Repository logo

Doctoral Degrees (Electronic Engineering)

Permanent URI for this collectionhttps://hdl.handle.net/10413/6867

Browse

Browsing Doctoral Degrees (Electronic Engineering) by Subject "Alamouti scheme."

Filter results by typing the first few letters
Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Deep learning-based space-time coding wireless MIMO receiver optimization.
    (2022) Mthethwa, Bhekisizwe Mzimkhulu.; Xu, Hongjun.
    With the high demand for high data throughput and reliable wireless links to cater for real-time or low latency mobile application services, the wireless research community has developed wireless multiple-input multiple-output (MIMO) architectures that cater to these stringent quality of service (QoS) requirements. For the case of wireless link reliability, spatial diversity in wireless MIMO architectures is used to increase the link reliability. Besides increasing link reliability using spatial diversity, space-time block coding schemes may be used to further increase the wireless link reliability by adding time diversity to the wireless link. Our research is centered around the optimization of resources used in decoding space-time block coded wireless signals. There are two categories of space-time block coding schemes namely the orthogonal and non-orthogonal space-time block codes (STBC). In our research, we concentrate on two non-orthogonal STBC schemes namely the uncoded space-time labeling diversity (USTLD) and the Golden code. These two non-orthogonal STBC schemes exhibit some advantages over the orthogonal STBC called Alamouti despite their non-linear optimal detection. Orthogonal STBC schemes have the advantage of simple linear optimal detection relative to the more complex non-linear optimal detection of non-orthogonal STBC schemes. Since our research concentrates on wireless MIMO STBC transmission, for detection to occur optimally at the receiver side of a space-time block coded wireless MIMO link, we need to optimally perform channel estimation and decoding. USTLD has a coding gain advantage over the Alamouti STBC scheme. This implies that the USTLD can deliver higher wireless link reliability relative to the Alamouti STBC for the same spectral efficiency. Despite this advantage of the USTLD, to the best of our knowledge, the literature has concentrated on USTLD wireless transmission under the assumption that the wireless receiver has full knowledge of the wireless channel without estimation errors. We thus perform research of the USTLD wireless MIMO transmission with imperfect channel estimation. The traditional least-squares (LS) and minimum mean squared error (MMSE) used in literature, for imperfect pilot-assisted channel estimation, require the full knowledge of the transmitted pilot symbols and/or wireless channel second order statistics which may not always be fully known. We, therefore, propose blind channel estimation facilitated by a deep learning model that makes it unnecessary to have prior knowledge of the wireless channel second order statistics, transmitted pilot symbols and/or average noise power. We also derive an optimal number of pilot symbols that maybe used for USTLD wireless MIMO channel estimation without compromising the wireless link reliability. It is shown from the Monte Carlo simulations that the error rate performance of the USTLD transmission is not compromised despite using only 20% of the required number of Zadoff-Chu sequence pilot symbols used by the traditional LS and MMSE channel estimators for both 16-QAM and 16-PSK baseband modulation. The Golden code is a STBC scheme with spatial multiplexing gain over the Alamouti scheme. This implies that the Golden code can deliver higher spectral efficiencies for the same link reliability with the Alamouti scheme. The Alamouti scheme has been implemented in the modern wireless standards because it adds time diversity, with low decoding complexity, to wireless MIMO links. The Golden code adds time diversity and improves wireless MIMO spectral efficiency but at the cost of much higher decoding complexity relative to the Alamouti scheme. Because of the high decoding complexity, the Golden code is not widely adopted in the modern wireless standards. We, therefore, propose analytical and deep learning-based sphere-decoding algorithms to lower the number of detection floating-point operations (FLOPS) and decoding latency of the Golden code under low- and high-density M-ary quadrature amplitude modulation (M-QAM) baseband transmissions whilst maintaining the near-optimal error rate performance. The proposed sphere-decoding algorithms achieve at most 99% reduction in Golden code detection FLOPS, at low SNR, relative to the sphere-decoder with sorted detection subsets (SD-SDS) whilst maintaining the error rate performance. For the case of high-density M-QAM Golden code transmission, the proposed analytical and deep learning sphere-decoders reduce decoding latency by at most 70%, relative to the SD-SDS decoder, without diminishing the error rate performance.