summaryrefslogtreecommitdiff
path: root/gnuradio-core/src
diff options
context:
space:
mode:
authorn4hy2008-08-18 22:36:19 +0000
committern4hy2008-08-18 22:36:19 +0000
commita3c5490e99dc21f24432c984cbdf0763e585a391 (patch)
tree5cec9c3543b987cea59d85da8339176578955fe1 /gnuradio-core/src
parente92e82dd65a5c91a6b3a9e098e8f58948c779fa5 (diff)
downloadgnuradio-a3c5490e99dc21f24432c984cbdf0763e585a391.tar.gz
gnuradio-a3c5490e99dc21f24432c984cbdf0763e585a391.tar.bz2
gnuradio-a3c5490e99dc21f24432c984cbdf0763e585a391.zip
New taps computation based on requested transition bandwidth and stopband attenuation. qa code testing each added
git-svn-id: http://gnuradio.org/svn/gnuradio/trunk@9316 221aa14e-8319-0410-a670-987f0aec2ac5
Diffstat (limited to 'gnuradio-core/src')
-rw-r--r--gnuradio-core/src/lib/general/gr_firdes.cc270
-rw-r--r--gnuradio-core/src/lib/general/gr_firdes.h51
-rw-r--r--gnuradio-core/src/lib/general/qa_gr_firdes.cc272
-rw-r--r--gnuradio-core/src/lib/general/qa_gr_firdes.h6
4 files changed, 587 insertions, 12 deletions
diff --git a/gnuradio-core/src/lib/general/gr_firdes.cc b/gnuradio-core/src/lib/general/gr_firdes.cc
index 06730b7f8..17c40d78a 100644
--- a/gnuradio-core/src/lib/general/gr_firdes.cc
+++ b/gnuradio-core/src/lib/general/gr_firdes.cc
@@ -51,6 +51,53 @@ static double Izero(double x)
//
vector<float>
+gr_firdes::low_pass_2(double gain,
+ double sampling_freq, // Hz
+ double cutoff_freq, // Hz BEGINNING of transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window_type,
+ double beta) // used only with Kaiser
+{
+ sanity_check_1f (sampling_freq, cutoff_freq, transition_width);
+
+ int ntaps = compute_ntaps_windes (sampling_freq, transition_width,
+ attenuation_dB);
+
+ // construct the truncated ideal impulse response
+ // [sin(x)/x for the low pass case]
+
+ vector<float> taps(ntaps);
+ vector<float> w = window (window_type, ntaps, beta);
+
+ int M = (ntaps - 1) / 2;
+ double fwT0 = 2 * M_PI * cutoff_freq / sampling_freq;
+ for (int n = -M; n <= M; n++){
+ if (n == 0)
+ taps[n + M] = fwT0 / M_PI * w[n + M];
+ else {
+ // a little algebra gets this into the more familiar sin(x)/x form
+ taps[n + M] = sin (n * fwT0) / (n * M_PI) * w[n + M];
+ }
+ }
+
+ // find the factor to normalize the gain, fmax.
+ // For low-pass, gain @ zero freq = 1.0
+
+ double fmax = taps[0 + M];
+ for (int n = 1; n <= M; n++)
+ fmax += 2 * taps[n + M];
+
+ gain /= fmax; // normalize
+
+ for (int i = 0; i < ntaps; i++)
+ taps[i] *= gain;
+
+
+ return taps;
+}
+
+vector<float>
gr_firdes::low_pass (double gain,
double sampling_freq,
double cutoff_freq, // Hz center of transition band
@@ -74,7 +121,7 @@ gr_firdes::low_pass (double gain,
for (int n = -M; n <= M; n++){
if (n == 0)
- taps[n + M] = fwT0 / M_PI * w[n + M];
+ taps[n + M] = fwT0 / M_PI * w[n + M];
else {
// a little algebra gets this into the more familiar sin(x)/x form
taps[n + M] = sin (n * fwT0) / (n * M_PI) * w[n + M];
@@ -96,22 +143,71 @@ gr_firdes::low_pass (double gain,
return taps;
}
+
//
// === High Pass ===
//
vector<float>
+gr_firdes::high_pass_2 (double gain,
+ double sampling_freq,
+ double cutoff_freq, // Hz center of transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window_type,
+ double beta) // used only with Kaiser
+{
+ sanity_check_1f (sampling_freq, cutoff_freq, transition_width);
+
+ int ntaps = compute_ntaps_windes (sampling_freq, transition_width,
+ attenuation_dB);
+
+ // construct the truncated ideal impulse response times the window function
+
+ vector<float> taps(ntaps);
+ vector<float> w = window (window_type, ntaps, beta);
+
+ int M = (ntaps - 1) / 2;
+ double fwT0 = 2 * M_PI * cutoff_freq / sampling_freq;
+
+ for (int n = -M; n <= M; n++){
+ if (n == 0)
+ taps[n + M] = (1 - (fwT0 / M_PI)) * w[n + M];
+ else {
+ // a little algebra gets this into the more familiar sin(x)/x form
+ taps[n + M] = -sin (n * fwT0) / (n * M_PI) * w[n + M];
+ }
+ }
+
+ // find the factor to normalize the gain, fmax.
+ // For high-pass, gain @ fs/2 freq = 1.0
+
+ double fmax = taps[0 + M];
+ for (int n = 1; n <= M; n++)
+ fmax += 2 * taps[n + M] * cos (n * M_PI);
+
+ gain /= fmax; // normalize
+
+ for (int i = 0; i < ntaps; i++)
+ taps[i] *= gain;
+
+
+ return taps;
+}
+
+
+vector<float>
gr_firdes::high_pass (double gain,
- double sampling_freq,
- double cutoff_freq, // Hz center of transition band
- double transition_width, // Hz width of transition band
- win_type window_type,
- double beta) // used only with Kaiser
+ double sampling_freq,
+ double cutoff_freq, // Hz center of transition band
+ double transition_width, // Hz width of transition band
+ win_type window_type,
+ double beta) // used only with Kaiser
{
sanity_check_1f (sampling_freq, cutoff_freq, transition_width);
int ntaps = compute_ntaps (sampling_freq, transition_width,
- window_type, beta);
+ window_type, beta);
// construct the truncated ideal impulse response times the window function
@@ -129,15 +225,15 @@ gr_firdes::high_pass (double gain,
taps[n + M] = -sin (n * fwT0) / (n * M_PI) * w[n + M];
}
}
-
+
// find the factor to normalize the gain, fmax.
// For high-pass, gain @ fs/2 freq = 1.0
-
+
double fmax = taps[0 + M];
for (int n = 1; n <= M; n++)
fmax += 2 * taps[n + M] * cos (n * M_PI);
- gain /= fmax; // normalize
+ gain /= fmax; // normalize
for (int i = 0; i < ntaps; i++)
taps[i] *= gain;
@@ -146,10 +242,58 @@ gr_firdes::high_pass (double gain,
}
//
-// === Band Pass ===
+// === Band Pass ===
//
vector<float>
+gr_firdes::band_pass_2 (double gain,
+ double sampling_freq,
+ double low_cutoff_freq, // Hz center of transition band
+ double high_cutoff_freq, // Hz center of transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window_type,
+ double beta) // used only with Kaiser
+{
+ sanity_check_2f (sampling_freq,
+ low_cutoff_freq,
+ high_cutoff_freq, transition_width);
+
+ int ntaps = compute_ntaps_windes (sampling_freq, transition_width,
+ attenuation_dB);
+
+ vector<float> taps(ntaps);
+ vector<float> w = window (window_type, ntaps, beta);
+
+ int M = (ntaps - 1) / 2;
+ double fwT0 = 2 * M_PI * low_cutoff_freq / sampling_freq;
+ double fwT1 = 2 * M_PI * high_cutoff_freq / sampling_freq;
+
+ for (int n = -M; n <= M; n++){
+ if (n == 0)
+ taps[n + M] = (fwT1 - fwT0) / M_PI * w[n + M];
+ else {
+ taps[n + M] = (sin (n * fwT1) - sin (n * fwT0)) / (n * M_PI) * w[n + M];
+ }
+ }
+
+ // find the factor to normalize the gain, fmax.
+ // For band-pass, gain @ center freq = 1.0
+
+ double fmax = taps[0 + M];
+ for (int n = 1; n <= M; n++)
+ fmax += 2 * taps[n + M] * cos (n * (fwT0 + fwT1) * 0.5);
+
+ gain /= fmax; // normalize
+
+ for (int i = 0; i < ntaps; i++)
+ taps[i] *= gain;
+
+ return taps;
+}
+
+
+vector<float>
gr_firdes::band_pass (double gain,
double sampling_freq,
double low_cutoff_freq, // Hz center of transition band
@@ -202,6 +346,47 @@ gr_firdes::band_pass (double gain,
//
vector<gr_complex>
+gr_firdes::complex_band_pass_2 (double gain,
+ double sampling_freq,
+ double low_cutoff_freq, // Hz center of transition band
+ double high_cutoff_freq, // Hz center of transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window_type,
+ double beta) // used only with Kaiser
+{
+ sanity_check_2f_c (sampling_freq,
+ low_cutoff_freq,
+ high_cutoff_freq, transition_width);
+
+ int ntaps = compute_ntaps_windes (sampling_freq, transition_width,
+ attenuation_dB);
+
+
+
+ vector<gr_complex> taps(ntaps);
+ vector<float> lptaps(ntaps);
+ vector<float> w = window (window_type, ntaps, beta);
+
+ lptaps = low_pass_2(gain,sampling_freq,(high_cutoff_freq - low_cutoff_freq)/2,transition_width,attenuation_dB,window_type,beta);
+
+ gr_complex *optr = &taps[0];
+ float *iptr = &lptaps[0];
+ float freq = M_PI * (high_cutoff_freq + low_cutoff_freq)/sampling_freq;
+ float phase=0;
+ if (lptaps.size() & 01) {
+ phase = - freq * ( lptaps.size() >> 1 );
+ } else phase = - freq/2.0 * ((1 + 2*lptaps.size()) >> 1);
+ for(unsigned int i=0;i<lptaps.size();i++) {
+ *optr++ = gr_complex(*iptr * cos(phase),*iptr * sin(phase));
+ iptr++, phase += freq;
+ }
+
+ return taps;
+}
+
+
+vector<gr_complex>
gr_firdes::complex_band_pass (double gain,
double sampling_freq,
double low_cutoff_freq, // Hz center of transition band
@@ -240,12 +425,60 @@ gr_firdes::complex_band_pass (double gain,
return taps;
}
-
//
// === Band Reject ===
//
vector<float>
+gr_firdes::band_reject_2 (double gain,
+ double sampling_freq,
+ double low_cutoff_freq, // Hz center of transition band
+ double high_cutoff_freq, // Hz center of transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window_type,
+ double beta) // used only with Kaiser
+{
+ sanity_check_2f (sampling_freq,
+ low_cutoff_freq,
+ high_cutoff_freq, transition_width);
+
+ int ntaps = compute_ntaps_windes (sampling_freq, transition_width,
+ attenuation_dB);
+
+ // construct the truncated ideal impulse response times the window function
+
+ vector<float> taps(ntaps);
+ vector<float> w = window (window_type, ntaps, beta);
+
+ int M = (ntaps - 1) / 2;
+ double fwT0 = 2 * M_PI * low_cutoff_freq / sampling_freq;
+ double fwT1 = 2 * M_PI * high_cutoff_freq / sampling_freq;
+
+ for (int n = -M; n <= M; n++){
+ if (n == 0)
+ taps[n + M] = 1.0 + ((fwT0 - fwT1) / M_PI * w[n + M]);
+ else {
+ taps[n + M] = (sin (n * fwT0) - sin (n * fwT1)) / (n * M_PI) * w[n + M];
+ }
+ }
+
+ // find the factor to normalize the gain, fmax.
+ // For band-reject, gain @ zero freq = 1.0
+
+ double fmax = taps[0 + M];
+ for (int n = 1; n <= M; n++)
+ fmax += 2 * taps[n + M];
+
+ gain /= fmax; // normalize
+
+ for (int i = 0; i < ntaps; i++)
+ taps[i] *= gain;
+
+ return taps;
+}
+
+vector<float>
gr_firdes::band_reject (double gain,
double sampling_freq,
double low_cutoff_freq, // Hz center of transition band
@@ -428,6 +661,19 @@ static const float width_factor[5] = { // indexed by win_type
};
int
+gr_firdes::compute_ntaps_windes(double sampling_freq,
+ double transition_width, // this is frequency, not relative frequency
+ double attenuation_dB)
+{
+ // Based on formula from Multirate Signal Processing for
+ // Communications Systems, fredric j harris
+ int ntaps = (int)(attenuation_dB*sampling_freq/(22.0*transition_width));
+ if ((ntaps & 1) == 0) // if even...
+ ntaps++; // ...make odd
+ return ntaps;
+}
+
+int
gr_firdes::compute_ntaps (double sampling_freq,
double transition_width,
win_type window_type,
diff --git a/gnuradio-core/src/lib/general/gr_firdes.h b/gnuradio-core/src/lib/general/gr_firdes.h
index f920398ad..bb62dad8f 100644
--- a/gnuradio-core/src/lib/general/gr_firdes.h
+++ b/gnuradio-core/src/lib/general/gr_firdes.h
@@ -68,6 +68,15 @@ class gr_firdes {
win_type window = WIN_HAMMING,
double beta = 6.76); // used only with Kaiser
+ static std::vector<float>
+ low_pass_2 (double gain,
+ double sampling_freq,
+ double cutoff_freq, // Hz beginning transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window = WIN_HAMMING,
+ double beta = 6.76); // used only with Kaiser
+
/*!
* \brief use "window method" to design a high-pass FIR filter
*
@@ -90,6 +99,15 @@ class gr_firdes {
win_type window = WIN_HAMMING,
double beta = 6.76); // used only with Kaiser
+ static std::vector<float>
+ high_pass_2 (double gain,
+ double sampling_freq,
+ double cutoff_freq, // Hz center of transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window = WIN_HAMMING,
+ double beta = 6.76); // used only with Kaiser
+
/*!
* \brief use "window method" to design a band-pass FIR filter
*
@@ -114,6 +132,15 @@ class gr_firdes {
win_type window = WIN_HAMMING,
double beta = 6.76); // used only with Kaiser
+ static std::vector<float>
+ band_pass_2 (double gain,
+ double sampling_freq,
+ double low_cutoff_freq, // Hz beginning transition band
+ double high_cutoff_freq, // Hz beginning transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window = WIN_HAMMING,
+ double beta = 6.76); // used only with Kaiser
/*!
* \brief use "window method" to design a complex band-pass FIR filter
@@ -140,6 +167,16 @@ class gr_firdes {
win_type window = WIN_HAMMING,
double beta = 6.76); // used only with Kaiser
+ static std::vector<gr_complex>
+ complex_band_pass_2 (double gain,
+ double sampling_freq,
+ double low_cutoff_freq, // Hz beginning transition band
+ double high_cutoff_freq, // Hz beginning transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window = WIN_HAMMING,
+ double beta = 6.76); // used only with Kaiser
+
/*!
* \brief use "window method" to design a band-reject FIR filter
@@ -166,6 +203,16 @@ class gr_firdes {
win_type window = WIN_HAMMING,
double beta = 6.76); // used only with Kaiser
+ static std::vector<float>
+ band_reject_2 (double gain,
+ double sampling_freq,
+ double low_cutoff_freq, // Hz beginning transition band
+ double high_cutoff_freq, // Hz beginning transition band
+ double transition_width, // Hz width of transition band
+ double attenuation_dB, // attenuation dB
+ win_type window = WIN_HAMMING,
+ double beta = 6.76); // used only with Kaiser
+
/*!\brief design a Hilbert Transform Filter
*
* \p ntaps: Number of taps, must be odd
@@ -221,6 +268,10 @@ private:
static int compute_ntaps (double sampling_freq,
double transition_width,
win_type window_type, double beta);
+
+ static int compute_ntaps_windes (double sampling_freq,
+ double transition_width,
+ double attenuation_dB);
};
#endif
diff --git a/gnuradio-core/src/lib/general/qa_gr_firdes.cc b/gnuradio-core/src/lib/general/qa_gr_firdes.cc
index 9184c8c92..5a41f118a 100644
--- a/gnuradio-core/src/lib/general/qa_gr_firdes.cc
+++ b/gnuradio-core/src/lib/general/qa_gr_firdes.cc
@@ -275,6 +275,211 @@ const static float t3_exp[107] = {
};
+const static float t4_exp[] = { // low pass
+ 0.001059958362,
+0.0002263929928,
+-0.001277606934,
+-0.0009675776237,
+ 0.001592264394,
+ 0.00243603508,
+-0.001451682881,
+-0.004769335967,
+5.281541594e-18,
+ 0.007567512803,
+ 0.003658855334,
+-0.009761494584,
+ -0.01011830103,
+ 0.009636915289,
+ 0.0193619132,
+-0.004935568199,
+ -0.03060629964,
+-0.007267376408,
+ 0.04236677289,
+ 0.03197422624,
+ -0.05274848267,
+ -0.0850463286,
+ 0.05989059806,
+ 0.31065014,
+ 0.4370569289,
+ 0.31065014,
+ 0.05989059806,
+ -0.0850463286,
+ -0.05274848267,
+ 0.03197422624,
+ 0.04236677289,
+-0.007267376408,
+ -0.03060629964,
+-0.004935568199,
+ 0.0193619132,
+ 0.009636915289,
+ -0.01011830103,
+-0.009761494584,
+ 0.003658855334,
+ 0.007567512803,
+5.281541594e-18,
+-0.004769335967,
+-0.001451682881,
+ 0.00243603508,
+ 0.001592264394,
+-0.0009675776237,
+-0.001277606934,
+0.0002263929928,
+ 0.001059958362,
+};
+
+
+const static float t5_exp[] = { //high pass
+-0.001062123571,
+-0.0002268554381,
+ 0.001280216733,
+ 0.000969554123,
+-0.001595516922,
+-0.002441011136,
+ 0.001454648213,
+ 0.004779078532,
+-5.292330097e-18,
+-0.007582970895,
+ -0.00366632943,
+ 0.009781434201,
+ 0.01013896987,
+-0.009656600654,
+ -0.01940146461,
+ 0.004945650231,
+ 0.03066881932,
+ 0.00728222169,
+ -0.04245331511,
+ -0.03203954175,
+ 0.05285623297,
+ 0.08522006124,
+ -0.06001294032,
+ -0.3112847209,
+ 0.5630782247,
+ -0.3112847209,
+ -0.06001294032,
+ 0.08522006124,
+ 0.05285623297,
+ -0.03203954175,
+ -0.04245331511,
+ 0.00728222169,
+ 0.03066881932,
+ 0.004945650231,
+ -0.01940146461,
+-0.009656600654,
+ 0.01013896987,
+ 0.009781434201,
+ -0.00366632943,
+-0.007582970895,
+-5.292330097e-18,
+ 0.004779078532,
+ 0.001454648213,
+-0.002441011136,
+-0.001595516922,
+ 0.000969554123,
+ 0.001280216733,
+-0.0002268554381,
+-0.001062123571,
+};
+
+const static float t6_exp[] = { // bandpass
+0.0002809273137,
+-0.001047327649,
+7.936541806e-05,
+-0.0004270860809,
+0.0007595835486,
+0.0008966081077,
+-0.0004236323002,
+0.0002423936094,
+-0.002212299034,
+0.0004807534278,
+0.0002620361629,
+ 0.001443728455,
+ 0.002229931997,
+-0.002720607212,
+5.731141573e-05,
+-0.004297634587,
+ 0.001878833398,
+ 0.003217151389,
+ 0.001357055153,
+ 0.003965090029,
+-0.008576190099,
+-0.0003257228818,
+-0.004805727862,
+ 0.004721920472,
+ 0.01007549558,
+-0.002688719891,
+ 0.004467967432,
+ -0.01837076992,
+5.119658377e-17,
+ 0.001125075156,
+ 0.008071650751,
+ 0.02113764361,
+ -0.01602453552,
+ 0.001618095324,
+ -0.03004053794,
+ 0.003163811285,
+ 0.0219683405,
+ 0.007950295694,
+ 0.03682873398,
+ -0.05142467469,
+ -0.00794606097,
+ -0.03965795785,
+ 0.01544955093,
+ 0.09681399167,
+ -0.01610304788,
+ 0.08297294378,
+ -0.2811714709,
+ -0.1094062924,
+ 0.5275565982,
+ -0.1094062924,
+ -0.2811714709,
+ 0.08297294378,
+ -0.01610304788,
+ 0.09681399167,
+ 0.01544955093,
+ -0.03965795785,
+ -0.00794606097,
+ -0.05142467469,
+ 0.03682873398,
+ 0.007950295694,
+ 0.0219683405,
+ 0.003163811285,
+ -0.03004053794,
+ 0.001618095324,
+ -0.01602453552,
+ 0.02113764361,
+ 0.008071650751,
+ 0.001125075156,
+5.119658377e-17,
+ -0.01837076992,
+ 0.004467967432,
+-0.002688719891,
+ 0.01007549558,
+ 0.004721920472,
+-0.004805727862,
+-0.0003257228818,
+-0.008576190099,
+ 0.003965090029,
+ 0.001357055153,
+ 0.003217151389,
+ 0.001878833398,
+-0.004297634587,
+5.731141573e-05,
+-0.002720607212,
+ 0.002229931997,
+ 0.001443728455,
+0.0002620361629,
+0.0004807534278,
+-0.002212299034,
+0.0002423936094,
+-0.0004236323002,
+0.0008966081077,
+0.0007595835486,
+-0.0004270860809,
+7.936541806e-05,
+-0.001047327649,
+0.0002809273137,
+};
+
void
qa_gr_firdes::t1 ()
{
@@ -341,4 +546,71 @@ qa_gr_firdes::t3 ()
void
qa_gr_firdes::t4 ()
{
+ vector<float> taps =
+ gr_firdes::low_pass_2 ( 1.0,
+ 8000,
+ 1750,
+ 500,
+ 66,
+ gr_firdes::WIN_HAMMING);
+
+ // std::cout << "ntaps: " << taps.size () << std::endl;
+ // print_taps (std::cout, taps);
+
+ CPPUNIT_ASSERT_EQUAL (NELEM (t4_exp), taps.size ());
+ for (unsigned int i = 0; i < taps.size (); i++)
+ CPPUNIT_ASSERT_DOUBLES_EQUAL (t4_exp[i], taps[i], 1e-9);
+
+
+ check_symmetry (taps);
+}
+
+void
+qa_gr_firdes::t5 ()
+{
+ vector<float> taps =
+ gr_firdes::high_pass_2 ( 1.0,
+ 8000,
+ 1750,
+ 500,
+ 66,
+ gr_firdes::WIN_HAMMING);
+
+ // std::cout << "ntaps: " << taps.size () << std::endl;
+ // print_taps (std::cout, taps);
+
+ CPPUNIT_ASSERT_EQUAL (NELEM (t5_exp), taps.size ());
+
+ for (unsigned int i = 0; i < taps.size (); i++)
+ CPPUNIT_ASSERT_DOUBLES_EQUAL (t5_exp[i], taps[i], 1e-9);
+
+ check_symmetry (taps);
+}
+
+void
+qa_gr_firdes::t6 ()
+{
+ vector<float> taps =
+ gr_firdes::band_pass_2 ( 1.0,
+ 20e6,
+ 5.75e6 - (5.28e6/2),
+ 5.75e6 + (5.28e6/2),
+ 0.62e6,
+ 66,
+ gr_firdes::WIN_HAMMING);
+
+ // std::cout << "ntaps: " << taps.size () << std::endl;
+ // print_taps (std::cout, taps);
+
+ CPPUNIT_ASSERT_EQUAL (NELEM (t6_exp), taps.size ());
+
+ for (unsigned int i = 0; i < taps.size (); i++)
+ CPPUNIT_ASSERT_DOUBLES_EQUAL (t6_exp[i], taps[i], 1e-7);
+
+ check_symmetry (taps);
+}
+
+void
+qa_gr_firdes::t7 ()
+{
}
diff --git a/gnuradio-core/src/lib/general/qa_gr_firdes.h b/gnuradio-core/src/lib/general/qa_gr_firdes.h
index cb57fd740..d8566454a 100644
--- a/gnuradio-core/src/lib/general/qa_gr_firdes.h
+++ b/gnuradio-core/src/lib/general/qa_gr_firdes.h
@@ -32,6 +32,9 @@ class qa_gr_firdes : public CppUnit::TestCase {
CPPUNIT_TEST (t2);
CPPUNIT_TEST (t3);
CPPUNIT_TEST (t4);
+ CPPUNIT_TEST (t5);
+ CPPUNIT_TEST (t6);
+ CPPUNIT_TEST (t7);
CPPUNIT_TEST_SUITE_END ();
private:
@@ -39,6 +42,9 @@ class qa_gr_firdes : public CppUnit::TestCase {
void t2 ();
void t3 ();
void t4 ();
+ void t5 ();
+ void t6 ();
+ void t7 ();
};