Machine learning for link adaptation in wireless networks
| dc.contributor.advisor | Heath, Robert W., Ph. D. | en |
| dc.contributor.committeeMember | Andrews, Jeffrey | en |
| dc.contributor.committeeMember | Nettles, Scott | en |
| dc.contributor.committeeMember | Caramanis, Constantine | en |
| dc.contributor.committeeMember | Qiu, Lili | en |
| dc.creator | Daniels, Robert C. | en |
| dc.date.accessioned | 2012-01-30T18:20:24Z | en |
| dc.date.accessioned | 2017-05-11T22:23:55Z | |
| dc.date.available | 2012-01-30T18:20:24Z | en |
| dc.date.available | 2017-05-11T22:23:55Z | |
| dc.date.issued | 2011-12 | en |
| dc.date.submitted | December 2011 | en |
| dc.date.updated | 2012-01-30T18:20:49Z | en |
| dc.description | text | en |
| dc.description.abstract | Link adaptation is an important component of contemporary wireless networks that require high spectral efficiency and service a variety of network applications/configurations. By exploiting information about the wireless channel, link adaptation strategically selects wireless communication transmission parameters in real-time to optimize performance. Link adaptation in practice has proven challenging due to impairments outside system models and analytical intractability in modern broadband networks with multiple antennas (MIMO), orthogonal frequency division multiplexing (OFDM), forward error correction, and bit-interleaving. The objective of this dissertation is to provide simple and flexible link adaptation algorithms with few link model assumptions that are amenable to modern wireless networks. First, a complete design and analysis of supervised learning for link adaptation in MIMO-OFDM is provided. This includes the construction of a publicly available data set, a new frame error rate bound leading to a new feature set, and IEEE 802.11n performance evaluation to verify that my design outperforms existing link quality metrics. Next, two supervised learning classification algorithms are designed to exploit information collected from packets transmitted and received over standard links in real time: database learning with nearest neighbor classifiers and support vector machines. Algorithms are also proposed to preserve diversity of feature sets in the database and to allow learning algorithms to seek out more information about the network. Finally, link adaptation with supervised learning is applied to MIMO-OFDM systems where the modulation order may be adapted per-stream. This leads to the analysis of the ordered SNR per stream and its connection to the cumulative distribution function of SNR on each stream. Decoupled link adaptation algorithms, which significantly reduce the complexity of non-uniform link adaptation algorithms, are proposed. New analysis of non-uniform link adaptation shows that the performance of decoupled link adaptation algorithms converge to the performance of joint (optimal) link adaptation algorithms as the number of modulation and coding options per-stream increase. This guides the construction of future standards to reduce the complexity of link adaptation in MIMO-OFDM. | en |
| dc.description.department | Electrical and Computer Engineering | en |
| dc.format.mimetype | application/pdf | en |
| dc.identifier.slug | 2152/ETD-UT-2011-12-4509 | en |
| dc.identifier.uri | http://hdl.handle.net/2152/ETD-UT-2011-12-4509 | en |
| dc.language.iso | eng | en |
| dc.subject | Wireless Communications | en |
| dc.subject | Link adaptation | en |
| dc.subject | Adaptive modulation and coding | en |
| dc.subject | Machine learning | en |
| dc.subject | MIMO | en |
| dc.subject | OFDM | en |
| dc.title | Machine learning for link adaptation in wireless networks | en |
| dc.type.genre | thesis | en |