利用於時間相關多輸入多輸出之基於幾何平均分解的差動回授預編碼器

Abstract

在本篇論文中，我們考慮在時間相關性多輸入多輸出系統之有限回授基於幾何平均分解的差動回授預編碼器，藉由時間相關性和差動回授預編碼器量化來有效利用有限的回饋位元，以改善系統之效能。由實驗結果顯示，當通道變化較慢的情況下，基於幾何平均分解的差動回授預編碼器量化有很突出的表現，即使是在回授的位元數很少的情況下。

In this thesis, we consider differential quantization of GMD(geometric mean decomposition) precoder for a time correlated channel. When the terminal speed is low and the channel is varying slowly, it is expected that the optimal precoders of consecutive channel uses are very close. In particular they can be related through a differential precoder that is close to the identity matrix. As the differential precoder is in the proximity of the identity matrix, it can quantized more efficiently. By using Frobenius norm as a distance measure, we have found through experiments the differential precoder lies in a small neighborhood of the identity matrix. The radius is proportional to a parameter determined by the time correlation of the channel. For the design of codebooks for quantizing the differential precoder, we show how to design random codewords within a prescribed average distance of the identity matrix. This is done by perturbing the identity matrix randomly and applying QR factorization on the perturbed identity matrix. Simulations are given to demonstrate that by using a small feedback rate the performance of the proposed differential GMD (dGMD) comes close to that of GMD with full channel information at the transmitter.