Deep Neural Network-Based Active User Detection for Grant-Free NOMA Systems

Abstract

As a means to support the access of massive machine-type communication devices, grant-free access and non-orthogonal multiple access (NOMA) have received great deal of attention in recent years. In the grant-free transmission, each device transmits information without the granting process so that the basestation needs to identify the active devices among all potential devices. This process, called an active user detection (AUD), is a challenging problem in the NOMA-based systems since it is difficult to identify active devices from the superimposed received signal. An aim of this paper is to put forth a new type of AUD based on deep neural network (DNN). By applying the training data in the properly designed DNN, the proposed AUD scheme learns the nonlinear mapping between the received NOMA signal and indices of active devices. As a result, the trained DNN can handle the whole AUD process, achieving an accurate detection of the active users. Numerical results demonstrate that the proposed AUD scheme outperforms the conventional approaches in both AUD success probability and computational complexity.By exploiting the fact that only a few active devices in a cell transmit the information concurrently (see Fig. 1), the AUD problem can be readily formulated as a sparse recovery problem [8], [9]. Since the transmit vector is sparse, compressed sensing (CS) technique has been popularly employed [10], [11]. In [8], the AUD problem is modeled as a single measurement vector (SMV) problem and MPA is used to solve the problem. In this CS-based AUD scheme, basestation detects devices based on the correlation between the received signal and device specific sequence. However, performance of the CS-based AUD is not that appealing when the columns of a system matrix (a.k.a. sensing matrix) are highly correlated and sparsity (the number of nonzero elements) of the underlying input vector increases. In fact, in the practical NOMAbased transmission, correlation among the NOMA sequences and also device activity (sparsity) are relatively high so that the CS-based AUD might not be effective. Indeed, it has been shown that the performance of the sparse recovery algorithm is degraded significantly when the mutual correlation and sparsity increase [11]. Therefore, it is of importance to come up with a new type of AUD scheme suitable for the overloaded yet less sparse access scenarios.An aim of this paper is to pursue an entirely different approach to detect active users in the grant-free NOMA scenario. For an efficient and accurate AUD, we exploit the deep neural network (DNN), a learning-based tool to approximate the complicated and nonlinear function.Over the years, DNN has been successfully applied in numerous applications such as image classification [12], machine translation [13], automatic speech recognition [14], and Go game [15].Recently, DNN has been also applied to various wireless systems such as multiple-input and multiple-output (MIMO) detection, wireless scheduling, and direction-of-arrival (DoA) estimation [16]. In these w...

Channel Estimation and User Activity Identification in Massive Grant-Free Multiple-Access

“…From ( 27) and ( 29) we can conceive that when the AP does not know the active UEs, it has to first carry out the symbol-based channel estimation, and then average the estimates obtained from the N P pilot symbols to give the final estimation. Furthermore, while AP knows b (p) i from the pilot sequences of UEs and it can also construct A A A i from the UEs' spreading sequences, AP has to know R R R a , in order to compute W W W i,p of (26) to fulfill the channel estimation based on (29). However, AP does not have the knowledge about the active UEs and even the number of them, R R R a is unable to be constructed using the knowledge available to AP, but has to be obtained from alternative approaches.…”

“…Owing to this, in recent years, various CSbased joint UAI, channel estimation and/or multiuser detection (MUD) algorithms have been developed and investigated in the context of mGFMA, when various sparsity structures are considered [6,7,10,11,[17][18][19][20][21][22][23][24][25][26][27][28]. While CS-based methods have some outstanding merits as claimed in references, they are not appearing for operation in the mGFMA systems where the number of active UEs is large [29]. This is because the recovery performance of CS-based methods is limited by the restricted isometric property (RIP) condition [30].…”

User Identification and Channel Estimation by Iterative DNN-Based Decoder on Multiple-Access Fading Channel

“…In the field of signal processing, algorithms based on ML have been applied to recover sparse signals [13], [14]. In related research on signature codes, by applying training data to a properly designed DNN, the scheme proposed in [15] learns the nonlinear mapping between the received signal and the support for detecting active users. Unlike traditional algorithms and ML methods based on conventional models, neural network-based solutions can improve the model's performance through a large amount of data, which is not present in traditional ML because of the influence of the standard model used.…”

Semi-Grant-Free Non-Orthogonal Multiple Access for Tactile Internet of Things

“…Although GF schemes are able to achieve the low-latency requirements, they lack in reliability in TIoT environments where a massive number of devices may need to be supported. To address this, GF-NOMA approaches have been presented and examined as possible solutions [13]- [17]. Particularly, in [13], the authors derived simplified expressions that approximate the outage probability and system throughout for both successive joint decoding (SJD) and successive interference cancellation (SIC), while in [14], Du et al developed an algorithm that exploits the block sparsity to effectively carry out the multi-user detection (MUD) process.…”

“…Jiang et al [15] presented a distributed transmission scheme that aims to mitigate collisions in massive machine-type communications (mMTC) scenarios, while, in [16], they developed a joint user and signal detection algorithm by leveraging the message passing principles of GF-NOMA systems. Finally, in [17], the authors utilized a deep neural network to develop an active user detection (AUD) method for mMTC.…”