Modified gate activation functions of Bi-LSTM-based SC-FDMA channel equalization
Open Access
- 1 August 2023
- journal article
- research article
- Published by Walter de Gruyter GmbH in Journal of Electrical Engineering
- Vol. 74 (4) , 256-266
Abstract
In recent years, artificial neural networks (ANNs) have grown a lot and helped solve numerous problems in wireless communication systems. We have evaluated the performance of the Bidirectional-Long-Short-Term-Memory (Bi-LSTM) recurrent neural networks (RNNs) for joint blind channel equalization and symbol detection using a variety of activation functions (Afs) for the gate units (sigmoid) of Bi-LSTMs without requiring any prior knowledge of channel state information (CSI). The performance of Bi-LSTM networks with different AFs found in the literature is compared. This comparison was carried out with the assistance of three different learning algorithms, namely Adam, rmsprop, and SGdm. The research findings clearly show that performance, as measured by equalization accuracy, can be improved. Furthermore, demonstrate that the sigmoid gate activation function (GAF), which is commonly used in Bi-LSTMs, does not significantly contribute to optimal network behavior. In contrast, there are a great many less well-known AFs that are capable of outperforming the ones that are most frequently utilized.Keywords
This publication has 31 references indexed in Scilit:
- Channel quality and utilization metric for interference estimation in Wireless Mesh NetworksComputers and Electrical Engineering, 2017
- A comparative performance analysis of different activation functions in LSTM networks for classificationNeural Computing and Applications, 2017
- An Introduction to Deep Learning for the Physical LayerIEEE Transactions on Cognitive Communications and Networking, 2017
- A new LSTM model by introducing biological cell state2016
- Hybrid speech recognition with Deep Bidirectional LSTM2013
- Optimization of the weights and asymmetric activation function family of neural network for time series forecastingExpert Systems with Applications, 2013
- Speech recognition with deep recurrent neural networks2013
- Channel Equalization Using Neural Networks: A ReviewIEEE Transactions on Systems, Man and Cybernetics, Part C (Applications and Reviews), 2010
- Framewise phoneme classification with bidirectional LSTM and other neural network architecturesNeural Networks, 2005
- A class +1 sigmoidal activation functions for FFANNsJournal of Economic Dynamics and Control, 2003