DNN-Aided Block Sparse Bayesian Learning for User Activity Detection and Channel Estimation in Grant-Free Non-Orthogonal Random Access

Abstract

In the upcoming Internet-of-Things (IoT) era, the communication is often featured by massive connection, sporadic transmission, and small-sized data packets, which poses new requirements on the delay expectation and resource allocation efficiency of the Random Access (RA) mechanisms of the IoT communication stack. A grant-free non-orthogonal random access (NORA) system is considered in this paper, which could simultaneously reduce the access delay and support more Machine Type Communication (MTC) devices with limited resources. In order to address the joint user activity detection (UAD) and channel estimation (CE) problem in the grant-free NORA system, we propose a deep neural network-aided message passing-based block sparse Bayesian learning (DNN-MP-BSBL) algorithm. In the DNN-MP-BSBL algorithm, the iterative message passing process is transferred from a factor graph to a deep neural network (DNN). Weights are imposed on the messages in the DNN and trained to minimize the estimation error. It is shown that the trained weights could alleviate the convergence problem of the MP-BSBL algorithm, especially on crowded RA scenarios. Simulation results show that the proposed DNN-MP-BSBL algorithm could improve the UAD and CE accuracy with a smaller number of iterations, indicating its advantages for low-latency grant-free NORA systems.

Detection Techniques for Massive Machine-Type Communications: Challenges and Solutions

“…Although the scenario matches with the previous ones, with N = 128, M = 64, MTCDs sporadically active with an activity probability drawn uniformly at random in [0.1,0.3] and the channel modelled as in (6), for both simulations, was necessary a training set with the size of 10 5 . The rest of the parameters considered specifically for each algorithm, are consistent with those cited in the original works, as for DNN-MP-BSBL [59], the threshold to decide the activity of the device is 0.1, the epoch number is 20, the learning rate is 10 −3 and 20 iterations. For BRNN, we used 10 6 samples for training, and the rest of the parameters followed the description in the section and the original paper.…”

“…By implicitly reproducing the behavior of the channel with the LSTM algorithm, this approach considers the greedy SISD algorithm [40] to perform activity and data detection. On the other hand, the works in [56] and [59] uses the DL approach to explicitly estimate the channels. Using the metadata vectors as the training set, both works map each received vector y as an input and considers as the loss function the mean square error (MSE) ĥ DNN − h 2 2 , whereĥ DNN is the estimated channel gain and h is the known channel gain in the training set.…”

“…The work in [59] transfers the iterative message-passing process of MP-BSBL [52], depicted in the last section, to a neural network. Motivated by the convergence speed of MP-BSBL, the authors of [59] impose weights on the Gaussian messages represented on the factor graph depicted in Fig. 14.…”

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

“…In [43], UE activity and signal detection were studied with mGFMA by designing preambles and exploiting the interleave-division multiple access (IDMA), when user signals are randomly and asynchronously transmitted. More recently, machine-learning (ML) approaches have been introduced to mGFMA [22,28,[44][45][46]. To be more specific, the authors of [22] investigated the joint UE activity detection and channel estimation by formulating them as a block sparse signal recovery problem, which is solved by a block sparse Bayesian learning (BSBL) algorithm.…”

“…In [44], the asynchronous sparse Bayesian learning algorithm was adopted for channel estimation while the support vector machine method was applied for UE activity detection, when the number of active users is assumed to be known to BS. To alleviate the convergence problem of the BSBL algorithm [22], a deep neural network-aided message passing-based block sparse Bayesian learning algorithm was proposed in [45] to achieve the joint UE activity detection and channel estimation in mGFMA scenarios. The joint optimization of finite-alphabet spreading sequences and multi-user detection were addressed in [46], where deep learning principle was introduced for the design of both encoder and decoder.…”