Toward Intelligent Reconfigurable Wireless Physical Layer (PHY)

Abstract

Next-generation wireless networks are getting significant attention because they promise 10factor enhancement in mobile broadband along with the potential to enable new heterogeneous services. Services include massive machine type communications desired for Industrial 4.0 along with ultra-reliable low latency services for remote healthcare and vehicular communications. In this article, we present the design of intelligent and reconfigurable physical layer (PHY) to bring these services to reality. First, we design and implement the reconfigurable PHY via a hardware-software co-design approach on system-on-chip consisting of the ARM processor and field-programmable gate array (FPGA). The reconfigurable PHY is then made intelligent by augmenting it with online machine learning (OML) based decision-making algorithm. Such PHY can learn the environment (for example, wireless channel) and dynamically adapt the transceivers' configuration (i.e., modulation scheme, word-length) and select the wireless channel on-the-fly. Since the environment is unknown and changes with time, we make the OML architecture reconfigurable to enable dynamic switch between various OML algorithms on-the-fly. We have demonstrated the functional correctness of the proposed architecture for different environments and word-lengths. The detailed throughput, latency, and complexity analysis validate the feasibility and importance of the proposed intelligent and reconfigurable PHY in next-generation networks.INDEX TERMS Intelligent reconfigurable architecture, zynq SoC, physical layer, machine learning, multiarmed bandit.

Hardware softwareSoCco‐design analysis and implementation ofMIMO‐OFDMfor4G/5G/6G eNodeBapplications

“…The drawback of the design is that it uses a lower-end FPGA with a static approach. 93 Dou and Zhang built OFDM on STM32F407, a microcontroller-based ARM Cortex-M4 processor, and reported the 256 FFT and IFFT with a floating-point execution time of 2.88 ms. The design has a lower microcontroller and is slow for computation.…”

“…Singh et al implemented OFDM with 16 QAM and 64 FFT, explored the reconfigurability, and observed an execution duration of 700 ms. The drawback of the design is that it uses a lower‐end FPGA with a static approach 93 . Dou and Zhang built OFDM on STM32F407, a microcontroller‐based ARM Cortex‐M4 processor, and reported the 256 FFT and IFFT with a floating‐point execution time of 2.88 ms.…”

Deep Neural Network Augmented Wireless Channel Estimation on System on Chip

“…Various studies and experiments have demonstrated that conventional frameworks such as Shannon theory [30], detection theory [31], and queuing theory [32] based wireless PHY suffer from performance degradation due to randomness and diversity of wireless environments. Instead of exploring approximate frameworks, recent advances in AI-MDL approaches and their ability to address hard-to-model problems offer a good alternative for making the wireless PHY robust [33] [34]. Numerous works have shown that traditional ML techniques have been successful in combating complex wireless environments and hardware non-linearities [21].…”