算法原理
训练序列结构 T=[A A],其中A表示复伪随机序列PN,进行N/2点ifft变换得到的符号序列
\[M(d)=\frac{\left | P(d) \right |}{R^{2}(d)}^{2}\]
\[P(d)=\sum_{m=0}^{L-1}r^{*}(d+m) r(d+m+L)\]
\[R(d)=\sum_{m=0}^{L-1}\left | r(d+m+L) \right |^{2}\]
\[L=N/2\]
所求得的d对应的是训练序列(不包含循环前缀)的开始位置。
★Schmidl:Schmidl算法利用一个由两端时域上完全相同的序列的前导来进行定时同步,但是这种方法得到的同步效果并不好,其同步度量函数曲线存在一个平顶,这使得定时同步估计存在偏差和不确定性。
参考文献
Schmidl T M,COX D C.Robust frequency and timing synchronization for OFDM[J].IEEE Trans.Commun.,1997,45(12):1613-1612.
%********************schmidl algorithm******************* %Example: % If % X = rand(2,3,4); % then % d = size(X) returns d = [2 3 4] % [m1,m2,m3,m4] = size(X) returns m1 = 2, m2 = 3, m3 = 4, m4 = 1 % [m,n] = size(X) returns m = 2, n = 12 % m2 = size(X,2) returns m2 = 3 close all; clear all; clc; %参数定义 N=256; %FFT/IFFT 变换的点数或者子载波个数(Nu=N) Ng=N/8; %循环前缀的长度 (保护间隔的长度) Ns=Ng+N; %包括循环前缀的符号长度 %************利用查表法生成复随机序列********************** QAMTable=[7+7i,-7+7i,-7-7i,7-7i]; buf=QAMTable(randi([0,3],N/2,1)+1); %加1是为了下标可能是0不合法 %*************在奇数子载波的位置插入零*********************zj:是偶数吧? x=zeros(N,1); index = 1; for n=1:2:N x(n)=buf(index); index=index+1; end; %**************利用IFFT变换生成Schmidl训练符号*************** sch = ifft(x); %[A A]的形式 %*****************添加一个空符号以及一个后缀符号************* src = QAMTable(randi([0,3],N,1)+1).‘; sym = ifft(src); sig =[zeros(N,1) sch sym]; %**********************添加循环前缀************************* tx =[sig(N - Ng +1:N,:);sig]; %***********************经过信道*************************** recv = reshape(tx,1,size(tx,1)*size(tx,2)); %size的1表示行,2表示列,从%前向后数,超过了为1 %recv1 = awgn(recv,1,‘measured‘); %recv2 = awgn(recv,5,‘measured‘); %recv3 = awgn(recv,10,‘measured‘); %*****************计算符号定时***************************** P=zeros(1,2*Ns); R=zeros(1,2*Ns); %P1=zeros(1,2*Ns); %R1=zeros(1,2*Ns); P2=zeros(1,2*Ns); R2=zeros(1,2*Ns); %P3=zeros(1,2*Ns); %R3=zeros(1,2*Ns); for d = Ns/2+1:1:2*Ns for m=0:1:N/2-1 P(d-Ns/2) = P(d-Ns/2) + conj(recv(d+m))*recv(d+N/2+m); R(d-Ns/2) = R(d-Ns/2) + power(abs(recv(d+N/2+m)),2); %P1(d-Ns/2) = P1(d-Ns/2) + conj(recv1(d+m))*recv1(d+N/2+m); %R1(d-Ns/2) = R1(d-Ns/2) + power(abs(recv1(d+N/2+m)),2); %P2(d-Ns/2) = P2(d-Ns/2) + conj(recv2(d+m))*recv2(d+N/2+m); %R2(d-Ns/2) = R2(d-Ns/2) + power(abs(recv2(d+N/2+m)),2); % P3(d-Ns/2) = P3(d-Ns/2) + conj(recv3(d+m))*recv3(d+N/2+m); % R3(d-Ns/2) = R3(d-Ns/2) + power(abs(recv3(d+N/2+m)),2); end end M=power(abs(P),2)./power(abs(R),2); %M1=power(abs(P1),2)./power(abs(R1),2); %M2=power(abs(P2),2)./power(abs(R2),2); %M3=power(abs(P3),2)./power(abs(R3),2); %**********************绘图****************************** figure(‘Color‘,‘w‘); d=1:1:400; figure(1); plot(d,M(d)); grid on; axis([0,400,0,1.1]); title(‘schmidl algorithm‘); xlabel(‘Time (sample)‘); ylabel(‘Timing Metric‘); %legend(‘no noise‘,‘SNR=1dB‘,‘SNR=5dB‘,‘SNR=10dB‘); hold on;
原文:https://www.cnblogs.com/jiandahao/p/9310924.html