Add caching on recurring calculations

Using a LRU cache. For now uses static parameters, in the future some configuration should be added.

Author: Eduardo Soares

commpy/channelcoding/convcode.py

@@ -5,6 +5,7 @@
 
 from __future__ import division
 
+import functools
 import math
 from warnings import warn
 
@@ -570,6 +571,19 @@ def _where_c(inarray, rows, cols, search_value, index_array):
     return number_found
 
 
+@functools.lru_cache(maxsize=128, typed=False)
+def _compute_branch_metrics(decoding_type, r_codeword, i_codeword_array):
+    if decoding_type == 'hard':
+        return hamming_dist(r_codeword.astype(int), i_codeword_array.astype(int))
+    elif decoding_type == 'soft':
+        neg_LL_0 = np.log(np.exp(r_codeword) + 1)  # negative log-likelihood to have received a 0
+        neg_LL_1 = neg_LL_0 - r_codeword  # negative log-likelihood to have received a 1
+        return np.where(i_codeword_array, neg_LL_1, neg_LL_0).sum()
+    elif decoding_type == 'unquantized':
+        i_codeword_array = 2 * i_codeword_array - 1
+        return euclid_dist(r_codeword, i_codeword_array)
+
+
 def _acs_traceback(r_codeword, trellis, decoding_type,
                    path_metrics, paths, decoded_symbols,
                    decoded_bits, tb_count, t, count,
@@ -605,15 +619,7 @@ def _acs_traceback(r_codeword, trellis, decoding_type,
             i_codeword_array = dec2bitarray(i_codeword, n)
 
             # Compute Branch Metrics
-            if decoding_type == 'hard':
-                branch_metric = hamming_dist(r_codeword.astype(int), i_codeword_array.astype(int))
-            elif decoding_type == 'soft':
-                neg_LL_0 = np.log(np.exp(r_codeword) + 1)  # negative log-likelihood to have received a 0
-                neg_LL_1 = neg_LL_0 - r_codeword  # negative log-likelihood to have received a 1
-                branch_metric = np.where(i_codeword_array, neg_LL_1, neg_LL_0).sum()
-            elif decoding_type == 'unquantized':
-                i_codeword_array = 2*i_codeword_array - 1
-                branch_metric = euclid_dist(r_codeword, i_codeword_array)
+            branch_metric = _compute_branch_metrics(decoding_type, tuple(r_codeword), tuple(i_codeword_array))
 
             # ADD operation: Add the branch metric to the
             # accumulated path metric and store it in the temporary array

commpy/utilities.py

@@ -17,6 +17,7 @@
    upsample             -- Upsample by an integral factor (zero insertion).
    signal_power         -- Compute the power of a discrete time signal.
 """
+import functools
 
 import numpy as np
 
@@ -47,13 +48,14 @@ def dec2bitarray(in_number, bit_width):
     """
 
     if isinstance(in_number, (np.integer, int)):
-        return decimal2bitarray(in_number, bit_width)
+        return decimal2bitarray(in_number, bit_width).copy()
     result = np.zeros(bit_width * len(in_number), np.int8)
     for pox, number in enumerate(in_number):
-        result[pox * bit_width:(pox + 1) * bit_width] = decimal2bitarray(number, bit_width)
+        result[pox * bit_width:(pox + 1) * bit_width] = decimal2bitarray(number, bit_width).copy()
     return result
 
 
+@functools.lru_cache(maxsize=128, typed=False)
 def decimal2bitarray(number, bit_width):
     """
     Converts a positive integer to NumPy array of the specified size containing bits (0 and 1). This version is slightly