KNN Based Classification of Digital Modulated Signals

Abstract

Demodulation process without the knowledge of modulation scheme requires Automatic Modulation Classification (AMC). When receiver has limited information about received signal then AMC become essential process. AMC finds important place in the field many civil and military fields such as modern electronic warfare, interfering source recognition, frequency management, link adaptation etc. In this paper we explore the use of K-nearest neighbor (KNN) for modulation classification with different distance measurement methods. Five modulation schemes are used for classification purpose which is Binary Phase Shift Keying (BPSK), Quadrature Phase Shift Keying (QPSK), Quadrature Amplitude Modulation (QAM), 16-QAM and 64-QAM. Higher order cummulants (HOC) are used as an input feature set to the classifier. Simulation results shows that proposed classification method provides better results for the considered modulation formats.

Performance of Feature-Based Techniques for Automatic Digital Modulation Recognition and Classification—A Review

“…HOS features are used to classify M-PSK and M-QAM modulation types, which are robust against AWGN [12,40,48,[51][52][53][54][55]. Furthermore, the multipath channel effects can be easily modelled using the HOS features [12,31,56,57], which are robust to frequency, phase offset, and timing errors [52,53,58]. In [48,49], the extracted features from ratio and absolute of HOC are used as a characteristic parameter for classifying between M-FSK and M-PSK, M-ASK, and 16 QAM, and between M-PSK and M-ASK [59].…”

“…In [102], eight digitally modulated signals, namely, 2FSK, 4FSK, MSK, BPSK, QPSK, 8PSK, 16QAM, and 64QAM, were classified using KNN. KNN has various disadvantages, such as nonparametric, lazy learner, cannot determine parameter k, and computationally greedy algorithm [12]. GP was used in conjunction with KNN in [54,103].…”

“…The FB approach can be considered a mapping relationship that conducts mapping of time-series signals to the feature field, and these featured parameters are used for signal recognition [8]. Feature extraction is based on various aspects of signals, such as instantaneous amplitude, phase, frequency, wavelet transform (WT), zero-crossing intervals, statistical features (e.g., higher-order moments and cumulants), cyclic cumulants [8,12], and signal spectra [6]. In feature selection, the most relevant features must be utilized to improve the classification accuracy.…”

“…The FB approach not only produces suboptimal performance but also has low computational costs. Therefore, this approach achieves near-optimal performance when it is systematically designed [12].…”

Genetic Algorithm Assisted Support Vector Machine for M-QAM Classification

“…The features selected for classification of M-QAM signals are as under[37]:20 = E [ 2 ( )] = cumm{y (n), y (n)} (5) 21 = E [| ( )| 2 ] = cumm{y (n), * ( )}(6)40 = 40 − 3 20 2 = cumm{y (n), y (n), y (n), y (n)} (7) 41 = 40 − 3 20 21 = cumm{y (n), y (n), y (n), * ( )} (8) 42 = 42 − | 20 | 2 − 2 21 = cumm {y (n), y (n), * ( ), * ( )} (9) =E [ − ( * ) ]…”