rs-fec-conv 0.4.5

Convolutional Encoder and Viterbi Decoder built with Rust bindings for integration with Python

rs_fec_conv

Getting Started

The package rs_fec_conv is a rust binding built with pyo3. rs_fec_conv is intended to be used in parallel with the scikit-dsp-comm package. The rust binding improve the processing time of the conv_encoder and viterbi_decoder algorithms.

Rust Install

Rust is not needed on the system to execute the binaries since the functions are already pre-compiled. Although, Rust can be downloaded online or installed on Windows Subsystem for Linux.

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

Package Requirements

This package requires Python 3.7.x.

rs_fec_conv Install

You can download the package rs_fec_conv from PyPi PyPi, or by the command

pip install rs_fec_conv

Note: The preferred method is to download from PyPi. If downloading directly from GitHub you will need to install Rust prior.

Results

BEP Simulation (EbN0=4,100000 bits) G, depth	Python Time (sec)	Rust Time (sec)	Rust Speed Factor Increase
('111', '101'), 10	39.88	0.79	50.24
('11111','11011','10101'), 25	675.00	21.32	31.66
('1111001','1011011'), 25	217.02	9.27	23.41

Tutorial

Convolutional Encoder

The function conv_encoder_rs can be implemented

import numpy as np
import matplotlib.pyplot as plt
import sk_dsp_comm.rs_fec_conv as rs_fec

# Generate random data
N = 20
x = randint(0,2,N)

# Initialize fec_conv object with either G length 2 or 3
G =('111','101')
# G = ('11110111','11011001','10010101')
cc1 = rs_fec.fec_conv(G,10)
state = '00'

# Convolutionally Encode Signal
y,state = cc1.conv_encoder(x,state)

# Plot input signal
subplot(211)
stem(x)
xlabel('Number of Samples')
ylabel('x')
title('Input Signal')

# Plot convolutionally encoded signal
subplot(212)
stem(y)
xlabel('Number of Samples')
ylabel('y')
title('Convolutionally Encoded Signal')
tight_layout()
savefig('conv_enc.png')

Viterbi Decoder

The function viterbi_decoder_rs can be implemented by

# Viterbi decode
z = cc1.viterbi_decoder(y.astype(int), 'hard', 3)

# Plot input signal
subplot(211)
stem(x[:11])
xlabel('Number of Samples')
ylabel('x')
title('Input Signal')
xlim([0,10])

# Plot viterbi decoded signal
subplot(212)
stem(z)
xlabel('Number of Samples')
ylabel('z')
title('Viterbi decoded Signal')
xlim([0,10])
tight_layout()
savefig('viterbi_dec.png')

Since there is no channel noise added to the signal the Viterbi decoded signal results in no bit errors from the original signal.

Channel Simulation

A simulation using AWGN can be done using by integrating with other functions provided in the scikit-dsp-comm toolbox

# Soft decision rate 1/2 simulation
N_bits_per_frame = 100000
EbN0 = 4
total_bit_errors = 0
total_bit_count = 0
cc1 = rs_fec.fec_conv(('11101','10011'),25)

# Encode with shift register starting state of '0000'
state = '0000'
while total_bit_errors < 100:
	# Create 100000 random 0/1 bits
	x = randint(0,2,N_bits_per_frame)
	y,state = cc1.conv_encoder(x,state)

	# Add channel noise to bits, include antipodal level shift to [-1,1]
	# Channel SNR is 3 dB less for rate 1/2
	yn_soft = dc.cpx_AWGN(2*y-1,EbN0-3,1) 
	yn_hard = ((np.sign(yn_soft.real)+1)/2).astype(int)
	z = cc1.viterbi_decoder(yn_hard,'hard')

	# Count bit errors
	bit_count, bit_errors = dc.bit_errors(x,z)
	total_bit_errors += bit_errors
	total_bit_count += bit_count
	print('Bits Received = %d, Bit errors = %d, BEP = %1.2e' %\
		  (total_bit_count, total_bit_errors,\
		   total_bit_errors/total_bit_count))

print('*****************************************************')
print('Bits Received = %d, Bit errors = %d, BEP = %1.2e' %\
	  (total_bit_count, total_bit_errors,\
	   total_bit_errors/total_bit_count))

Rate 1/2 Object

kmax = 0, taumax = 0

Bits Received = 99976, Bit errors = 845, BEP = 8.45e-03

---

Bits Received = 99976, Bit errors = 845, BEP = 8.45e-03