| Summary: | 博士 === 國立臺北科技大學 === 電資國際專班 === 107 === This dissertation aims to design an inte lligent non-orthogonal cognitive radio (NOCR) system with cyclostationarity based joint spectrum sensing and interference cancellation scheme. The operation of the NOCR system is divided into two primary stages, transmitter (TX) side sub-Nyquist multi-band spectrum sensing and receiver (RX) side neuromorphic cyclic frequency shift (FRESH) filter based joint equalization and signal separation.
The transmitter side spectrum sensing can be performed using Nyquist rate approach or sub-Nyquist rate multi-stage spectrum estimation and signal detection approach. In this dissertation, firstly, cyclostationarity based joint spectrum sensing and wireless channel equalization strategy for discrete wavelet packet based block-filtered OFDM system is designed, which demonstrates good bit error rate (BER) performance under frequency selective fading channel.
Next, a neuromorphic cyclostationary interference cancellation strategy using deep de-noising auto-encoder (DDA) for NOCR system comprising fast convolution
filter-bank OFDM (FC-f-OFDM) and conventional CP-OFDM is proposed. The system is shown to operate well under noisy interference limited conditions. Interference cancellation performance of the proposed system is superior to that of conventional blind RLS based interference canceller.
Next, spatial modulation based virtual antenna array cognitive radio (CR) receiver is proposed. Null steering beamforming principle is applied which helps to avoid causing harmful interference to the licensed users of the spectral band. The CR system is simulated with GFDM waveform, while the licensed user systems use OFDM waveform. Finally, the idea is extended to sub-Nyquist sampling based CR systems.
In this final work, the most popular sub-Nyquist sampling system known as the modulated wideband converter (MWC) is applied along with orthogonal matching pursuit (OMP) based compressive signal (CS) reconstruction algorithm. The purpose is to identify the active sub-bands of the multi-band FC-f-OFDM system, followed by cyclostationarity pattern classification. A 3-band FC-f-OFDM signal with 20 MHz bandwidth each is simulated with the MWC based system. A graph based wavelet filter-bank is employed for de-noising sparse cyclostationary features recovered. The de-noised features are then processed by a deep auto-encoder for active system recognition. At low SNR conditions, classification performance achieved by proposed system is higher than only-auto-encoder based recognition.
The work presented in this dissertation primarily covers filter-bank multicarrier waveform design based on fast convolution block digital filtering principle for CR physical (PHY) layer, cyclostationary spectrum sensing of the proposed waveform using a combination of adaptive line enhancement and FREQuency Shift (FRESH) filtering, FRESH equalization based successive interference cancellation (SIC) and detection of NOCR signals and compressive cyclostationary feature processing using graph filter-bank and deep neural network. Although most of the work done here concern improvements of existing waveform design and processing schemes, the sensing and interference cancellation in NOCR and sub-Nyquist cyclostationary feature processing using graph filter-bank are some novel strategies towards improvement of waveform design and processing in next generation wireless systems.
|
|---|
