gnss-sdr/src/algorithms/acquisition/gnuradio_blocks/galileo_e5a_noncoherent_iq_...

817 lines
40 KiB
C++

/*!
* \file galileo_e5a_noncoherent_iq_acquisition_caf_cc.cc
* \brief Adapts a PCPS acquisition block to an AcquisitionInterface for
* Galileo E5a data and pilot Signals
* \author Marc Sales, 2014. marcsales92(at)gmail.com
* \based on work from:
* <ul>
* <li> Javier Arribas, 2011. jarribas(at)cttc.es
* <li> Luis Esteve, 2012. luis(at)epsilon-formacion.com
* <li> Marc Molina, 2013. marc.molina.pena@gmail.com
* </ul>
*
* -------------------------------------------------------------------------
*
* Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
*
* GNSS-SDR is a software defined Global Navigation
* Satellite Systems receiver
*
* This file is part of GNSS-SDR.
*
* GNSS-SDR is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* GNSS-SDR is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
*
* -------------------------------------------------------------------------
*/
#include "galileo_e5a_noncoherent_iq_acquisition_caf_cc.h"
#include <sstream>
#include <gnuradio/io_signature.h>
#include <glog/logging.h>
#include <volk/volk.h>
#include <volk_gnsssdr/volk_gnsssdr.h>
#include "control_message_factory.h"
using google::LogMessage;
galileo_e5a_noncoherentIQ_acquisition_caf_cc_sptr galileo_e5a_noncoherentIQ_make_acquisition_caf_cc(
unsigned int sampled_ms,
unsigned int max_dwells,
unsigned int doppler_max, long freq, long fs_in,
int samples_per_ms, int samples_per_code,
bool bit_transition_flag,
gr::msg_queue::sptr queue, bool dump,
std::string dump_filename,
bool both_signal_components_,
int CAF_window_hz_,
int Zero_padding_)
{
return galileo_e5a_noncoherentIQ_acquisition_caf_cc_sptr(
new galileo_e5a_noncoherentIQ_acquisition_caf_cc(sampled_ms, max_dwells, doppler_max, freq, fs_in, samples_per_ms,
samples_per_code, bit_transition_flag, queue, dump, dump_filename, both_signal_components_, CAF_window_hz_, Zero_padding_));
}
galileo_e5a_noncoherentIQ_acquisition_caf_cc::galileo_e5a_noncoherentIQ_acquisition_caf_cc(
unsigned int sampled_ms,
unsigned int max_dwells,
unsigned int doppler_max, long freq, long fs_in,
int samples_per_ms, int samples_per_code,
bool bit_transition_flag,
gr::msg_queue::sptr queue, bool dump,
std::string dump_filename,
bool both_signal_components_,
int CAF_window_hz_,
int Zero_padding_) :
gr::block("galileo_e5a_noncoherentIQ_acquisition_caf_cc",
gr::io_signature::make(1, 1, sizeof(gr_complex)),
gr::io_signature::make(0, 0, sizeof(gr_complex)))
{
d_sample_counter = 0; // SAMPLE COUNTER
d_active = false;
d_state = 0;
d_queue = queue;
d_freq = freq;
d_fs_in = fs_in;
d_samples_per_ms = samples_per_ms;
d_samples_per_code = samples_per_code;
d_max_dwells = max_dwells;
d_well_count = 0;
d_doppler_max = doppler_max;
if (Zero_padding_ > 0)
{
d_sampled_ms = 1;
}
else
{
d_sampled_ms = sampled_ms;
}
d_fft_size = sampled_ms * d_samples_per_ms;
d_mag = 0;
d_input_power = 0.0;
d_num_doppler_bins = 0;
d_bit_transition_flag = bit_transition_flag;
d_buffer_count = 0;
d_both_signal_components = both_signal_components_;
d_CAF_window_hz = CAF_window_hz_;
d_inbuffer = static_cast<gr_complex*>(volk_malloc(d_fft_size * sizeof(gr_complex), volk_get_alignment()));
d_fft_code_I_A = static_cast<gr_complex*>(volk_malloc(d_fft_size * sizeof(gr_complex), volk_get_alignment()));
d_magnitudeIA = static_cast<float*>(volk_malloc(d_fft_size * sizeof(float), volk_get_alignment()));
if (d_both_signal_components == true)
{
d_fft_code_Q_A = static_cast<gr_complex*>(volk_malloc(d_fft_size * sizeof(gr_complex), volk_get_alignment()));
d_magnitudeQA = static_cast<float*>(volk_malloc(d_fft_size * sizeof(float), volk_get_alignment()));
}
else
{
d_fft_code_Q_A = 0;
d_magnitudeQA = 0;
}
// IF COHERENT INTEGRATION TIME > 1
if (d_sampled_ms > 1)
{
d_fft_code_I_B = static_cast<gr_complex*>(volk_malloc(d_fft_size * sizeof(gr_complex), volk_get_alignment()));
d_magnitudeIB = static_cast<float*>(volk_malloc(d_fft_size * sizeof(float), volk_get_alignment()));
if (d_both_signal_components == true)
{
d_fft_code_Q_B = static_cast<gr_complex*>(volk_malloc(d_fft_size * sizeof(gr_complex), volk_get_alignment()));
d_magnitudeQB = static_cast<float*>(volk_malloc(d_fft_size * sizeof(float), volk_get_alignment()));
}
else
{
d_fft_code_Q_B = 0;
d_magnitudeQB = 0;
}
}
else
{
d_fft_code_I_B = 0;
d_magnitudeIB = 0;
d_fft_code_Q_B = 0;
d_magnitudeQB = 0;
}
// Direct FFT
d_fft_if = new gr::fft::fft_complex(d_fft_size, true);
// Inverse FFT
d_ifft = new gr::fft::fft_complex(d_fft_size, false);
// For dumping samples into a file
d_dump = dump;
d_dump_filename = dump_filename;
d_doppler_resolution = 0;
d_threshold = 0;
d_doppler_step = 250;
d_grid_doppler_wipeoffs = 0;
d_gnss_synchro = 0;
d_code_phase = 0;
d_doppler_freq = 0;
d_test_statistics = 0;
d_CAF_vector = 0;
d_CAF_vector_I = 0;
d_CAF_vector_Q = 0;
d_channel = 0;
d_gr_stream_buffer = 0;
}
galileo_e5a_noncoherentIQ_acquisition_caf_cc::~galileo_e5a_noncoherentIQ_acquisition_caf_cc()
{
if (d_num_doppler_bins > 0)
{
for (unsigned int i = 0; i < d_num_doppler_bins; i++)
{
volk_free(d_grid_doppler_wipeoffs[i]);
}
delete[] d_grid_doppler_wipeoffs;
}
volk_free(d_inbuffer);
volk_free(d_fft_code_I_A);
volk_free(d_magnitudeIA);
if (d_both_signal_components == true)
{
volk_free(d_fft_code_Q_A);
volk_free(d_magnitudeQA);
}
// IF INTEGRATION TIME > 1
if (d_sampled_ms > 1)
{
volk_free(d_fft_code_I_B);
volk_free(d_magnitudeIB);
if (d_both_signal_components == true)
{
volk_free(d_fft_code_Q_B);
volk_free(d_magnitudeQB);
}
}
if (d_CAF_window_hz > 0)
{
volk_free(d_CAF_vector);
volk_free(d_CAF_vector_I);
if (d_both_signal_components == true)
{
volk_free(d_CAF_vector_Q);
}
}
delete d_fft_if;
delete d_ifft;
if (d_dump)
{
d_dump_file.close();
}
}
void galileo_e5a_noncoherentIQ_acquisition_caf_cc::set_local_code(std::complex<float> * codeI, std::complex<float> * codeQ )
{
// DATA SIGNAL
// Three replicas of data primary code. CODE A: (1,1,1)
memcpy(d_fft_if->get_inbuf(), codeI, sizeof(gr_complex)*d_fft_size);
d_fft_if->execute(); // We need the FFT of local code
//Conjugate the local code
volk_32fc_conjugate_32fc(d_fft_code_I_A,d_fft_if->get_outbuf(),d_fft_size);
// SAME FOR PILOT SIGNAL
if (d_both_signal_components == true)
{
// Three replicas of pilot primary code. CODE A: (1,1,1)
memcpy(d_fft_if->get_inbuf(), codeQ, sizeof(gr_complex)*d_fft_size);
d_fft_if->execute(); // We need the FFT of local code
//Conjugate the local code
volk_32fc_conjugate_32fc(d_fft_code_Q_A,d_fft_if->get_outbuf(),d_fft_size);
}
// IF INTEGRATION TIME > 1 code, we need to evaluate the other possible combination
// Note: max integration time allowed = 3ms (dealt in adapter)
if (d_sampled_ms > 1)
{
// DATA CODE B: First replica is inverted (0,1,1)
volk_32fc_s32fc_multiply_32fc(&(d_fft_if->get_inbuf())[0],
&codeI[0], gr_complex(-1,0),
d_samples_per_code);
d_fft_if->execute(); // We need the FFT of local code
//Conjugate the local code
volk_32fc_conjugate_32fc(d_fft_code_I_B,d_fft_if->get_outbuf(),d_fft_size);
if (d_both_signal_components == true)
{
// PILOT CODE B: First replica is inverted (0,1,1)
volk_32fc_s32fc_multiply_32fc(&(d_fft_if->get_inbuf())[0],
&codeQ[0], gr_complex(-1,0),
d_samples_per_code);
d_fft_if->execute(); // We need the FFT of local code
//Conjugate the local code
volk_32fc_conjugate_32fc(d_fft_code_Q_B,d_fft_if->get_outbuf(),d_fft_size);
}
}
}
void galileo_e5a_noncoherentIQ_acquisition_caf_cc::init()
{
d_gnss_synchro->Flag_valid_acquisition = false;
d_gnss_synchro->Flag_valid_symbol_output = false;
d_gnss_synchro->Flag_valid_pseudorange = false;
d_gnss_synchro->Flag_valid_word = false;
d_gnss_synchro->Flag_preamble = false;
d_gnss_synchro->Acq_delay_samples = 0.0;
d_gnss_synchro->Acq_doppler_hz = 0.0;
d_gnss_synchro->Acq_samplestamp_samples = 0;
d_mag = 0.0;
d_input_power = 0.0;
const double GALILEO_TWO_PI = 6.283185307179600;
// Count the number of bins
d_num_doppler_bins = 0;
for (int doppler = static_cast<int>(-d_doppler_max);
doppler <= static_cast<int>(d_doppler_max);
doppler += d_doppler_step)
{
d_num_doppler_bins++;
}
// Create the carrier Doppler wipeoff signals
d_grid_doppler_wipeoffs = new gr_complex*[d_num_doppler_bins];
for (unsigned int doppler_index = 0; doppler_index < d_num_doppler_bins; doppler_index++)
{
d_grid_doppler_wipeoffs[doppler_index] = static_cast<gr_complex*>(volk_malloc(d_fft_size * sizeof(gr_complex), volk_get_alignment()));
int doppler = -static_cast<int>(d_doppler_max) + d_doppler_step * doppler_index;
float phase_step_rad = GALILEO_TWO_PI * (d_freq + doppler) / static_cast<float>(d_fs_in);
float _phase[1];
_phase[0] = 0;
volk_gnsssdr_s32f_sincos_32fc(d_grid_doppler_wipeoffs[doppler_index], - phase_step_rad, _phase, d_fft_size);
}
/* CAF Filtering to resolve doppler ambiguity. Phase and quadrature must be processed
* separately before non-coherent integration */
// if (d_CAF_filter)
if (d_CAF_window_hz > 0)
{
d_CAF_vector = static_cast<float*>(volk_malloc(d_num_doppler_bins * sizeof(float), volk_get_alignment()));
d_CAF_vector_I = static_cast<float*>(volk_malloc(d_num_doppler_bins * sizeof(float), volk_get_alignment()));
if (d_both_signal_components == true)
{
d_CAF_vector_Q = static_cast<float*>(volk_malloc(d_num_doppler_bins * sizeof(float), volk_get_alignment()));
}
}
}
void galileo_e5a_noncoherentIQ_acquisition_caf_cc::set_state(int state)
{
d_state = state;
if (d_state == 1)
{
d_gnss_synchro->Acq_delay_samples = 0.0;
d_gnss_synchro->Acq_doppler_hz = 0.0;
d_gnss_synchro->Acq_samplestamp_samples = 0;
d_well_count = 0;
d_mag = 0.0;
d_input_power = 0.0;
d_test_statistics = 0.0;
}
else if (d_state == 0)
{}
else
{
LOG(ERROR) << "State can only be set to 0 or 1";
}
}
int galileo_e5a_noncoherentIQ_acquisition_caf_cc::general_work(int noutput_items,
gr_vector_int &ninput_items, gr_vector_const_void_star &input_items,
gr_vector_void_star &output_items __attribute__((unused)))
{
/*
* By J.Arribas, L.Esteve, M.Molina and M.Sales
* Acquisition strategy (Kay Borre book + CFAR threshold):
* 1. Compute the input signal power estimation
* 2. Doppler serial search loop
* 3. Perform the FFT-based circular convolution (parallel time search)
* 4. OPTIONAL: CAF filter to avoid doppler ambiguity
* 5. Record the maximum peak and the associated synchronization parameters
* 6. Compute the test statistics and compare to the threshold
* 7. Declare positive or negative acquisition using a message queue
*/
int acquisition_message = -1; //0=STOP_CHANNEL 1=ACQ_SUCCEES 2=ACQ_FAIL
/* States: 0 Stop Channel
* 1 Load the buffer until it reaches fft_size
* 2 Acquisition algorithm
* 3 Positive acquisition
* 4 Negative acquisition
*/
switch (d_state)
{
case 0:
{
if (d_active)
{
//restart acquisition variables
d_gnss_synchro->Acq_delay_samples = 0.0;
d_gnss_synchro->Acq_doppler_hz = 0.0;
d_gnss_synchro->Acq_samplestamp_samples = 0;
d_well_count = 0;
d_mag = 0.0;
d_input_power = 0.0;
d_test_statistics = 0.0;
d_state = 1;
}
d_sample_counter += ninput_items[0]; // sample counter
consume_each(ninput_items[0]);
break;
}
case 1:
{
const gr_complex *in = (const gr_complex *)input_items[0]; //Get the input samples pointer
unsigned int buff_increment;
if (ninput_items[0] + d_buffer_count <= d_fft_size)
{
buff_increment = ninput_items[0];
}
else
{
buff_increment = (d_fft_size - d_buffer_count);
}
memcpy(&d_inbuffer[d_buffer_count], in, sizeof(gr_complex) * buff_increment);
// If buffer will be full in next iteration
if (d_buffer_count >= d_fft_size - d_gr_stream_buffer)
{
d_state = 2;
}
d_buffer_count += buff_increment;
d_sample_counter += buff_increment; // sample counter
consume_each(buff_increment);
break;
}
case 2:
{
// Fill last part of the buffer and reset counter
const gr_complex *in = (const gr_complex *)input_items[0]; //Get the input samples pointer
if (d_buffer_count < d_fft_size)
{
memcpy(&d_inbuffer[d_buffer_count], in, sizeof(gr_complex)*(d_fft_size-d_buffer_count));
}
d_sample_counter += d_fft_size-d_buffer_count; // sample counter
// initialize acquisition algorithm
int doppler;
unsigned int indext = 0;
unsigned int indext_IA = 0;
unsigned int indext_IB = 0;
unsigned int indext_QA = 0;
unsigned int indext_QB = 0;
float magt = 0.0;
float magt_IA = 0.0;
float magt_IB = 0.0;
float magt_QA = 0.0;
float magt_QB = 0.0;
float fft_normalization_factor = static_cast<float>(d_fft_size) * static_cast<float>(d_fft_size);
d_input_power = 0.0;
d_mag = 0.0;
d_well_count++;
DLOG(INFO) << "Channel: " << d_channel
<< " , doing acquisition of satellite: " << d_gnss_synchro->System << " "<< d_gnss_synchro->PRN
<< " ,sample stamp: " << d_sample_counter << ", threshold: "
<< d_threshold << ", doppler_max: " << d_doppler_max
<< ", doppler_step: " << d_doppler_step;
// 1- Compute the input signal power estimation
volk_32fc_magnitude_squared_32f(d_magnitudeIA, d_inbuffer, d_fft_size);
volk_32f_accumulator_s32f(&d_input_power, d_magnitudeIA, d_fft_size);
d_input_power /= static_cast<float>(d_fft_size);
// 2- Doppler frequency search loop
for (unsigned int doppler_index = 0; doppler_index < d_num_doppler_bins; doppler_index++)
{
// doppler search steps
doppler = -static_cast<int>(d_doppler_max) + d_doppler_step * doppler_index;
volk_32fc_x2_multiply_32fc(d_fft_if->get_inbuf(), d_inbuffer,
d_grid_doppler_wipeoffs[doppler_index], d_fft_size);
// 3- Perform the FFT-based convolution (parallel time search)
// Compute the FFT of the carrier wiped--off incoming signal
d_fft_if->execute();
// CODE IA
// Multiply carrier wiped--off, Fourier transformed incoming signal
// with the local FFT'd code reference using SIMD operations with VOLK library
volk_32fc_x2_multiply_32fc(d_ifft->get_inbuf(),
d_fft_if->get_outbuf(), d_fft_code_I_A, d_fft_size);
// compute the inverse FFT
d_ifft->execute();
// Search maximum
volk_32fc_magnitude_squared_32f(d_magnitudeIA, d_ifft->get_outbuf(), d_fft_size);
volk_32f_index_max_16u(&indext_IA, d_magnitudeIA, d_fft_size);
// Normalize the maximum value to correct the scale factor introduced by FFTW
magt_IA = d_magnitudeIA[indext_IA] / (fft_normalization_factor * fft_normalization_factor);
if (d_both_signal_components == true)
{
// REPEAT FOR ALL CODES. CODE_QA
volk_32fc_x2_multiply_32fc(d_ifft->get_inbuf(),
d_fft_if->get_outbuf(), d_fft_code_Q_A, d_fft_size);
d_ifft->execute();
volk_32fc_magnitude_squared_32f(d_magnitudeQA, d_ifft->get_outbuf(), d_fft_size);
volk_32f_index_max_16u(&indext_QA, d_magnitudeQA, d_fft_size);
magt_QA = d_magnitudeQA[indext_QA] / (fft_normalization_factor * fft_normalization_factor);
}
if (d_sampled_ms > 1) // If Integration time > 1 code
{
// REPEAT FOR ALL CODES. CODE_IB
volk_32fc_x2_multiply_32fc(d_ifft->get_inbuf(),
d_fft_if->get_outbuf(), d_fft_code_I_B, d_fft_size);
d_ifft->execute();
volk_32fc_magnitude_squared_32f(d_magnitudeIB, d_ifft->get_outbuf(), d_fft_size);
volk_32f_index_max_16u(&indext_IB, d_magnitudeIB, d_fft_size);
magt_IB = d_magnitudeIB[indext_IB] / (fft_normalization_factor * fft_normalization_factor);
if (d_both_signal_components == true)
{
// REPEAT FOR ALL CODES. CODE_QB
volk_32fc_x2_multiply_32fc(d_ifft->get_inbuf(),
d_fft_if->get_outbuf(), d_fft_code_Q_B, d_fft_size);
d_ifft->execute();
volk_32fc_magnitude_squared_32f(d_magnitudeQB, d_ifft->get_outbuf(), d_fft_size);
volk_32f_index_max_16u(&indext_QB, d_magnitudeQB, d_fft_size);
magt_QB = d_magnitudeIB[indext_QB] / (fft_normalization_factor * fft_normalization_factor);
}
}
// Integrate noncoherently the two best combinations (I² + Q²)
// and store the result in the I channel.
// If CAF filter to resolve doppler ambiguity is needed,
// peak is stored before non-coherent integration.
if (d_sampled_ms > 1) // T_integration > 1 code
{
if (magt_IA >= magt_IB)
{
// if (d_CAF_filter) {d_CAF_vector_I[doppler_index] = magt_IA;}
if (d_CAF_window_hz > 0) {d_CAF_vector_I[doppler_index] = d_magnitudeIA[indext_IA];}
if (d_both_signal_components)
{
// Integrate non-coherently I+Q
if (magt_QA >= magt_QB)
{
// if (d_CAF_filter) {d_CAF_vector_Q[doppler_index] = magt_QA;}
if (d_CAF_window_hz > 0) {d_CAF_vector_Q[doppler_index] = d_magnitudeQA[indext_QA];}
for (unsigned int i=0; i<d_fft_size; i++)
{
d_magnitudeIA[i] += d_magnitudeQA[i];
}
}
else
{
// if (d_CAF_filter) {d_CAF_vector_Q[doppler_index] = magt_QB;}
if (d_CAF_window_hz > 0) {d_CAF_vector_Q[doppler_index] = d_magnitudeQB[indext_QB];}
for (unsigned int i=0; i<d_fft_size; i++)
{
d_magnitudeIA[i] += d_magnitudeQB[i];
}
}
}
volk_32f_index_max_16u(&indext, d_magnitudeIA, d_fft_size);
magt = d_magnitudeIA[indext] / (fft_normalization_factor * fft_normalization_factor);
}
else
{
// if (d_CAF_filter) {d_CAF_vector_I[doppler_index] = magt_IB;}
if (d_CAF_window_hz > 0) {d_CAF_vector_I[doppler_index] = d_magnitudeIB[indext_IB];}
if (d_both_signal_components)
{
// Integrate non-coherently I+Q
if (magt_QA >= magt_QB)
{
//if (d_CAF_filter) {d_CAF_vector_Q[doppler_index] = magt_QA;}
if (d_CAF_window_hz > 0) {d_CAF_vector_Q[doppler_index] = d_magnitudeQA[indext_QA];}
for (unsigned int i=0; i<d_fft_size; i++)
{
d_magnitudeIB[i] += d_magnitudeQA[i];
}
}
else
{
// if (d_CAF_filter) {d_CAF_vector_Q[doppler_index] = magt_QB;}
if (d_CAF_window_hz > 0) {d_CAF_vector_Q[doppler_index] = d_magnitudeQB[indext_QB];}
for (unsigned int i=0; i<d_fft_size; i++)
{
d_magnitudeIB[i] += d_magnitudeQB[i];
}
}
}
volk_32f_index_max_16u(&indext, d_magnitudeIB, d_fft_size);
magt = d_magnitudeIB[indext] / (fft_normalization_factor * fft_normalization_factor);
}
}
else
{
// if (d_CAF_filter) {d_CAF_vector_I[doppler_index] = magt_IA;}
if (d_CAF_window_hz > 0) {d_CAF_vector_I[doppler_index] = d_magnitudeIA[indext_IA];}
if (d_both_signal_components)
{
// if (d_CAF_filter) {d_CAF_vector_Q[doppler_index] = magt_QA;}
if (d_CAF_window_hz > 0) {d_CAF_vector_Q[doppler_index] = d_magnitudeQA[indext_QA];}
// NON-Coherent integration of only 1 code
for (unsigned int i=0; i<d_fft_size; i++)
{
d_magnitudeIA[i] += d_magnitudeQA[i];
}
}
volk_32f_index_max_16u(&indext, d_magnitudeIA, d_fft_size);
magt = d_magnitudeIA[indext] / (fft_normalization_factor * fft_normalization_factor);
}
// 4- record the maximum peak and the associated synchronization parameters
if (d_mag < magt)
{
d_mag = magt;
// In case that d_bit_transition_flag = true, we compare the potentially
// new maximum test statistics (d_mag/d_input_power) with the value in
// d_test_statistics. When the second dwell is being processed, the value
// of d_mag/d_input_power could be lower than d_test_statistics (i.e,
// the maximum test statistics in the previous dwell is greater than
// current d_mag/d_input_power). Note that d_test_statistics is not
// restarted between consecutive dwells in multidwell operation.
if (d_test_statistics < (d_mag / d_input_power) || !d_bit_transition_flag)
{
d_gnss_synchro->Acq_delay_samples = static_cast<double>(indext % d_samples_per_code);
d_gnss_synchro->Acq_doppler_hz = static_cast<double>(doppler);
d_gnss_synchro->Acq_samplestamp_samples = d_sample_counter;
// 5- Compute the test statistics and compare to the threshold
d_test_statistics = d_mag / d_input_power;
}
}
// Record results to file if required
if (d_dump)
{
std::stringstream filename;
std::streamsize n = sizeof(float) * (d_fft_size); // noncomplex file write
filename.str("");
filename << "../data/test_statistics_E5a_sat_"
<< d_gnss_synchro->PRN << "_doppler_" << doppler << ".dat";
d_dump_file.open(filename.str().c_str(), std::ios::out | std::ios::binary);
if (d_sampled_ms > 1) // If integration time > 1 code
{
if (magt_IA >= magt_IB)
{
d_dump_file.write((char*)d_magnitudeIA, n);
}
else
{
d_dump_file.write((char*)d_magnitudeIB, n);
}
}
else
{
d_dump_file.write((char*)d_magnitudeIA, n);
}
d_dump_file.close();
}
}
// std::cout << "d_mag " << d_mag << ".d_sample_counter " << d_sample_counter << ". acq delay " << d_gnss_synchro->Acq_delay_samples<< " indext "<< indext << std::endl;
// 6 OPTIONAL: CAF filter to avoid Doppler ambiguity in bit transition.
if (d_CAF_window_hz > 0)
{
int CAF_bins_half;
float* accum = static_cast<float*>(volk_malloc(sizeof(float), volk_get_alignment()));
CAF_bins_half = d_CAF_window_hz / (2 * d_doppler_step);
float weighting_factor;
weighting_factor = 0.5 / static_cast<float>(CAF_bins_half);
// weighting_factor = 0;
// std::cout << "weighting_factor " << weighting_factor << std::endl;
// Initialize first iterations
for (int doppler_index = 0; doppler_index < CAF_bins_half; doppler_index++)
{
d_CAF_vector[doppler_index] = 0;
// volk_32f_accumulator_s32f_a(&d_CAF_vector[doppler_index], d_CAF_vector_I, CAF_bins_half+doppler_index+1);
for (int i = 0; i < CAF_bins_half + doppler_index + 1; i++)
{
d_CAF_vector[doppler_index] += d_CAF_vector_I[i] * (1 - weighting_factor * static_cast<unsigned int>((doppler_index - i)));
}
// d_CAF_vector[doppler_index] /= CAF_bins_half+doppler_index+1;
d_CAF_vector[doppler_index] /= 1 + CAF_bins_half + doppler_index - weighting_factor * CAF_bins_half * (CAF_bins_half + 1) / 2 - weighting_factor * doppler_index * (doppler_index + 1) / 2; // triangles = [n*(n+1)/2]
if (d_both_signal_components)
{
accum[0] = 0;
// volk_32f_accumulator_s32f_a(&accum[0], d_CAF_vector_Q, CAF_bins_half+doppler_index+1);
for (int i = 0; i < CAF_bins_half + doppler_index + 1; i++)
{
accum[0] += d_CAF_vector_Q[i] * (1 - weighting_factor * static_cast<unsigned int>(abs(doppler_index - i)));
}
// accum[0] /= CAF_bins_half+doppler_index+1;
accum[0] /= 1 + CAF_bins_half + doppler_index - weighting_factor * CAF_bins_half * (CAF_bins_half + 1) / 2 - weighting_factor * doppler_index * (doppler_index + 1) / 2; // triangles = [n*(n+1)/2]
d_CAF_vector[doppler_index] += accum[0];
}
}
// Body loop
for (unsigned int doppler_index = CAF_bins_half; doppler_index < d_num_doppler_bins - CAF_bins_half; doppler_index++)
{
d_CAF_vector[doppler_index] = 0;
// volk_32f_accumulator_s32f_a(&d_CAF_vector[doppler_index], &d_CAF_vector_I[doppler_index-CAF_bins_half], 2*CAF_bins_half+1);
for (int i = doppler_index - CAF_bins_half; i < static_cast<int>(doppler_index + CAF_bins_half + 1); i++)
{
d_CAF_vector[doppler_index] += d_CAF_vector_I[i] * (1 - weighting_factor * static_cast<unsigned int>((doppler_index - i)));
}
// d_CAF_vector[doppler_index] /= 2*CAF_bins_half+1;
d_CAF_vector[doppler_index] /= 1 + 2 * CAF_bins_half - 2 * weighting_factor * CAF_bins_half * (CAF_bins_half + 1) / 2;
if (d_both_signal_components)
{
accum[0] = 0;
// volk_32f_accumulator_s32f_a(&accum[0], &d_CAF_vector_Q[doppler_index-CAF_bins_half], 2*CAF_bins_half);
for (int i = doppler_index-CAF_bins_half; i < static_cast<int>(doppler_index + CAF_bins_half + 1); i++)
{
accum[0] += d_CAF_vector_Q[i] * (1 - weighting_factor * static_cast<unsigned int>((doppler_index - i)));
}
// accum[0] /= 2*CAF_bins_half+1;
accum[0] /= 1 + 2 * CAF_bins_half - 2 * weighting_factor * CAF_bins_half * (CAF_bins_half + 1) / 2;
d_CAF_vector[doppler_index] += accum[0];
}
}
// Final iterations
for (int doppler_index = d_num_doppler_bins - CAF_bins_half; doppler_index < static_cast<int>(d_num_doppler_bins); doppler_index++)
{
d_CAF_vector[doppler_index] = 0;
// volk_32f_accumulator_s32f_a(&d_CAF_vector[doppler_index], &d_CAF_vector_I[doppler_index-CAF_bins_half], CAF_bins_half + (d_num_doppler_bins-doppler_index));
for (int i = doppler_index - CAF_bins_half; i < static_cast<int>(d_num_doppler_bins); i++)
{
d_CAF_vector[doppler_index] += d_CAF_vector_I[i] * (1 - weighting_factor * (abs(doppler_index - i)));
}
// d_CAF_vector[doppler_index] /= CAF_bins_half+(d_num_doppler_bins-doppler_index);
d_CAF_vector[doppler_index] /= 1 + CAF_bins_half + (d_num_doppler_bins - doppler_index - 1) - weighting_factor * CAF_bins_half * (CAF_bins_half + 1) / 2 - weighting_factor * (d_num_doppler_bins - doppler_index - 1) * (d_num_doppler_bins - doppler_index) / 2;
if (d_both_signal_components)
{
accum[0] = 0;
// volk_32f_accumulator_s32f_a(&accum[0], &d_CAF_vector_Q[doppler_index-CAF_bins_half], CAF_bins_half + (d_num_doppler_bins-doppler_index));
for (int i = doppler_index-CAF_bins_half; i < static_cast<int>(d_num_doppler_bins); i++)
{
accum[0] += d_CAF_vector_Q[i] * (1 - weighting_factor * (abs(doppler_index - i)));
}
// accum[0] /= CAF_bins_half+(d_num_doppler_bins-doppler_index);
accum[0] /= 1 + CAF_bins_half + (d_num_doppler_bins - doppler_index - 1) - weighting_factor * CAF_bins_half * (CAF_bins_half + 1) / 2 - weighting_factor * (d_num_doppler_bins - doppler_index - 1) * (d_num_doppler_bins - doppler_index) / 2;
d_CAF_vector[doppler_index] += accum[0];
}
}
// Recompute the maximum doppler peak
volk_32f_index_max_16u(&indext, d_CAF_vector, d_num_doppler_bins);
doppler = -static_cast<int>(d_doppler_max) + d_doppler_step * indext;
d_gnss_synchro->Acq_doppler_hz = static_cast<double>(doppler);
// Dump if required, appended at the end of the file
if (d_dump)
{
std::stringstream filename;
std::streamsize n = sizeof(float) * (d_num_doppler_bins); // noncomplex file write
filename.str("");
filename << "../data/test_statistics_E5a_sat_" << d_gnss_synchro->PRN << "_CAF.dat";
d_dump_file.open(filename.str().c_str(), std::ios::out | std::ios::binary);
d_dump_file.write((char*)d_CAF_vector, n);
d_dump_file.close();
}
volk_free(accum);
}
if (d_well_count == d_max_dwells)
{
if (d_test_statistics > d_threshold)
{
d_state = 3; // Positive acquisition
}
else
{
d_state = 4; // Negative acquisition
}
}
else
{
d_state = 1;
}
consume_each(d_fft_size - d_buffer_count);
d_buffer_count = 0;
break;
}
case 3:
{
// 7.1- Declare positive acquisition using a message queue
DLOG(INFO) << "positive acquisition";
DLOG(INFO) << "satellite " << d_gnss_synchro->System << " " << d_gnss_synchro->PRN;
DLOG(INFO) << "sample_stamp " << d_sample_counter;
DLOG(INFO) << "test statistics value " << d_test_statistics;
DLOG(INFO) << "test statistics threshold " << d_threshold;
DLOG(INFO) << "code phase " << d_gnss_synchro->Acq_delay_samples;
DLOG(INFO) << "doppler " << d_gnss_synchro->Acq_doppler_hz;
DLOG(INFO) << "magnitude " << d_mag;
DLOG(INFO) << "input signal power " << d_input_power;
d_active = false;
d_state = 0;
acquisition_message = 1;
this->message_port_pub(pmt::mp("events"), pmt::from_long(acquisition_message));
d_sample_counter += ninput_items[0]; // sample counter
consume_each(ninput_items[0]);
break;
}
case 4:
{
// 7.2- Declare negative acquisition using a message queue
DLOG(INFO) << "negative acquisition";
DLOG(INFO) << "satellite " << d_gnss_synchro->System << " " << d_gnss_synchro->PRN;
DLOG(INFO) << "sample_stamp " << d_sample_counter;
DLOG(INFO) << "test statistics value " << d_test_statistics;
DLOG(INFO) << "test statistics threshold " << d_threshold;
DLOG(INFO) << "code phase " << d_gnss_synchro->Acq_delay_samples;
DLOG(INFO) << "doppler " << d_gnss_synchro->Acq_doppler_hz;
DLOG(INFO) << "magnitude " << d_mag;
DLOG(INFO) << "input signal power " << d_input_power;
d_active = false;
d_state = 0;
d_sample_counter += ninput_items[0]; // sample counter
consume_each(ninput_items[0]);
acquisition_message = 2;
this->message_port_pub(pmt::mp("events"), pmt::from_long(acquisition_message));
break;
}
}
return noutput_items;
}