Abstract:Deep learning-based symbol detector gains increasing attention due to the simple algorithm design than the traditional model-based algorithms such as Viterbi and BCJR. The supervised learning framework is often employed to predict the input symbols, where training symbols are used to train the model. There are two major limitations in the supervised approaches: a) a model needs to be retrained from scratch when new train symbols come to adapt to a new channel status, and b) the length of the training symbols needs to be longer than a certain threshold to make the model generalize well on unseen symbols. To overcome these challenges, we propose a meta-learning-based self-supervised symbol detector named MetaSSD. Our contribution is two-fold: a) meta-learning helps the model adapt to a new channel environment based on experience with various meta-training environments, and b) self-supervised learning helps the model to use relatively less supervision than the previously suggested learning-based detectors. In experiments, MetaSSD outperforms OFDM-MMSE with noisy channel information and shows comparable results with BCJR. Further ablation studies show the necessity of each component in our framework.

What problem does this paper attempt to address?

The problems that this paper attempts to solve mainly focus on two major limitations of deep - learning symbol detectors in wireless communications: 1. **Difficulty in model adaptation to new environments**: When new training symbols arrive to adapt to new channel states, the supervised learning method needs to retrain the model from scratch. This is not only time - consuming but also difficult to achieve in a rapidly changing environment. 2. **Dependence on a large amount of supervised data**: In order for the model to perform well on unseen symbols, the supervised learning framework requires that the length of the training symbols must exceed a certain threshold. This means that a large amount of training data is required, which poses a challenge to the efficiency and real - time performance of the communication system. To solve these problems, the authors propose a symbol detector based on meta - learning and self - supervised learning, called MetaSSD (Meta - Learned Self - Supervised Detection). Specifically: - **Meta - learning** helps the model quickly adapt to new channel environments through the experience accumulated in various meta - training environments. It can adjust to new environments within a small number of update steps. - **Self - supervised learning** reduces the dependence on supervised data, enabling the model to learn effectively with fewer training symbols. Symbols are predicted by minimizing the reconstruction error of the channel output. ### Formula Summary - **Conditional Distribution**: \[ p(y|x)=\prod_{i = 1}^N p(y_i|x_{i - L+1:i}) \] where \(y_i\) is the channel output and \(x_{i - L+1:i}\) is the input symbol. - **Cross - Entropy Loss Function**: \[ L_x(\theta)=\sum_{i = 1}^P CE(x_i,\hat{p}(x_i)) \] where \(P\) is the length of the training symbols and \(CE\) is the cross - entropy loss. - **Self - supervised Loss Function**: \[ L_y(\theta)=\sum_{i = 1}^N \ell(y_i,g_{\hat{h}}(\hat{x}_{i - L+1:i})) \] where \(\ell\) is the loss function and \(g_{\hat{h}}\) is the estimation function based on the channel information \(\hat{h}\). - **Total Loss Function**: \[ L_D(\theta)=L_x(\theta)+\alpha L_y(\theta) \] where \(\alpha\) controls the importance of the self - supervised loss. - **Meta - learning Optimization Objective**: \[ \theta^*=\arg\min_\theta\sum_{t = 1}^T L^{(t)}_D\left(\theta-\lambda\nabla_\theta L^{(t)}_D(\theta)\right) \] Through these methods, MetaSSD can quickly adapt to new channel environments while reducing the need for supervised data, thereby improving the performance and robustness of symbol detection.

MetaSSD: Meta-Learned Self-Supervised Detection

Spatial Likelihood Voting with Self-Knowledge Distillation for Weakly Supervised Object Detection.

Real-time Machine Learning for Symbol Detection in MIMO-OFDM Systems

Deep Learning-Based Symbol Detection for Time-Varying Nonstationary Channels

Meta-ViterbiNet: Online Meta-Learned Viterbi Equalization for Non-Stationary Channels

Deep Neural Network-Based Symbol Detection for Highly Dynamic Channels

Deep-Learning-Based Adaptive Symbol Decision for Visual MIMO System with Variable Channel Modeling

Research on Signal Detection of OFDM Systems Based on the LSTM Network Optimized by the Improved Chameleon Swarm Algorithm

On Unsupervised Deep Learning Solutions for Coherent MU-SIMO Detection in Fading Channels

Self-Supervised Learning for Anomalous Sound Detection

Symbol-Level Online Channel Tracking for Deep Receivers

MM-FSOD: Meta and metric integrated few-shot object detection

Knowledge-driven Meta-learning for CSI Feedback

An Intelligent Signal Detection Method Based On Dnn For Mbm System

DSV: An Alignment Validation Loss for Self-supervised Outlier Model Selection

DeepIoT : Deep Learning Based Symbol Detection for Spatially Undersampled Internet of Things

Sparse Bayesian Learning Based Symbol Detection for Generalised Spatial Modulation in Large-Scale MIMO Systems.

On the Robustness of Deep Learning-aided Symbol Detectors to Varying Conditions and Imperfect Channel Knowledge

Untrained Neural Network based Bayesian Detector for OTFS Modulation Systems

Semi-Data-Aided Channel Estimation for MIMO Systems via Reinforcement Learning

Symbol Detection Aided Minimum Mean Square Error Interference Alignment