基于隐马尔可夫模型的调频广播信号整点报时特性识别方法

由于整点报时与不报时是区分正常广播与“黑广播”的重要指标之一，因此文章提出一种基于隐马尔可夫模型的调频广播信号整点报时特性识别方法。利用静音识别算法将调频广播整点时刻频谱数据表示为0-1序列并在0-1序列中分析调频广播信号的静音序列变化规律，得到调频广播频谱数据的隐藏状态和观察状态的随机序列；通过子区间划分，分别获得初始隐藏状态转移矩阵和初始观察状态概率矩阵；采用Baum-Welch迭代学习算法，在调频广播整点时刻频谱数据训练样本集上对隐藏状态转移矩阵和观察状态概率矩阵进行训练学习，分别得到调频广播整点报时和不报时的隐马尔可夫模型，并给出调频广播信号整点报时特性识别算法。采用实测调频广播整点报时频谱数据进行整点报时特性识别实验，其结果表明，该方法可有效地识别调频广播信号整点报时特性，其整点报时特性的平均识别率为87.5%。该方法在快速发现“黑广播”方面具有一定的实用价值。

Time Announcing Feature Recognition of Frequency Modulation Broadcast Signals Based on Hidden Markov Model

As whether it has the time signal in broadcasting or not is one of the important indicators to distinguish the legal radios from the illegal radios, this paper proposes a recognition method of the time announcing feature of the frequency modulation broadcast signals based on the Hidden Markov Model theory. With mute recognition algorithm, the time spectrum data of the frequency modulation broadcast is transformed as a 0-1 sequence, and the mute sequences are analyzed in 0-1 sequence, and hide states and observable states random sequences are final obtained. After the sub intervals are divided, initial hide states transformation matrix and observable states probability matrix are achieved. By learning hide states transformation and observable states probability matrixes based on Baum-Welch algorithm and training samples, Hidden Markov Model of the time telling and non-time telling of frequency modulation broadcast signals are established and the time announcing feature recognition algorithm is provided. The practical monitoring data of frequency modulation broadcast are utilized to perform the time announcing feature recognition experiment. The results indicate that the proposed method is capable of recognizing the FM broadcasting time announcing feature, and the average rate of recognition is equal to 87.5%. The proposed method has practical value in detecting the frequencies that are illegally occupied efficiently.