| Summary: | 碩士 === 國立中興大學 === 電機工程學系所 === 103 === In recent years, Multiple-Input and Multiple-output (MIMO) Orthogonal Frequency Division Multiplexing (OFDM) systems has been widely used in a variety of wireless communication systems. The closed-loop MIMO-OFDM system is enhanced by the introduction of pre-coding technology, which feedbacks the channel state information obtained from the receiver to the transmitter in order to pre-code the data for transmission. The advantage of pre-coding is that the channel effect can be pre-compensated, thereby reducing multipath interference for a transmission channel consisting of multiple independent parallel sub-channels. The complexity of the MIMO signal detection can be reduced due to the precoding measure as well. Most linear pre-coding schemes are based on the method of channel matrix decomposition. Popular schemes include Singular Value Decomposition (SVD) , Geometric Mean Decomposition (GMD) and so on. In this thesis, we focus on the most efficient GMD scheme.
In GMD, the channel matrix is decomposed into the product of three matrices, i.e., the pre-coding matrix, the equivalent channel matrix and the receiver matrix used in signal detection. Among them, the precoding matrix information must be fedback to the transmitter. The amount of feedback data, however, will occupy considerable communication bandwdith. Therefore, it is essential to compress the pre-coding matrix information before feedback, i.e.,. “parameterization” of the pre-coding matrix. This calls for additional matrix decomposition and parameter quantization. In this thesis, we present a new parameterization algorithm and its hardware implementation. The parameterization is achieved by perfroming the QR decomposition via Givens rotations first and using the rotation angles of the Givens rotations as the feedback to reconstruct the precoding matrix. Since low complexity coordinate rotations digital computer (CORDIC) modules are employed to implement the Givens rotations, the micro-rotation sequence, instead of the real rotation angle information, is recorded. We can thus reduce the amount of feedback data by 75% and the saving is in line with the spec of IEEE 802.11ac.
Based on the proposed parameterization scheme, the corresponding hardware circuit design and chip implementation are also developed. The design can support a MIMO system with the number of 4×4 antennas and has a high throughput performance. We follow the specs of IEEE 802.11ac to develop the design. A highly parallel architecture and low complexity CORDIC modules are adopted to enhance the throughput while confining the hardware complexity. Various circuit optimization skills are also introduced to achieve a high operating frequency and a low complexity chip implementation. The 4×4 pre-coding matrix parameterization chip is realized in a TSMC 90nm CMOS process technology. The implementation result indicates that the design, with a chip area of 1.23 mm2 and a working frequency up to 194.93MHz, can parameterize a 4×4 precoding matrix every 4 clock cycles. This corresponds to a throughput of 48.73M matrix parameterizations per second. The total power consumption is 28.3 mW. The gate count efficiency (throughput divided by the gate count) is 644.57, which is much superior to the numbers of similar designs reported in the literatures.
|
|---|
