Module polarcodes.decoder_utils
Expand source code
import numpy as np
from polarcodes.utils import *
def hard_decision(y):
"""
Hard decision of a log-likelihood.
"""
if y >= 0:
return 0
else:
return 1
def upper_llr(l1, l2):
"""
Update top branch LLR in the log-domain.
This function supports shortening by checking for infinite LLR cases.
Parameters
----------
l1: float
input LLR corresponding to the top branch
l2: float
input LLR corresponding to the bottom branch
Returns
----------
float
the top branch LLR at the next stage of the decoding tree
"""
# check for infinite LLR, used in shortening
if l1 == np.inf and l2 != np.inf:
return l2
elif l1 != np.inf and l2 == np.inf:
return l1
elif l1 == np.inf and l2 == np.inf:
return np.inf
else: # principal decoding equation
return logdomain_sum(l1 + l2, 0) - logdomain_sum(l1, l2)
def lower_llr(l1, l2, b):
"""
Update bottom branch LLR in the log-domain.
This function supports shortening by checking for infinite LLR cases.
Parameters
----------
l1: float
input LLR corresponding to the top branch
l2: float
input LLR corresponding to the bottom branch
b: int
the decoded bit of the top branch
Returns
----------
float, np.nan
the bottom branch LLR at the next stage of the decoding tree
"""
if b == 0: # check for infinite LLRs, used in shortening
if l1 == np.inf or l2 == np.inf:
return np.inf
else: # principal decoding equation
return l1 + l2
elif b == 1: # principal decoding equation
return l1 - l2
return np.nan
def active_llr_level(i, n):
"""
Find the first 1 in the binary expansion of i.
"""
mask = 2**(n-1)
count = 1
for k in range(n):
if (mask & i) == 0:
count += 1
mask >>= 1
else:
break
return min(count, n)
def active_bit_level(i, n):
"""
Find the first 0 in the binary expansion of i.
"""
mask = 2**(n-1)
count = 1
for k in range(n):
if (mask & i) > 0:
count += 1
mask >>= 1
else:
break
return min(count, n)Functions
def active_bit_level(i, n)-
Find the first 0 in the binary expansion of i.
Expand source code
def active_bit_level(i, n): """ Find the first 0 in the binary expansion of i. """ mask = 2**(n-1) count = 1 for k in range(n): if (mask & i) > 0: count += 1 mask >>= 1 else: break return min(count, n) def active_llr_level(i, n)-
Find the first 1 in the binary expansion of i.
Expand source code
def active_llr_level(i, n): """ Find the first 1 in the binary expansion of i. """ mask = 2**(n-1) count = 1 for k in range(n): if (mask & i) == 0: count += 1 mask >>= 1 else: break return min(count, n) def hard_decision(y)-
Hard decision of a log-likelihood.
Expand source code
def hard_decision(y): """ Hard decision of a log-likelihood. """ if y >= 0: return 0 else: return 1 def lower_llr(l1, l2, b)-
Update bottom branch LLR in the log-domain. This function supports shortening by checking for infinite LLR cases.
Parameters
l1:float- input LLR corresponding to the top branch
l2:float- input LLR corresponding to the bottom branch
b:int- the decoded bit of the top branch
Returns
float, np.nan- the bottom branch LLR at the next stage of the decoding tree
Expand source code
def lower_llr(l1, l2, b): """ Update bottom branch LLR in the log-domain. This function supports shortening by checking for infinite LLR cases. Parameters ---------- l1: float input LLR corresponding to the top branch l2: float input LLR corresponding to the bottom branch b: int the decoded bit of the top branch Returns ---------- float, np.nan the bottom branch LLR at the next stage of the decoding tree """ if b == 0: # check for infinite LLRs, used in shortening if l1 == np.inf or l2 == np.inf: return np.inf else: # principal decoding equation return l1 + l2 elif b == 1: # principal decoding equation return l1 - l2 return np.nan def upper_llr(l1, l2)-
Update top branch LLR in the log-domain. This function supports shortening by checking for infinite LLR cases.
Parameters
l1:float- input LLR corresponding to the top branch
l2:float- input LLR corresponding to the bottom branch
Returns
float- the top branch LLR at the next stage of the decoding tree
Expand source code
def upper_llr(l1, l2): """ Update top branch LLR in the log-domain. This function supports shortening by checking for infinite LLR cases. Parameters ---------- l1: float input LLR corresponding to the top branch l2: float input LLR corresponding to the bottom branch Returns ---------- float the top branch LLR at the next stage of the decoding tree """ # check for infinite LLR, used in shortening if l1 == np.inf and l2 != np.inf: return l2 elif l1 != np.inf and l2 == np.inf: return l1 elif l1 == np.inf and l2 == np.inf: return np.inf else: # principal decoding equation return logdomain_sum(l1 + l2, 0) - logdomain_sum(l1, l2)