March 2024 - gerrit-log - lists.osmocom.org

[L] Change in osmo-trx[master]: ms: add demod test tool and data

by Hoernchen

Hoernchen has submitted this change. ( https://gerrit.osmocom.org/c/osmo-trx/+/36047?usp=email ) Change subject: ms: add demod test tool and data ...................................................................... ms: add demod test tool and data It just uses the viterbi equalizer and the sigproclib to generate and demodulate bursts and prints the bits, only useful for development. Change-Id: I852e34d9667d1f12f235f8b3da1fcc0d738b2db9 --- M .checkpatch.conf M .gitignore M configure.ac M utils/Makefile.am A utils/va-test/Makefile.am A utils/va-test/burst-gen.cpp A utils/va-test/demodbits_tsc7.s8 A utils/va-test/nb_chunk_tsc7.cfile 8 files changed, 566 insertions(+), 0 deletions(-) Approvals: dexter: Looks good to me, but someone else must approve; Verified laforge: Looks good to me, but someone else must approve Jenkins Builder: Verified Hoernchen: Looks good to me, approved diff --git a/.checkpatch.conf b/.checkpatch.conf index 1e4b479..1699801 100644 --- a/.checkpatch.conf +++ b/.checkpatch.conf @@ -1,4 +1,5 @@ --exclude osmocom-bb/.* --exclude .*h --exclude Transceiver52M/grgsm_vitac/.* +--exclude utils/va-test/.* --ignore FUNCTION_WITHOUT_ARGS \ No newline at end of file diff --git a/.gitignore b/.gitignore index 0b33e0f..581b08f 100644 --- a/.gitignore +++ b/.gitignore @@ -83,6 +83,7 @@ !contrib/osmo-trx.spec.in utils/osmo-prbs-tool +utils/va-test/osmo-burst-gen /.qtc_clangd/* /.cache/* /.vscode/* diff --git a/configure.ac b/configure.ac index 69055fd..8aecb27 100644 --- a/configure.ac +++ b/configure.ac @@ -374,6 +374,7 @@ tests/CommonLibs/Makefile \ tests/Transceiver52M/Makefile \ utils/Makefile \ + utils/va-test/Makefile \ doc/Makefile \ doc/examples/Makefile \ contrib/Makefile \ diff --git a/utils/Makefile.am b/utils/Makefile.am index bac9a7f..1818093 100644 --- a/utils/Makefile.am +++ b/utils/Makefile.am @@ -1,6 +1,8 @@ AM_CPPFLAGS = $(LIBOSMOCODING_CFLAGS) AM_CFLAGS = -Wall +DIST_SUBDIRS = va-test + EXTRA_DIST = clockdump.sh matlab noinst_PROGRAMS = osmo-prbs-tool diff --git a/utils/va-test/Makefile.am b/utils/va-test/Makefile.am new file mode 100644 index 0000000..d5a72e5 --- /dev/null +++ b/utils/va-test/Makefile.am @@ -0,0 +1,17 @@ +include $(top_srcdir)/Makefile.common + +noinst_PROGRAMS = osmo-burst-gen + +osmo_burst_gen_SOURCES = burst-gen.cpp \ + ${top_srcdir}/Transceiver52M/grgsm_vitac/grgsm_vitac.cpp \ + ${top_srcdir}/Transceiver52M/grgsm_vitac/viterbi_detector.cc +osmo_burst_gen_LDADD = \ + ${top_srcdir}/Transceiver52M/libtransceiver_common.la \ + $(ARCH_LA) \ + $(GSM_LA) \ + $(COMMON_LA) +osmo_burst_gen_CPPFLAGS = -Wall $(STD_DEFINES_AND_INCLUDES) \ + -I${top_srcdir}/Transceiver52M/arch/common \ + -I${top_srcdir}/Transceiver52M/device/common \ + -I${top_srcdir}/Transceiver52M + diff --git a/utils/va-test/burst-gen.cpp b/utils/va-test/burst-gen.cpp new file mode 100644 index 0000000..bdd0c4b --- /dev/null +++ b/utils/va-test/burst-gen.cpp @@ -0,0 +1,531 @@ +/* + * (C) 2023 by sysmocom s.f.m.c. GmbH <info(a)sysmocom.de> + * All Rights Reserved + * + * Author: Eric Wild <ewild(a)sysmocom.de> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU Affero General Public License as published by + * the Free Software Foundation; either version 3 of the License, or + * (at your option) any later version. + * + * This program 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 Affero General Public License for more details. + * + * You should have received a copy of the GNU Affero General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + * + */ + +// this allows messing with the demod to check the detecton offset impact, +// not intended for actual automated tests. + +#include "sigProcLib.h" + +extern "C" { +#include "convert.h" +#include <convolve.h> +} + +#define _CRT_SECURE_NO_WARNINGS +#include <algorithm> +#include <string.h> +#include <iomanip> +#include <numeric> +#include <memory> +#include <iostream> +#include <fstream> + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <grgsm_vitac/grgsm_vitac.h> + +#define DO_RACH + +const int SAMPLE_SCALE_FACTOR = 1; + +template <typename DST_T, typename SRC_T, typename ST> +void convert_and_scale(void *dst, void *src, unsigned int src_len, ST scale) +{ + for (unsigned int i = 0; i < src_len; i++) + reinterpret_cast<DST_T *>(dst)[i] = static_cast<DST_T>((reinterpret_cast<SRC_T *>(src)[i]) * scale); +} +template <typename DST_T, typename SRC_T> +void convert_and_scale_default(void *dst, void *src, unsigned int src_len) +{ + return convert_and_scale<DST_T, SRC_T>(dst, src, src_len, SAMPLE_SCALE_FACTOR); +} + +static const unsigned int txFullScale = (float)(1 << 14) - 1; +// static const unsigned int rxFullScale = (float)(1 << 14) - 1; + +static const BitVector + gRACHBurstx("0011101001001011011111111001100110101010001111000110111101111110000111001001010110011000"); + +static const BitVector gTrainingSequencex[] = { + BitVector("00100101110000100010010111"), BitVector("00101101110111100010110111"), + BitVector("01000011101110100100001110"), BitVector("01000111101101000100011110"), + BitVector("00011010111001000001101011"), BitVector("01001110101100000100111010"), + BitVector("10100111110110001010011111"), BitVector("11101111000100101110111100"), +}; + +struct mrv { + std::vector<char> bits; + signalVector *rvbuf; + std::unique_ptr<std::vector<std::complex<float>>> convolved; + // mrv(): bits(), demod_bits() {} + CorrType ct; +}; + +static mrv genRandNormalBurstx(int tsc, int sps, int tn) +{ + mrv retstruct; + int i = 0; + BitVector bits(148); + + /* Tail bits */ + for (; i < 3; i++) + bits[i] = 0; + + /* Random bits */ + for (int j = 0; i < 60; i++, j++) + bits[i] = rand() % 2; + + /* Stealing bit */ + bits[i++] = 0; + + /* Training sequence */ + for (int n = 0; i < 87; i++, n++) + bits[i] = gTrainingSequencex[tsc][n]; + + /* Stealing bit */ + bits[i++] = 0; + + /* Random bits */ + for (; i < 145; i++) + bits[i] = rand() % 2; + + /* Tail bits */ + for (; i < 148; i++) + bits[i] = 0; + + int guard = 8 + !(tn % 4); + auto r = modulateBurst(bits, guard, sps); + + retstruct.rvbuf = r; + for (size_t i = 0; i < bits.size(); i++) + retstruct.bits.push_back(bits.bit(i) ? 1 : 0); + return retstruct; +} + +static mrv genRandAccessBurstx(int delay, int sps, int tn) +{ + mrv retstruct; + int i = 0; + BitVector bits(88 + delay); + + /* delay */ + for (; i < delay; i++) + bits[i] = 0; + + /* head and synch bits */ + for (int n = 0; i < 49 + delay; i++, n++) + bits[i] = gRACHBurstx[n]; + + /* Random bits */ + for (int j = 0; i < 85 + delay; i++, j++) + bits[i] = rand() % 2; + + for (; i < 88 + delay; i++) + bits[i] = 0; + + int guard = 68 - delay + !(tn % 4); + auto r = modulateBurst(bits, guard, sps); + + retstruct.rvbuf = r; + for (size_t i = 0; i < bits.size(); i++) + retstruct.bits.push_back(bits.bit(i) ? 1 : 0); + return retstruct; +} + +extern gr_complex d_acc_training_seq[N_ACCESS_BITS]; ///<encoded training sequence of a SCH burst +extern gr_complex d_sch_training_seq[N_SYNC_BITS]; ///<encoded training sequence of a SCH burst +extern gr_complex d_norm_training_seq[TRAIN_SEQ_NUM] + [N_TRAIN_BITS]; ///<encoded training sequences of a normal and dummy burst + +void sv_write_helper(signalVector *burst, std::string fname) +{ + auto start = burst->begin(); + auto n = burst->bytes(); + char *data = reinterpret_cast<char *>(start); + + const int len_in_real = burst->size() * 2; + auto cvrtbuf_tx_a = new int16_t[len_in_real]; + convert_float_short(cvrtbuf_tx_a, (float *)burst->begin(), float(txFullScale), len_in_real); + + std::ofstream fout; + fout.open(fname + ".cfile", std::ios::binary | std::ios::out); + fout.write(data, n); + fout.close(); + + fout.open(fname + ".cs16", std::ios::binary | std::ios::out); + fout.write((char *)cvrtbuf_tx_a, len_in_real * sizeof(uint16_t)); + fout.close(); + delete[] cvrtbuf_tx_a; +} + +// borrowed from a real world burst.. +static std::vector<std::complex<float>> chan_im_resp = { + { 4.1588e-05 + -0.000361925 }, { 0.000112728 + -0.000289796 }, { 0.000162952 + -0.000169028 }, + { 0.000174185 + -2.54575e-05 }, { 0.000142947 + 0.000105992 }, { 8.65919e-05 + 0.000187041 }, + { 4.15799e-05 + 0.000184346 }, { 5.30207e-05 + 7.84921e-05 }, { 0.000158877 + -0.000128058 }, + { 0.000373956 + -0.000407954 }, { 0.000680606 + -0.000712065 }, { 0.00102929 + -0.000979604 }, + { 0.00135049 + -0.00115333 }, { 0.00157434 + -0.0011948 }, { 0.00165098 + -0.00109534 }, + { 0.00156519 + -0.000878794 }, { 0.0013399 + -0.000594285 }, { 0.00102788 + -0.00030189 }, + { 0.000694684 + -5.58912e-05 }, { 0.000399328 + 0.000109463 } +}; + +// as above, downsampled to 1sps + just magnitude +static std::vector<float> chan_im_resp_trunc = { 1., 0.20513351, 0.10020305, 0.11490235 }; + +template <typename A, typename B> +auto conv(const std::vector<A> &a, const std::vector<B> &b) -> std::unique_ptr<std::vector<A>> +{ + int data_len = a.size(); + int conv_len = b.size(); + int conv_size = conv_len + data_len - 1; + auto retv = std::make_unique<std::vector<A>>(conv_size); + + for (int i = 0; i < data_len; ++i) { + for (int j = 0; j < conv_len; ++j) { + (*retv)[i + j] += a[i] * b[j]; + } + } + return retv; +} + +template <typename A> +static auto conv(const A *a, int len, std::vector<float> &b) +{ + std::vector<A> aa(len); + std::copy_n(a, len, aa.begin()); + std::reverse(b.begin(), b.end()); + return conv(aa, b); +} +template <typename A> +static auto conv(const A *a, int len, std::vector<A> &b) +{ + std::vector<A> aa(len); + std::copy_n(a, len, aa.begin()); + std::reverse(b.begin(), b.end()); + return conv(aa, b); +} + +// signalvector is owning despite claiming not to, but we can pretend, too.. +static void dummy_free(void *wData){}; +static void *dummy_alloc(size_t newSize) +{ + return 0; +}; + +template <typename T> +size_t read_from_file(std::string path, std::vector<T> &outvec) +{ + std::ifstream infile; + infile.open(path, std::ios::in | std::ios::binary); + if (infile.fail()) { + std::cout << " not found: " << path << std::endl; + exit(0); + } + infile.seekg(0, std::ios_base::end); + size_t fsize = infile.tellg(); + auto fsize_in_T = fsize / sizeof(T); + infile.seekg(0, std::ios_base::beg); + + outvec.resize(fsize_in_T); + infile.read(reinterpret_cast<char *>(&outvec[0]), fsize); + infile.close(); + std::cout << "Read " << fsize << " from " << path << std::endl; + return fsize; +} +void demod_real_burst(int num = 0) +{ + auto path = "./nb_chunk_tsc7.cfile"; + auto bitfile = "./demodbits_tsc7.s8"; + + std::vector<std::complex<float>> burstdata; + std::vector<char> bitdata; + read_from_file(path, burstdata); + read_from_file(bitfile, bitdata); + + // print "known good" burst bits + std::cerr << "known bits:" << std::endl; + std::cerr << std::setw(5) << 0 << " - "; + for (auto i : bitdata) + std::cout << (i > 0 ? "1" : "0"); + std::cerr << std::endl; + std::cerr << "demod tests sigproclib:" << std::endl; + + auto ct = CorrType::TSC; + auto delay = 0; + auto tsc = 7; + int offset = 0; + auto cplx = reinterpret_cast<complex *>(&burstdata[offset]); + auto stdcplx = reinterpret_cast<std::complex<float> *>(&burstdata[offset]); + signalVector sv(&cplx[0], 0, burstdata.size() - offset, dummy_alloc, dummy_free); + + struct estim_burst_params ebp; + auto rc = detectAnyBurst(sv, tsc, BURST_THRESH, 4, ct, 40, &ebp); + + auto rxBurst = std::unique_ptr<SoftVector>(demodAnyBurst(sv, (CorrType)rc, 4, &ebp)); + // print osmotrx sigproclib demod result + std::cerr << std::setw(5) << int(ebp.toa) << " o "; + for (ssize_t i = 0 + delay; i < 148 + delay; i++) + std::cout << (rxBurst->bit(i) ? "1" : "0"); + std::cerr << std::endl; + + std::cerr << "demod test va:" << std::endl; + std::complex<float> chan_imp_resp[CHAN_IMP_RESP_LENGTH * d_OSR]; + float ncmax; + char demodded_softbits[444]; + + // demod at known offset + { + auto inp = &stdcplx[29]; // known offset + auto normal_burst_startX = get_norm_chan_imp_resp(inp, &chan_imp_resp[0], &ncmax, tsc); + detect_burst_nb(inp, &chan_imp_resp[0], normal_burst_startX, demodded_softbits); + + std::cerr << std::setw(5) << normal_burst_startX << " v "; + for (size_t i = 0; i < 148; i++) + std::cerr << (demodded_softbits[i] < 0 ? "1" : "0"); + std::cerr << std::endl; + } + { + std::cerr << "-- va start offset loop --" << std::endl; + std::cerr << "offset/det offset/#errors/known^demod bits" << std::endl; + for (int i = 0; i < 34; i++) { + auto inp = &stdcplx[i]; + auto conved_beg = inp; + + auto me = get_norm_chan_imp_resp(conved_beg, &chan_imp_resp[0], &ncmax, tsc); + detect_burst_nb(conved_beg, &chan_imp_resp[0], me, demodded_softbits); + auto bitdiffarr = std::make_unique<char[]>(148); + for (size_t i = 0; i < 148; i++) + bitdiffarr.get()[i] = (demodded_softbits[i] < 0 ? 1 : 0) ^ (bitdata[i] > 0 ? 1 : 0); + auto ber = std::accumulate(bitdiffarr.get(), bitdiffarr.get() + 148, 0); + + std::cerr << std::setw(3) << i << ": " << std::setw(3) << me << " v " << std::setw(3) << ber + << " "; + for (size_t i = 0; i < 148; i++) + std::cerr << (bitdiffarr[i] ? "1" : "0"); + std::cerr << std::endl; + // std::cerr << std::setw(4) << i << " (" << std::setw(4) << 29 - i << "):" << std::setw(4) << org + // << " " << std::setw(4) << me << " y " << std::endl; + } + } +} + +auto gen_burst(CorrType t, int delay, int tsc) +{ + mrv rs; + if (t == CorrType::RACH) { + rs = genRandAccessBurstx(delay, 4, tsc); + + } else if (t == CorrType::TSC) { + rs = genRandNormalBurstx(tsc, 4, 0); + } else { + std::cerr << "wtf?" << std::endl; + exit(0); + } + rs.ct = t; + + signalVector *burst = rs.rvbuf; + // sv_write_helper(burst, std::to_string(num)); + // scaleVector(*burst, {1, 0}); + const int len_in_real = burst->size() * 2; + auto cvrtbuf_tx_a = std::make_unique<short[]>(len_in_real); + auto cvrtbuf_rx_a = std::make_unique<float[]>(len_in_real); + auto rx_cfloat = reinterpret_cast<std::complex<float> *>(&cvrtbuf_rx_a[0]); + + convert_float_short(cvrtbuf_tx_a.get(), (float *)burst->begin(), float(txFullScale), len_in_real); + convert_short_float(cvrtbuf_rx_a.get(), cvrtbuf_tx_a.get(), len_in_real); + for (int i = 0; i < len_in_real; i++) // scale properly! + cvrtbuf_rx_a[i] *= 1. / txFullScale; + auto conved = conv(rx_cfloat, burst->size(), chan_im_resp); + + std::cerr << "-- generated " << (t == CorrType::RACH ? "RACH" : "TSC") << " burst --" << std::endl; + for (size_t i = 0; i < rs.bits.size(); i++) + std::cerr << (rs.bits[i] ? "1" : "0"); + std::cerr << std::endl; + delete burst; + rs.convolved = std::move(conved); + return rs; +} + +void demod_generated_burst(CorrType t) +{ + int tsc = 0; + int delay = 0; + auto rs = gen_burst(t, delay, tsc); + auto conved_beg = &(*rs.convolved)[0]; + + if (rs.ct == CorrType::RACH) { + std::complex<float> chan_imp_resp[CHAN_IMP_RESP_LENGTH * d_OSR]; + float ncmax; + char demodded_softbits[444]; + int normal_burst_start = 0; + normal_burst_start = get_access_imp_resp(conved_beg, &chan_imp_resp[0], &ncmax, 0); + normal_burst_start = std::max(normal_burst_start, 0); + for (int j = 0; j < 4; j++) { + for (int start_val = 0; start_val < 16; start_val++) { + auto bitdiffarr = std::make_unique<char[]>(rs.bits.size()); + detect_burst_ab(conved_beg, &chan_imp_resp[0], normal_burst_start + j, + demodded_softbits, start_val); + + for (size_t i = 0; i < rs.bits.size(); i++) + bitdiffarr.get()[i] = (demodded_softbits[i] < 0 ? 1 : 0) ^ rs.bits[i]; + auto ber = std::accumulate(bitdiffarr.get(), bitdiffarr.get() + rs.bits.size(), 0); + + std::cerr << "ber " << std::setw(4) << ber << " bo:" << std::setw(4) << j + << " vas:" << std::setw(4) << start_val << " "; + // for (size_t i = 0; i < rs.num_bits; i++) + // std::cerr << (demodded_softbits[i] < 0 ? "1" : "0"); + // std::cerr << std::endl; + // std::cerr << "d " << std::setw(4) << ber << " "; + for (size_t i = 0; i < rs.bits.size(); i++) + std::cerr << (bitdiffarr.get()[i] ? "1" : "0"); + std::cerr << std::endl; + + // std::cerr << "v " << std::setw(4) << j << std::setw(4) << start_val << " "; + // for (size_t i = 0; i < rs.num_bits; i++) + // std::cerr << (demodded_softbits[i] < 0 ? "1" : "0"); + // std::cerr << std::endl; + // std::cerr << "d " << std::setw(4) << ber << " "; + // for (size_t i = 0; i < rs.num_bits; i++) + // std::cerr << (ptr.get()[i] ? "1" : "0"); + // std::cerr << std::endl; + } + } + + } else { + std::complex<float> chan_imp_resp[CHAN_IMP_RESP_LENGTH * d_OSR]; + float ncmax; + char demodded_softbits[444]; + + auto normal_burst_start = get_norm_chan_imp_resp(conved_beg, &chan_imp_resp[0], &ncmax, tsc); + detect_burst_nb(conved_beg, &chan_imp_resp[0], normal_burst_start + 0, demodded_softbits); + std::cerr << "toa " << std::setprecision(2) << normal_burst_start << std::endl; + + std::cerr << "vita "; + for (size_t i = 0; i < rs.bits.size(); i++) + std::cerr << (demodded_softbits[i] < 0 ? "1" : "0"); + std::cerr << std::endl; + std::cerr << "diff "; + for (size_t i = 0; i < rs.bits.size(); i++) + std::cerr << ((demodded_softbits[i] < 0 ? 1 : 0) ^ rs.bits[i] ? "1" : "0"); + std::cerr << std::endl; + } + + struct estim_burst_params ebp; + char demodded_softbits[444]; + complex *rx_sigproc_cfloat = reinterpret_cast<complex *>(conved_beg); + signalVector sv(rx_sigproc_cfloat, 0, rs.convolved->size(), dummy_alloc, dummy_free); + + auto rc = detectAnyBurst(sv, tsc, BURST_THRESH, 4, rs.ct, 40, &ebp); + auto rxBurst = std::unique_ptr<SoftVector>(demodAnyBurst(sv, (CorrType)rc, 4, &ebp)); + + std::cerr << "toa " << std::setprecision(2) << ebp.toa << std::endl; + + for (ssize_t i = 0; i < delay; i++) // maybe pad rach op? + demodded_softbits[i] = 0; + for (size_t i = 0 + delay; i < rs.bits.size() + delay; i++) + demodded_softbits[i] = (rxBurst->bit(i) ? 1 : 0); + + std::cerr << "sigp "; + for (size_t i = 0; i < rs.bits.size(); i++) + std::cerr << (demodded_softbits[i] ? "1" : "0"); + std::cerr << std::endl; + + std::cerr << "diff "; + for (size_t i = 0; i < rs.bits.size(); i++) + std::cerr << (demodded_softbits[i] ^ rs.bits[i] ? "1" : "0"); + std::cerr << std::endl; +} + +void demod_test_offsets() +{ + const int tsc = 0; + const int delaybuffer_realoffset = 100; + + { + auto rs = gen_burst(CorrType::RACH, 0, tsc); + typeof(*rs.convolved) delay_buffer(rs.convolved->size() * 2); // plenty of space.. + + for (int delay = -10; delay < 60; delay++) { + std::fill(delay_buffer.begin(), delay_buffer.end(), 0); + std::copy(rs.convolved->begin(), rs.convolved->end(), + delay_buffer.begin() + delaybuffer_realoffset + delay); + + auto conved_beg = &delay_buffer[delaybuffer_realoffset]; + + std::complex<float> chan_imp_resp[CHAN_IMP_RESP_LENGTH * d_OSR]; + float ncmax; + auto va_burst_start = get_access_imp_resp(conved_beg, &chan_imp_resp[0], &ncmax, 60); + + complex *rx_sigproc_cfloat = reinterpret_cast<complex *>(conved_beg); + struct estim_burst_params ebp; + signalVector sv(rx_sigproc_cfloat, 0, rs.convolved->size(), dummy_alloc, dummy_free); + detectAnyBurst(sv, tsc, BURST_THRESH, 4, rs.ct, 60, &ebp); + std::cerr << "delay:" << std::setw(3) << std::setprecision(2) << delay; + std::cerr << " va: " << std::setw(3) << std::setprecision(2) << va_burst_start; + std::cerr << " sg: " << std::setw(3) << std::setprecision(2) << ebp.toa; + std::cerr << " d: " << std::setw(3) << std::setprecision(2) << (ebp.toa * 4) - va_burst_start; + std::cerr << " ! " << float(va_burst_start + 13) / 4 << std::endl; + } + } + { + auto rs = gen_burst(CorrType::TSC, 0, tsc); + typeof(*rs.convolved) delay_buffer(rs.convolved->size() * 2); // plenty of space.. + + for (int delay = -10; delay < 10; delay++) { + std::fill(delay_buffer.begin(), delay_buffer.end(), 0); + std::copy(rs.convolved->begin(), rs.convolved->end(), + delay_buffer.begin() + delaybuffer_realoffset + delay); + + auto conved_beg = &delay_buffer[delaybuffer_realoffset]; + + std::complex<float> chan_imp_resp[CHAN_IMP_RESP_LENGTH * d_OSR]; + float ncmax; + auto va_burst_start = get_norm_chan_imp_resp(conved_beg, &chan_imp_resp[0], &ncmax, tsc); + + complex *rx_sigproc_cfloat = reinterpret_cast<complex *>(conved_beg); + struct estim_burst_params ebp; + signalVector sv(rx_sigproc_cfloat, 0, rs.convolved->size(), dummy_alloc, dummy_free); + detectAnyBurst(sv, tsc, BURST_THRESH, 4, rs.ct, 60, &ebp); + std::cerr << "delay:" << std::setw(3) << std::setprecision(2) << delay; + std::cerr << " va: " << std::setw(3) << std::setprecision(2) << va_burst_start; + std::cerr << " sg: " << std::setw(3) << std::setprecision(2) << ebp.toa; + std::cerr << " d: " << std::setw(3) << std::setprecision(2) << (ebp.toa * 4) - va_burst_start; + std::cerr << " ! " << float(va_burst_start + 19) / 4 << std::endl; + } + } +} + +int main() +{ + convolve_init(); + convert_init(); + sigProcLibSetup(); + initvita(); + + for (int i = 0; i < 1; i++) { + demod_real_burst(i); + demod_generated_burst(CorrType::RACH); + demod_generated_burst(CorrType::TSC); + demod_test_offsets(); + } +} diff --git a/utils/va-test/demodbits_tsc7.s8 b/utils/va-test/demodbits_tsc7.s8 new file mode 100644 index 0000000..c91cd6c --- /dev/null +++ b/utils/va-test/demodbits_tsc7.s8 Binary files differ diff --git a/utils/va-test/nb_chunk_tsc7.cfile b/utils/va-test/nb_chunk_tsc7.cfile new file mode 100644 index 0000000..67e577e --- /dev/null +++ b/utils/va-test/nb_chunk_tsc7.cfile Binary files differ -- To view, visit https://gerrit.osmocom.org/c/osmo-trx/+/36047?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-Change-Id: I852e34d9667d1f12f235f8b3da1fcc0d738b2db9 Gerrit-Change-Number: 36047 Gerrit-PatchSet: 3 Gerrit-Owner: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: dexter <pmaier(a)sysmocom.de> Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-CC: pespin <pespin(a)sysmocom.de> Gerrit-MessageType: merged

1 year, 5 months

1
0
0 0

[L] Change in osmo-trx[master]: ms: add demod test tool and data

by Hoernchen

Attention is currently required from: pespin. Hoernchen has posted comments on this change. ( https://gerrit.osmocom.org/c/osmo-trx/+/36047?usp=email ) Change subject: ms: add demod test tool and data ...................................................................... Patch Set 3: Code-Review+2 -- To view, visit https://gerrit.osmocom.org/c/osmo-trx/+/36047?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-Change-Id: I852e34d9667d1f12f235f8b3da1fcc0d738b2db9 Gerrit-Change-Number: 36047 Gerrit-PatchSet: 3 Gerrit-Owner: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: dexter <pmaier(a)sysmocom.de> Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-CC: pespin <pespin(a)sysmocom.de> Gerrit-Attention: pespin <pespin(a)sysmocom.de> Gerrit-Comment-Date: Mon, 04 Mar 2024 20:05:26 +0000 Gerrit-HasComments: No Gerrit-Has-Labels: Yes Gerrit-MessageType: comment

1 year, 5 months

1
0
0 0

[L] Change in osmo-trx[master]: ms: add sigproclib demod

by Hoernchen

Hoernchen has submitted this change. ( https://gerrit.osmocom.org/c/osmo-trx/+/36046?usp=email ) Change subject: ms: add sigproclib demod ...................................................................... ms: add sigproclib demod This is basically a fixed version of ttsous ancient branch that can be used instead of the VA. Required config option part of a future patchset. Change-Id: I6558992bd69f18526be5ebe7d424ca00ceb67772 --- M GSM/GSMCommon.cpp M GSM/GSMCommon.h M Transceiver52M/Resampler.cpp M Transceiver52M/ms/ms_rx_lower.cpp M Transceiver52M/ms/ms_upper.cpp M Transceiver52M/sigProcLib.cpp M Transceiver52M/sigProcLib.h 7 files changed, 419 insertions(+), 11 deletions(-) Approvals: Jenkins Builder: Verified pespin: Looks good to me, but someone else must approve laforge: Looks good to me, but someone else must approve Hoernchen: Looks good to me, approved diff --git a/GSM/GSMCommon.cpp b/GSM/GSMCommon.cpp index 5e9e4ae..a9e2bb1 100644 --- a/GSM/GSMCommon.cpp +++ b/GSM/GSMCommon.cpp @@ -55,12 +55,15 @@ }; const BitVector GSM::gDummyBurst("0001111101101110110000010100100111000001001000100000001111100011100010111000101110001010111010010100011001100111001111010011111000100101111101010000"); +const BitVector GSM::gDummyBurstTSC("01110001011100010111000101"); /* 3GPP TS 05.02, section 5.2.7 "Access burst (AB)", synch. sequence bits */ const BitVector GSM::gRACHSynchSequenceTS0("01001011011111111001100110101010001111000"); /* GSM, GMSK (default) */ const BitVector GSM::gRACHSynchSequenceTS1("01010100111110001000011000101111001001101"); /* EGPRS, 8-PSK */ const BitVector GSM::gRACHSynchSequenceTS2("11101111001001110101011000001101101110111"); /* EGPRS, GMSK */ +const BitVector GSM::gSCHSynchSequence("1011100101100010000001000000111100101101010001010111011000011011"); + // |-head-||---------midamble----------------------||--------------data----------------||t| const BitVector GSM::gRACHBurst("0011101001001011011111111001100110101010001111000110111101111110000111001001010110011000"); diff --git a/GSM/GSMCommon.h b/GSM/GSMCommon.h index 48723b4..aa059c2 100644 --- a/GSM/GSMCommon.h +++ b/GSM/GSMCommon.h @@ -52,11 +52,16 @@ /** C0T0 filler burst, GSM 05.02, 5.2.6 */ extern const BitVector gDummyBurst; +extern const BitVector gDummyBurstTSC; /** Random access burst synch. sequence */ extern const BitVector gRACHSynchSequenceTS0; extern const BitVector gRACHSynchSequenceTS1; extern const BitVector gRACHSynchSequenceTS2; + +/** Synchronization burst sync sequence */ +extern const BitVector gSCHSynchSequence; + /** Random access burst synch. sequence, GSM 05.02 5.2.7 */ extern const BitVector gRACHBurst; diff --git a/Transceiver52M/Resampler.cpp b/Transceiver52M/Resampler.cpp index 910c7ff..841c3a9 100644 --- a/Transceiver52M/Resampler.cpp +++ b/Transceiver52M/Resampler.cpp @@ -32,7 +32,7 @@ #define M_PI 3.14159265358979323846264338327f #endif -#define MAX_OUTPUT_LEN 4096 +#define MAX_OUTPUT_LEN 4096*4 using namespace std; diff --git a/Transceiver52M/ms/ms_rx_lower.cpp b/Transceiver52M/ms/ms_rx_lower.cpp index c2adda3..d894e96 100644 --- a/Transceiver52M/ms/ms_rx_lower.cpp +++ b/Transceiver52M/ms/ms_rx_lower.cpp @@ -19,6 +19,8 @@ * */ +#include "sigProcLib.h" +#include "signalVector.h" #include <atomic> #include <cassert> #include <complex> @@ -155,12 +157,12 @@ auto current_gsm_time = timekeeper.gsmtime(); const auto buf_len = is_first_sch_acq ? SCH_LEN_SPS : ONE_TS_BURST_LEN; const auto which_in_buffer = is_first_sch_acq ? first_sch_buf : burst_copy_buffer; + memset((void *)&sch_acq_buffer[0], 0, sizeof(sch_acq_buffer)); +#if 1 const auto which_out_buffer = is_first_sch_acq ? sch_acq_buffer : &sch_acq_buffer[40 * 2]; const auto ss = reinterpret_cast<std::complex<float> *>(which_out_buffer); std::complex<float> channel_imp_resp[CHAN_IMP_RESP_LENGTH * d_OSR]; - int start; - memset((void *)&sch_acq_buffer[0], 0, sizeof(sch_acq_buffer)); convert_and_scale(which_out_buffer, which_in_buffer, buf_len * 2, 1.f / float(rxFullScale)); if (is_first_sch_acq) { float max_corr = 0; @@ -173,9 +175,22 @@ detect_burst_nb(&ss[start], &channel_imp_resp[0], 0, sch_demod_bits); auto sch_decode_success = decode_sch(sch_demod_bits, is_first_sch_acq); +#if 0 + auto burst = new signalVector(buf_len, 50); + const auto corr_type = is_first_sch_acq ? sch_detect_type::SCH_DETECT_BUFFER : sch_detect_type::SCH_DETECT_FULL; + struct estim_burst_params ebp; + // scale like uhd, +-2k -> +-32k + convert_and_scale(burst->begin(), which_in_buffer, buf_len * 2, SAMPLE_SCALE_FACTOR); + + auto rv = detectSCHBurst(*burst, 4, 4, corr_type, &ebp); + + int howmuchdelay = ebp.toa * 4; + std::cerr << "ooffs: " << howmuchdelay << " " << std::endl; + std::cerr << "voffs: " << start << " " << sch_decode_success << std::endl; +#endif if (sch_decode_success) { - const auto ts_offset_symb = 0; + const auto ts_offset_symb = 4; if (is_first_sch_acq) { // update ts to first sample in sch buffer, to allow delay calc for current ts first_sch_ts_start = first_sch_buf_rcv_ts + start - (ts_offset_symb * 4) - 1; @@ -190,6 +205,97 @@ DBGLG2() << "L SCH : \x1B[31m decode fail \033[0m @ toa:" << start << " " << current_gsm_time.FN() << ":" << current_gsm_time.TN() << std::endl; } +#else + const auto ts_offset_symb = 4; + auto burst = new signalVector(buf_len, 50); + const auto corr_type = is_first_sch_acq ? sch_detect_type::SCH_DETECT_BUFFER : sch_detect_type::SCH_DETECT_FULL; + struct estim_burst_params ebp; + + // scale like uhd, +-2k -> +-32k + convert_and_scale(burst->begin(), which_in_buffer, buf_len * 2, SAMPLE_SCALE_FACTOR); + + auto rv = detectSCHBurst(*burst, 4, 4, corr_type, &ebp); + + int howmuchdelay = ebp.toa * 4; + + if (!rv) { + delete burst; + DBGLG() << "SCH : \x1B[31m detect fail \033[0m NOOOOOOOOOOOOOOOOOO toa:" << ebp.toa << " " + << current_gsm_time.FN() << ":" << current_gsm_time.TN() << std::endl; + return false; + } + + SoftVector *bits; + if (is_first_sch_acq) { + // can't be legit with a buf size spanning _at least_ one SCH but delay that implies partial sch burst + if (howmuchdelay < 0 || (buf_len - howmuchdelay) < ONE_TS_BURST_LEN) { + delete burst; + return false; + } + + struct estim_burst_params ebp2; + // auto sch_chunk = new signalVector(ONE_TS_BURST_LEN, 50); + // auto sch_chunk_start = sch_chunk->begin(); + // memcpy(sch_chunk_start, sch_buf_f.data() + howmuchdelay, sizeof(std::complex<float>) * ONE_TS_BURST_LEN); + + auto delay = delayVector(burst, NULL, -howmuchdelay); + + scaleVector(*delay, (complex)1.0 / ebp.amp); + + auto rv2 = detectSCHBurst(*delay, 4, 4, sch_detect_type::SCH_DETECT_FULL, &ebp2); + DBGLG() << "FIRST SCH : " << (rv2 ? "yes " : " ") << "Timing offset " << ebp2.toa << " symbols" + << std::endl; + + bits = demodAnyBurst(*delay, SCH, 4, &ebp2); + delete delay; + } else { + bits = demodAnyBurst(*burst, SCH, 4, &ebp); + } + + delete burst; + + // clamp to +-1.5 because +-127 softbits scaled by 64 after -0.5 can be at most +-1.5 + clamp_array(bits->begin(), 148, 1.5f); + + float_to_sbit(&bits->begin()[0], (signed char *)&sch_demod_bits[0], 62, 148); + // float_to_sbit(&bits->begin()[106], &data[39], 62, 39); + + if (decode_sch((char *)sch_demod_bits, is_first_sch_acq)) { + auto current_gsm_time_updated = timekeeper.gsmtime(); + if (is_first_sch_acq) { + // update ts to first sample in sch buffer, to allow delay calc for current ts + first_sch_ts_start = first_sch_buf_rcv_ts + howmuchdelay - (ts_offset_symb * 4); + } else { + // continuous sch tracking, only update if off too much + auto diff = [](float x, float y) { return x > y ? x - y : y - x; }; + + auto d = diff(ebp.toa, ts_offset_symb); + if (abs(d) > 0.3) { + if (ebp.toa < ts_offset_symb) + ebp.toa = d; + else + ebp.toa = -d; + temp_ts_corr_offset += ebp.toa * 4; + + DBGLG() << "offs: " << ebp.toa << " " << temp_ts_corr_offset << std::endl; + } + } + + auto a = gsm_sch_check_fn(current_gsm_time_updated.FN() - 1); + auto b = gsm_sch_check_fn(current_gsm_time_updated.FN()); + auto c = gsm_sch_check_fn(current_gsm_time_updated.FN() + 1); + DBGLG() << "L SCH : Timing offset " << rv << " " << ebp.toa << " " << a << b << c << "fn " + << current_gsm_time_updated.FN() << ":" << current_gsm_time_updated.TN() << std::endl; + + delete bits; + return true; + } else { + DBGLG2() << "L SCH : \x1B[31m decode fail \033[0m @ toa:" << ebp.toa << " " << current_gsm_time.FN() + << ":" << current_gsm_time.TN() << std::endl; + } + + delete bits; +#endif return false; } diff --git a/Transceiver52M/ms/ms_upper.cpp b/Transceiver52M/ms/ms_upper.cpp index c5664cd..2e8bc11 100644 --- a/Transceiver52M/ms/ms_upper.cpp +++ b/Transceiver52M/ms/ms_upper.cpp @@ -199,6 +199,7 @@ return true; } +#if 1 convert_and_scale(ss, e.burst, ONE_TS_BURST_LEN * 2, 1.f / float(rxFullScale)); pow = energyDetect(sv, 20 * 4 /*sps*/); @@ -232,6 +233,42 @@ // detect_burst(ss, &chan_imp_resp2[0], dummy_burst_start, outbin); #endif } +#else + + // lower layer sch detection offset, easy to verify by just printing the detected value using both the va+sigproc code. + convert_and_scale(ss + 16, e.burst, ONE_TS_BURST_LEN * 2, 15); + + pow = energyDetect(sv, 20 * 4 /*sps*/); + if (pow < -1) { + LOG(ALERT) << "Received empty burst"; + return false; + } + + avg = sqrt(pow); + + /* Detect normal or RACH bursts */ + CorrType type = CorrType::TSC; + struct estim_burst_params ebp; + auto rc = detectAnyBurst(sv, mTSC, 3, 4, type, 48, &ebp); + if (rc > 0) { + type = (CorrType)rc; + } + + if (rc < 0) { + std::cerr << "UR : \x1B[31m rx fail \033[0m @ toa:" << ebp.toa << " " << e.gsmts.FN() << ":" + << e.gsmts.TN() << std::endl; + return false; + } + SoftVector *bits = demodAnyBurst(sv, type, 4, &ebp); + + SoftVector::const_iterator burstItr = bits->begin(); + // invert and fix to +-127 sbits + for (int ii = 0; ii < 148; ii++) { + demodded_softbits[ii] = *burstItr++ > 0.0f ? -127 : 127; + } + delete bits; + +#endif RSSI = (int)floor(20.0 * log10(rxFullScale / avg)); // FIXME: properly handle offset, sch/nb alignment diff? handled by lower anyway... timingOffset = (int)round(0); diff --git a/Transceiver52M/sigProcLib.cpp b/Transceiver52M/sigProcLib.cpp index df87f94..5fac365 100644 --- a/Transceiver52M/sigProcLib.cpp +++ b/Transceiver52M/sigProcLib.cpp @@ -129,6 +129,8 @@ static CorrelationSequence *gMidambles[] = {NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL}; static CorrelationSequence *gEdgeMidambles[] = {NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL}; static CorrelationSequence *gRACHSequences[] = {NULL,NULL,NULL}; +static CorrelationSequence *gSCHSequence = NULL; +static CorrelationSequence *gDummySequence = NULL; static PulseSequence *GSMPulse1 = NULL; static PulseSequence *GSMPulse4 = NULL; @@ -151,6 +153,12 @@ gRACHSequences[i] = NULL; } + delete gSCHSequence; + gSCHSequence = NULL; + + delete gDummySequence; + gDummySequence = NULL; + delete GMSKRotation1; delete GMSKReverseRotation1; delete GMSKRotation4; @@ -315,6 +323,7 @@ append = true; break; case CUSTOM: + // FIXME: x->getstart? if (start < h->size() - 1) { head = h->size() - start; append = true; @@ -1289,6 +1298,77 @@ return status; } +static bool generateDummyMidamble(int sps) +{ + bool status = true; + float toa; + complex *data = NULL; + signalVector *autocorr = NULL, *midamble = NULL; + signalVector *midMidamble = NULL, *_midMidamble = NULL; + + delete gDummySequence; + + /* Use middle 16 bits of each TSC. Correlation sequence is not pulse shaped */ + midMidamble = modulateBurst(gDummyBurstTSC.segment(5,16), 0, sps, true); + if (!midMidamble) + return false; + + /* Simulated receive sequence is pulse shaped */ + midamble = modulateBurst(gDummyBurstTSC, 0, sps, false); + if (!midamble) { + status = false; + goto release; + } + + // NOTE: Because ideal TSC 16-bit midamble is 66 symbols into burst, + // the ideal TSC has an + 180 degree phase shift, + // due to the pi/2 frequency shift, that + // needs to be accounted for. + // 26-midamble is 61 symbols into burst, has +90 degree phase shift. + scaleVector(*midMidamble, complex(-1.0, 0.0)); + scaleVector(*midamble, complex(0.0, 1.0)); + + conjugateVector(*midMidamble); + + /* For SSE alignment, reallocate the midamble sequence on 16-byte boundary */ + data = (complex *) convolve_h_alloc(midMidamble->size()); + _midMidamble = new signalVector(data, 0, midMidamble->size(), convolve_h_alloc, free); + _midMidamble->setAligned(true); + midMidamble->copyTo(*_midMidamble); + + autocorr = convolve(midamble, _midMidamble, NULL, NO_DELAY); + if (!autocorr) { + status = false; + goto release; + } + + gDummySequence = new CorrelationSequence; + gDummySequence->sequence = _midMidamble; + gDummySequence->gain = peakDetect(*autocorr, &toa, NULL); + + /* For 1 sps only + * (Half of correlation length - 1) + midpoint of pulse shape + remainder + * 13.5 = (16 / 2 - 1) + 1.5 + (26 - 10) / 2 + */ + if (sps == 1) + gDummySequence->toa = toa - 13.5; + else + gDummySequence->toa = 0; + +release: + delete autocorr; + delete midamble; + delete midMidamble; + + if (!status) { + delete _midMidamble; + free(data); + gDummySequence = NULL; + } + + return status; +} + static CorrelationSequence *generateEdgeMidamble(int tsc) { complex *data = NULL; @@ -1384,6 +1464,69 @@ return status; } +bool generateSCHSequence(int sps) +{ + bool status = true; + float toa; + complex *data = NULL; + signalVector *autocorr = NULL; + signalVector *seq0 = NULL, *seq1 = NULL, *_seq1 = NULL; + + delete gSCHSequence; + + seq0 = modulateBurst(gSCHSynchSequence, 0, sps, false); + if (!seq0) + return false; + + seq1 = modulateBurst(gSCHSynchSequence, 0, sps, true); + if (!seq1) { + status = false; + goto release; + } + + conjugateVector(*seq1); + + /* For SSE alignment, reallocate the midamble sequence on 16-byte boundary */ + data = (complex *) convolve_h_alloc(seq1->size()); + _seq1 = new signalVector(data, 0, seq1->size(), convolve_h_alloc, free); + _seq1->setAligned(true); + seq1->copyTo(*_seq1); + + autocorr = convolve(seq0, _seq1, autocorr, NO_DELAY); + if (!autocorr) { + status = false; + goto release; + } + + gSCHSequence = new CorrelationSequence; + gSCHSequence->sequence = _seq1; + gSCHSequence->buffer = data; + gSCHSequence->gain = peakDetect(*autocorr, &toa, NULL); + + /* For 1 sps only + * (Half of correlation length - 1) + midpoint of pulse shaping filer + * 20.5 = (64 / 2 - 1) + 1.5 + */ + if (sps == 1) + gSCHSequence->toa = toa - 32.5; + else + gSCHSequence->toa = 0.0; + +release: + delete autocorr; + delete seq0; + delete seq1; + + if (!status) { + delete _seq1; + free(data); + gSCHSequence = NULL; + } + + return status; +} + + /* * Peak-to-average computation +/- range from peak in symbols */ @@ -1441,14 +1584,15 @@ return energy/windowLength; } -static signalVector *downsampleBurst(const signalVector &burst) +static signalVector *downsampleBurst(const signalVector &burst, int in_len = DOWNSAMPLE_IN_LEN, + int out_len = DOWNSAMPLE_OUT_LEN) { - signalVector in(DOWNSAMPLE_IN_LEN, dnsampler->len()); - signalVector *out = new signalVector(DOWNSAMPLE_OUT_LEN); - burst.copyToSegment(in, 0, DOWNSAMPLE_IN_LEN); + signalVector in(in_len, dnsampler->len()); + // gSCHSequence->sequence->size(), ensure next conv has no realloc + signalVector *out = new signalVector(out_len, 64); + burst.copyToSegment(in, 0, in_len); - if (dnsampler->rotate((float *) in.begin(), DOWNSAMPLE_IN_LEN, - (float *) out->begin(), DOWNSAMPLE_OUT_LEN) < 0) { + if (dnsampler->rotate((float *)in.begin(), in_len, (float *)out->begin(), out_len) < 0) { delete out; out = NULL; } @@ -1469,6 +1613,12 @@ /* Integer position where the sequence starts */ const int ps = start + 1 - N + (int)roundf(toa); + if(ps < 0) // might be -22 for toa 40 with N=64, if off by a lot during sch ms sync + return 0; + + if (ps + N > (int)burst->size()) + return 0; + /* Estimate Signal power */ S = 0.0f; for (int i=0, j=ps; i<(int)N; i++,j++) @@ -1652,6 +1802,80 @@ return rc; } +int detectSCHBurst(signalVector &burst, + float thresh, + int sps, + sch_detect_type state, struct estim_burst_params *ebp) +{ + int rc, start, target, head, tail, len; + complex _amp; + CorrelationSequence *sync; + + if ((sps != 1) && (sps != 4)) + return -1; + + target = 3 + 39 + 64; + + switch (state) { + case sch_detect_type::SCH_DETECT_NARROW: + head = 4; + tail = 4; + break; + case sch_detect_type::SCH_DETECT_BUFFER: + target = 1; + head = 0; + tail = (12 * 8 * 625) / 4; // 12 frames, downsampled /4 to 1 sps + break; + case sch_detect_type::SCH_DETECT_FULL: + default: + head = target - 1; + tail = 39 + 3 + 9; + break; + } + + start = (target - head) * 1 - 1; + len = (head + tail) * 1; + sync = gSCHSequence; + signalVector corr(len); + + signalVector *dec = downsampleBurst(burst, len * 4, len); + rc = detectBurst(*dec, corr, sync, thresh, 1, start, len, ebp); + delete dec; + + if (rc < 0) { + return -1; + } else if (!rc) { + ebp->amp = 0.0f; + ebp->toa = 0.0f; + return 0; + } + + if (state == sch_detect_type::SCH_DETECT_BUFFER) + ebp->toa = ebp->toa - (3 + 39 + 64); + else { + /* Subtract forward search bits from delay */ + ebp->toa = ebp->toa - head; + } + + return rc; +} + +static int detectDummyBurst(const signalVector &burst, float threshold, + int sps, unsigned max_toa, struct estim_burst_params *ebp) +{ + int rc, target, head, tail; + CorrelationSequence *sync; + + target = 3 + 58 + 16 + 5; + head = 10; + tail = 6 + max_toa; + sync = gDummySequence; + + ebp->tsc = 0; + rc = detectGeneralBurst(burst, threshold, sps, target, head, tail, sync, ebp); + return rc; +} + /* * Normal burst detection * @@ -1670,7 +1894,7 @@ return -SIGERR_UNSUPPORTED; target = 3 + 58 + 16 + 5; - head = 6; + head = 10; tail = 6 + max_toa; sync = gMidambles[tsc]; @@ -1719,6 +1943,9 @@ case RACH: rc = detectRACHBurst(burst, threshold, sps, max_toa, type == EXT_RACH, ebp); break; + case IDLE: + rc = detectDummyBurst(burst, threshold, sps, max_toa, ebp); + break; default: LOG(ERR) << "Invalid correlation type"; } @@ -1921,6 +2148,9 @@ generateRACHSequence(&gRACHSequences[1], gRACHSynchSequenceTS1, 1); generateRACHSequence(&gRACHSequences[2], gRACHSynchSequenceTS2, 1); + generateSCHSequence(1); + generateDummyMidamble(1); + for (int tsc = 0; tsc < 8; tsc++) { generateMidamble(1, tsc); gEdgeMidambles[tsc] = generateEdgeMidamble(tsc); diff --git a/Transceiver52M/sigProcLib.h b/Transceiver52M/sigProcLib.h index 0c3c7c6..39c8ddd 100644 --- a/Transceiver52M/sigProcLib.h +++ b/Transceiver52M/sigProcLib.h @@ -31,6 +31,7 @@ TSC, ///< timeslot should contain a normal burst EXT_RACH, ///< timeslot should contain an extended access burst RACH, ///< timeslot should contain an access burst + SCH, EDGE, ///< timeslot should contain an EDGE burst IDLE ///< timeslot is an idle (or dummy) burst }; @@ -93,6 +94,8 @@ void scaleVector(signalVector &x, complex scale); +signalVector *delayVector(const signalVector *in, signalVector *out, float delay); + /** Rough energy estimator. @param rxBurst A GSM burst. @@ -133,6 +136,17 @@ unsigned max_toa, struct estim_burst_params *ebp); +enum class sch_detect_type { + SCH_DETECT_FULL, + SCH_DETECT_NARROW, + SCH_DETECT_BUFFER, +}; + +int detectSCHBurst(signalVector &rxBurst, + float detectThreshold, + int sps, + sch_detect_type state, struct estim_burst_params *ebp); + /** Demodulate burst basde on type and output soft bits */ SoftVector *demodAnyBurst(const signalVector &burst, CorrType type, int sps, struct estim_burst_params *ebp); -- To view, visit https://gerrit.osmocom.org/c/osmo-trx/+/36046?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-Change-Id: I6558992bd69f18526be5ebe7d424ca00ceb67772 Gerrit-Change-Number: 36046 Gerrit-PatchSet: 2 Gerrit-Owner: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-Reviewer: pespin <pespin(a)sysmocom.de> Gerrit-MessageType: merged

1 year, 5 months

1
0
0 0

[L] Change in osmo-trx[master]: ms: add sigproclib demod

by Hoernchen

Attention is currently required from: pespin. Hoernchen has posted comments on this change. ( https://gerrit.osmocom.org/c/osmo-trx/+/36046?usp=email ) Change subject: ms: add sigproclib demod ...................................................................... Patch Set 2: Code-Review+2 -- To view, visit https://gerrit.osmocom.org/c/osmo-trx/+/36046?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-Change-Id: I6558992bd69f18526be5ebe7d424ca00ceb67772 Gerrit-Change-Number: 36046 Gerrit-PatchSet: 2 Gerrit-Owner: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-Reviewer: pespin <pespin(a)sysmocom.de> Gerrit-Attention: pespin <pespin(a)sysmocom.de> Gerrit-Comment-Date: Mon, 04 Mar 2024 20:05:15 +0000 Gerrit-HasComments: No Gerrit-Has-Labels: Yes Gerrit-MessageType: comment

1 year, 5 months

1
0
0 0

[S] Change in osmo-trx[master]: ms: hard preswapped VA gsm bits

by Hoernchen

Hoernchen has submitted this change. ( https://gerrit.osmocom.org/c/osmo-trx/+/36045?usp=email ) Change subject: ms: hard preswapped VA gsm bits ...................................................................... ms: hard preswapped VA gsm bits small * 127 can still be small enough to end up 0 after casting which breaks everything. Change-Id: I44b95dced64208eebfb5214da034a5d49abdd0df --- M Transceiver52M/grgsm_vitac/grgsm_vitac.cpp 1 file changed, 13 insertions(+), 1 deletion(-) Approvals: Jenkins Builder: Verified laforge: Looks good to me, but someone else must approve Hoernchen: Looks good to me, approved pespin: Looks good to me, but someone else must approve diff --git a/Transceiver52M/grgsm_vitac/grgsm_vitac.cpp b/Transceiver52M/grgsm_vitac/grgsm_vitac.cpp index 5779d9d..2016541 100644 --- a/Transceiver52M/grgsm_vitac/grgsm_vitac.cpp +++ b/Transceiver52M/grgsm_vitac/grgsm_vitac.cpp @@ -95,7 +95,7 @@ viterbi_detector(filtered_burst, burst_size, rhh, start_state, stop_states, 2, output); for (unsigned int i = 0; i < burst_size; i++) - output_binary[i] = (char)(output[i] * -127); // pre flip bits! + output_binary[i] = output[i] > 0 ? -127 : 127; // pre flip bits! } NO_UBSAN void detect_burst_nb(const gr_complex *input, gr_complex *chan_imp_resp, int burst_start, char *output_binary, -- To view, visit https://gerrit.osmocom.org/c/osmo-trx/+/36045?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-Change-Id: I44b95dced64208eebfb5214da034a5d49abdd0df Gerrit-Change-Number: 36045 Gerrit-PatchSet: 1 Gerrit-Owner: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-Reviewer: pespin <pespin(a)sysmocom.de> Gerrit-MessageType: merged

1 year, 5 months

1
0
0 0

[S] Change in osmo-trx[master]: ms: hard preswapped VA gsm bits

by Hoernchen

Attention is currently required from: pespin. Hoernchen has posted comments on this change. ( https://gerrit.osmocom.org/c/osmo-trx/+/36045?usp=email ) Change subject: ms: hard preswapped VA gsm bits ...................................................................... Patch Set 1: Code-Review+2 -- To view, visit https://gerrit.osmocom.org/c/osmo-trx/+/36045?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-Change-Id: I44b95dced64208eebfb5214da034a5d49abdd0df Gerrit-Change-Number: 36045 Gerrit-PatchSet: 1 Gerrit-Owner: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-Reviewer: pespin <pespin(a)sysmocom.de> Gerrit-Attention: pespin <pespin(a)sysmocom.de> Gerrit-Comment-Date: Mon, 04 Mar 2024 20:05:09 +0000 Gerrit-HasComments: No Gerrit-Has-Labels: Yes Gerrit-MessageType: comment

1 year, 5 months

1
0
0 0

[L] Change in osmo-trx[master]: ms: get rid of std::thread

by Hoernchen

Hoernchen has submitted this change. ( https://gerrit.osmocom.org/c/osmo-trx/+/36044?usp=email ) Change subject: ms: get rid of std::thread ...................................................................... ms: get rid of std::thread 2fc2b594da6e329577b195cb2543a8dd9e1b9ed0 changed std::thread to pthread for proper affinity to circumvent startup issues, so just stick to pthread instead of mixing std::thread and pthread, which made tracking thread creation difficult due to different functions. Change-Id: I0ba2fd958530394b9d99ed82111064d428c5870f --- M Transceiver52M/Makefile.am M Transceiver52M/ms/bladerf_specific.h M Transceiver52M/ms/ms.cpp M Transceiver52M/ms/ms.h M Transceiver52M/ms/ms_upper.cpp M Transceiver52M/ms/threadpool.h A Transceiver52M/ms/threadsched.cpp A Transceiver52M/ms/threadsched.h M Transceiver52M/ms/uhd_specific.h 9 files changed, 248 insertions(+), 141 deletions(-) Approvals: Hoernchen: Looks good to me, approved Jenkins Builder: Verified pespin: Looks good to me, but someone else must approve laforge: Looks good to me, but someone else must approve diff --git a/Transceiver52M/Makefile.am b/Transceiver52M/Makefile.am index e82111e..dadfde9 100644 --- a/Transceiver52M/Makefile.am +++ b/Transceiver52M/Makefile.am @@ -88,6 +88,7 @@ MS_LOWER_SRC = \ ms/sch.c \ ms/ms.cpp \ + ms/threadsched.cpp \ ms/ms_rx_lower.cpp \ grgsm_vitac/grgsm_vitac.cpp \ grgsm_vitac/viterbi_detector.cc @@ -101,6 +102,7 @@ noinst_HEADERS += \ ms/ms.h \ + ms/threadsched.h \ ms/bladerf_specific.h \ ms/uhd_specific.h \ ms/ms_upper.h \ diff --git a/Transceiver52M/ms/bladerf_specific.h b/Transceiver52M/ms/bladerf_specific.h index 57aae75..f4abe57 100644 --- a/Transceiver52M/ms/bladerf_specific.h +++ b/Transceiver52M/ms/bladerf_specific.h @@ -428,10 +428,12 @@ auto get_rx_burst_handler_fn(bh_fn_t burst_handler) { - auto fn = [this] { + using thist = decltype(this); + auto fn = [](void *args) -> void * { + thist t = reinterpret_cast<thist>(args); int status = 0; if (!stop_lower_threads_flag) - status = bladerf_stream(rx_stream, BLADERF_RX_X1); + status = bladerf_stream(t->rx_stream, BLADERF_RX_X1); if (status < 0) std::cerr << "rx stream error! " << bladerf_strerror(status) << std::endl; @@ -441,10 +443,12 @@ } auto get_tx_burst_handler_fn(bh_fn_t burst_handler) { - auto fn = [this] { + using thist = decltype(this); + auto fn = [](void *args) -> void * { + thist t = reinterpret_cast<thist>(args); int status = 0; if (!stop_lower_threads_flag) - status = bladerf_stream(tx_stream, BLADERF_TX_X1); + status = bladerf_stream(t->tx_stream, BLADERF_TX_X1); if (status < 0) std::cerr << "rx stream error! " << bladerf_strerror(status) << std::endl; diff --git a/Transceiver52M/ms/ms.cpp b/Transceiver52M/ms/ms.cpp index b8710a6..4ae8668 100644 --- a/Transceiver52M/ms/ms.cpp +++ b/Transceiver52M/ms/ms.cpp @@ -35,6 +35,8 @@ #include "sch.h" } +#include "threadsched.h" + dummylog ms_trx::dummy_log; #ifdef DBGXX @@ -83,13 +85,11 @@ if (stop_lower_threads_flag) return; auto fn = get_rx_burst_handler_fn(rx_bh()); - lower_rx_task = std::thread(fn); - set_name_aff_sched(lower_rx_task.native_handle(), sched_params::thread_names::RXRUN); + lower_rx_task = spawn_worker_thread(sched_params::thread_names::RXRUN, fn, this); usleep(1000); auto fn2 = get_tx_burst_handler_fn(tx_bh()); - lower_tx_task = std::thread(fn2); - set_name_aff_sched(lower_tx_task.native_handle(), sched_params::thread_names::TXRUN); + lower_tx_task = spawn_worker_thread(sched_params::thread_names::TXRUN, fn2, this); actually_enable_streams(); } @@ -105,9 +105,9 @@ stop_lower_threads_flag = true; close_device(); std::cerr << "dev closed..." << std::endl; - lower_rx_task.join(); + pthread_join(lower_rx_task, nullptr); std::cerr << "L rx dead..." << std::endl; - lower_tx_task.join(); + pthread_join(lower_tx_task, nullptr); std::cerr << "L tx dead..." << std::endl; } diff --git a/Transceiver52M/ms/ms.h b/Transceiver52M/ms/ms.h index 18a6954..2ad78de 100644 --- a/Transceiver52M/ms/ms.h +++ b/Transceiver52M/ms/ms.h @@ -26,7 +26,7 @@ #include <cstdint> #include <mutex> #include <iostream> -#include <thread> +// #include <thread> #if defined(BUILDBLADE) #include "bladerf_specific.h" @@ -42,6 +42,7 @@ #include "GSMCommon.h" #include "itrq.h" #include "threadpool.h" +#include "threadsched.h" const unsigned int ONE_TS_BURST_LEN = (3 + 58 + 26 + 58 + 3 + 8.25) * 4 /*sps*/; const unsigned int SCH_LEN_SPS = (ONE_TS_BURST_LEN * 8 /*ts*/ * 12 /*frames*/); @@ -216,43 +217,23 @@ } }; -static struct sched_params { - enum thread_names { U_CTL = 0, U_RX, U_TX, SCH_SEARCH, MAIN, LEAKCHECK, RXRUN, TXRUN, _THRD_NAME_COUNT }; - enum target { ODROID = 0, PI4 }; - const char *name; - int core; - int schedtype; - int prio; -} schdp[][sched_params::_THRD_NAME_COUNT]{ - { - { "upper_ctrl", 2, SCHED_RR, sched_get_priority_max(SCHED_RR) }, - { "upper_rx", 2, SCHED_RR, sched_get_priority_max(SCHED_RR) - 5 }, - { "upper_tx", 2, SCHED_RR, sched_get_priority_max(SCHED_RR) - 1 }, - - { "sch_search", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 5 }, - { "main", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 5 }, - { "leakcheck", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 10 }, - - { "rxrun", 4, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 2 }, - { "txrun", 5, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 1 }, - }, - { - { "upper_ctrl", 1, SCHED_RR, sched_get_priority_max(SCHED_RR) }, - { "upper_rx", 1, SCHED_RR, sched_get_priority_max(SCHED_RR) - 5 }, - { "upper_tx", 1, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 1 }, - - { "sch_search", 1, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 5 }, - { "main", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 5 }, - { "leakcheck", 1, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 10 }, - - { "rxrun", 2, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 2 }, - { "txrun", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 1 }, - }, -}; - using ts_hitter_q_t = spsc_cond<64, GSM::Time, true, false>; -struct ms_trx : public BASET { +// used to globally initialize the sched/hw information +struct sched_hw_info { + int hw_cpus; + sched_params::target hw_target; + + sched_hw_info() + { + hw_cpus = std::thread::hardware_concurrency(); + hw_target = hw_cpus > 4 ? sched_params::target::ODROID : sched_params::target::PI4; + set_sched_target(hw_target); + std::cerr << "scheduling for: " << (hw_cpus > 4 ? "odroid" : "pi4") << std::endl; + } +}; + +struct ms_trx : public BASET, public sched_hw_info { using base = BASET; static dummylog dummy_log; unsigned int mTSC; @@ -260,8 +241,8 @@ int timing_advance; bool do_auto_gain; - std::thread lower_rx_task; - std::thread lower_tx_task; + pthread_t lower_rx_task; + pthread_t lower_tx_task; // provides bursts to upper rx thread rx_queue_t rxqueue; @@ -277,9 +258,7 @@ int64_t first_sch_ts_start = -1; time_keeper timekeeper; - int hw_cpus; - sched_params::target hw_target; - single_thread_pool worker_thread; + single_thread_pool worker_thread; // uses base class sched target hw info void start_lower_ms(); std::atomic<bool> upper_is_ready; @@ -301,12 +280,8 @@ : mTSC(0), mBSIC(0), timing_advance(0), do_auto_gain(false), rxqueue(), first_sch_buf(new blade_sample_type[SCH_LEN_SPS]), burst_copy_buffer(new blade_sample_type[ONE_TS_BURST_LEN]), first_sch_buf_rcv_ts(0), - rcv_done{ false }, sch_thread_done{ false }, hw_cpus(std::thread::hardware_concurrency()), - hw_target(hw_cpus > 4 ? sched_params::target::ODROID : sched_params::target::PI4), - upper_is_ready(false) + rcv_done{ false }, sch_thread_done{ false }, upper_is_ready(false) { - std::cerr << "scheduling for: " << (hw_cpus > 4 ? "odroid" : "pi4") << std::endl; - set_name_aff_sched(worker_thread.get_handle(), sched_params::thread_names::SCH_SEARCH); } virtual ~ms_trx() @@ -323,73 +298,4 @@ assert(val > -127 && val < 128); timing_advance = val * 4; } - - void set_name_aff_sched(sched_params::thread_names name) - { - set_name_aff_sched(pthread_self(), name); - } - - void set_name_aff_sched(std::thread::native_handle_type h, sched_params::thread_names name) - { - auto tgt = schdp[hw_target][name]; - // std::cerr << "scheduling for: " << tgt.name << ":" << tgt.core << std::endl; - set_name_aff_sched(h, tgt.name, tgt.core, tgt.schedtype, tgt.prio); - } - - using pt_sig = void *(*)(void *); - - pthread_t spawn_worker_thread(sched_params::thread_names name, pt_sig fun, void *arg) - { - auto tgt = schdp[hw_target][name]; - // std::cerr << "scheduling for: " << tgt.name << ":" << tgt.core << " prio:" << tgt.prio << std::endl; - return do_spawn_thr(tgt.name, tgt.core, tgt.schedtype, tgt.prio, fun, arg); - } - - private: - void set_name_aff_sched(std::thread::native_handle_type h, const char *name, int cpunum, int schedtype, - int prio) - { - pthread_setname_np(h, name); - - cpu_set_t cpuset; - - CPU_ZERO(&cpuset); - CPU_SET(cpunum, &cpuset); - - if (pthread_setaffinity_np(h, sizeof(cpuset), &cpuset) < 0) { - std::cerr << name << " affinity: errreur! " << std::strerror(errno); - return exit(0); - } - - sched_param sch_params; - sch_params.sched_priority = prio; - if (pthread_setschedparam(h, schedtype, &sch_params) < 0) { - std::cerr << name << " sched: errreur! " << std::strerror(errno); - return exit(0); - } - } - - pthread_t do_spawn_thr(const char *name, int cpunum, int schedtype, int prio, pt_sig fun, void *arg) - { - pthread_t thread; - - pthread_attr_t attr; - pthread_attr_init(&attr); - - sched_param sch_params; - sch_params.sched_priority = prio; - cpu_set_t cpuset; - CPU_ZERO(&cpuset); - CPU_SET(cpunum, &cpuset); - auto a = pthread_attr_setaffinity_np(&attr, sizeof(cpu_set_t), &cpuset); - a |= pthread_attr_setschedpolicy(&attr, schedtype); - a |= pthread_attr_setschedparam(&attr, &sch_params); - a |= pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED); - if(a) - std::cerr << "thread arg rc:" << a << std::endl; - pthread_create(&thread, &attr, fun, arg); - pthread_setname_np(thread, name); - pthread_attr_destroy(&attr); - return thread; - } }; diff --git a/Transceiver52M/ms/ms_upper.cpp b/Transceiver52M/ms/ms_upper.cpp index db15226..c5664cd 100644 --- a/Transceiver52M/ms/ms_upper.cpp +++ b/Transceiver52M/ms/ms_upper.cpp @@ -50,6 +50,7 @@ #include "ms_trxcon_if.h" #include "ms_upper.h" +#include "threadsched.h" extern bool trxc_l1ctl_init(void *tallctx); struct trxcon_inst *g_trxcon; @@ -457,7 +458,7 @@ std::cerr << "Error initializing hardware, quitting.." << std::endl; return -1; } - trx->set_name_aff_sched(sched_params::thread_names::MAIN); + set_name_aff_sched(sched_params::thread_names::MAIN); if (!trxc_l1ctl_init(tall_trxcon_ctx)) { std::cerr << "Error initializing l1ctl, quitting.." << std::endl; diff --git a/Transceiver52M/ms/threadpool.h b/Transceiver52M/ms/threadpool.h index 2180c68..4b1eefd 100644 --- a/Transceiver52M/ms/threadpool.h +++ b/Transceiver52M/ms/threadpool.h @@ -20,13 +20,12 @@ * */ -#include <functional> -#include <thread> #include <atomic> #include <vector> #include <future> #include <mutex> #include <queue> +#include "threadsched.h" struct single_thread_pool { std::mutex m; @@ -34,7 +33,7 @@ std::atomic<bool> stop_flag; std::atomic<bool> is_ready; std::deque<std::function<void()>> wq; - std::thread worker_thread; + pthread_t worker_thread; template <class F> void add_task(F &&f) @@ -45,19 +44,23 @@ return; } - single_thread_pool() : stop_flag(false), is_ready(false), worker_thread(std::thread([this] { thread_loop(); })) + single_thread_pool() : stop_flag(false), is_ready(false) { + worker_thread = spawn_worker_thread( + sched_params::thread_names::SCH_SEARCH, + [](void *args) -> void * { + using thist = decltype(this); + thist t = reinterpret_cast<thist>(args); + t->thread_loop(); + return 0; + }, + this); } ~single_thread_pool() { stop(); } - std::thread::native_handle_type get_handle() - { - return worker_thread.native_handle(); - } - private: void stop() { @@ -67,7 +70,7 @@ stop_flag = true; cv.notify_one(); } - worker_thread.join(); + pthread_join(worker_thread, nullptr); } void thread_loop() diff --git a/Transceiver52M/ms/threadsched.cpp b/Transceiver52M/ms/threadsched.cpp new file mode 100644 index 0000000..ba5cae7 --- /dev/null +++ b/Transceiver52M/ms/threadsched.cpp @@ -0,0 +1,104 @@ +/* + * (C) 2023 by sysmocom s.f.m.c. GmbH <info(a)sysmocom.de> + * All Rights Reserved + * + * Author: Eric Wild <ewild(a)sysmocom.de> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU Affero General Public License as published by + * the Free Software Foundation; either version 3 of the License, or + * (at your option) any later version. + * + * This program 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 Affero General Public License for more details. + * + * You should have received a copy of the GNU Affero General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + * + */ + +#include <cerrno> +#include <cstdlib> +#include <cstring> +#include <iostream> +#include <thread> + +extern "C" { +#include <pthread.h> +} + +#include "threadsched.h" + +sched_params::target scheduling_target; + +void set_sched_target(sched_params::target t) +{ + scheduling_target = t; +} + +void set_name_aff_sched(std::thread::native_handle_type h, const char *name, int cpunum, int schedtype, int prio) +{ + pthread_setname_np(h, name); + + cpu_set_t cpuset; + + CPU_ZERO(&cpuset); + CPU_SET(cpunum, &cpuset); + + if (pthread_setaffinity_np(h, sizeof(cpuset), &cpuset) < 0) { + std::cerr << name << " affinity: errreur! " << std::strerror(errno); + return exit(0); + } + + sched_param sch_params; + sch_params.sched_priority = prio; + if (pthread_setschedparam(h, schedtype, &sch_params) < 0) { + std::cerr << name << " sched: errreur! " << std::strerror(errno); + return exit(0); + } +} + +static pthread_t do_spawn_thr(const char *name, int cpunum, int schedtype, int prio, worker_func_sig fun, void *arg) +{ + pthread_t thread; + + pthread_attr_t attr; + pthread_attr_init(&attr); + + sched_param sch_params; + sch_params.sched_priority = prio; + cpu_set_t cpuset; + CPU_ZERO(&cpuset); + CPU_SET(cpunum, &cpuset); + auto a = pthread_attr_setaffinity_np(&attr, sizeof(cpu_set_t), &cpuset); + a |= pthread_attr_setschedpolicy(&attr, schedtype); + a |= pthread_attr_setschedparam(&attr, &sch_params); + a |= pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED); + if (a) + std::cerr << "thread arg rc:" << a << std::endl; + pthread_create(&thread, &attr, fun, arg); + pthread_setname_np(thread, name); + pthread_attr_destroy(&attr); + return thread; +} + +void set_name_aff_sched(std::thread::native_handle_type h, sched_params::thread_names name) +{ + auto tgt = schdp[scheduling_target][name]; + // std::cerr << "scheduling for: " << tgt.name << ":" << tgt.core << std::endl; + set_name_aff_sched(h, tgt.name, tgt.core, tgt.schedtype, tgt.prio); +} + +void set_name_aff_sched(sched_params::thread_names name) +{ + set_name_aff_sched(pthread_self(), name); +} + +pthread_t spawn_worker_thread(sched_params::thread_names name, worker_func_sig fun, void *arg) +{ + auto tgt = schdp[scheduling_target][name]; + // std::cerr << "scheduling for: " << tgt.name << ":" << tgt.core << " prio:" << tgt.prio << std::endl; + return do_spawn_thr(tgt.name, tgt.core, tgt.schedtype, tgt.prio, fun, arg); +} diff --git a/Transceiver52M/ms/threadsched.h b/Transceiver52M/ms/threadsched.h new file mode 100644 index 0000000..7cc9176 --- /dev/null +++ b/Transceiver52M/ms/threadsched.h @@ -0,0 +1,68 @@ +#pragma once +/* + * (C) 2023 by sysmocom s.f.m.c. GmbH <info(a)sysmocom.de> + * All Rights Reserved + * + * Author: Eric Wild <ewild(a)sysmocom.de> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU Affero General Public License as published by + * the Free Software Foundation; either version 3 of the License, or + * (at your option) any later version. + * + * This program 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 Affero General Public License for more details. + * + * You should have received a copy of the GNU Affero General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + * + */ + +extern "C" { +#include <pthread.h> +#include <sched.h> +} + +static struct sched_params { + enum thread_names { U_CTL = 0, U_RX, U_TX, SCH_SEARCH, MAIN, LEAKCHECK, RXRUN, TXRUN, _THRD_NAME_COUNT }; + enum target { ODROID = 0, PI4 }; + const char *name; + int core; + int schedtype; + int prio; +} schdp[][sched_params::_THRD_NAME_COUNT]{ + { + { "upper_ctrl", 2, SCHED_RR, sched_get_priority_max(SCHED_RR) }, + { "upper_rx", 2, SCHED_RR, sched_get_priority_max(SCHED_RR) - 5 }, + { "upper_tx", 2, SCHED_RR, sched_get_priority_max(SCHED_RR) - 1 }, + + { "sch_search", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 5 }, + { "main", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 5 }, + { "leakcheck", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 10 }, + + { "rxrun", 4, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 2 }, + { "txrun", 5, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 1 }, + }, + { + { "upper_ctrl", 1, SCHED_RR, sched_get_priority_max(SCHED_RR) }, + { "upper_rx", 1, SCHED_RR, sched_get_priority_max(SCHED_RR) - 5 }, + { "upper_tx", 1, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 1 }, + + { "sch_search", 1, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 5 }, + { "main", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 5 }, + { "leakcheck", 1, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 10 }, + + { "rxrun", 2, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 2 }, + { "txrun", 3, SCHED_FIFO, sched_get_priority_max(SCHED_FIFO) - 1 }, + }, +}; + +void set_sched_target(sched_params::target t); + +using worker_func_sig = void *(*)(void *); + +void set_name_aff_sched(sched_params::thread_names name); + +pthread_t spawn_worker_thread(sched_params::thread_names name, worker_func_sig fun, void *arg); diff --git a/Transceiver52M/ms/uhd_specific.h b/Transceiver52M/ms/uhd_specific.h index 5723fd4..151c002 100644 --- a/Transceiver52M/ms/uhd_specific.h +++ b/Transceiver52M/ms/uhd_specific.h @@ -231,24 +231,29 @@ auto get_rx_burst_handler_fn(bh_fn_t burst_handler) { - auto fn = [this, burst_handler] { + // C cb -> ghetto closure capture, which is fine, the args never change. + static auto rx_burst_cap_this = this; + static auto rx_burst_cap_bh = burst_handler; + auto fn = [](void *args) -> void * { pthread_setname_np(pthread_self(), "rxrun"); uhd::stream_cmd_t stream_cmd(uhd::stream_cmd_t::STREAM_MODE_START_CONTINUOUS); stream_cmd.stream_now = true; stream_cmd.time_spec = uhd::time_spec_t(); - rx_stream->issue_stream_cmd(stream_cmd); + rx_burst_cap_this->rx_stream->issue_stream_cmd(stream_cmd); - while (!stop_lower_threads_flag) { - rx_cb(burst_handler); + while (!rx_burst_cap_this->stop_lower_threads_flag) { + rx_burst_cap_this->rx_cb(rx_burst_cap_bh); } + return 0; }; return fn; } auto get_tx_burst_handler_fn(bh_fn_t burst_handler) { - auto fn = [] { + auto fn = [](void *args) -> void * { // dummy + return 0; }; return fn; } -- To view, visit https://gerrit.osmocom.org/c/osmo-trx/+/36044?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-Change-Id: I0ba2fd958530394b9d99ed82111064d428c5870f Gerrit-Change-Number: 36044 Gerrit-PatchSet: 1 Gerrit-Owner: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-Reviewer: pespin <pespin(a)sysmocom.de> Gerrit-MessageType: merged

1 year, 5 months

1
0
0 0

[L] Change in osmo-trx[master]: ms: get rid of std::thread

by Hoernchen

Hoernchen has posted comments on this change. ( https://gerrit.osmocom.org/c/osmo-trx/+/36044?usp=email ) Change subject: ms: get rid of std::thread ...................................................................... Patch Set 1: Code-Review+2 -- To view, visit https://gerrit.osmocom.org/c/osmo-trx/+/36044?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-Change-Id: I0ba2fd958530394b9d99ed82111064d428c5870f Gerrit-Change-Number: 36044 Gerrit-PatchSet: 1 Gerrit-Owner: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Hoernchen <ewild(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-Reviewer: pespin <pespin(a)sysmocom.de> Gerrit-Comment-Date: Mon, 04 Mar 2024 20:05:01 +0000 Gerrit-HasComments: No Gerrit-Has-Labels: Yes Gerrit-MessageType: comment

1 year, 5 months

1
0
0 0

[S] Change in libosmo-netif[master]: osmo_stream_cli_write(): fix double-free of msgb

by laforge

Attention is currently required from: daniel, fixeria, jolly, pespin. laforge has posted comments on this change. ( https://gerrit.osmocom.org/c/libosmo-netif/+/36159?usp=email ) The change is no longer submittable: Code-Review is unsatisfied now. Change subject: osmo_stream_cli_write(): fix double-free of msgb ...................................................................... Patch Set 1: -Code-Review (1 comment) Patchset: PS1: I guess it may be that we actually want to return at the end of that 'ret < 0' clause? not sure if reconnect and then write_disable makes sense? -- To view, visit https://gerrit.osmocom.org/c/libosmo-netif/+/36159?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: libosmo-netif Gerrit-Branch: master Gerrit-Change-Id: I954678acd90593e5311d22a10033c7741f280be3 Gerrit-Change-Number: 36159 Gerrit-PatchSet: 1 Gerrit-Owner: fixeria <vyanitskiy(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: daniel <dwillmann(a)sysmocom.de> Gerrit-Reviewer: jolly <andreas(a)eversberg.eu> Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-Reviewer: pespin <pespin(a)sysmocom.de> Gerrit-Attention: jolly <andreas(a)eversberg.eu> Gerrit-Attention: pespin <pespin(a)sysmocom.de> Gerrit-Attention: fixeria <vyanitskiy(a)sysmocom.de> Gerrit-Attention: daniel <dwillmann(a)sysmocom.de> Gerrit-Comment-Date: Mon, 04 Mar 2024 19:49:35 +0000 Gerrit-HasComments: Yes Gerrit-Has-Labels: Yes Gerrit-MessageType: comment

1 year, 5 months

1
0
0 0

[S] Change in libosmo-netif[master]: osmo_stream_cli_write(): fix double-free of msgb

by laforge

Attention is currently required from: fixeria. laforge has posted comments on this change. ( https://gerrit.osmocom.org/c/libosmo-netif/+/36159?usp=email ) Change subject: osmo_stream_cli_write(): fix double-free of msgb ...................................................................... Patch Set 1: Code-Review+2 -- To view, visit https://gerrit.osmocom.org/c/libosmo-netif/+/36159?usp=email To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: libosmo-netif Gerrit-Branch: master Gerrit-Change-Id: I954678acd90593e5311d22a10033c7741f280be3 Gerrit-Change-Number: 36159 Gerrit-PatchSet: 1 Gerrit-Owner: fixeria <vyanitskiy(a)sysmocom.de> Gerrit-Reviewer: Jenkins Builder Gerrit-Reviewer: laforge <laforge(a)osmocom.org> Gerrit-Attention: fixeria <vyanitskiy(a)sysmocom.de> Gerrit-Comment-Date: Mon, 04 Mar 2024 19:48:02 +0000 Gerrit-HasComments: No Gerrit-Has-Labels: Yes Gerrit-MessageType: comment

1 year, 5 months

1
0
0 0

2025

2024

2023

2022

gerrit-log March 2024