Cognitive Manager in Wireless Communication Networks

Cognitive manager in wireless communications is a cognitive mechanism that adjusts parameters of a user or a network according to highly dynamic environments. The decisions for the varying environment are based on learning patterns from history experiences in order to achieve more efficient communication performance.The cognitive cycle is shown in the figure below.

This project focuses on two main parts. First, patterns that deserve consideration of learning are determined, i.e. the patterns which are repeatable and can be utilized for future optimization and adaptation. Second, optimization algorithms that improve the network performance with the aid of learning patterns are developed.

The scenarios of this project include but not limited to the following:

• Long-Term-Window Resource Allocation
  • Pattern: large scale fading parameter (SINR map)
  • Application: resource allocation
  Average SINR in the cellular coverage depends on the environment. This position-dependent variable can be computed with large scale fading parameter. With this geographical-overlay knowledge (SINR map) and mobility prediction methods the near-future channel conditions of mobile users can be predicted. Therefore, a new resource allocation algorithm aiming at assigning resources to users in a near-future-time window is proposed. It considers the channel variation in both time and frequency dimensions and improves the resource utilization efficiency.
  This algorithm can be extended to multicell/heterogeneous network which    considers cooperation among base stations (BS).

• Channel Correlation Map
  • Pattern: delay spread and angular spread
  • Application: adaptive channel estimation, adaptive channel prediction, adaptive scheduling
  Channel frequency correlation and spatial correlation depend on delay spread and angular energy distribution respectively. These two parameters are environment-dependent and can be learned. The channel correlation knowledge can be used for adaptive channel estimation and prediction without online tracking and feedback the correlation parameters. What is more, due to the feedback delay, the scheduling at the BS suffers performance degradation using out-dated CSI (channel state information). A better channel prediction with the a priori correlation information can be used to combat the influence of the feedback delay.
  Quantized feedback channel decides the granularity of CSI. More feedback, less data transmission for limited bandwidth. But more feedback leads to better scheduling performance. The trade-off can be investigated.

• Secrecy Communication
  • Pattern: to be determined
  • Application: Optimization of secrecy communication
  Secrecy communication utilizes beamforming or/and artificial noise to guarantee the transmission between legitimate nodes and suppress the performance of eavesdroppers' links to a certain level. The locations of eavesdroppers can benefit the optimization of secrecy network. The "secrecy region" is a potential way to make it feasible. The patterns for learning "secrecy region"should be determined.