Purdue University Graduate School
Dissertation_Tolunay_Seyfi.pdf (2.57 MB)


Download (2.57 MB)
posted on 2020-05-07, 18:19 authored by Tolunay SeyfiTolunay Seyfi
Interference management is necessary to meet the growth in demand for wireless data services. The problem was studied in previous work by assuming a fixed channel connectivity model, while network topologies tend to change frequently in practice.

The associations between cell edge mobile terminals and base stations in a wireless interference network that is backed by cooperative communication schemes is investigated and association decisions are identified that are information-theoretically optimal when taking the uplink-downlink average. Then, linear wireless networks are evaluated from a statistical point of view, where the associations between base stations and mobile terminals are fixed and channel fluctuations exist due to shadow fading. Moreover, the considered fading model is formed by having links in the wireless network, each subject independently to erasure with a known probability.

Throughout the information theoretic analysis, it is assumed that the network topology is known to the cooperating transmitting nodes. This assumption may not hold in practical wireless networks, particularly Ad-Hoc ones, where decentralized mobile nodes form a temporary network. Further, communication in many next generation networks, including cellular, is envisioned to take place over different wireless technologies, similar to the co-existence of Bluetooth, ZigBee, and WiFi in the 2.4 GHz ISM-Band. The competition of these wireless technologies for scarce spectrum resources confines their coexistence. It is hence elementary for collaborative interference management strategies to identify the channel type and index of a wireless signal, that is received, to promote intelligent use of available frequency bands. It is shown that deep learning based approaches can be used to identify interference between the wireless technologies of the 2.4 GHz ISM-Band effectively, which is compulsory for identifying the channel topology. The value of using deep neural network architectures such as CNN, CLDNN, LSTM, ResNet and DenseNet for this problem of Wireless Channel Identification is investigated. Here, the major focus is on minimizing the time, that takes for training, and keeping a high classification accuracy of the different network architectures through band and training SNR selection, Principal Component Analysis (PCA) and different sub-Nyquist sampling techniques.
Finally, a number theoretic approach for fast discovery of the network topology is proposed. More precisely, partial results on the simulation of the message passing model are utilized to present a model for discovering the network topology. Specifically, the minimum number of communication rounds needed to discover the network topology is examined. Here, a single-hop network is considered that is restricted to interference-avoidance, i.e., a message is successfully delivered if and only if the transmitting node is the only active transmitter connected to its receiving node. Then, the interference avoidance restriction is relaxed by assuming that receivers can eliminate interference emanating from already discovered transmitters. Finally, it is explored how the network size and the number of interfering transmitters per user adjust the sum of observations.


Degree Type

  • Doctor of Philosophy


  • Electrical and Computer Engineering

Campus location

  • West Lafayette

Advisor/Supervisor/Committee Chair

Aly El Gamal

Additional Committee Member 2

Mark R. Bell

Additional Committee Member 3

David J. Love

Additional Committee Member 4

Michael D. Zoltowski