Learning Joint Detection, Equalization and Decoding for Short-Packet Communications
Sebastian Dörner, Jannis Clausius, Sebastian Cammerer, and Stephan ten Brink | University of Stuttgart
IEEE Transactions on Communications, vol. 71, no. 2, pp. 837-850 | February, 2023
Ambiguity Subspace Resolution for Digital Twin Parameter Estimation
Wolfgang Zirwas, Brenda Vilas Boas, and Lars Thiele | Nokia
IEEE ICC Workshops | 2023
Machine Learning Based Channel Prediction for NR Type II CSI Reporting
Brenda Vilas Boas, Wolfgang Zirwas, and Martin Haardt | Nokia
IEEE ICC | 2023
The application of artificial intelligence and machine learning (AI/ML) into the wireless physical layer is under discussion at 3GPP. Channel state information (CSI) prediction is among the sub use cases being studied. In this work, we propose an AI/ML CSI predictor that aims to compensate the scheduling delays at the base station. The AI/ML CSI predictor operates at the user equipment side and generates the channel reporting based on its prediction. Our AI/ML CSI predictor is designed for the intended prediction time, e.g., 5 ms, by collecting a few past measurements at the input. Our architecture is flexible regarding the number of physical resource blocks and can be used by all user equipments within the cell. Our results show that the proposed AI/ML CSI predictor has the 90 % normalized squared error performance around −13 dB and less than 1.4 % of the predicted eigenvectors have a squared generalized cosine similarity below 0.9, which is much better than zero order hold.
Deep-LaRGE: Higher-Order SVD and Deep Learning for Model Order Selection in MIMO OFDM Systems
Brenda Vilas Boas, Wolfgang Zirwas, and Martin Haardt | Nokia
WSA | 2023
Despite the large volume of research on the field of model order selection, finding the correct rank number can still be challenging. Propagation environments with many scatters may generate channel multipath components (MPCs) which are closely spaced. This clustering of MPCs in addition to noise makes the model order selection task difficult for wireless channels which can directly impact user equipment (UE) throughput, e.g., wrong lower rank approximation for channel estimation via Unitary ESPRIT. In this paper, we exploit the multidimensional characteristics of MIMO orthogonal frequency division multiplexing (OFDM) systems and propose an artificial intelligence and machine learning (AI/ML) method capable of determining the number of MPCs with a higher accuracy than state of the art methods in almost coherent scenarios. Moreover, our results show that our proposed AI/ML method has an enhanced reliability as the threshold for signal singular value selection is 80 %.
A Hybrid Approach combining ANN-based and Conventional Demapping in Communication for Efficient FPGA-Implementation
Jonas Ney, Bilal Hammoud, and Norbert Wehn | TU Kaiserslautern
29th Reconfigurable Architectures Workshop (RAW 2022) | May, 2022
In communication systems, Autoencoder (AE) refersto the concept of replacing parts of the transmitter and receiver by artificial neural networks (ANNs) to train the system end-to-end over a channel model. This approach aims to improve communication performance, especially for varying channel conditions, with the cost of high computational complexity for training and inference. Field-programmable gate arrays (FPGAs) have been shown to be a suitable platform for energy-efficient ANN implementation. However, the high number of operations and the large model size of ANNs limit the performance on the resource-constrained devices, which is critical for low-latency and high-throughput communication systems. To tackle his challenge,we propose a novel approach for efficient ANN-based demapping on FPGAs, which combines the adaptability of the AE with the efficiency of conventional demapping algorithms. After adaption to channel conditions, the channel characteristics, implicitly learned by the ANN, are extracted to enable the use of optimized conventional demapping algorithms for inference. We validate the hardware efficiency of our approach by providing FPGA implementation results and by comparing the communication performance to that of conventional systems. Our work opens a door for the practical application of ANN-based communication algorithms on FPGAs.
Efficient FPGA Implementation of an ANN-Based Demapper using Cross-Layer Analysis
Jonas Ney, Bilal Hammoud, Sebastian Dörner, Matthias Herrmann, Jannis Clausius, Stephan ten Brink, and Norbert Wehn | TU Kaiserslautern, University of Stuttgart
MDPI Electronics, Special Issue “Applications of FPGAs and Reconfigurable Computing: Current Trends and Future Perspectives” | April, 2022
In the field of communication, autoencoder (AE) refers to a system that replaces parts of the traditional transmitter and receiver with artificial neural networks (ANNs). To meet the system performance requirements, it is necessary for the AE to adapt to the changing wireless-channel conditions at runtime. Thus, online fine-tuning in the form of ANN-retraining is of great importance. Many algorithms on the ANN layer are developed to improve the AE’s performance at the communication layer. Yet, the link of the system performance and the ANN topology to the hardware layer is not fully explored. In this paper, we analyze the relations between the design layers and present a hardware implementation of an AE-based demapper that enables fine-tuning to adapt to varying channel conditions. As a platform, we selected field-programmable gate arrays (FPGAs) which provide high flexibility and allow to satisfy the low-power and low-latency requirements of embedded communication systems. Furthermore, our cross-layer approach leverages the flexibility of FPGAs to dynamically adapt the degree of parallelism (DOP) to satisfy the system-level requirements and to ensure environmental adaptation. Our solution achieves 2000× higher throughput than a high-performance graphics processor unit (GPU), draws 5× less power than an embedded central processing unit (CPU) and is 5800× more energy efficient compared to an embedded GPU for small batch size. To the best of our knowledge, such a cross-layer design approach combined with FPGA implementation is unprecedented.
On End-to-End Learning of Joint Detection and Decoding for Short-Packet Communications
Jannis Clausius, Sebastian Dörner, Sebastian Cammerer, and Stephan ten Brink | University of Stuttgart
IEEE Globecom Workshops (GC Wkshps), Rio de Janeiro, Brazil, 2022, pp. 377-382 | 2022
DRLLA: Deep Reinforcement Learning for Link Adaptation
Florian Geiser, Daniel Wessel, Matthias Hummert, Andreas Weber, Dirk Wübben, Armin Dekorsky, and Alberto Viseras | Motius, University of Bremen, Nokia
Telecom 2022, 3(4), 692-705 | November, 2022
Link adaptation (LA) matches transmission parameters to conditions on the radio link, and therefore plays a major role in telecommunications. Improving LA is within the requirements for next-generation mobile telecommunication systems, and by refining link adaptation, a higher channel efficiency can be achieved (i.e., an increased data rate thanks to lower required bandwidth). Furthermore, by replacing traditional LA algorithms, radio transmission systems can better adapt themselves to a dynamic environment. There are several drawbacks to current state-of-the-art approaches, including predefined and static decision boundaries or relying on a single, low-dimensional metric. Nowadays, a broadly used approach to handle a variety of related input variables is a neural network (NN). NNs are able to make use of multiple inputs, and when combined with reinforcement learning (RL), the so-called deep reinforcement learning (DRL) approach emerges. Using DRL, more complex parameter relationships can be considered in order to recommend the modulation and coding scheme (MCS) used in LA. Hence, this work examines the potential of DRL and includes experiments on different channels. The main contribution of this work lies in using DRL algorithms for LA, optimized for throughput based on a subcarrier observation matrix and a packet success rate feedback system. We apply Natural Actor-Critic (NAC) and Proximal Policy Optimization (PPO) algorithms on simulated channels with a subsequent feasibility study on a prerecorded real-world channel. Empirical results produced by experiments on the examined channels hint that Deep Reinforcement Learning for Link Adaptation (DRLLA) offers good performance indicated by a promising data rate on the additive white Gaussian noise (AWGN) channel, the non-line-of-sight (NLOS) channel, and a prerecorded real-world channel. No matter the channel impairment, the agent is able to respond to changing signal-to-interference-plus-noise-ratio (SINR) levels, as exhibited by expected changes in the effective data rate.
DOI: 10.3390/telecom3040037
Minimum-Integer Computation Finite Alphabet Message Passing Decoder: From Theory to Decoder Implementations towards 1 Tb/s
Tobias Monsees, Oliver Griebel, Matthias Herrmann, Dirk Wübben, Armin Dekorsy, and Norbert Wehn | University of Bremen and TU Kaiserslautern
AMDPI Entropy, Special Issue “Theory and Application of the Information Bottleneck Method” | October, 2022
In Message Passing (MP) decoding of Low-Density Parity Check (LDPC) codes, extrinsic information is exchanged between Check Node (CNs) and Variable Node (VNs). In a practical implementation, this information exchange is limited by quantization using only a small number of bits. In recent investigations, a novel class of Finite Alphabet Message Passing (FA-MP) decoders are designed to maximize the Mutual Information (MI) using only a small number of bits per message (e.g., 3 or 4 bits) with a communication performance close to high-precision Belief Propagation (BP) decoding. In contrast to the conventional BP decoder, operations are given as discrete-input discrete-output mappings which can be described by multidimensional LUT (mLUTs). A common approach to avoid exponential increases in the size of mLUTs with the node degree is given by the sequential LUT (sLUT) design approach, i.e., by using a sequence of two-dimensional Lookup-Table (LUTs) for the design, leading to a slight performance degradation. Recently, approaches such as Reconstruction-Computation-Quantization (RCQ) and Mutual Information-Maximizing Quantized Belief Propagation (MIM-QBP) have been proposed to avoid the complexity drawback of using mLUTs by using pre-designed functions that require calculations over a computational domain. It has been shown that these calculations are able to represent the mLUT mapping exactly by executing computations with infinite precision over real numbers. Based on the framework of MIM-QBP and RCQ, the Minimum-Integer Computation (MIC) decoder design generates low-bit integer computations that are derived from the Log-Likelihood Ratio (LLR) separation property of the information maximizing quantizer to replace the mLUT mappings either exactly or approximately. We derive a novel criterion for the bit resolution that is required to represent the mLUT mappings exactly. Furthermore, we show that our MIC decoder has exactly the communication performance of the corresponding mLUT decoder, but with much lower implementation complexity. We also perform an objective comparison between the state-of-the-art Min-Sum (MS) and the FA-MP decoder implementations for throughput towards 1 Tb/s in a state-of-the-art 28 nm Fully-Depleted Silicon-on-Insulator (FD-SOI) technology. Furthermore, we demonstrate that our new MIC decoder implementation outperforms previous FA-MP decoders and MS decoders in terms of reduced routing complexity, area efficiency and energy efficiency.
Optimum Quantization of Memoryless Channels with N-ary Input
Tobias Monsees, Dirk Wübben and Armin Dekorsy | University of Bremen
Asilomar Conference on Signals, Systems, and Computers 2022 | October, 2022
This paper considers channel quantization of memoryless channels with N-ary input x and Mutual Information (MI) as fidelity criterion. We make use of an equivalent formulation of the quantization problem that transforms the channel output y into a N − 1 dimensional probability-simplex by using the posterior-distribution p(x|y). By using Burshtein’s optimality theorem, it is possible to show that there exist an optimal solution that is obtained by separating hyperplane cuts in this probability-simplex. We show that for practically relevant real valued input/output channels, the posterior-distribution p(x|y) is located on a smooth curve in the N − 1 dimensional probability-simplex. Under mild conditions, the optimality theorem provides the existence of an optimal solution that is obtained by separating connected segments of this curve. For this case, we provide further insights into the underlying optimization problem and motivate a Dynamic Programming (DP) approach for finding the global optimal quantizer mapping that maximizes the end-2-end MI for the given cardinality of the quantizer output. Numerical investigation with N-ASK input and real valued Additive White Gaussian Noise (AWGN) show that this approach is superior to common design approaches which only converge to a local optimal quantizer mapping.
Machine Learning for CSI Recreation in the Digital Twin Based on Prior Knowledge
Brenda Vilas Boas, Wolfgang Zirwas, and Martin Haardt | Nokia
IEEE Open Journal of the Communications Society | September, 2022
Knowledge of channel state information (CSI) is fundamental to many functionalities for mobile communication systems. With the advance of machine learning (ML) and digital maps, i.e., digital twins, we have a big opportunity to learn the propagation environment and design novel methods to derive and report CSI. In this work, we propose to combine untrained neural networks (UNNs) and conditional generative adversarial networks (cGANs) for MIMO channel recreation based on prior knowledge. The UNNs learn the prior-CSI for some locations which are used to build the input to a cGAN. Based on the prior-CSI estimates, their locations and the location of the desired channel, the cGAN is trained to output the channel expected at the desired location. This combined approach can be used for low overhead CSI reporting as, after training, we only need to report the desired location. Our results show that our CSI recreation method is successful in modelling the wireless channel under different configurations of prior-CSI spatial sampling. In addition, the results consider a real world measurement campaign for indoor line of sight and non-line of sight channels. The signal to noise ratio (SNR) achieved by our CSI recreation is better than the SNR reported by the measured campaign providers. Moreover, our CSI recreation provides means for low overhead CSI reporting as the UNN structure is underparameterized compared to the full explicit CSI, and only the desired location is needed for the cGAN to recreate the desired CSI.
DOI: 10.1109/OJCOMS.2022.3208323
Partial Order-Based Decoding of Rate-1 Nodes in Fast Simplified Successive-Cancellation List Decoders for Polar Codes
Lucas Johannsen, Claus Kestel, Oliver Griebel, Timo Vogt, and Norbert Wehn | TU Kaiserslautern
ACM/SIGDA International Symposium on Field-Programmable GMDPI Electronics, Special Issue “VLSI Architectures for Wireless Communications and Digital Signal Processing” | February, 2022
Polar codes are the first family of error-correcting codes that can achieve channel capacity. Among the known decoding algorithms, Successive-Cancellation List (SCL) decoding supported by a Cyclic Redundancy Check (CRC) shows the best error-correction performance at the cost of a high decoding complexity. The decoding of Rate-1 nodes belongs to the most complex tasks in SCL decoding. In this paper, we present a new algorithm that largely reduces the number of considered candidates in a Rate-1 node and generate all required candidates in parallel. For this purpose, we use a partial order of the candidate paths to prove that only a specified number of candidates needs to be considered. Further complexity reductions are achieved by an extended threshold-based path exclusion scheme at the cost of negligible error-correction performance loss. We present detailed Application-Specific Integrated Circuit (ASIC) implementation data on a 28 nm Fully Depleted Silicon on Insulator (FD-SOI) Complementary Metal-Oxide-Semiconductor (CMOS) technology for decoders with code length 128. We show that the new decoders outperform state-of-the-art reference decoders. For list size 8, improvements of up to 158.8% and 62.5% in area and energy efficiency are observed, respectively.
FPGA-based Trainable Autoencoder for Communication Systems
Jonas Ney, Sebastian Dörner, Matthias Herrmann, Mohammad Hassani Sadi, Jannis Clausius, Stephan ten Brink, and Norbert Wehn | TU Kaiserslautern, University of Stuttgart
ACM/SIGDA International Symposium on Field-Programmable Gate Arrays | February, 2022
In communication systems, autoencoder refers to a system that replaces parts of the traditional transmitter and receiver of the baseband processing chain with artificial neural networks (ANNs). This allows to jointly train the system for an underlying channel model by reconstructing the input symbols at the output. Since the actual behavior of a real communication channel cannot be perfectly reproduced by an abstract model, it is necessary for the autoencoder to adapt to the changing conditions at runtime. Thus, online fine-tuning, in the form of ANN-retraining is of great importance. A platform able to satisfy the low-latency and low-power requirements of embedded communication systems are Field-programmable gate arrays (FPGAs). In this paper, we present an online-trainable low-power FPGA architecture for the receiver of an autoencoder-based communication chain. The architecture is embedded into an exploration framework that automatically determines the optimal degree of parallelism to minimize latency or power consumption. Our solutions achieve 2000×higher throughput than a high-performance GPU, draw 5×less power than an embedded CPU and are 5800×more energy efficient compared to an embedded GPU, for a batch size of one. To the best of our knowledge, this is the first FPGA-based autoencoder implementation for communication systems.
Transfer Learning Capabilities of Untrained Neural Networks for MIMO CSI Recreation
Brenda Vilas Boas, Wolfgang Zirwas, and Martin Haardt | Nokia
ICC 2022 – IEEE International Conference on Communications | 2022
Machine learning (ML) applications for wireless communications have gained momentum on the standardization discussions for 5G advanced and beyond. One of the biggest challenges for real world ML deployment is the need for labeled signals and big measurement campaigns. To overcome those problems, we propose the use of untrained neural networks (UNNs) for MIMO channel recreation/estimation and low over-head reporting. The UNNs learn the propagation environment by fitting a few channel measurements and we exploit their learned prior to provide higher channel estimation gains. Moreover, we present a UNN for simultaneous channel recreation for multiple users, or multiple user equipment (UE) positions, in which we have a trade-off between the estimated channel gain and the number of parameters. Our results show that transfer learning techniques are effective in accessing the learned prior on the environment structure as they provide higher channel gain for neighbouring users. Moreover, the proposed UNN channel state information (CSI) estimators are under-parameterized and can further enable low-overhead CSI reporting.
DOI: 10.1109/ICC45855.2022.9838738
Combining AI/ML and PHY Layer Rule Based Inference – Some First Results
Brenda Vilas Boas, Wolfgang Zirwas, and Martin Haardt | Nokia
23rd IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC 2022) | 2022
Knowledge of channel state information (CSI) is fundamental to many functionalities for Machine learning (ML) applications for wireless communications have gained momentum In 3GPP New Radio (NR) Release 18 we see the first study item starting in May 2022, which will evaluate the potential of artificial intelligence and machine learning (AI/ML) methods for Radio Access Network (RAN) 1, i.e., for mobile radio PHY and MAC layer applications. We use the profiling method for an accurate iterative estimation of the parameters of the dominant multipath components, as it promises a large channel prediction horizon. We investigate options to partly or fully replace some functionalities of rule based PHY layer algorithms by AI/ML inferences, with the goal to achieve either a higher performance, lower latency, or reduced processing complexity. We provide first results for noise reduction, then a combined scheme for model order selection, compare options to infer multipath component start parameters, and provide an outlook on a possible channel prediction framework.
DOI: 10.1109/ICC45855.210.1109/SPAWC51304.2022.9833980
Finite-Alphabet Message Passing using only Integer Operations for Highly Parallel LDPC Decoders
Tobias Monsees, Dirk Wübben, Armin Dekorsy, Oliver Griebel, Matthias Herrmann, Norbert Wehn | University of Bremen
23rd IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC 2022)
Relative Entropy based Message Combining for Exploiting Diversity in Information Optimized Processing
Tobias Monsees, Dirk Wübben and Armin Dekorsy | University of Bremen
25th International ITG Workshop on Smart Antennas (WSA 2021) | November, 2021
Maximizing information has become a powerful technique for the design of efficient receiver components with very low bit-resolution. In the established information processing approach, the algorithmic tasks are executed on discrete messages and the processing steps are designed to optimize the mutual information. In this paper, we extend the concept of information optimized processing for exploiting diversity. To that end, we propose the Relative Entropy based Message Combining (REMC) approach in order to merge discrete messages with different underlying distributions, e.g., stemming from different diversity branches. We exemplary evaluate the proposed REMC for a single-user uplink-model with multiple Radio Access Points (RAPs) and higher order modulation schemes. The numerical results show that message combining is required to consider the reliability of different diversity branches that leverages diversity gains in an information optimized receiver.
Forward-Aware Information Bottleneck-Based Vector Quantization: Multiterminal Extensions for Parallel and Successive Retrieval
Shayan Hassanpour, Dirk Wübben, Armin Dekorsy | University of Bremen
IEEE Transactions on Communications | 2021 (Early Access)
Consider the following setup: Through a joint design, multiple observations of a remote data source shall be locally compressed before getting transmitted via several error-prone, rate-limited forward links to a (distant) processing unit. For addressing this specific instance of multiterminal Joint Source-Channel Coding problem, in this article, the foundational principle of the Information Bottleneck method is fully extended to obtain purely statistical design approaches, enjoying the Mutual Information as their fidelity criterion. Specifically, the forms of stationary points for two types of distributed compression schemes are characterized here. Subsequently, those acquired solutions are utilized as the centerpiece of the proposed generic, iterative algorithm, termed the Multiterminal Forward-Aware Vector Information Bottleneck (M-FAVIB), for addressing the design optimizations. Leveraging an unfolding trick, it will be proven that both distributed compression schemes fall into the category of Successive Upper-Bound Minimization, ensuring their convergence to a stationary point. Eventually, the effectiveness of the proposed compression schemes will be substantiated as well by means of numerical investigations over some typical transmission scenarios.
Neural Network-based Forecasting of Decodability for Early ARQ
Matthias Hummert, Dirk Wübben, Armin Dekorsy | University of Bremen
17th International Symposium on Wireless Communication Systems (ISWCS 2021) | September, 2021
Forecasting the decodability of a received packet of a given decoder is a hard task as many State-of-the-Art (SoTA) decoders are of high complexity and not easy to analyse in an analytical fashion. Gathering this forecast on the other hand would enable to save computational complexity and latency as a decoder execution can be saved if it is unlikely that the received packet is decoded correctly. On top, we can provide early feedback for Automatic Repeat Request (ARQ) schemes before actually running the decoding chain. Guided by this motivation, several approaches of classifying the received packet before the actual decoding process have been discussed. We propose to use neural networks (NN) in the context of forecasting received packets for a given receiver chain. We evaluate the performance of the NN by evaluating different performance metrics and perform an efficiency analysis of ARQ.
DOI: 10.1109/ISWCS49558.2021.9562182
Machine Learning Scaled Belief Propagation for Short Codes
Matthias Hummert, Dirk Wübben, Armin Dekorsy | University of Bremen
IEEE 94th Vehicular Technology Conference (VTC2021-Fall) | September, 2021
The problem of finding good error correcting codes for short block lenghts and its corresponding decoders is an open research topic. A frequently applied soft decoder is the Belief Propagation (BP) decoder, however with degraded performance in case of short loops in the Tanner graph. This is especially problematic for short length codes as loops of small length are more likely to occur. In this paper, we propose the Machine Learning Scaled Belief Propagation (MLS-BP) to mitigate the performance loss of BP decoding for short length codes by introducing a learned scaling factor for the receive signals. The key point of this approach is the fact that the implementation of the BP decoder is not changed and the simple scaling leads to performance results comparable to other proposed BP improvements.
DOI: 10.1109/VTC2021-Fall52928.2021.9625308
Wiener Filter versus Recurrent Neural Network-based 2D-Channel Estimation for V2X Communications
Moritz Benedikt Fischer, Sebastian Dörner, Sebastian Cammerer, Takayuki Shimizu, Bin Cheng, Hongsheng Lu, Stephan ten Brink | University of Stuttgart
32nd IEEE Intelligent Vehicles Symposium | May, 2021
We compare the potential of neural network (NN)-based channel estimation with classical linear minimum mean square error (LMMSE)-based estimators, also known as Wiener filtering. For this, we propose a low-complexity recurrent neural network (RNN)-based estimator that allows channel equalization of a sequence of channel observations based on independent time- and frequency-domain long short-term memory (LSTM) cells. Motivated by Vehicle-to-Everything (V2X) applications, we simulate time- and frequency-selective channels with orthogonal frequency division multiplex (OFDM) and extend our channel models in such a way that a continuous degradation from line-of-sight (LoS) to non-line-of-sight (NLoS) conditions can be emulated. It turns out that the NN-based system cannot just compete with the LMMSE equalizer, but it also can be trained w.r.t. resilience against system parameter mismatch. We thereby showcase the conceptual simplicity of such a data-driven system design, as this not only enables more robustness against, e.g., signal-to-noise-ratio (SNR) or Doppler spread estimation mismatches, but also allows to use the same equalizer over a wider range of input parameters without the need of re-building (or re-estimating) the filter coefficients. Particular attention has been paid to ensure compatibility with the existing IEEE 802.11p piloting scheme for V2X communications. Finally, feeding the payload data symbols as additional equalizer input unleashes further performance gains. We show significant gains over the conventional LMMSE equalization for highly dynamic channel conditions if such a data-augmented equalization scheme is used.
Two-step Machine Learning Approach for Channel Estimation with Mixed Resolution RF Chains
Brenda Vilas Boas, Wolfgang Zirwas, and Martin Haardt | Nokia
2021 IEEE International Conference on Communications Workshops (ICC Workshops) | 2021
Massive MIMO is one of the main features of 5G mobile radio systems. However, it often leads to high cost, size and power consumption. To overcome these issues, the use of constrained radio frequency (RF) frontends has been proposed, as well as novel precoders, e.g., a multi-antenna, greedy, iterative and quantized precoding algorithm (MAGIQ). Nevertheless, the best performance of MAGIQ assumes accurate channel knowledge per antenna element, for example, from uplink sounding reference signals. In this context, we propose an efficient uplink channel estimator by applying machine learning (ML) algorithms. In a first step a conditional generative adversarial network (cGAN) predicts the radio channels from a limited set of full resolution RF chains to the rest of the low resolution RF chain antenna elements. A long-short term memory (LSTM) neural network extracts further phase information from the low resolution RF chain antenna elements. Our results indicate that our proposed ML approach is competitive with traditional Unitary Tensor-ESPRIT in scenarios with various closely spaced multipath components (MPCs).
DOI: 10.1109/ICCWorkshops50388.2021.9473491