دانلود رایگان مقاله دیدگاه عملی عملکرد و آنالیز بهره وری الگوریتم THP

abstract

As an attractive interference cancellation (IC) technique, Tomlinson–Harashima precoding (THP) has been investigated thoroughly in theory. Several high performance THP variants have been proposed, e.g., sorted QR decomposition (SQRD), Cholesky decomposition, vertical Bell Laboratories space time (V-BLAST) and lattice reduction aided THPs. From a practical perspective, however, limited hardware implementations have been reported in the literature so far. To bridge the progress gap between the theory and the practice, we present a comprehensive analysis of these THP variants in terms of performance and implementation efficiency in this paper. We first evaluate their bit-error rate (BER) performance under perfect and imperfect channel state information (CSI) scenarios. Subsequently, the emphasis is put on their implementation efficiency, including the computational complexity, the numerical precision requirement and the parallelism potential. Our analysis shows a wide trade-off space exists between the performance and the implementation efficiency in different THP variants, which is especially valuable for hardware designers to implement cost-efficient architectures biased towards practical systems.

6. Discussion

In summary, according to our analysis, we find three interesting facts related to the implementation of THP, namely, 1, the lattice reduction aided algorithm is of better numerical stability but it is more sensitive to imperfect CSI; 2, the computational complexity of some non-linear precoding algorithms (SQRD and RSiegel) is not always greater than their linear pre-equalization counterparts at the preprocessing stage, which is an advantage under fast fading scenarios; 3, the parallelism potential of these THP algorithms diverse significantly, which implies their implementation efforts diverse significantly as well. Overall, SQRD and MMSE are more implementation efficient than V-BLAST and RSiegel. Moreover, the trade-off space of different THP algorithms is summarized in Table 4. Some general guidelines can be drawn from this table. For example, due to its better BER performance and lower implementation effort, SQRD is more suitable to be employed under fast fading scenarios where the preprocessing has to be performed frequently. Under slow fading scenarios, however, the computation load of preprocessing phase is negligible compared to the IC phase and the transmitter is more likely to know close to perfect CSI. Therefore, it is advisable to employ RSiegel and V-BLAST to achieve high performance in low and high SNR region respectively.