Bi-Directional LSTM-Assisted Active User Detection for Uplink Grant-Free Code-Domain NOMA
1 Botsa Siddartha Gowtham
UG Student
Department of ECE
GMR Institute of Technology (GMRIT) – Deemed to be University
Rajam, Andhra Pradesh, India
22341A0434@gmrit.edu.in
2 Kella Bhanu Sathwik Naidu
UG Student
Department of ECE
GMR Institute of Technology (GMRIT) – Deemed to be University
Rajam, Andhra Pradesh, India
22341A0481@gmrit.edu.in
3 Gangu Anusha
UG Student
Department of ECE
GMR Institute of Technology (GMRIT) – Deemed to be University
Rajam, Andhra Pradesh, India
22341A0458@gmrit.edu.in
4 Sunkari Karthik
UG Student
Department of ECE
GMR Institute of Technology (GMRIT) – Deemed to be University
Rajam, Andhra Pradesh, India
22341A04H0@gmrit.edu.in
*5 Dr. G. Nooka Raju
Senior Assistant Professor
Department of ECE
GMR Institute of Technology (GMRIT) – Deemed to be University
Rajam, Andhra Pradesh, India
Email: nookaraju.g@gmrit.edu.in
Abstract— GF-NOMA is a good access technology for massive IoT networks because many devices transmit their data without any request or permission, and the base station does not know the active users in the system. Instead, it receives a signal from multiple active users, so it must first identify the active users and then recover the symbols transmitted by the users from the mixed signal it receives. To identify the active users, a bidirectional LSTM network is used, considering the patterns of active users in multiple time slots. The input to this network is provided by the features of the signal, calculated by using the correlation operation. Once the active users are identified, the symbols transmitted by the users are estimated using the minimum mean square error method. This method provides better results in terms of the detection probability and bit error rate compared to the traditional method using the LSTM network.
Keywords— Grant-Free NOMA, Active User Detection, Bidirectional LSTM, MMSE Detection, Sparse Uplink Access, Frequency-Selective Fading.
