Some SM functions

still need testing + examples
BruninLima · Nov 10, 2021 · 3920789 · 3920789
1 parent 0765338
commit 3920789
Show file tree

Hide file tree

Showing 7 changed files with 282 additions and 23 deletions.
diff --git a/examples/Jupyter Notebooks/LMS-example.ipynb b/examples/Jupyter Notebooks/LMS-example.ipynb
@@ -35,16 +35,6 @@
     "import pydaptivefiltering as pdf"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "bd8ad797",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "pdf.main.AdaptiveFilter.adapt_LM"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 2,
@@ -73,7 +63,7 @@
   {
    "cell_type": "code",
    "execution_count": 3,
-   "id": "128e247d",
+   "id": "eee8ad93",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -94,7 +84,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "061c17b2",
+   "id": "3027dced",
    "metadata": {},
    "source": [
     "### desired signal"
@@ -103,7 +93,7 @@
   {
    "cell_type": "code",
    "execution_count": 5,
-   "id": "4d548220",
+   "id": "78036c04",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -119,7 +109,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "1612256e",
+   "id": "2798d8d4",
    "metadata": {},
    "source": [
     "### Istanciating Adaptive Filter"
@@ -128,7 +118,7 @@
   {
    "cell_type": "code",
    "execution_count": 6,
-   "id": "85edb896",
+   "id": "dc0f4344",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -137,7 +127,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "a809666e",
+   "id": "a9b728bd",
    "metadata": {},
    "source": [
     "### Adapting with the LMS Algorithm   "
@@ -146,7 +136,7 @@
   {
    "cell_type": "code",
    "execution_count": 7,
-   "id": "026a7d78",
+   "id": "1a857ecf",
    "metadata": {},
    "outputs": [
     {
@@ -169,7 +159,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "cd20c36d",
+   "id": "6c8d3b3b",
    "metadata": {},
    "source": [
     "## Plotting"
@@ -178,7 +168,7 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "50f18f56",
+   "id": "48b87b98",
    "metadata": {},
    "outputs": [],
    "source": [

diff --git a/examples/Python Scripts/LMS based algorithms/example_system_LMS .py b/examples/Python Scripts/LMS based algorithms/example_system_LMS .py
@@ -81,21 +81,22 @@ def LMS_example():
     plt.gca().set_title('Learning Curve for MSE [dB]')
     MSE = [abs(err)**2 for err in Output['errors']]
     plt.gca().semilogy(MSE)
-
+    plt.grid()
     plt.subplot(222)
     plt.gca().set_title('Learning Curve for MSEmin [dB]')
     MSEmin = [abs(n)**2 for n in ComplexNoise]
     plt.gca().semilogy(MSEmin)
-
+    plt.grid()
     plt.subplot(223)
     plt.gca().set_title('Evolution of the Coefficients (Real Part)')
     real_part = [coef.real for coef in Output['coefficients']]
     plt.gca().plot(real_part)
-
+    plt.grid()
     plt.subplot(224)
     plt.gca().set_title('Evolution of the Coefficients (Imaginary Part)')
     imag_part = [coef.imag for coef in Output['coefficients']]
     plt.gca().plot(imag_part)
+    plt.grid()
 
     plt.tight_layout(pad=4.0)
     plt.show()
diff --git a/pydaptivefiltering/SM/AP.py b/pydaptivefiltering/SM/AP.py
@@ -0,0 +1,136 @@
+#  SM.AP.py
+#
+#      Implements the Set-membership Affine-Projection algorithm for COMPLEX valued data.
+#      (Algorithm 6.2 - book: Adaptive Filtering: Algorithms and Practical
+#                                                       Implementation, Diniz)
+#
+#      Authors:
+#       . Bruno Ramos Lima Netto        - [email protected]  & [email protected]
+#       . Guilherme de Oliveira Pinto   - [email protected] & [email protected]
+#       . Markus Vinícius Santos Lima   - mvsl20@gmailcom           & [email protected]
+#       . Wallace Alves Martins         - [email protected]     & [email protected]
+#       . Luiz Wagner Pereira Biscainho - [email protected]          & [email protected]
+#       . Paulo Sergio Ramirez Diniz    -                             [email protected]
+
+# Imports
+import numpy as np
+from time import time
+
+
+def AP(Filter, desired_signal: np.ndarray, input_signal: np.ndarray, gamma_bar: float, gamma_barVector: np.ndarray, gamma: float, memoryLength: int, verbose: bool = False) -> dict:
+    """
+    Description
+    -----------
+        Implements the Set-membership Affine-Projection algorithm for COMPLEX valued data.
+        (Algorithm 6.2 - book: Adaptive Filtering: Algorithms and Practical Implementation, Diniz)
+
+    Syntax
+    ------
+    OutputDictionary = SM.AP(Filter, desired_signal, input_signal, step, verbose)
+
+    Inputs
+    -------
+        filter  : Adaptive Filter                       filter object
+        desired : Desired signal                        numpy array (row vector)
+        input   : Input signal to feed filter           numpy array (row vector)
+        gamma_bar : Upper bound for the error modulus.
+        gamma_barVector       : Upper bound vector for the error modulus. (COLUMN vector)
+        gamma                 : Regularization factor.
+        memoryLength          : Reuse data factor.
+        verbose : Verbose boolean                       bool
+
+    Outputs
+    -------
+        dictionary:
+            outputs      : Store the estimated output of each iteration.        numpy array (collumn vector)
+            errors       : Store the error for each iteration.                  numpy array (collumn vector)
+            coefficients : Store the estimated coefficients for each iteration. numpy array (collumn vector)
+            nUpdates     : Number of filter coefficient updates.
+
+    Main Variables
+    --------- 
+        regressor
+        outputs_vector[k] represents the output errors at iteration k    
+        FIR error vectors. 
+        error_vector[k] represents the output errors at iteration k.
+
+    Misc Variables
+    --------------
+        tic
+        nIterations
+
+
+    Comments:
+            Set-membership filtering implies that the (adaptive) filter coefficients are only
+        updated if the magnitude of the error is greater than S.gamma_bar. In practice, we
+        choose S.gamma_bar as a function of the noise variance (sigma_n2). A commom choice
+        is S.gamma_bar = sqrt(5 * sigma_n2).
+            When SM-AP updates, it makes the a posteriori error equal to S.gamma_barVector and
+        therefore, each of the (S.memoryLength+1) elements of S.gamma_barVector must be less
+        or equal to S.gamma_bar (in modulus).
+
+    Authors
+    -------
+        . Bruno Ramos Lima Netto        - [email protected]  & [email protected]
+        . Guilherme de Oliveira Pinto   - [email protected] & [email protected]
+        . Markus Vinícius Santos Lima   - mvsl20@gmailcom           & [email protected]
+        . Wallace Alves Martins         - [email protected]     & [email protected]
+        . Luiz Wagner Pereira Biscainho - [email protected]          & [email protected]
+        . Paulo Sergio Ramirez Diniz    -                             [email protected]
+
+    """
+
+    # Initialization
+    tic = time()
+    nIterations = desired_signal.size
+
+    regressor = np.zeros(
+        (Filter.filter_order+1, memoryLength+1), dtype=input_signal.dtype)
+    error_vector = np.array([])
+    outputs_vector = np.array([])
+    nUpdates = 0
+    # Main Loop
+
+    # assert len(gamma_barVector) == memoryLength + 1
+    # assert gamma_barVector .<= gamma_bar
+
+    prefixedInput = np.concatenate(
+        (np.zeros(Filter.filter_order+1), input_signal))
+    prefixedDesired = np.concatenate((np.zeros(memoryLength+1), input_signal))
+
+    for it in range(nIterations):
+
+        regressor[:, 2:memoryLength+1] = regressor[:, 1:memoryLength]
+
+        regressor[:, 1] = prefixedInput[it+(Filter.filter_order):-1]
+
+        coefficients = Filter.coefficients
+
+        output_it = np.dot(regressor.conj(), coefficients)
+
+        error_it = prefixedDesired[it+memoryLength].conj() - output_it
+
+        if abs(error_it) > gamma_bar:
+            nUpdates += 1
+
+            auxTerm = (regressor.T.conj()*regressor +
+                       gamma*np.eye(memoryLength+1))
+
+            coefficients = coefficients + regressor * \
+                (1/auxTerm) * (error_it - gamma_barVector.conj())
+
+            Filter.coefficients_history.append(coefficients)
+
+        error_vector = np.append(error_vector, error_it)
+        outputs_vector = np.append(outputs_vector, output_it)
+
+        Filter.coefficients = coefficients
+
+    if verbose == True:
+        print(" ")
+        print('Total runtime {:.03} ms'.format((time() - tic)*1000))
+
+    return {'outputs': outputs_vector,
+            'errors': error_vector, 'coefficients': Filter.coefficients_history}, nUpdates
+
+#   EOF
diff --git a/pydaptivefiltering/SM/NLMS.py b/pydaptivefiltering/SM/NLMS.py
@@ -0,0 +1,127 @@
+#  SM.NLMS.py
+#
+#      Implements the Set-membership Normalized LMS algorithm for COMPLEX valued data.
+#      (Algorithm 6.1 - book: Adaptive Filtering: Algorithms and Practical
+#                                                       Implementation, Diniz)
+#
+#      Authors:
+#       . Bruno Ramos Lima Netto        - [email protected]  & [email protected]
+#       . Guilherme de Oliveira Pinto   - [email protected] & [email protected]
+#       . Markus Vinícius Santos Lima   - mvsl20@gmailcom           & [email protected]
+#       . Wallace Alves Martins         - [email protected]     & [email protected]
+#       . Luiz Wagner Pereira Biscainho - [email protected]          & [email protected]
+#       . Paulo Sergio Ramirez Diniz    -                             [email protected]
+
+# Imports
+import numpy as np
+from time import time
+
+
+def NLMS(Filter, desired_signal: np.ndarray, input_signal: np.ndarray, gamma_bar: float, gamma:float, verbose: bool = False) -> dict:
+    """
+    Description
+    -----------
+        Implements the Set-membership Normalized LMS algorithm for COMPLEX valued data.
+        (Algorithm 6.1 - book: Adaptive Filtering: Algorithms and Practical Implementation, Diniz)
+
+    Syntax
+    ------
+    OutputDictionary = SM.NLMS(Filter, desired_signal, input_signal, verbose)
+
+    Inputs
+    -------
+        filter  : Adaptive Filter                       filter object
+        desired : Desired signal                        numpy array (row vector)
+        input   : Input signal to feed filter           numpy array (row vector)
+        gamma_bar : Upper bound for the error modulus.
+        gamma_barVector       : Upper bound vector for the error modulus. (COLUMN vector)
+        gamma                 : Regularization factor.
+        memoryLength          : Reuse data factor.
+        verbose : Verbose boolean                       bool
+
+    Outputs
+    -------
+        dictionary:
+            outputs      : Store the estimated output of each iteration.        numpy array (collumn vector)
+            errors       : Store the error for each iteration.                  numpy array (collumn vector)
+            coefficients : Store the estimated coefficients for each iteration. numpy array (collumn vector)
+
+    Main Variables
+    --------- 
+        regressor
+        outputs_vector[k] represents the output errors at iteration k    
+        FIR error vectors. 
+        error_vector[k] represents the output errors at iteration k.
+
+    Misc Variables
+    --------------
+        tic
+        nIterations
+
+
+    Comments:
+            Set-membership filtering implies that the (adaptive) filter coefficients are only
+        updated if the magnitude of the error is greater than S.gamma_bar. In practice, we
+        choose S.gamma_bar as a function of the noise variance (sigma_n2). A commom choice
+        is S.gamma_bar = sqrt(5 * sigma_n2).
+
+    Authors
+    -------
+        . Bruno Ramos Lima Netto        - [email protected]  & [email protected]
+        . Guilherme de Oliveira Pinto   - [email protected] & [email protected]
+        . Markus Vinícius Santos Lima   - mvsl20@gmailcom           & [email protected]
+        . Wallace Alves Martins         - [email protected]     & [email protected]
+        . Luiz Wagner Pereira Biscainho - [email protected]          & [email protected]
+        . Paulo Sergio Ramirez Diniz    -                             [email protected]
+
+    """
+
+    # Initialization
+    tic = time()
+    nIterations = desired_signal.size
+
+    regressor = np.zeros(Filter.filter_order+1, dtype=input_signal.dtype)
+    error_vector = np.array([])
+    outputs_vector = np.array([])
+    nUpdates = 0
+
+    # Main Loop
+
+    prefixedInput = np.concatenate(
+        (np.zeros(Filter.filter_order), input_signal))
+
+    for it in range(nIterations):
+
+        regressor = prefixedInput[it+(Filter.filter_order):-1]
+
+        coefficients = Filter.coefficients
+
+        output_it = np.dot(coefficients.conj(), regressor)
+
+        error_it = desired_signal[it] - output_it
+
+        if abs(error_it) > gamma_bar:
+
+            nUpdates += 1
+
+            mu = 1 - (gamma_bar/abs(error_it))
+
+        else:
+
+            mu = 0
+
+        error_vector = np.append(error_vector, error_it)
+        outputs_vector = np.append(outputs_vector, output_it)
+
+        coefficients = coefficients + mu / (gamma + regressor.T.conj()*regressor)*(error_it.conj()*regressor)
+
+        Filter.coefficients = coefficients
+
+    if verbose == True:
+        print(" ")
+        print('Total runtime {:.03} ms'.format((time() - tic)*1000))
+
+    return {'outputs': outputs_vector,
+            'errors': error_vector, 'coefficients': Filter.coefficients_history}, nUpdates
+
+#   EOF
diff --git a/pydaptivefiltering/SM/__init__.py b/pydaptivefiltering/SM/__init__.py
@@ -0,0 +1,5 @@
+from pydaptivefiltering.SM.AP import AP
+from pydaptivefiltering.SM.NLMS import NLMS
+# from pydaptivefiltering.SM.BNLMS import LMS_Newton
+# from pydaptivefiltering.SM.sAP import LMS_Newton
+# from pydaptivefiltering.SM.sPUAP import LMS_Newton
diff --git a/pydaptivefiltering/Set-membership_Algorithms/SM_AP.py b/pydaptivefiltering/Set-membership_Algorithms/SM_AP.py