Deep Neural Network based Channel Estimation Scheme in UWA FBMC-OQAM System

  • Sushma U Bhover

Student thesis: Master's ThesisMaster of Science by Research (MScR)


Underwater Acoustic (UWA) communication has grown in importance due to rising scientific, military, and commercial uses. Remotely operated underwater vehicles (ROVs) are utilized to prevent human deaths in several of these applications. Many underwater applications need real-time video communication. Frequency dependent attenuation and high ambient noise make communication over the Underwater Acoustic Channel (UAC) problematic.
So low data rates in the tens of kbps are now possible across vast distances. This is insufficient for the real-time transmission of large volumes of data. This thesis analyses low bit rate error in both horizontal and vertical UACs. The thesis has two stages:
The first stage is to investigate the best modulation strategy for UWA communication: Physical layer waveforms like orthogonal frequency-division multiplexing (OFDM) and Filterbank Multicarrier (FBMC) modulation make optimal use of the limited acoustic bandwidth while mitigating detrimental propagation phenomena like multipath distortions and Doppler effect. The prototype filter’s excellent temporal and frequency localization properties have demonstrated that FBMC outperforms OFDM in doubly-dispersive UACs. Due to the absence of the Cyclic Prefix (CP), FBMC systems based on OFDM-OQAM achieve 100% bandwidth efficiency. FEC codes, like Turbo codes, substantially enhance the error performance of different systems in both horizontal and vertical UACs.
Second stage is to identify and investigate the following issues related to channel estimation in the UWA system: OFDM and FBMC are possible modulation methods in UWA. Channel estimate is needed to manage the numerous distortions and interferences that can occur in a given channel. However, standard UWA-OFDM systems can’t employ channel estimate methods like least square (LS) or minimum mean square error (MMSE) because of the intricate multipath channels. For the same reason, the channel estimation algorithms established for Orthogonal Frequency Division Multiplexing cannot be simply used to FBMC systems. This difficulty can be alleviated by using deep neural network (DNN) channel estimators and a new training approach.
In the next phase, the proposed DNN models provide the estimated channel impulse responses based on the appropriate channel impulse responses and the received piolt symbols. In terms of bit error rate, the proposed Deep Learning-Channel estimation (DL-CE) based approaches outperform the MMSE algorithm in OFDM. The Deep Learning-Channel estimation (DL-CE) outperforms the Interference Approximation Methods (IAM)-R and IAM-C in FBMC.
This effective implementation indicates that deep learning techniques may be used to OFDM and FBMC systems and is an interesting issue deserving of future investigation.
Date of Award2 Dec 2021
Original languageEnglish
Awarding Institution
  • University of Bristol
SupervisorAngela Doufexi (Supervisor) & Dimitris Agrafiotis (Supervisor)


  • UWA Communication, OFDM, FBMC,
  • Channel estimation, Deep learning,

Cite this