Shi, Yun Q. (Committee co-chair)
Su, Wei (Committee co-chair)
Haddad, Richard A. (Committee member)
Kosinski, John (Committee member)
Personick, Stewart D. (Committee member)
Simeone, Osvaldo (Committee member)
|
With the emergence of software defined radios (SDRs), an adaptive receiver is needed that can configure various parameters, such as carrier frequency, bandwidth, symbol timing, and signal to noise ratio (SNR), and automatically identify modulation schemes. In this dissertation research, several fundamental SDR tasks for analog modulations are investigated, since analog radios are often used by civil government agencies and some unconventional military forces. Hence, the detection and recognition of "old technology" analog modulations remain an important task both for civil and military electronic support systems and for notional cognitive radios.
In this dissertation, a Cyclostationarity-Based Decision Tree classifier is developed to separate between analog modulations and digital modulations, and classify signals into several subsets of modulation types. In order to further recognize the specific modulation type of analog signals, more effort and work are, however, needed. For this purpose, two general methods for automatic modulation classification (AMC), feature- based method and likelihood-based method, are investigated in this dissertation for analog modulation schemes. For feature-based method, a multi-class SVM-based AMC classifier is developed. After training, the developed classifier can achieve high classification accuracy in a wide range of SNR. While the likelihood-based methods for digital modulation types have been well developed, it is noted that the likelihood-based methods for analog modulation types are seldom explored in the literature. Average-Likelihood-Ratio-Testing based AMC algorithms have been developed to automatically classify AM, DSB and FM signals in both coherent and non-coherent situations In addition, the Non-Data-Aided SNR estimation algorithms are investigated, which can be used to estimate the signal power and noise power either before or after modulation classification.
|