thus, further improvements of the Cadambe-Jafar scheme are demanded to enable interference alignment in more realistic settings. In this work, we propose a new precoding vector optimization that improves bit error rates (BER) using zero-forcing at the receivers. Furthermore, we compare and combine our approach with existing performance enhancement techniques for interference alignment such as orthonormalizing the precoding matrices or using lattice decoding instead of zero-forcing at the receiver. Finally, we implement interference alignment with symbol extension in the frequency domain along with the presented BER enhancement techniques on a software defined radio platform to validate our approaches. Both simulation results and testbed measurements show significant BER improvements for different M-QAM schemes compared to the original interference alignment mechanism. Moreover, our precoding optimization scheme based on zero-forcing outperforms lattice decoding in practical systems due to its lower sensitivity to real-world effects.