DescriptionThis thesis presents simulation and an experimental investigation of a decentralized spectrum allocation algorithm for secondary user co-existence in a cognitive radio network. Conventional centralized approaches use a Base Station (BS) or a centralized server with a global view of spectrum availability to carry out spectrum allocation. To keep this global view updated in real-time requires significant amount of signaling between the users and the BS, requiring additional spectrum. On the other hand, a decentralized spectrum allocation has the advantage of low-overhead for control information signaling and is typically more appropriate for ad-hoc network deployments. Various frameworks for decentralized resource allocation have already been proposed (Game theoretic framework, markov decision theoretic framework etc.). However, a full system implementation of any such technique has not been attempted. Inspired by the TCP congestion control mechanism, an Additive Increase Multiplicative Decrease (AIMD) based decentralized spectral allocation algorithm is proposed. It leads to a network-level fairness and efficient spectral usage in a cognitive radio network. A network consisting of only secondary users is considered for this problem and each secondary user uses a Non-Contiguous OFDM (NC-OFDM) transceiver system for data transmission. The GNU Radio-USRP2 cognitive radio platform is used for implementation purposes. This thesis also presents an experimental evaluation of an active interference suppression scheme used along with NC-OFDM to enhance the spectral quality of the transmitted waveform and, thereby, improving the overall spectral efficiency of the network. The RF transmission quality of the IA-PFT based NC-OFDM transmission using GNU Radio-USRP2 has been verified with reference to simulation results. Some system level throughput measurement results for Primary- Secondary Co-existence experiments are also discussed in the thesis. The functioning of the decentralized spectrum allocation algorithm has been validated using MATLAB simulations.