Godkända
Analysis and Implementation of Linear MIMO Signal Detection Algorithms
Dan Liu (2009)
Start
2011-12-19
Presentation
2014-12-16 14:00
Plats:
E:3139
Avslutat:
2014-12-18
Examensrapport:
Sammanfattning
Nowadays the demanding for wireless communication increases dramatically across the world. Multiple-input multiple-output (MIMO) is a wireless communication technology with great prospect. For baseband receivers, various sophisticated MIMO detection algorithms have been proposed in literature to exploit the gains provided by MIMO. However, it is challenging to implement algorithms on hardware platforms under different system operating scenarios. In this thesis, two kinds of linear MMSE MIMO detection algorithms are studied for MIMO systems with the same transmitting and receiving antenna numbers: squared MMSE and square-root MMSE. Specifically, three matrix inversion approaches are introduced to the former algorithm, including analytic or block-wise inversion, direct matrix inversion (DMI), and QR-decomposition (QRD) based inversion. As to the latter algorithm, QRD method is adopted. First of all, computational complexity of the four schemes mentioned above are considered on two platforms which are application-specific integrated circuit (ASIC) and reconfigurable cell array (RCA) architecture. Comparisons show that QRD-based square-root MMSE requires the least operation amount on both platforms. Note that no matter which scheme is used, the operation amount on RCAs is higher than that on ASICs. However, the difference of complexity between two platforms decreases when antenna number increases. Next, computation accuracy of the four aforementioned schemes is evaluated. Because of lower computational complexity, two QRD-based schemes are compared in 16-bit fixed-point format. Results show that square-root MMSE algorithm has a better performance. Finally, due to the high computational accuracy and low complexity, QRD-based square-root MMSE is implemented using ASIC approach and demonstrated on FPGA using 16-bit fixed-point format for a 4x4MIMO system. Three different architectures are designed, including two pipeline versions and a time-multiplexed scheme. The throughput of the first pipeline version is 3 times faster than the second one, while its device utilization is twice as much as the latter. Time-multiplexed design has the lowest resource utilization; however, it reveals a lowest throughput because of the high data dependency in the QRD algorithm.
Handledare: Chenxin Zhang (EIT)
Examinator: Joachim Rodrigues (EIT)