Deep Learning-Aided Tabu Search Detection for Large MIMO Systems

Abstract

In this study, we consider the application of deep learning (DL) to tabu search (TS) detection in large multiple-input multiple-output (MIMO) systems. First, we propose a deep neural network architecture for symbol detection, termed the fast-convergence sparsely connected detection network (FS-Net), which is obtained by optimizing the prior detection networks called DetNet and ScNet. Then, we propose the DL-aided TS algorithm, in which the initial solution is approximated by the proposed FS-Net.Furthermore, in this algorithm, an adaptive early termination algorithm and a modified searching process are performed based on the predicted approximation error, which is determined from the FS-Net-based initial solution, so that the optimal solution can be reached earlier. The simulation results show that the proposed algorithm achieves approximately 90% complexity reduction for a 32 × 32 MIMO system with QPSK with respect to the existing TS algorithms, while maintaining almost the same performance. Index TermsMIMO, deep learning, deep neural network, tabu search.

Intelligent Radio Signal Processing: A Survey

“…Consequently, the computational complexity of DetNet is extremely high. Furthermore, although the performance of DetNet is shown to be good for the case N M , subsequent work [11], [120] showed the network performance to be far from optimal for square systems, i.e., N ≈ M .…”

“…For instance, many DL-based algorithms have been proposed to improve and accelerate near-optimal detection schemes. Nguyen et al [11] employed a DL model, namely FS-Net, to initialize the highly reliable solution for the tabu search (TS) detection scheme, and also proposed an early termination scheme to further accelerate the optimization process. Compared with the original TS scheme, the DL-aided TS detector can reduce the computational complexity by approximately 90% at an SNR of 20 dB with similar performance.…”

“…• Training approaches: Unlike the classical methods in which a hand-engineered detection scheme is applied in an online method, an MLD needs to train an ML model before it is used for signal detection. For example, in a DNN-based detector, the weights and biases of the DNN are trained to minimize the distance between the MLbased solution and the labels, i.e.,ŝ ML and s [11], [49], [118], [120]. In particular, the training can be carried out either offline [115]- [120], [122]- [130], [133], [135] or online [114], [121], [132].…”

“…Intuitively, ( 8) can be solved by the iterative PGD optimization method. Motivated by this, a series of unfolding MLDs, including the detection network (DetNet) [49], [118], sparsely-connected DNN (ScNet) [120], fast-convergence sparsely connected DNN (FS-Net) [11], multilayer DNN (Twin-DNN) [119], and Cascade DNN (Cascade-Net) [135] were proposed for signal detection. In these schemes,ŝ ML is updated over L layers of the DNN as follows:…”

An Orthogonal-SGD Based Learning Approach for MIMO Detection under Multiple Channel Models

“…[5]- [8]), which can take advantage of the model knowledge to mitigate the curse of dimensionality problem inherent in the deep learning procedure. Moreover, ML and hand-engineered approaches can work together to form a synergy when conducting the signal detection [9]- [15].…”