gr-digital: Improve PLL Loops and Clock Recovery

These updates incorporate the same inlining of loop functions and the newly pushed fast_cc_multiply function to ensure consistent performance improvements across systems that do not support the cx-limited-range compiler parameter (Macs and Windows).
author: ghostop14 <ghostop14@gmail.com> 2020-02-14 11:47:37 -0500
committer: Marcus Müller <marcus@hostalia.de> 2020-02-15 14:25:57 +0100
commit: 504ab76d9e4fc3e9f497b9fd34f01c1bad346a5d (patch)
tree: c64edca089a87217e56d8840a38382d7e2b4ec56
parent: a56c3a7f4908133dc6ceb93ffd3b2f9e1dda8e0e (diff)
13 files changed, 135 insertions, 161 deletions
diff --git a/gr-analog/lib/pll_carriertracking_cc_impl.cc b/gr-analog/lib/pll_carriertracking_cc_impl.cc
index 81c04aa01b..f403cf3528 100644
--- a/gr-analog/lib/pll_carriertracking_cc_impl.cc
+++ b/gr-analog/lib/pll_carriertracking_cc_impl.cc
@@ -14,7 +14,6 @@
 
 #include "pll_carriertracking_cc_impl.h"
 #include <gnuradio/io_signature.h>
-#include <gnuradio/math.h>
 #include <gnuradio/sincos.h>
 
 #include <cmath>
@@ -44,29 +43,6 @@ pll_carriertracking_cc_impl::pll_carriertracking_cc_impl(float loop_bw,
 
 pll_carriertracking_cc_impl::~pll_carriertracking_cc_impl() {}
 
-float pll_carriertracking_cc_impl::mod_2pi(float in)
-{
-    if (in > GR_M_PI)
-        return in - (2.0 * GR_M_PI);
-    else if (in < -GR_M_PI)
-        return in + (2.0 * GR_M_PI);
-    else
-        return in;
-}
-
-float pll_carriertracking_cc_impl::phase_detector(gr_complex sample, float ref_phase)
-{
-    float sample_phase;
-    //  sample_phase = atan2(sample.imag(),sample.real());
-    sample_phase = gr::fast_atan2f(sample.imag(), sample.real());
-    return mod_2pi(sample_phase - ref_phase);
-}
-
-bool pll_carriertracking_cc_impl::lock_detector(void)
-{
-    return (fabsf(d_locksig) > d_lock_threshold);
-}
-
 bool pll_carriertracking_cc_impl::squelch_enable(bool set_squelch)
 {
     return d_squelch_enable = set_squelch;
@@ -89,7 +65,8 @@ int pll_carriertracking_cc_impl::work(int noutput_items,
 
     for (int i = 0; i < noutput_items; i++) {
         gr::sincosf(d_phase, &t_imag, &t_real);
-        optr[i] = iptr[i] * gr_complex(t_real, -t_imag);
+
+        fast_cc_multiply(optr[i], iptr[i], gr_complex(t_real, -t_imag));
 
         error = phase_detector(iptr[i], d_phase);
 
diff --git a/gr-analog/lib/pll_carriertracking_cc_impl.h b/gr-analog/lib/pll_carriertracking_cc_impl.h
index 3cb7146da9..82a2f1c21c 100644
--- a/gr-analog/lib/pll_carriertracking_cc_impl.h
+++ b/gr-analog/lib/pll_carriertracking_cc_impl.h
@@ -12,6 +12,7 @@
 #define INCLUDED_ANALOG_PLL_CARRIERTRACKING_CC_IMPL_H
 
 #include <gnuradio/analog/pll_carriertracking_cc.h>
+#include <gnuradio/math.h>
 
 namespace gr {
 namespace analog {
@@ -22,14 +23,30 @@ private:
     float d_locksig, d_lock_threshold;
     bool d_squelch_enable;
 
-    float mod_2pi(float in);
-    float phase_detector(gr_complex sample, float ref_phase);
+    float mod_2pi(float in)
+    {
+        if (in > GR_M_PI)
+            return in - (2.0 * GR_M_PI);
+        else if (in < -GR_M_PI)
+            return in + (2.0 * GR_M_PI);
+        else
+            return in;
+    }
+
+    float phase_detector(gr_complex sample, float ref_phase)
+    {
+        float sample_phase;
+        //  sample_phase = atan2(sample.imag(),sample.real());
+        sample_phase = gr::fast_atan2f(sample.imag(), sample.real());
+        return mod_2pi(sample_phase - ref_phase);
+    }
 
 public:
     pll_carriertracking_cc_impl(float loop_bw, float max_freq, float min_freq);
     ~pll_carriertracking_cc_impl();
 
-    bool lock_detector(void);
+    bool lock_detector(void) { return (fabsf(d_locksig) > d_lock_threshold); }
+
     bool squelch_enable(bool);
     float set_lock_threshold(float);
 
diff --git a/gr-analog/lib/pll_freqdet_cf_impl.cc b/gr-analog/lib/pll_freqdet_cf_impl.cc
index f460441e92..baeecc7739 100644
--- a/gr-analog/lib/pll_freqdet_cf_impl.cc
+++ b/gr-analog/lib/pll_freqdet_cf_impl.cc
@@ -37,23 +37,6 @@ pll_freqdet_cf_impl::pll_freqdet_cf_impl(float loop_bw, float max_freq, float mi
 
 pll_freqdet_cf_impl::~pll_freqdet_cf_impl() {}
 
-float pll_freqdet_cf_impl::mod_2pi(float in)
-{
-    if (in > GR_M_PI)
-        return in - (2.0 * GR_M_PI);
-    else if (in < -GR_M_PI)
-        return in + (2.0 * GR_M_PI);
-    else
-        return in;
-}
-
-float pll_freqdet_cf_impl::phase_detector(gr_complex sample, float ref_phase)
-{
-    float sample_phase;
-    sample_phase = gr::fast_atan2f(sample.imag(), sample.real());
-    return mod_2pi(sample_phase - ref_phase);
-}
-
 int pll_freqdet_cf_impl::work(int noutput_items,
                               gr_vector_const_void_star& input_items,
                               gr_vector_void_star& output_items)
diff --git a/gr-analog/lib/pll_freqdet_cf_impl.h b/gr-analog/lib/pll_freqdet_cf_impl.h
index de5c03192b..2110bcc720 100644
--- a/gr-analog/lib/pll_freqdet_cf_impl.h
+++ b/gr-analog/lib/pll_freqdet_cf_impl.h
@@ -12,6 +12,7 @@
 #define INCLUDED_ANALOG_PLL_FREQDET_CF_IMPL_H
 
 #include <gnuradio/analog/pll_freqdet_cf.h>
+#include <gnuradio/math.h>
 
 namespace gr {
 namespace analog {
@@ -19,14 +20,28 @@ namespace analog {
 class pll_freqdet_cf_impl : public pll_freqdet_cf
 {
 private:
-    float phase_detector(gr_complex sample, float ref_phase);
+    float mod_2pi(float in)
+    {
+        if (in > GR_M_PI)
+            return in - (2.0 * GR_M_PI);
+        else if (in < -GR_M_PI)
+            return in + (2.0 * GR_M_PI);
+        else
+            return in;
+    }
+
+    float phase_detector(gr_complex sample, float ref_phase)
+    {
+        float sample_phase;
+        sample_phase = gr::fast_atan2f(sample.imag(), sample.real());
+        return mod_2pi(sample_phase - ref_phase);
+    }
+
 
 public:
     pll_freqdet_cf_impl(float loop_bw, float max_freq, float min_freq);
     ~pll_freqdet_cf_impl();
 
-    float mod_2pi(float in);
-
     void set_loop_bandwidth(float bw);
     void set_damping_factor(float df);
     void set_alpha(float alpha);
diff --git a/gr-analog/lib/pll_refout_cc_impl.cc b/gr-analog/lib/pll_refout_cc_impl.cc
index 067e650de5..9824d12178 100644
--- a/gr-analog/lib/pll_refout_cc_impl.cc
+++ b/gr-analog/lib/pll_refout_cc_impl.cc
@@ -14,7 +14,6 @@
 
 #include "pll_refout_cc_impl.h"
 #include <gnuradio/io_signature.h>
-#include <gnuradio/math.h>
 #include <gnuradio/sincos.h>
 #include <math.h>
 
@@ -37,23 +36,6 @@ pll_refout_cc_impl::pll_refout_cc_impl(float loop_bw, float max_freq, float min_
 
 pll_refout_cc_impl::~pll_refout_cc_impl() {}
 
-float pll_refout_cc_impl::mod_2pi(float in)
-{
-    if (in > GR_M_PI)
-        return in - (2.0 * GR_M_PI);
-    else if (in < -GR_M_PI)
-        return in + (2.0 * GR_M_PI);
-    else
-        return in;
-}
-
-float pll_refout_cc_impl::phase_detector(gr_complex sample, float ref_phase)
-{
-    float sample_phase;
-    sample_phase = gr::fast_atan2f(sample.imag(), sample.real());
-    return mod_2pi(sample_phase - ref_phase);
-}
-
 int pll_refout_cc_impl::work(int noutput_items,
                              gr_vector_const_void_star& input_items,
                              gr_vector_void_star& output_items)
diff --git a/gr-analog/lib/pll_refout_cc_impl.h b/gr-analog/lib/pll_refout_cc_impl.h
index da9be48b16..dc8bf2dc9d 100644
--- a/gr-analog/lib/pll_refout_cc_impl.h
+++ b/gr-analog/lib/pll_refout_cc_impl.h
@@ -12,6 +12,7 @@
 #define INCLUDED_ANALOG_PLL_REFOUT_CC_IMPL_H
 
 #include <gnuradio/analog/pll_refout_cc.h>
+#include <gnuradio/math.h>
 
 namespace gr {
 namespace analog {
@@ -19,8 +20,23 @@ namespace analog {
 class pll_refout_cc_impl : public pll_refout_cc
 {
 private:
-    float mod_2pi(float in);
-    float phase_detector(gr_complex sample, float ref_phase);
+    float mod_2pi(float in)
+    {
+        if (in > GR_M_PI)
+            return in - (2.0 * GR_M_PI);
+        else if (in < -GR_M_PI)
+            return in + (2.0 * GR_M_PI);
+        else
+            return in;
+    }
+
+    float phase_detector(gr_complex sample, float ref_phase)
+    {
+        float sample_phase;
+        sample_phase = gr::fast_atan2f(sample.imag(), sample.real());
+        return mod_2pi(sample_phase - ref_phase);
+    }
+
 
 public:
     pll_refout_cc_impl(float loop_bw, float max_freq, float min_freq);
diff --git a/gr-digital/lib/clock_recovery_mm_cc_impl.cc b/gr-digital/lib/clock_recovery_mm_cc_impl.cc
index 0c1b7e0892..4f2f4c0c58 100644
--- a/gr-digital/lib/clock_recovery_mm_cc_impl.cc
+++ b/gr-digital/lib/clock_recovery_mm_cc_impl.cc
@@ -74,27 +74,6 @@ void clock_recovery_mm_cc_impl::forecast(int noutput_items,
             (int)ceil((noutput_items * d_omega) + d_interp->ntaps()) + FUDGE;
 }
 
-gr_complex clock_recovery_mm_cc_impl::slicer_0deg(gr_complex sample)
-{
-    float real = 0, imag = 0;
-
-    if (sample.real() > 0)
-        real = 1;
-    if (sample.imag() > 0)
-        imag = 1;
-    return gr_complex(real, imag);
-}
-
-gr_complex clock_recovery_mm_cc_impl::slicer_45deg(gr_complex sample)
-{
-    float real = -1, imag = -1;
-    if (sample.real() > 0)
-        real = 1;
-    if (sample.imag() > 0)
-        imag = 1;
-    return gr_complex(real, imag);
-}
-
 void clock_recovery_mm_cc_impl::set_omega(float omega)
 {
     d_omega = omega;
@@ -135,8 +114,8 @@ int clock_recovery_mm_cc_impl::general_work(int noutput_items,
             d_c_1T = d_c_0T;
             d_c_0T = slicer_0deg(d_p_0T);
 
-            x = (d_c_0T - d_c_2T) * conj(d_p_1T);
-            y = (d_p_0T - d_p_2T) * conj(d_c_1T);
+            fast_cc_multiply(x, d_c_0T - d_c_2T, conj(d_p_1T));
+            fast_cc_multiply(y, d_p_0T - d_p_2T, conj(d_c_1T));
             u = y - x;
             mm_val = u.real();
             out[oo++] = d_p_0T;
@@ -169,8 +148,8 @@ int clock_recovery_mm_cc_impl::general_work(int noutput_items,
             d_c_1T = d_c_0T;
             d_c_0T = slicer_0deg(d_p_0T);
 
-            x = (d_c_0T - d_c_2T) * conj(d_p_1T);
-            y = (d_p_0T - d_p_2T) * conj(d_c_1T);
+            fast_cc_multiply(x, d_c_0T - d_c_2T, conj(d_p_1T));
+            fast_cc_multiply(y, d_p_0T - d_p_2T, conj(d_c_1T));
             u = y - x;
             mm_val = u.real();
             out[oo++] = d_p_0T;
diff --git a/gr-digital/lib/clock_recovery_mm_cc_impl.h b/gr-digital/lib/clock_recovery_mm_cc_impl.h
index cb444020c3..79402f5896 100644
--- a/gr-digital/lib/clock_recovery_mm_cc_impl.h
+++ b/gr-digital/lib/clock_recovery_mm_cc_impl.h
@@ -61,8 +61,26 @@ private:
     gr_complex d_p_2T, d_p_1T, d_p_0T;
     gr_complex d_c_2T, d_c_1T, d_c_0T;
 
-    gr_complex slicer_0deg(gr_complex sample);
-    gr_complex slicer_45deg(gr_complex sample);
+    gr_complex slicer_0deg(gr_complex sample)
+    {
+        float real = 0.0f, imag = 0.0f;
+
+        if (sample.real() > 0.0f)
+            real = 1.0f;
+        if (sample.imag() > 0.0f)
+            imag = 1.0f;
+        return gr_complex(real, imag);
+    }
+
+    gr_complex slicer_45deg(gr_complex sample)
+    {
+        float real = -1.0f, imag = -1.0f;
+        if (sample.real() > 0.0f)
+            real = 1.0f;
+        if (sample.imag() > 0.0f)
+            imag = 1.0f;
+        return gr_complex(real, imag);
+    }
 };
 
 } /* namespace digital */
diff --git a/gr-digital/lib/clock_recovery_mm_ff_impl.cc b/gr-digital/lib/clock_recovery_mm_ff_impl.cc
index 1c35c3c44e..ceff90a8e5 100644
--- a/gr-digital/lib/clock_recovery_mm_ff_impl.cc
+++ b/gr-digital/lib/clock_recovery_mm_ff_impl.cc
@@ -60,8 +60,6 @@ void clock_recovery_mm_ff_impl::forecast(int noutput_items,
             (int)ceil((noutput_items * d_omega) + d_interp->ntaps());
 }
 
-static inline float slice(float x) { return x < 0 ? -1.0F : 1.0F; }
-
 void clock_recovery_mm_ff_impl::set_omega(float omega)
 {
     d_omega = omega;
diff --git a/gr-digital/lib/clock_recovery_mm_ff_impl.h b/gr-digital/lib/clock_recovery_mm_ff_impl.h
index 4a2b127134..f0361a8074 100644
--- a/gr-digital/lib/clock_recovery_mm_ff_impl.h
+++ b/gr-digital/lib/clock_recovery_mm_ff_impl.h
@@ -57,6 +57,8 @@ private:
     filter::mmse_fir_interpolator_ff* d_interp;
 
     bool d_verbose;
+
+    float slice(float x) { return x < 0 ? -1.0F : 1.0F; }
 };
 
 } /* namespace digital */
diff --git a/gr-digital/lib/clock_tracking_loop.cc b/gr-digital/lib/clock_tracking_loop.cc
index 6c04d2d87e..2cb7c996c9 100644
--- a/gr-digital/lib/clock_tracking_loop.cc
+++ b/gr-digital/lib/clock_tracking_loop.cc
@@ -101,60 +101,6 @@ void clock_tracking_loop::update_gains()
     set_beta(beta);
 }
 
-void clock_tracking_loop::advance_loop(float error)
-{
-    // So the loop can be reverted one step, if needed.
-    d_prev_avg_period = d_avg_period;
-    d_prev_inst_period = d_inst_period;
-    d_prev_phase = d_phase;
-
-    // Integral arm of PI filter
-    d_avg_period = d_avg_period + d_beta * error;
-    // Limit the integral arm output here, as a large negative
-    // error input can lead to a negative d_avg_period, which
-    // will cause an infitine loop in the phase wrap method.
-    period_limit();
-
-    // Proportional arm of PI filter and final sum of PI filter arms
-    d_inst_period = d_avg_period + d_alpha * error;
-    // Limit the filter output here, for the errant case of a large
-    // negative error input, that can lead to a negative d_inst_period,
-    // which results in an incorrect phase increment, as it is assumed
-    // to be moving forward to the next symbol.
-    if (d_inst_period <= 0.f)
-        d_inst_period = d_avg_period;
-
-    // Compute the new, unwrapped clock phase
-    d_phase = d_phase + d_inst_period;
-}
-
-void clock_tracking_loop::revert_loop()
-{
-    d_avg_period = d_prev_avg_period;
-    d_inst_period = d_prev_inst_period;
-    d_phase = d_prev_phase;
-}
-
-void clock_tracking_loop::phase_wrap()
-{
-    float period = d_avg_period; // One could argue d_inst_period instead
-    float limit = period / 2.0f;
-
-    while (d_phase > limit)
-        d_phase -= period;
-
-    while (d_phase <= -limit)
-        d_phase += period;
-}
-
-void clock_tracking_loop::period_limit()
-{
-    if (d_avg_period > d_max_avg_period)
-        d_avg_period = d_max_avg_period;
-    else if (d_avg_period < d_min_avg_period)
-        d_avg_period = d_min_avg_period;
-}
-
 /*******************************************************************
  * SET FUNCTIONS
  *******************************************************************/
diff --git a/gr-digital/lib/clock_tracking_loop.h b/gr-digital/lib/clock_tracking_loop.h
index f215316a3e..981114117f 100644
--- a/gr-digital/lib/clock_tracking_loop.h
+++ b/gr-digital/lib/clock_tracking_loop.h
@@ -360,7 +360,32 @@ public:
     /*! \brief Advance the loop based on the current gain
      *  settings and the input error signal.
      */
-    void advance_loop(float error);
+    void advance_loop(float error)
+    {
+        // So the loop can be reverted one step, if needed.
+        d_prev_avg_period = d_avg_period;
+        d_prev_inst_period = d_inst_period;
+        d_prev_phase = d_phase;
+
+        // Integral arm of PI filter
+        d_avg_period = d_avg_period + d_beta * error;
+        // Limit the integral arm output here, as a large negative
+        // error input can lead to a negative d_avg_period, which
+        // will cause an infitine loop in the phase wrap method.
+        period_limit();
+
+        // Proportional arm of PI filter and final sum of PI filter arms
+        d_inst_period = d_avg_period + d_alpha * error;
+        // Limit the filter output here, for the errant case of a large
+        // negative error input, that can lead to a negative d_inst_period,
+        // which results in an incorrect phase increment, as it is assumed
+        // to be moving forward to the next symbol.
+        if (d_inst_period <= 0.f)
+            d_inst_period = d_avg_period;
+
+        // Compute the new, unwrapped clock phase
+        d_phase = d_phase + d_inst_period;
+    }
 
     /*! \brief Undo a single, prior advance_loop() call.
      *
@@ -373,7 +398,12 @@ public:
      * perform correctly given the constraints of GNURadio's streaming
      * engine, interpolation filtering, and tag propagation.
      */
-    void revert_loop();
+    void revert_loop()
+    {
+        d_avg_period = d_prev_avg_period;
+        d_inst_period = d_prev_inst_period;
+        d_phase = d_prev_phase;
+    }
 
     /*! \brief
      * Keep the clock phase estimate, tau, between -T_avg/2 and T_avg/2.
@@ -391,7 +421,17 @@ public:
      * clock, and this class doesn't actually use the phase at all,
      * so calling this is optional.
      */
-    void phase_wrap();
+    void phase_wrap()
+    {
+        float period = d_avg_period; // One could argue d_inst_period instead
+        float limit = period / 2.0f;
+
+        while (d_phase > limit)
+            d_phase -= period;
+
+        while (d_phase <= -limit)
+            d_phase += period;
+    }
 
     /*! \brief
      * Keep the estimated average clock period, T_avg, between T_avg_min
@@ -409,7 +449,13 @@ public:
      * It is set as a separate virtual method in case another way is desired
      * as this is fairly heavy-handed.
      */
-    virtual void period_limit();
+    virtual void period_limit()
+    {
+        if (d_avg_period > d_max_avg_period)
+            d_avg_period = d_max_avg_period;
+        else if (d_avg_period < d_min_avg_period)
+            d_avg_period = d_min_avg_period;
+    }
 
     /*******************************************************************
      * SET FUNCTIONS
diff --git a/gr-dtv/lib/atsc/atsc_fpll_impl.cc b/gr-dtv/lib/atsc/atsc_fpll_impl.cc
index b85a38207d..88d9940615 100644
--- a/gr-dtv/lib/atsc/atsc_fpll_impl.cc
+++ b/gr-dtv/lib/atsc/atsc_fpll_impl.cc
@@ -55,12 +55,7 @@ int atsc_fpll_impl::work(int noutput_items,
         d_nco.sincos(&a_sin, &a_cos); // compute cos and sin
 
         // Mix out carrier and output I-only signal
-
-        // PR Merge Note: Once the Costas Optimization PR #3076 merges, this
-        // line below helps with performance when cx_limited_range is not available
-        // such as on Macs and Windows.  Once the merge happens I'll push an update.
-        // gr::fast_cc_multiply(result, in[k], gr_complex(a_sin, a_cos));
-        result = in[k] * gr_complex(a_sin, a_cos);
+        gr::fast_cc_multiply(result, in[k], gr_complex(a_sin, a_cos));
 
         out[k] = result.real();
author	ghostop14 <ghostop14@gmail.com>	2020-02-14 11:47:37 -0500
committer	Marcus Müller <marcus@hostalia.de>	2020-02-15 14:25:57 +0100
commit	504ab76d9e4fc3e9f497b9fd34f01c1bad346a5d (patch)
tree	c64edca089a87217e56d8840a38382d7e2b4ec56
parent	a56c3a7f4908133dc6ceb93ffd3b2f9e1dda8e0e (diff)