Cognitive radio methodology, physical layer policies and machine learning
In a method of cognitive communication for non-interfering transmission, wherein the improvement comprises the step of conducting radio scene analysis to find not just the spectrum holes or White spaces; but also to use the signal classification, machine learning and prediction information to learn more things about the existing signals and its underlying protocols, to find the Gray space, hence utilizing the signal space, consisting of space, time, frequency (spectrum), code and location more efficiently.
1. A method for permitting maximum utilization of a frequency spectrum utilizing a cognitive communication system, comprising the steps of:
conducting an autonomous radio scene analysis of the signal space for non-interfering signal transmissions where the signal space may consist of at least one of time, frequency, space and code;
performing radio scene analysis or spectrum sensing using higher order statistics to detect the presence of signals, such that if no signal is detected, then that space is determined to be White space which may be used for signal transmission, whereas if signal is detected, then that space is determined to be Gray space, the analysis including
classifying the detected signal through feature extraction in which:
the time and frequency domain behavior is analyzed using spectrogram followed by a single linkage clustering to identify the clusters belonging to the same signal,
time width, bandwidth, carrier frequency of clusters is used to distinguish between various signal types,
standard deviation of the carrier frequency of clusters is used to distinguish between non frequency hopping and frequency hopping signals,
singular value decomposition of the detected signals is performed to identify the code space,
connectionist classification is used to distinguish between various signal types,
incremental un-supervised learning is used to predict signal patterns,
the time frequency detection ratio is used to classify the signal as single carrier or multi-carrier,
the signal space is divided into: broadband or narrowband, frequency hopping or non-frequency hopping, broad pulse or narrow pulse, single carrier or multi-carrier, one signal type or multiple signal types,
time or arrival prediction is performed using incremental learning over time since last pulse to time to next pulse, and
competitive or non-competitive policy sets are defined for signal transmission thus to complete the radio scene analysis, wherein
OFDM signal is transmitted in the White Space, a narrow band signal or another DSSS with an orthogonal spreading code is transmitted in the Gray space if DSSS signal is detected, signal is transmitted on unused sub-carriers and un-used bands if an OFDM signal is detected,
signals are transmitted in bands that are never used, OFDM is used to fill up the un-used spectrum, temporal spectral holes are used for signal transmission for a TDMA signal, adaptive beam-forming is used for signals that can be separated in space for signal transmission, and DSSS or FHSS signals are transmitted on unused hands, and when time and frequency prediction of the next hop is carried out, signal is transmitted in one or more sub-bands that are predicted to be vacant if an FHSS signal is detected.