You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
srsRAN_4G/test/phy/dummy_gnb_stack.h

751 lines
27 KiB
C++

/**
*
* \section COPYRIGHT
*
* Copyright 2013-2021 Software Radio Systems Limited
*
* By using this file, you agree to the terms and conditions set
* forth in the LICENSE file which can be found at the top level of
* the distribution.
*
*/
#ifndef SRSRAN_DUMMY_GNB_STACK_H
#define SRSRAN_DUMMY_GNB_STACK_H
#include "dummy_rx_harq_proc.h"
#include "dummy_tx_harq_proc.h"
#include "srsenb/test/common/rlc_test_dummy.h"
#include "srsgnb/hdr/stack/common/test/dummy_nr_classes.h"
#include "srsgnb/hdr/stack/mac/mac_nr.h"
#include "srsgnb/hdr/stack/mac/sched_nr.h"
#include "srsgnb/src/stack/mac/test/sched_nr_cfg_generators.h"
#include "srsran/srslog/srslog.h"
#include <mutex>
#include <set>
#include <srsenb/hdr/stack/mac/common/mac_metrics.h>
#include <srsran/adt/circular_array.h>
#include <srsran/common/phy_cfg_nr.h>
#include <srsran/common/standard_streams.h>
#include <srsran/common/string_helpers.h>
#include <srsran/interfaces/gnb_interfaces.h>
class gnb_dummy_stack : public srsenb::stack_interface_phy_nr
{
const static uint32_t NUMEROLOGY_IDX = 0;
public:
struct prach_metrics_t {
uint32_t count;
float avg_ta = 0.0f;
float min_ta = +INFINITY;
float max_ta = -INFINITY;
};
struct pucch_metrics_t {
float epre_db_avg = 0.0f;
float epre_db_min = +INFINITY;
float epre_db_max = -INFINITY;
float rsrp_db_avg = 0.0f;
float rsrp_db_min = +INFINITY;
float rsrp_db_max = -INFINITY;
float snr_db_avg = 0.0f;
float snr_db_min = +INFINITY;
float snr_db_max = -INFINITY;
float ta_us_avg = 0.0f;
float ta_us_min = +INFINITY;
float ta_us_max = -INFINITY;
uint32_t count = 0;
};
struct metrics_t {
std::map<uint32_t, prach_metrics_t> prach = {}; ///< PRACH metrics indexed with premable index
srsenb::mac_ue_metrics_t mac = {}; ///< MAC metrics
uint32_t sr_count = 0; ///< SR counter
uint32_t cqi_count = 0; ///< CQI opportunity counter
uint32_t cqi_valid_count = 0; ///< Valid CQI counter
pucch_metrics_t pucch = {};
pucch_metrics_t pusch = {};
};
private:
srslog::basic_logger& logger = srslog::fetch_basic_logger("GNB STK");
bool use_dummy_mac = true;
const uint16_t rnti = 0x1234;
struct {
srsran::circular_array<srsran_dci_location_t, SRSRAN_NOF_SF_X_FRAME> dci_location = {};
uint32_t mcs = 0;
uint32_t freq_res = 0;
std::set<uint32_t> slots = {};
} dl, ul;
srsran::circular_array<uint32_t, SRSRAN_NOF_SF_X_FRAME> dl_data_to_ul_ack;
uint32_t ss_id = 0;
srsran::phy_cfg_nr_t phy_cfg = {};
bool valid = false;
srsran::task_scheduler task_sched;
srsenb::rrc_nr_dummy rrc_obj;
srsenb::rlc_dummy rlc_obj;
std::unique_ptr<srsenb::mac_nr> mac;
srslog::basic_logger& sched_logger;
bool autofill_sch_bsr = false;
bool wait_preamble = false;
std::atomic<bool> enable_user_sched = {false};
std::mutex metrics_mutex;
metrics_t metrics = {};
// HARQ feedback
class pending_ack_t
{
private:
std::mutex mutex;
srsran_pdsch_ack_nr_t ack = {};
public:
pending_ack_t() = default;
void push_ack(srsran_harq_ack_resource_t& ack_resource)
{
// Prepare ACK information
srsran_harq_ack_m_t ack_m = {};
ack_m.resource = ack_resource;
ack_m.present = true;
std::unique_lock<std::mutex> lock(mutex);
ack.nof_cc = 1;
srsran_harq_ack_insert_m(&ack, &ack_m);
}
srsran_pdsch_ack_nr_t get_ack()
{
std::unique_lock<std::mutex> lock(mutex);
srsran_pdsch_ack_nr_t ret = ack;
ack = {};
return ret;
}
uint32_t get_dai()
{
std::unique_lock<std::mutex> lock(mutex);
return ack.cc[0].M % 4;
}
};
std::array<pending_ack_t, TTIMOD_SZ> pending_ack = {};
// PUSCH state
class pending_pusch_t
{
private:
std::mutex mutex;
srsran_sch_cfg_nr_t pusch = {};
bool valid = false;
uint32_t pid = 0;
public:
pending_pusch_t() = default;
void push(const uint32_t& pid_, const srsran_sch_cfg_nr_t& pusch_)
{
std::unique_lock<std::mutex> lock(mutex);
pusch = pusch_;
pid = pid_;
valid = true;
}
bool pop(uint32_t& pid_, srsran_sch_cfg_nr_t& pusch_)
{
std::unique_lock<std::mutex> lock(mutex);
bool ret = valid;
pusch_ = pusch;
pid_ = pid;
valid = false;
return ret;
}
};
std::array<pending_pusch_t, TTIMOD_SZ> pending_pusch = {};
dummy_tx_harq_entity tx_harq_proc;
dummy_rx_harq_entity rx_harq_proc;
bool schedule_pdsch(const srsran_slot_cfg_t& slot_cfg, dl_sched_t& dl_sched)
{
if (dl.slots.count(SRSRAN_SLOT_NR_MOD(srsran_subcarrier_spacing_15kHz, slot_cfg.idx)) == 0 or
not enable_user_sched) {
return true;
}
// Instantiate PDCCH and PDSCH
pdcch_dl_t pdcch = {};
pdsch_t pdsch = {};
// Second TB is not used
pdsch.data[1] = nullptr;
// Fill DCI configuration
pdcch.dci_cfg = phy_cfg.get_dci_cfg();
// Fill DCI context
if (not phy_cfg.get_dci_ctx_pdsch_rnti_c(ss_id, dl.dci_location[slot_cfg.idx], rnti, pdcch.dci.ctx)) {
logger.error("Error filling PDSCH DCI context");
return false;
}
uint32_t harq_feedback = dl_data_to_ul_ack[slot_cfg.idx];
uint32_t harq_ack_slot_idx = TTI_ADD(slot_cfg.idx, harq_feedback);
// Fill DCI fields
srsran_dci_dl_nr_t& dci = pdcch.dci;
dci.freq_domain_assigment = dl.freq_res;
dci.time_domain_assigment = 0;
dci.mcs = dl.mcs;
dci.rv = 0;
dci.ndi = (slot_cfg.idx / SRSRAN_NOF_SF_X_FRAME) % 2;
dci.pid = slot_cfg.idx % SRSRAN_NOF_SF_X_FRAME;
dci.dai = pending_ack[harq_ack_slot_idx % pending_ack.size()].get_dai();
dci.tpc = 1;
dci.pucch_resource = 0;
if (dci.ctx.format == srsran_dci_format_nr_1_0) {
dci.harq_feedback = dl_data_to_ul_ack[slot_cfg.idx] - 1;
} else {
dci.harq_feedback = slot_cfg.idx;
}
// Create PDSCH configuration
if (not phy_cfg.get_pdsch_cfg(slot_cfg, dci, pdsch.sch)) {
logger.error("Error converting DCI to grant");
return false;
}
// Set TBS
// Select grant and set data
pdsch.data[0] = tx_harq_proc[slot_cfg.idx].get_tb(pdsch.sch.grant.tb[0].tbs);
// Set softbuffer
pdsch.sch.grant.tb[0].softbuffer.tx = &tx_harq_proc[slot_cfg.idx].get_softbuffer(dci.ndi);
// Reset Tx softbuffer always
srsran_softbuffer_tx_reset(pdsch.sch.grant.tb[0].softbuffer.tx);
// Push scheduling results
dl_sched.pdcch_dl.push_back(pdcch);
dl_sched.pdsch.push_back(pdsch);
// Generate PDSCH HARQ Feedback
srsran_harq_ack_resource_t ack_resource = {};
if (not phy_cfg.get_pdsch_ack_resource(dci, ack_resource)) {
logger.error("Error getting ack resource");
return false;
}
// Calculate PUCCH slot and push resource
pending_ack[harq_ack_slot_idx % pending_ack.size()].push_ack(ack_resource);
return true;
}
bool schedule_pusch(const srsran_slot_cfg_t& slot_cfg, dl_sched_t& dl_sched)
{
if (ul.slots.count(SRSRAN_SLOT_NR_MOD(srsran_subcarrier_spacing_15kHz, slot_cfg.idx + 4)) == 0 or
not enable_user_sched) {
return true;
}
// Instantiate PDCCH
pdcch_ul_t pdcch = {};
// Fill DCI configuration
pdcch.dci_cfg = phy_cfg.get_dci_cfg();
// Fill DCI context
if (not phy_cfg.get_dci_ctx_pusch_rnti_c(ss_id, ul.dci_location[slot_cfg.idx], rnti, pdcch.dci.ctx)) {
logger.error("Error filling PDSCH DCI context");
return false;
}
// Fill DCI fields
srsran_dci_ul_nr_t& dci = pdcch.dci;
dci.freq_domain_assigment = ul.freq_res;
dci.time_domain_assigment = 0;
dci.freq_hopping_flag = 0;
dci.mcs = ul.mcs;
dci.rv = 0;
dci.ndi = (slot_cfg.idx / SRSRAN_NOF_SF_X_FRAME) % 2;
dci.pid = slot_cfg.idx % SRSRAN_NOF_SF_X_FRAME;
dci.tpc = 1;
// Create PDSCH configuration
srsran_sch_cfg_nr_t pusch_cfg = {};
if (not phy_cfg.get_pusch_cfg(slot_cfg, dci, pusch_cfg)) {
logger.error("Error converting DCI to grant");
return false;
}
// Set softbuffer
pusch_cfg.grant.tb[0].softbuffer.rx = &rx_harq_proc[dci.pid].get_softbuffer(dci.ndi, pusch_cfg.grant.tb[0].tbs);
// Push scheduling results
dl_sched.pdcch_ul.push_back(pdcch);
// Set pending PUSCH
pending_pusch[TTI_TX(slot_cfg.idx) % pending_pusch.size()].push(dci.pid, pusch_cfg);
return true;
}
bool handle_uci_data(const srsran_uci_cfg_nr_t& cfg, const srsran_uci_value_nr_t& value)
{
std::unique_lock<std::mutex> lock(metrics_mutex);
// Process HARQ-ACK
for (uint32_t i = 0; i < cfg.ack.count; i++) {
const srsran_harq_ack_bit_t* ack_bit = &cfg.ack.bits[i];
bool is_ok = (value.ack[i] == 1) and value.valid;
uint32_t tb_count = (ack_bit->tb0 ? 1 : 0) + (ack_bit->tb1 ? 1 : 0);
metrics.mac.tx_brate += tx_harq_proc[ack_bit->pid].get_tbs();
metrics.mac.tx_pkts += tb_count;
if (not is_ok) {
metrics.mac.tx_errors += tb_count;
logger.debug("NACK received!");
}
}
// Process SR
if (value.valid and value.sr > 0) {
metrics.sr_count++;
}
// Process CQI
for (uint32_t i = 0; i < cfg.nof_csi; i++) {
// Increment CQI opportunity
metrics.cqi_count++;
// Skip if invalid or not supported CSI report
if (not value.valid or cfg.csi[i].cfg.quantity != SRSRAN_CSI_REPORT_QUANTITY_CRI_RI_PMI_CQI or
cfg.csi[i].cfg.freq_cfg != SRSRAN_CSI_REPORT_FREQ_WIDEBAND) {
continue;
}
// Add statistics
metrics.mac.dl_cqi =
SRSRAN_VEC_SAFE_CMA(value.csi->wideband_cri_ri_pmi_cqi.cqi, metrics.mac.dl_cqi, metrics.cqi_count);
metrics.cqi_valid_count++;
}
return true;
}
public:
struct args_t {
srsran::phy_cfg_nr_t phy_cfg; ///< Physical layer configuration
std::string use_dummy_mac = "dummymac"; ///< Use dummy or real NR scheduler
bool wait_preamble = false; ///< Whether a UE is created automatically or the stack waits for a PRACH
uint16_t rnti = 0x1234; ///< C-RNTI
uint32_t ss_id = 1; ///< Search Space identifier
uint32_t pdcch_aggregation_level = 0; ///< PDCCH aggregation level
uint32_t pdcch_dl_candidate = 0; ///< PDCCH DL DCI candidate index
uint32_t pdcch_ul_candidate = 1; ///< PDCCH UL DCI candidate index
struct {
uint32_t rb_start = 0; ///< Start frequency domain resource block
uint32_t rb_length = 10; ///< Number of frequency domain resource blocks
uint32_t mcs = 10; ///< Modulation code scheme
std::string slots = ""; ///< Slot list, empty string means no scheduling
} pdsch, pusch;
std::string log_level = "warning";
};
gnb_dummy_stack(const args_t& args) :
rnti(args.rnti),
phy_cfg(args.phy_cfg),
ss_id(args.ss_id),
use_dummy_mac(args.use_dummy_mac == "dummymac"),
sched_logger(srslog::fetch_basic_logger("MAC")),
wait_preamble(args.wait_preamble)
{
logger.set_level(srslog::str_to_basic_level(args.log_level));
sched_logger.set_level(srslog::str_to_basic_level(args.log_level));
srslog::fetch_basic_logger("MAC-NR").set_level(srslog::str_to_basic_level(args.log_level));
autofill_sch_bsr = args.pdsch.slots != "" and args.pdsch.slots != "none";
// create sched object
mac.reset(new srsenb::mac_nr{&task_sched});
srsenb::mac_nr_args_t mac_args{};
mac_args.sched_cfg.pdsch_enabled = args.pdsch.slots != "" and args.pdsch.slots != "none";
mac_args.sched_cfg.pusch_enabled = args.pusch.slots != "" and args.pusch.slots != "none";
mac_args.sched_cfg.fixed_dl_mcs = args.pdsch.mcs;
mac_args.sched_cfg.fixed_ul_mcs = args.pusch.mcs;
mac->init(mac_args, nullptr, nullptr, &rlc_obj, &rrc_obj);
std::vector<srsenb::sched_nr_cell_cfg_t> cells_cfg = srsenb::get_default_cells_cfg(1, phy_cfg);
mac->cell_cfg(cells_cfg);
dl.mcs = args.pdsch.mcs;
ul.mcs = args.pusch.mcs;
if (args.pdsch.slots != "none" and not args.pdsch.slots.empty()) {
if (args.pdsch.slots == "all") {
for (uint32_t n = 1; n < SRSRAN_NSLOTS_PER_FRAME_NR(phy_cfg.carrier.scs); n++) {
dl.slots.insert(n);
}
} else {
srsran::string_parse_list(args.pdsch.slots, ',', dl.slots);
}
}
if (args.pusch.slots != "none" and not args.pusch.slots.empty()) {
if (args.pusch.slots == "all") {
for (uint32_t n = 0; n < SRSRAN_NSLOTS_PER_FRAME_NR(phy_cfg.carrier.scs); n++) {
ul.slots.insert(n);
}
} else {
srsran::string_parse_list(args.pusch.slots, ',', ul.slots);
}
}
// Select DCI locations
for (uint32_t slot = 0; slot < SRSRAN_NOF_SF_X_FRAME; slot++) {
srsran::bounded_vector<srsran_dci_location_t, SRSRAN_SEARCH_SPACE_MAX_NOF_CANDIDATES_NR> locations;
if (not phy_cfg.get_dci_locations(slot, rnti, args.ss_id, args.pdcch_aggregation_level, locations)) {
logger.error(
"Error generating locations for slot %d and aggregation level %d", slot, args.pdcch_aggregation_level);
return;
}
// DCI DL
if (args.pdcch_dl_candidate >= locations.size()) {
logger.error("Candidate index %d exceeds the number of candidates %d for aggregation level %d",
args.pdcch_dl_candidate,
(uint32_t)locations.size(),
args.pdcch_aggregation_level);
return;
}
dl.dci_location[slot] = locations[args.pdcch_dl_candidate];
// DCI UL
if (args.pdcch_ul_candidate >= locations.size()) {
logger.error("Candidate index %d exceeds the number of candidates %d for aggregation level %d",
args.pdcch_ul_candidate,
(uint32_t)locations.size(),
args.pdcch_aggregation_level);
return;
}
ul.dci_location[slot] = locations[args.pdcch_ul_candidate];
}
// Select DL frequency domain resources
dl.freq_res = srsran_ra_nr_type1_riv(args.phy_cfg.carrier.nof_prb, args.pdsch.rb_start, args.pdsch.rb_length);
// Select DL frequency domain resources
ul.freq_res = srsran_ra_nr_type1_riv(args.phy_cfg.carrier.nof_prb, args.pusch.rb_start, args.pusch.rb_length);
// Setup DL Data to ACK timing
for (uint32_t i = 0; i < SRSRAN_NOF_SF_X_FRAME; i++) {
if (args.phy_cfg.duplex.mode == SRSRAN_DUPLEX_MODE_TDD) {
dl_data_to_ul_ack[i] = args.phy_cfg.harq_ack.dl_data_to_ul_ack[i % args.phy_cfg.duplex.tdd.pattern1.period_ms];
} else {
dl_data_to_ul_ack[i] = args.phy_cfg.harq_ack.dl_data_to_ul_ack[i % args.phy_cfg.harq_ack.nof_dl_data_to_ul_ack];
}
}
// If reached this point the configuration is valid
valid = true;
}
~gnb_dummy_stack() = default;
void stop()
{
if (not use_dummy_mac) {
mac->stop();
}
}
bool is_valid() const { return valid; }
void start_scheduling()
{
// add UE to scheduler
if (not use_dummy_mac and not wait_preamble) {
srsenb::sched_nr_ue_cfg_t ue_cfg = srsenb::get_default_ue_cfg(1, phy_cfg);
ue_cfg.lc_ch_to_add.emplace_back();
ue_cfg.lc_ch_to_add.back().lcid = 4;
ue_cfg.lc_ch_to_add.back().cfg.direction = srsenb::mac_lc_ch_cfg_t::BOTH;
mac->reserve_rnti(0, ue_cfg);
}
enable_user_sched = true;
}
int slot_indication(const srsran_slot_cfg_t& slot_cfg) override { return 0; }
dl_sched_t* get_dl_sched(const srsran_slot_cfg_t& slot_cfg) override
{
logger.set_context(slot_cfg.idx);
sched_logger.set_context(slot_cfg.idx);
if (not use_dummy_mac) {
if (autofill_sch_bsr) {
mac->rlc_buffer_state(rnti, 0, 100000, 0);
mac->ul_bsr(rnti, 0, 100000);
}
dl_sched_t* dl_res = mac->get_dl_sched(slot_cfg);
if (dl_res == nullptr) {
return nullptr;
}
for (pdsch_t& pdsch : dl_res->pdsch) {
// Set TBS
// Select grant and set data
pdsch.data[0] = tx_harq_proc[slot_cfg.idx].get_tb(pdsch.sch.grant.tb[0].tbs);
pdsch.data[1] = nullptr;
}
return dl_res;
}
dl_sched_t& dl_sched = dl_scheds[slot_cfg.idx];
dl_sched = {};
// Check if it is TDD DL slot and PDSCH mask, if no PDSCH shall be scheduled, do not set any grant and skip
if (not srsran_duplex_nr_is_dl(&phy_cfg.duplex, phy_cfg.carrier.scs, slot_cfg.idx)) {
return nullptr;
}
if (not schedule_pdsch(slot_cfg, dl_sched)) {
logger.error("Error scheduling PDSCH");
return nullptr;
}
// Schedule SSB before UL
for (uint32_t ssb_idx = 0; ssb_idx < SRSRAN_SSB_NOF_CANDIDATES; ssb_idx++) {
if (phy_cfg.ssb.position_in_burst[ssb_idx]) {
srsran_mib_nr_t mib = {};
mib.ssb_idx = ssb_idx;
mib.sfn = slot_cfg.idx / SRSRAN_NSLOTS_PER_FRAME_NR(phy_cfg.carrier.scs);
mib.hrf = (slot_cfg.idx % SRSRAN_NSLOTS_PER_FRAME_NR(phy_cfg.carrier.scs)) >=
SRSRAN_NSLOTS_PER_FRAME_NR(phy_cfg.carrier.scs) / 2;
mac_interface_phy_nr::ssb_t ssb = {};
if (srsran_pbch_msg_nr_mib_pack(&mib, &ssb.pbch_msg) < SRSRAN_SUCCESS) {
logger.error("Error Packing MIB in slot %d", slot_cfg.idx);
continue;
}
ssb.pbch_msg.ssb_idx = (uint32_t)ssb_idx;
dl_sched.ssb.push_back(ssb);
}
}
// Check if the UL slot is valid, if not skip UL scheduling
if (not srsran_duplex_nr_is_ul(&phy_cfg.duplex, phy_cfg.carrier.scs, TTI_TX(slot_cfg.idx))) {
return &dl_sched;
}
if (not schedule_pusch(slot_cfg, dl_sched)) {
logger.error("Error scheduling PUSCH");
return nullptr;
}
// Schedule NZP-CSI-RS, iterate all NZP-CSI-RS sets
for (const srsran_csi_rs_nzp_set_t& set : phy_cfg.pdsch.nzp_csi_rs_sets) {
// For each NZP-CSI-RS resource available in the set
for (uint32_t i = 0; i < set.count; i++) {
// Select resource
const srsran_csi_rs_nzp_resource_t& nzp_csi_resource = set.data[i];
// Check if the resource is scheduled for this slot
if (srsran_csi_rs_send(&nzp_csi_resource.periodicity, &slot_cfg)) {
dl_sched.nzp_csi_rs.push_back(nzp_csi_resource);
}
}
}
return &dl_sched;
}
ul_sched_t* get_ul_sched(const srsran_slot_cfg_t& slot_cfg) override
{
logger.set_context(slot_cfg.idx);
sched_logger.set_context(slot_cfg.idx);
if (not use_dummy_mac) {
ul_sched_t* ul_res = mac->get_ul_sched(slot_cfg);
return ul_res;
}
ul_sched_t& ul_sched = ul_scheds[slot_cfg.idx];
ul_sched.pucch.clear();
ul_sched.pusch.clear();
// Get ACK information
srsran_pdsch_ack_nr_t ack = pending_ack[slot_cfg.idx % pending_ack.size()].get_ack();
bool has_ack = ack.nof_cc > 0;
if (has_ack) {
if (logger.debug.enabled()) {
std::array<char, 512> str = {};
if (srsran_harq_ack_info(&ack, str.data(), (uint32_t)str.size()) > 0) {
logger.debug("HARQ feedback:\n%s", str.data());
}
}
}
mac_interface_phy_nr::pusch_t pusch = {};
bool has_pusch = pending_pusch[slot_cfg.idx % pending_pusch.size()].pop(pusch.pid, pusch.sch);
srsran_uci_cfg_nr_t uci_cfg = {};
if (not phy_cfg.get_uci_cfg(slot_cfg, ack, uci_cfg)) {
logger.error("Error getting UCI configuration");
return nullptr;
}
// Schedule PUSCH
if (has_pusch) {
// If has PUSCH, no SR shall be received
uci_cfg.o_sr = 0;
// Put UCI configuration in PUSCH config
if (not phy_cfg.get_pusch_uci_cfg(slot_cfg, uci_cfg, pusch.sch)) {
logger.error("Error setting UCI configuration in PUSCH");
return nullptr;
}
ul_sched.pusch.push_back(pusch);
} else if ((uci_cfg.ack.count > 0 || uci_cfg.nof_csi > 0 || uci_cfg.o_sr > 0) and enable_user_sched) {
// If any UCI information is triggered, schedule PUCCH
ul_sched.pucch.emplace_back();
uci_cfg.pucch.rnti = rnti;
mac_interface_phy_nr::pucch_t& pucch = ul_sched.pucch.back();
pucch.candidates.emplace_back();
pucch.candidates.back().uci_cfg = uci_cfg;
if (not phy_cfg.get_pucch_uci_cfg(slot_cfg, uci_cfg, pucch.pucch_cfg, pucch.candidates.back().resource)) {
logger.error("Error getting UCI CFG");
return nullptr;
}
// If this slot has a SR opportunity and the selected PUCCH format is 1, consider positive SR.
if (uci_cfg.o_sr > 0 and uci_cfg.ack.count > 0 and
pucch.candidates.back().resource.format == SRSRAN_PUCCH_NR_FORMAT_1) {
// Set SR negative
if (uci_cfg.o_sr > 0) {
uci_cfg.sr_positive_present = false;
}
// Append new resource
pucch.candidates.emplace_back();
pucch.candidates.back().uci_cfg = uci_cfg;
if (not phy_cfg.get_pucch_uci_cfg(slot_cfg, uci_cfg, pucch.pucch_cfg, pucch.candidates.back().resource)) {
logger.error("Error getting UCI CFG");
return nullptr;
}
}
}
return &ul_sched;
}
int pucch_info(const srsran_slot_cfg_t& slot_cfg, const pucch_info_t& pucch_info) override
{
if (not use_dummy_mac) {
mac->pucch_info(slot_cfg, pucch_info);
} else {
// Handle UCI data
if (not handle_uci_data(pucch_info.uci_data.cfg, pucch_info.uci_data.value)) {
logger.error("Error handling UCI data from PUCCH reception");
return SRSRAN_ERROR;
}
}
// Skip next steps if uci data is invalid
if (not pucch_info.uci_data.value.valid) {
return SRSRAN_SUCCESS;
}
// Handle PHY metrics
std::unique_lock<std::mutex> lock(metrics_mutex);
metrics.pucch.epre_db_avg = SRSRAN_VEC_CMA(pucch_info.csi.epre_dB, metrics.pucch.epre_db_avg, metrics.pucch.count);
metrics.pucch.epre_db_min = SRSRAN_MIN(metrics.pucch.epre_db_min, pucch_info.csi.epre_dB);
metrics.pucch.epre_db_max = SRSRAN_MAX(metrics.pucch.epre_db_max, pucch_info.csi.epre_dB);
metrics.pucch.rsrp_db_avg = SRSRAN_VEC_CMA(pucch_info.csi.rsrp_dB, metrics.pucch.rsrp_db_avg, metrics.pucch.count);
metrics.pucch.rsrp_db_min = SRSRAN_MIN(metrics.pucch.rsrp_db_min, pucch_info.csi.rsrp_dB);
metrics.pucch.rsrp_db_max = SRSRAN_MAX(metrics.pucch.rsrp_db_max, pucch_info.csi.rsrp_dB);
metrics.pucch.snr_db_avg = SRSRAN_VEC_CMA(pucch_info.csi.snr_dB, metrics.pucch.snr_db_avg, metrics.pucch.count);
metrics.pucch.snr_db_min = SRSRAN_MIN(metrics.pucch.snr_db_min, pucch_info.csi.snr_dB);
metrics.pucch.snr_db_max = SRSRAN_MAX(metrics.pucch.snr_db_max, pucch_info.csi.snr_dB);
metrics.pucch.ta_us_avg = SRSRAN_VEC_CMA(pucch_info.csi.delay_us, metrics.pucch.ta_us_avg, metrics.pucch.count);
metrics.pucch.ta_us_min = SRSRAN_MIN(metrics.pucch.ta_us_min, pucch_info.csi.delay_us);
metrics.pucch.ta_us_max = SRSRAN_MAX(metrics.pucch.ta_us_max, pucch_info.csi.delay_us);
metrics.pucch.count++;
return SRSRAN_SUCCESS;
}
int pusch_info(const srsran_slot_cfg_t& slot_cfg, pusch_info_t& pusch_info) override
{
if (not use_dummy_mac) {
mac->pusch_info(slot_cfg, pusch_info);
} else {
// Handle UCI data
if (not handle_uci_data(pusch_info.uci_cfg, pusch_info.pusch_data.uci)) {
logger.error("Error handling UCI data from PUCCH reception");
return SRSRAN_ERROR;
}
// Handle UL-SCH metrics
std::unique_lock<std::mutex> lock(metrics_mutex);
if (not pusch_info.pusch_data.tb[0].crc) {
metrics.mac.rx_errors++;
}
metrics.mac.rx_brate += rx_harq_proc[pusch_info.pid].get_tbs();
metrics.mac.rx_pkts++;
}
// Handle PHY metrics
metrics.pusch.epre_db_avg = SRSRAN_VEC_CMA(pusch_info.csi.epre_dB, metrics.pusch.epre_db_avg, metrics.pusch.count);
metrics.pusch.epre_db_min = SRSRAN_MIN(metrics.pusch.epre_db_min, pusch_info.csi.epre_dB);
metrics.pusch.epre_db_max = SRSRAN_MAX(metrics.pusch.epre_db_max, pusch_info.csi.epre_dB);
metrics.pusch.rsrp_db_avg = SRSRAN_VEC_CMA(pusch_info.csi.rsrp_dB, metrics.pusch.rsrp_db_avg, metrics.pusch.count);
metrics.pusch.rsrp_db_min = SRSRAN_MIN(metrics.pusch.rsrp_db_min, pusch_info.csi.rsrp_dB);
metrics.pusch.rsrp_db_max = SRSRAN_MAX(metrics.pusch.rsrp_db_max, pusch_info.csi.rsrp_dB);
metrics.pusch.snr_db_avg = SRSRAN_VEC_CMA(pusch_info.csi.snr_dB, metrics.pusch.snr_db_avg, metrics.pusch.count);
metrics.pusch.snr_db_min = SRSRAN_MIN(metrics.pusch.snr_db_min, pusch_info.csi.snr_dB);
metrics.pusch.snr_db_max = SRSRAN_MAX(metrics.pusch.snr_db_max, pusch_info.csi.snr_dB);
metrics.pusch.ta_us_avg = SRSRAN_VEC_CMA(pusch_info.csi.delay_us, metrics.pusch.ta_us_avg, metrics.pusch.count);
metrics.pusch.ta_us_min = SRSRAN_MIN(metrics.pusch.ta_us_min, pusch_info.csi.delay_us);
metrics.pusch.ta_us_max = SRSRAN_MAX(metrics.pusch.ta_us_max, pusch_info.csi.delay_us);
metrics.pusch.count++;
return SRSRAN_SUCCESS;
}
void rach_detected(const rach_info_t& rach_info) override
{
if (not use_dummy_mac) {
mac->rach_detected(rach_info);
task_sched.run_pending_tasks();
}
std::unique_lock<std::mutex> lock(metrics_mutex);
prach_metrics_t& prach_metrics = metrics.prach[rach_info.preamble];
prach_metrics.avg_ta = SRSRAN_VEC_SAFE_CMA((float)rach_info.time_adv, prach_metrics.avg_ta, prach_metrics.count);
prach_metrics.min_ta = SRSRAN_MIN((float)rach_info.time_adv, prach_metrics.min_ta);
prach_metrics.max_ta = SRSRAN_MAX((float)rach_info.time_adv, prach_metrics.max_ta);
prach_metrics.count++;
}
metrics_t get_metrics()
{
std::unique_lock<std::mutex> lock(metrics_mutex);
if (not use_dummy_mac) {
srsenb::mac_metrics_t mac_metrics;
mac->get_metrics(mac_metrics);
metrics.mac = mac_metrics.ues[0];
}
return metrics;
}
private:
srsran::circular_array<dl_sched_t, TTIMOD_SZ> dl_scheds;
srsran::circular_array<ul_sched_t, TTIMOD_SZ> ul_scheds;
};
#endif // SRSRAN_DUMMY_GNB_STACK_H