QCLDPC.py

import math
import matplotlib.pyplot as plt
import numpy as np
from QCEncode import QCEncode
from LDPCDecode import LDPCDecode
from QCLDPCBaseH import QCLDPCBaseH
from util import zero

CodeLength=576
CodeRate=1/2
qcH,Hb=QCLDPCBaseH(CodeLength,CodeRate) #generate check matrix qcH
row,col=qcH.shape
rowb,colb=Hb.shape
block=CodeLength/colb
NewsLength = int(CodeRate*CodeLength) #length of each weight to be encoded
print(' The unit message length is:'+ str(NewsLength))

# Function to clear the kth bit of n, k start from 1, ex: '10101' has k = 1 -> '10100' )
def clear_nBit(num, n):
    # num and n must be integer
    return (num >> n) << n

clear_nBit_np = np.vectorize(clear_nBit)

def Encode(InputData):#InputData is 9 model weights, each weight is 32 bits
    # InputData = np.array[3213,685,6513,351,351,3513,51,351,35135]

    DataOri = np.trunc(InputData*100000).astype(int) #As the clear_nBit function needs interger input so here I change InputData to interger first. 
    Pn=10 # Set least significant Pn bit to 0
    DataPn = clear_nBit_np(DataOri, Pn) 

    PerData=np.trunc(np.zeros(NewsLength)).astype(int) #initialize PerData
    PerData=bin(DataPn)  #PerData should be binary sequence of 288 bits
    
    #partial encoding
    PerCode=np.zeros(CodeLength)
    PerCode=QCEncode(PerData,qcH,Hb)#PerData should be binary sequence of 288 bits
    
    #through the channel
    snr=2
    SNR=10^(snr/10)
    sigma=math.sqrt(1/(2*CodeRate*SNR))
    Noise=np.random.randn(PerCode.shape)*sigma
    ChannelNoise=(-(2*PerCode-1))+Noise
    return ChannelNoise


def bitExtracted(number, k, p):  
    return (((1 << k) - 1)  &  (number >> (p-1)))


def separate_encoded_data(encoded_data, CodeRate, CodeLength):
    num_1 = CodeRate*CodeLength
    num_2 = CodeLength - num_1
    Codeword1 = bitExtracted(encoded_data,num_1, num_2+1)
    Codeword2 = bitExtracted(encoded_data,num_2, 1)
    return Codeword1,Codeword2

def concat_binary(Codeword1,Codeword2, CodeRate, CodeLength):
    num_2 = CodeLength*(1-CodeRate)
    return Codeword1 << int(num_2) | Codeword2




def Decode(encoded_data):
    CodeNum=len(encoded_data) #number of weights

    ITERNum=10 #Decoding iterations

   # for i in range(CodeNum) :

    snr=2
    SNR=10^(snr/10)
    sigma=math.sqrt(1/(2*CodeRate*SNR))
        
    #BP decoding
    Codeword1,Codeword2 = separate_encoded_data(encoded_data)
    codeword1_tail = bitExtracted(Codeword1, 5, 1)
    Codeword1= clear_nBit(Codeword1,5) #change the least significant 5 bits of first part of OutputChannelOri to 0
    OutputDecode = concat_binary(Codeword1, Codeword2, CodeRate, CodeLength)
    BPOutputdecodeOri=LDPCDecode(OutputDecode,rowb,colb,block,qcH,sigma,ITERNum)
    BPOutputdecode = clear_nBit(BPOutputdecodeOri,5) | codeword1_tail #change the least significant 5 bits of first part of BPOutputdecodeOri to its noisy type

    return BPOutputdecode