作者:桂。
时间:2017年12月19日20:43:04
链接:http://www.cnblogs.com/xingshansi/p/8067839.html
前言
主要记录基本的频分复用原理,以及仿真实现。
一、频分复用原理
频分复用FDM:
通常x1..4(t)可以是同一个序列的串并转化,也可以是不同序列,频分复用示意图:
主要包含三个操作:1)上采样(up-sample); 2)滤波(fir);3)累加(sum)。
频分复用:将多个不同频段的信号拼接为一个宽带信号,主要包含三个操作:1)上采样(up-sample); 2)滤波(fir);3)累加(sum)。
- 上采样
T1/T2 = 4,故上采样倍数为4,上采样有原数据保持、插值、补零等方法,这里采用最基本的补零方法。不失一般性,X0(n)、X1(n)、X2(n)、X3(n)分别按不同频率的正弦信号处理。
以x1(n) 的频域变换:
4倍插值后的频谱:
可以看出插值后出现了多个重复周期,因此需要借助低通滤波以保留单一周期,如下图所示:
因此需要构造不同频段的滤波器,四个蓝色阴影部分拼接(累加)即可。
- 滤波器
这里prototype滤波器:
共构造8个滤波器,分成四组,输出y(n)为:
Y(n) = y0(n)+ y1(n)+ y2(n)+ y3(n)
ym(n):
Ym(n) = xm_interpl(n)*[ha (n) exp((-im*2*pi*(m*n))/8)+ ha(n) exp((-im*2*pi*((8-m)*n))/8)]
= 2*xm_interpl(n)*[ha(n)cos((-2*pi*(m*n))/8)]
其中ha (n) = h(n)* exp((-im*pi*n)/8)为prototype filter,至此便完成了信号的频分多路复用(FDM)理论推导.
- 累加
滤波后的各个输出累加,即得到调制的y(n),仿真图如图所示:
结果与上文一致。
二、仿真结果
频分复用的接收端是发射的逆过程,分别利用 基本滤波器、多相滤波器实现:
基本滤波器:
%recovery signal: x clc;clear all;close all; load fir2.mat; fir = fir2; B = 4000;%4KHz fs1 = 2*B; D = 4; t1 = 0:1/fs1:(128-1)/fs1; f = [800 1600 2200 2800];%frequency x0 = sin(2*pi*t1*f(1)); x1 = sin(2*pi*t1*f(2)); x2 = sin(2*pi*t1*f(3)); x3 = sin(2*pi*t1*f(4)); x_shape = [x0;x1;x2;x3]; %% interp x0_interp = [x0;zeros(3,length(t1))]; x0_interp = x0_interp(:)'; x1_interp = [x1;zeros(3,length(t1))]; x1_interp = x1_interp(:)'; x2_interp = [x2;zeros(3,length(t1))]; x2_interp = x2_interp(:)'; x3_interp = [x3;zeros(3,length(t1))]; x3_interp = x3_interp(:)'; %%prototype filter x_all = [x0_interp;x1_interp;x2_interp;x3_interp;flipud([x0_interp;x1_interp;x2_interp;x3_interp])]; im = sqrt(-1); iseq = 1:length(fir); for j = 1:D h_channel(j,:) = fir.*cos((2*pi*((j-1/2)*(iseq-1)))/8); % h_channel(j,:) = fir.*exp((1j*2*pi*((j-1/2)*(iseq-1)))/8); end %%add signal yn = zeros(1,length(x3_interp)); for i = 1:D yn = filter(h_channel(i,:),1,x_all(i,:))+yn; end %%demultiplex x_channel = zeros(D,length(yn)/D); for i = 1:D x_channel(i,:) = downsample(filter(h_channel(i,:),1,yn),D); end figure() for i = 1:D subplot(2,2,i) plot(linspace(0,fs1,length(t1)),abs(fft(x_channel(i,:)))); xlabel('frequency(Hz)');ylabel('amplitude');title('direct filter -> x'); end %%plot mse figure() for i = 1:4 x_channel(i,:) = x_channel(i,:)/max(abs( x_channel(i,:))); subplot (2,2,i) plot(linspace(0,fs1,length(t1)),x_channel(i,:));hold on; plot(linspace(0,fs1,length(t1)),x_shape(i,:),'r--');hold on; % plot(linspace(0,fs1,length(t1)),abs(x_shape(i,:)-x_channel(i,:)).^2,'k'); xlabel('frequency(Hz)');title('MSE'); % legend('recovery','orignal','MSE'); end
多相滤波器,推导:
令l = iD+p,D表示分解后信号路数,此处D = 4:
再将结果取实部即可得解。
%recovery signal by polyphase filter: x clc;clear all;close all; load fir2.mat; fir = fir2; B = 4000;%4KHz fs1 = 2*B; D = 4; t1 = 0:1/fs1:(128-1)/fs1; f = [800 1600 2200 2800];%frequency x0 = sin(2*pi*t1*f(1)); x1 = sin(2*pi*t1*f(2)); x2 = sin(2*pi*t1*f(3)); x3 = sin(2*pi*t1*f(4)); x_shape = [x0;x1;x2;x3]; %% interp x0_interp = [x0;zeros(3,length(t1))]; x0_interp = x0_interp(:)'; x1_interp = [x1;zeros(3,length(t1))]; x1_interp = x1_interp(:)'; x2_interp = [x2;zeros(3,length(t1))]; x2_interp = x2_interp(:)'; x3_interp = [x3;zeros(3,length(t1))]; x3_interp = x3_interp(:)'; %%prototype filter x_all = [x0_interp;x1_interp;x2_interp;x3_interp;flipud([x0_interp;x1_interp;x2_interp;x3_interp])]; im = sqrt(-1); iseq = 1:length(fir); for j = 1:D h_channel(j,:) = fir.*cos((-2*pi*((j-1/2)*(iseq-1)))/8); % h_channel(j,:) = fir.*exp((1j*2*pi*((j-1/2)*(iseq-1)))/8); end %%add signal yn = zeros(1,length(x3_interp)); for i = 1:D yn = filter(h_channel(i,:),1,x_all(i,:))+yn; end %%demultiplex %prototype filter h0 = fir.*exp((-1j*2*pi*((-1/2)*(iseq-1)))/8); h_py = fliplr(reshape(h0,D,length(h0)/D)); y_py = (reshape(yn,D,length(yn)/D)); x_channel = zeros(D,length(yn)/D); for i = 1:D x_channel(i,:) = filter(h_py(i,:),1,y_py(i,:)); end x_channel = real(ifft(x_channel)); x_channel = x_channel([1,4,2,3],:); %%plot mse figure() for i = 1:4 x_channel(i,:) = x_channel(i,:)/max(abs( x_channel(i,:))); subplot (2,2,i) plot(linspace(0,fs1,length(t1)),x_channel(i,:));hold on; plot(linspace(0,fs1,length(t1)),x_shape(i,:),'r--');hold on; % plot(linspace(0,fs1,length(t1)),abs(x_shape(i,:)-x_channel(i,:)).^2,'k'); xlabel('frequency(Hz)');title('MSE'); % legend('recovery','orignal','MSE'); end
三、其他
原型滤波器信道化思路:
信道化与频分复用略有不同,频分复用主要是余弦函数,理论上相邻无衰减,得到的余弦曲线并不理想:
当有一定的过渡带时,余弦曲线:
可见此时应该有一个过渡带才更加合理,而不是像信道化体系常用的约束:相邻信道无缝连接。