equalizer: fix gain handling and remove dead code

amodem/equalizer.py

@@ -1,7 +1,9 @@
 import numpy as np
 from numpy.linalg import lstsq
 
-from amodem import dsp, config
+from amodem import dsp
+from amodem import config
+from amodem import sampling
 
 import itertools
 import random
@@ -40,7 +42,7 @@ def equalize(signal, symbols, order):
     assert symbols.shape[1] == Nfreq
     length = symbols.shape[0]
 
-    matched = np.array(carriers) * Nfreq / (0.5*Nsym)
+    matched = np.array(carriers) / (0.5*Nsym)
     matched = matched[:, ::-1].transpose().conj()
     y = dsp.lfilter(x=signal, b=matched, a=[1])
 

amodem/sampling.py

@@ -30,7 +30,7 @@ class Interpolator(object):
 class Sampler(object):
     def __init__(self, src, interp=None):
         self.freq = 1.0
-        self.gain = 1.0
+        self.equalizer = lambda x: x
         if interp is not None:
             self.interp = interp
             self.resolution = self.interp.resolution
@@ -71,7 +71,7 @@ class Sampler(object):
         except StopIteration:
             pass
 
-        return frame[:count] * self.gain
+        return self.equalizer(frame[:count])
 
 
 def resample(src, dst, df=0.0):

amodem/train.py

@@ -1,16 +1,3 @@
-import itertools
-
 prefix = [1]*400 + [0]*50
-
+equalizer_length = 500
-
+silence_length = 100
-def _equalizer_sequence():
-    res = []
-
-    symbols = [1, 1j, -1, -1j]
-    for s in itertools.islice(itertools.cycle(symbols), 100):
-        res.extend([s]*1 + [0]*1)
-
-    res.extend([0]*20)
-    return res
-
-equalizer = _equalizer_sequence()

tests/test_equalizer.py

@@ -1,7 +1,8 @@
 from numpy.linalg import norm
 import numpy as np
 
-from amodem import train, dsp, config
+from amodem import dsp
+from amodem import config
 from amodem import equalizer
 
 
@@ -9,29 +10,6 @@ def assert_approx(x, y, e=1e-12):
     assert norm(x - y) < e * norm(x)
 
 
-def test_fir():
-    a = [1, 0.8, -0.1, 0, 0]
-    tx = train.equalizer
-    rx = dsp.lfilter(x=tx, b=[1], a=a)
-    h_ = dsp.estimate(x=rx, y=tx, order=len(a))
-    tx_ = dsp.lfilter(x=rx, b=h_, a=[1])
-    assert_approx(h_, a)
-    assert_approx(tx, tx_)
-
-
-def test_iir():
-    alpha = 0.1
-    b = [1, -alpha]
-    tx = train.equalizer
-    rx = dsp.lfilter(x=tx, b=b, a=[1])
-    h_ = dsp.estimate(x=rx, y=tx, order=20)
-    tx_ = dsp.lfilter(x=rx, b=h_, a=[1])
-
-    h_expected = np.array([alpha ** i for i in range(len(h_))])
-    assert_approx(h_, h_expected)
-    assert_approx(tx, tx_)
-
-
 def test_training():
     L = 1000
     t1 = equalizer.train_symbols(L)
@@ -68,8 +46,8 @@ def test_isi():
     gain = float(config.Nfreq)
 
     symbols = equalizer.train_symbols(length=length)
-    x = equalizer.modulator(symbols)
+    x = equalizer.modulator(symbols) * gain
-    assert_approx(equalizer.demodulator(gain * x, size=length), symbols)
+    assert_approx(equalizer.demodulator(x, size=length), symbols)
 
     den = np.array([1, -0.6, 0.1])
     num = np.array([0.5])
@@ -79,5 +57,5 @@ def test_isi():
     assert_approx(h, den / num)
 
     y = dsp.lfilter(x=y, b=h, a=[1])
-    z = equalizer.demodulator(gain * y, size=length)
+    z = equalizer.demodulator(y, size=length)
     assert_approx(z, symbols)