Çoklu Alıcılar Kullanarak Çevrimiçi ve Zaman Uyumsuz Doğrusal Kipleme Sınıflandırması

Abstract

In this thesis, online and asynchronous modulation classification algorithms for multiple receivers are developed for recognition of linear modulation formats. The term asynchronous refers to the lack of timing synchronization between receiver and transmitter and can be modelled by a time offset. Besides the unknown time offset, complex channel coefficient which embodies the channel gain and channel phase between transmitter and receivers is also assumed unknown.

In the proposed modulation classification techniques, first above mentioned unknown parameters are required to be estimated and then classification decision is made by using estimates of the unknown parameters. To that end, hybrid likelihood based approach is adopted. In the hybrid likelihood based approach, likelihood function (LF) of the received signal is averaged separately over the distributions of the constellation points of every candidate modulation formats. Then, averaged LFs are maximized w.r.t. unknown parameters for each candidate modulation format. The classification decision is made in favor of the modulation format with highest likelihood function score.

An online parameter estimation approach is adopted in the designed modulation classification techniques. In this approach, Titterington's online Expectation Maximization (EM) algorithm based parameter estimation is considered. In the proposed parameter estimation techniques, first the received signal is sampled with the current estimate of the time offset. Then, \emph{a posteriori} probabilities are computed by using the obtained sample. Next, estimates of the unknown parameters are updated by using computed \emph{a posteriori} probabilities. This procedure continues as the new waveforms are collected. After enough samples are collected the classification decision is made in favor of the modulation format with highest likelihood score.

The proposed parameter estimation approach is highly susceptive to the channel state due to its online processing. By using only last sample for the parameter update results in highly irregular parameter estimation in the channels with high noise content. To avoid this, a multiple receiver approach is adopted to benefit from the Signal to Noise Ratio (SNR) diversity due to independent channels between the transmitter and receivers. In a multiple receiver network, even if the channel conditions of the one of the receivers are sub optimal, correct classification decisions can be made due to the other receivers which experience more favorable channel conditions.

In the literature, asynchronous modulation classification algorithms are based on the batch processing of the waveform which increase the memory requirements and computational complexity of the algorithms. The proposed modulation classification algorithms by adopting an online parameter approach, reduce computational complexity and the memory requirements of the proposed algorithms significantly.

In this thesis, besides the proposed EM algorithm based methods, another asynchronous modulation classification technique which utilizes Particle Swarm Optimization (PSO) algorithm for the parameter estimation is proposed.

Performance evaluations of the proposed modulation classification techniques are done by the help of simulations for different average received SNRs, number of receivers and initial distances of the estimates of the unknown parameters to their true values. It is observed that, proposed algorithms attain comparable performances with the clairvoyant classifier which has the perfect information of the unknown parameters in the mid to high SNR region.