1.Department of Electronics and Optical Engineering, Army Engineering University, Shijiazhuang 050003, China 2.School of Mathematics and Statistics, Xi’an Jiaotong University, Xi’an 710049, China 3.The Troop of 66389, Shijiazhang 050000, China
Fund Project:Project supported by the Natural Science Foundation of Hebei Province, China (Grant No. F2017506006)
Received Date:19 June 2019
Accepted Date:05 August 2019
Available Online:01 November 2019
Published Online:05 November 2019
Abstract:Aiming at the reception of the intermediate frequency signal of sine wave of radio and communication system at extremely low signal-to-noise ratio (SNR), a quadratic polynomial receiving scheme for sine signals enhanced by stochastic resonance (SR) is proposed. Through analyzing the mechanism of sine signals enhanced by SR and introducing the decision time, the analytic periodic stable solution with time parameters of the Fokker-Planck Equation (FPE) is obtained through converting the non-autonomous FPE into an autonomous equation. Based on the probability density function of the particle of SR output, a quadratic polynomial receiving scheme is proposed by analyzing the feature of energy detector and matching filter receiver. By maximizng the deflection coefficient, the binomial coefficients and the test statistic are obtained. For further reducing the bit error, by combining the thought of " the average of N samples”, a quadratic polynomial receiving scheme for sine signals enhanced by SR is proposed through the hypothesis under Gaussian distribution approximation of the law of large N. And the conclusion is obtained as follows. When N is 500 and the SNR is greater than –17 dB, the bit error rate is less than 2.2 × 10–2, under the constraint of the parameters of the optimally matched SR. Keywords:stochastic resonance/ reception of sine signal/ quadratic polynomial receiving scheme/ Fokker-Planck equation/ deflection coefficient
混合强噪声的正弦信号时域、频域波形分别见图2(a)和图2(b), 经SR系统增强输出后的时域、频域波形分别见图2(c)和图2(d). 当输入SNR = –18 dB时, 时域图(图2(a))和频域图(图2(b))呈现出杂乱的、无规律的, 无法看到1 kHz正弦信号分量的任何特征; 然而经SR系统处理后, 时域图(图2(c))出现周期性特征, 说明存在周期分量信号; 通过频域图(图2(d))观察到1 kHz出现明显的信号分量(采样点数20000个, 频率分辨率10 Hz, 所以峰值出现在99.95处), 且输出全局SNR为–14.0957 dB, 提高了3.9043 dB. 这是因为SR单元对信号的处理可相当于非线性低通滤波, 会增强低频区的某些频率分量, 减弱其他的频率分量. 但与线性滤波器只改变输出频谱结构, 不改变输出各频率的概率密度的性质不同的是, SR单元在改变频谱结构的同时, 也改变了输出粒子的概率密度(见图3). 经过SR系统后, 平坦分布的高斯白噪声将向低频区聚集, 使低频区能量变大, 和低频正弦信号一起驱动粒子在双稳态势垒之间跃迁, 时域信号出现一定的周期特性, 频域观察更显著, 改变了含噪信号的频谱结构, 宏观上表现为SNR增大. 图 2 正弦信号经SR系统增强前后的时频域波形(输入SNR = –18 dB, 噪声功率${\sigma ^2} = 4$, 信号幅度$A = 0.25$, SR系统参数$a = 1 \times {10^4}$, $b = 3.3856 \times {10^{12}}$) (a) 输入信号时域波形; (b)输入信号频域幅值谱; (c)输出信号时域波形; (d)输出信号频域幅值谱 Figure2. Waveform of time and frequency zone of sine wave enhanced by SR (input SNR = –18 dB, the noise intensity${\sigma ^2} = 4$, signal amplitude $A = 0.25$, parameters of system $a = 1 \times {10^4}$, $b = 3.3856 \times {10^{12}}$): (a) The waveform of input signal in time zone; (b) the amplitude of input signal in frequency zone; (c) the waveform of output signal in time zone; (d) the amplitude of output signal in frequency zone
图 3 粒子处于不同位置时的概率密度(输入SNR = –14 dB dB, 噪声功率${\sigma ^2} = 4$, 信号幅度$A = 0.4$, SR系统参数$a = 1 \times {10^4}$, $b = 2.6406 \times {10^{12}}$) (a)未经SR处理的粒子的分布概率; (b)经SR处理后粒子的分布概率; (c)经SR处理后粒子的分布概率局部图 Figure3. Probability density function of particles of SR (input SNR = –14 dB, the noise intensity${\sigma ^2} = 4$, signal amplitude $A = 0.4$, parameters of system $a = 1 \times {10^4}$, $b = 2.6406 \times {10^{12}}$): (a) The probability density of particles before SR processed; (b) the probability density of particles after SR processed; (c) the partial of probability density of particles after SR processed
不同判决点数N时系统在判决时刻的输出仿真值如图4所示, 图4(a)为单次检验统计量$g\left( {{x_0}} \right)$的波形, 图4(b)为$g\left( {{x_0}} \right)$的多次(N = 10)平均波形. 可以看出不同N时, 对于检验统计量$g\left( {{x_0}} \right)$的发散程度不同. 这是因为判决时刻时, 粒子的运动大部分集中于势阱内, 其位置由势阱决定; 但同时受到噪声影响, 粒子不会在势阱内静止不动, 而是在一定范围内抖动, 出现图4(a)中情形, 其平衡位置就是势阱的大概位置. 图4(b)为经过N次平均之后的波形, 进一步体现了平衡位置所在, 同时减小了随机变量$g\left( {{x_0}} \right)$的抖动范围. 从而证明了N次平均统计量更有利于提高判决准确性. 图 4 不同N时$g\left( {{x_0}} \right)$的输出值(输入SNR = –18 dB, 噪声功率${\sigma ^2} = 4$, 信号幅度$A = 0.25$, SR系统参数$a = 1 \times {10^4}$, $b = 3.3856 \times {10^{12}}$) (a) N = 1时检验统计量的时域波形; (b) N = 10时检验统计量的时域波形 Figure4. Output of $g\left( {{x_0}} \right)$ at different N (input SNR = –18 dB, the noise intensity${\sigma ^2} = 4$, signal amplitude $A = 0.25$, parameters of system $a = 1 \times {10^4}$, $b = 3.3856 \times {10^{12}}$): (a) The waveform of test statistics when N = 1; (b) the waveform of test statistics when N = 10
不同判决点数N时系统的输出概率密度理论值和仿真值如图5所示, 理论值为由$E\left[ {g\left( {{x_0}} \right)} \right]$和$\sigma \left[ {g\left( {{x_0}} \right)} \right]$确定的高斯分布, 仿真值为系统的输出频率. 从图5(a)和图5(b)可以看出, 在N不符合中心极限定理条件时, 系统输出的频率与高斯分布的概率密度相差较大; 从图5(c)和图5(d)可以看出, 当N > 50时, 符合中心极限定理条件, 系统输出的频率与相应的高斯分布符合较好; 且随着N的增大, 系统输出的方差${{{\sigma ^2}} / N}$减小, 粒子的聚集性更加集中, 在均值附近出现的概率更大, 更加有利于区分两种不同的假设, 理论和仿真均说明这一点. 图 5 不同N时g(x0)的输出概率密度(输入SNR = –14 dB, 噪声功率${\sigma ^2} = 4$, 信号幅度A = 0.4, SR系统参数$a = 1 \;\times $ 104, $b = 2.6406 \times {10^{12}}$, $g\left( x \right) = {x^2} + 0.0701x$) (a) N = 1时粒子的分布概率; (b) N = 10时粒子的分布概率; (c) N = 50时粒子的分布概率; (d) N = 100时粒子的分布概率 Figure5. Output probability density function of $g\left( {{x_0}} \right)$ at different N (input SNR = –14 dB, the noise intensity ${\sigma ^2} = 4$, signal amplitude $A = 0.4$, parameters of system $a = 1 \times $ 104, $b = 2.6406 \times {10^{12}}$, $g\left( x \right) = {x^2} + 0.0701x$): (a) The output probability density when N = 1; (b) the output probability density when N = 10; (c) the output probability density when N = 50; (d) the output probability density when N = 100
24.4.不同接收结构的系统输出误码率 -->
4.4.不同接收结构的系统输出误码率
不同接收结构的系统输出误码率如图6所示. 可以看到, 直接N次累积平均接收结构的误码率最大, 其次是能量接收结构, 二次多项式$g\left( x \right)$接收结构的误码率最小; 验证了二次多项式接收结构的性能优于能量接收和直接累积平均接收结构, 但三者性能相差不大. 这是因为N次累积平均接收结构与包络接收一致, 而能量接收与包络接收性能相差无几. 根据“3.2节 接收算法流程”中第二步: “根据(10)式和(14)式确定$g\left( x \right)$的系数${l_1}$, ${l_2}$”, 同时依据偏移系数d的线性偏移不变性, 可确定此时二次多项式$g\left( x \right)$接收结构的最优系数为: 二次项系统为1, 一次项系数分别为[0.4616, 0.3312, 0.2291, 0.1421, 0.1002, 0.0701], 常数项无影响, 一次项系数较二次项系统相差较大, 起主要作用的是二次项系数, 所以和能量接收性能相差不大. 同时, 不同接收结构主要改变的是不同假设情况下系统输出的期望, 对于方差影响不大; 而在高斯分布时系统输出的误码率主要由方差决定; 所以三者接收性能差别不大. 图 6 不同接收结构的系统输出误码率 Figure6. Output bit error ratio of different receiving structure