Low-Complexity Beamforming for MISO Interference Channel: A Unified Unsupervised Learning Approach

We consider the problem of designing low-complexity beamforming vectors for the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$K$</tex>-user multiple input single output (MISO) interference channel (IFC). We focus on three prominent design problems: sum-rate maximization, transmit power minimization, and maximization of the worst-user rate. To that end, we propose a unified design framework for tackling all three optimization problems. Our proposed technique employs unsupervised deep neural network (DNN). Specifically, by intricate manipulation of the training loss function, we exploit the DNN to solve the beamforming optimization problem. The proposed scheme alleviates the problem of generating labeled datasets, especially when the optimal beamformer is unknown. We compare our approach against state-of-the-art benchmarks, such as weighted minimum mean square error (WMMSE), zero-forcing (ZF), maximum-ratio combining (MRT) and hybrid ZF-MRT. Our numerical results show that our proposed scheme matches or outperforms the stated benchmarks in the high SNR regime. In the testing phase, our scheme finds the optimal beamformer in almost 6% of the time needed for the next benchmark contender, which shows the potential of applying our work in real-time settings. Furthermore, our proposed scheme is more robust against imperfect channel state information (CSI), i.e., shows less error in designing the beamforming vector with noisy CSI, compared to hybrid ZF-MRT.