python 实现基于16QAM的OFDM模拟

未完待续…先发过来存一下,防止丢掉(很可能有错误的地方,自己对这些理解也很浅显)

import numpy as np
import random
import pylab as plt

n = 2000                                                                 #生成二进制数的数量
_db_ = 17                                                                #信噪比db
bit_list = []
for i in range(n):
    bit_list.append(random.randint(0,1))

input = np.array(bit_list)
reshaped_input = np.reshape(input,(int(n/4),4))

def _trans_(_a, _b, _c, _d):                                          
    return 8*_a + 4*_b + 2*_c + _d

_map_ = dict([('0',(3,3)),('1',(1,3)),('2',(-3,3)),('3',(-1,3)),        
              ('7',(-1,1)),('6',(-3,1)),('5',(1,1)),('4',(3,1)),
              ('8',(3,-3)),('9',(1,-3)),('10',(-3,-3)),('11',(-1,-3)),
              ('15',(-1,-1)),('14',(-3,-1)),('13',(1,-1)),('12',(3,-1))])

qam_output_list = []                     
    
for i in range(int(n/4)):
    num = _trans_(reshaped_input[i][0],reshaped_input[i][1],reshaped_input[i][2],reshaped_input[i][3]) 

    transed_num = complex(_map_[str(num)][0],_map_[str(num)][1])        #complex函数转换成复数形式,代表I Q两路
    qam_output_list.append(transed_num)

qam_output_array = np.array(qam_output_list)                            #相当于完成了串并转换

qam_output_real_array = np.zeros(int(n/4))
qam_output_imag_array = np.zeros(int(n/4))

for i in range(int(n/4)):
    qam_output_real_array[i] = qam_output_array[i].real
    qam_output_imag_array[i] = qam_output_array[i].imag
    
ifft_output = np.fft.ifft(qam_output_array)                            #傅里叶逆变换

real_array = np.zeros(int(n/4))                                        #存储实部
imag_array = np.zeros(int(n/4))                                        #存储虚部

for i in range(int(n/4)):
    real_array[i] = ifft_output[i].real
    imag_array[i] = ifft_output[i].imag

def wgn(x, snr):                                   
    snr = 10**(snr/10.0)
    xpower = np.sum(x**2)/len(x)
    npower = xpower / snr
    return np.random.randn(len(x)) * np.sqrt(npower)

real_array_ = real_array + wgn(real_array,_db_)                          #加入高斯白噪声后的信号
imag_array_ = imag_array + wgn(imag_array,_db_)

fft_input_list = []

for i in range(int(n/4)):                                                #将加入噪声后的信号转换为复数形式
    transed_num_ = complex(real_array_[i],imag_array_[i])
    fft_input_list.append(transed_num_)
fft_input_array = np.array(fft_input_list)

fft_output = np.fft.fft(fft_input_array)                                  #fft

i_output = np.zeros(int(n/4))
q_output = np.zeros(int(n/4))

for i in range(int(n/4)):
    i_output[i] = fft_output[i].real
    q_output[i] = fft_output[i].imag    

plt.figure(figsize=(10,16))     
plt.subplot(211)
plt.scatter(qam_output_real_array[0:int(n/4)],qam_output_imag_array[0:int(n/4)])
plt.title('plot img 1')
plt.subplot(212)
plt.scatter(i_output[0:int(n/4)],q_output[0:int(n/4)])
plt.title('plot img 2')
plt.show()

nn = np.array([(-3,-3),(-3,-1),(-1,-3),(-1,-1),(3,3),(3,1),(1,3),(1,1),
               (-3,3),(-3,1),(-1,3),(-1,1),(3,-3),(3,-1),(1,-3),(1,-1)])

for _n_ in range(100):
    _i_ = i_output[_n_]
    _q_ = i_output[_n_]
    
    d_list = []
    for i in range(16):
        d = (_i_ - nn[i][0])**2 + (_q_ - nn[i][1])**2
        d_list.append(d)
        
    print(d_list.index(min(d_list)))
评论 4
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值