Commit 41b18e21 authored by Francesco Mani's avatar Francesco Mani

UE SSB periodicity (hardcoded)

parent aa12c954
...@@ -123,7 +123,7 @@ void phy_procedures_nrUE_TX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t e ...@@ -123,7 +123,7 @@ void phy_procedures_nrUE_TX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t e
@param r_type indicates the relaying operation: 0: no_relaying, 1: unicast relaying type 1, 2: unicast relaying type 2, 3: multicast relaying @param r_type indicates the relaying operation: 0: no_relaying, 1: unicast relaying type 1, 2: unicast relaying type 2, 3: multicast relaying
@param phy_vars_rn pointer to RN variables @param phy_vars_rn pointer to RN variables
*/ */
int phy_procedures_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eNB_id,uint8_t do_pdcch_flag,runmode_t mode); int phy_procedures_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eNB_id,uint8_t do_pdcch_flag,runmode_t mode,fapi_nr_pbch_config_t pbch_config);
int phy_procedures_slot_parallelization_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eNB_id,uint8_t abstraction_flag,uint8_t do_pdcch_flag,runmode_t mode,relaying_type_t r_type); int phy_procedures_slot_parallelization_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eNB_id,uint8_t abstraction_flag,uint8_t do_pdcch_flag,runmode_t mode,relaying_type_t r_type);
......
...@@ -4958,18 +4958,20 @@ int phy_procedures_slot_parallelization_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_pr ...@@ -4958,18 +4958,20 @@ int phy_procedures_slot_parallelization_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_pr
int phy_procedures_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eNB_id, int phy_procedures_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eNB_id,
uint8_t do_pdcch_flag,runmode_t mode) { uint8_t do_pdcch_flag,runmode_t mode,
fapi_nr_pbch_config_t pbch_config) {
int l,l2; int l,l2;
int pilot1; int pilot1;
int frame_rx = proc->frame_rx; int frame_rx = proc->frame_rx;
int nr_tti_rx = proc->nr_tti_rx; int nr_tti_rx = proc->nr_tti_rx;
NR_UE_PDCCH *pdcch_vars = ue->pdcch_vars[ue->current_thread_id[nr_tti_rx]][0]; NR_UE_PDCCH *pdcch_vars = ue->pdcch_vars[ue->current_thread_id[nr_tti_rx]][0];
uint16_t nb_symb_sch = 8; // to be updated by higher layer uint16_t nb_symb_sch = 8; // to be updated by higher layer
uint8_t nb_symb_pdcch = pdcch_vars->coreset[0].duration; uint8_t nb_symb_pdcch = pdcch_vars->coreset[0].duration;
uint8_t ssb_periodicity = 20; // TODO hardcoded to be taken from upper layers?
uint8_t ssb_frame_periodicity;
LOG_D(PHY," ****** start RX-Chain for Frame.Slot %d.%d ****** \n", frame_rx%1024, nr_tti_rx); LOG_D(PHY," ****** start RX-Chain for Frame.Slot %d.%d ****** \n", frame_rx%1024, nr_tti_rx);
...@@ -5094,18 +5096,16 @@ int phy_procedures_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eN ...@@ -5094,18 +5096,16 @@ int phy_procedures_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eN
VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_PDSCH_PROC_RA, VCD_FUNCTION_OUT); VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_PDSCH_PROC_RA, VCD_FUNCTION_OUT);
} }
ssb_frame_periodicity = ssb_periodicity/10 ; // 10ms is the frame length
// looking for pbch only in frames according to ssb periodicity
if ( (nr_tti_rx == 0) && (ue->decode_MIB == 1)) if ( (nr_tti_rx == 0) && (ue->decode_MIB == 1) && !((frame_rx-(pbch_config.system_frame_number))%ssb_frame_periodicity))
{ {
LOG_D(PHY," ------ PBCH ChannelComp/LLR: frame.slot %d.%d ------ \n", frame_rx%1024, nr_tti_rx); LOG_D(PHY," ------ PBCH ChannelComp/LLR: frame.slot %d.%d ------ \n", frame_rx%1024, nr_tti_rx);
uint8_t i_ssb = ue->rx_ind.rx_indication_body[0].mib_pdu.ssb_index;
uint8_t n_hf = (((ue->rx_ind.rx_indication_body[0].mib_pdu.additional_bits)>>4)&0x01);
for (int i=1; i<4; i++) { for (int i=1; i<4; i++) {
nr_slot_fep(ue, nr_slot_fep(ue,
(ue->symbol_offset+i), //mu=1 case B (ue->symbol_offset+i),
nr_tti_rx, nr_tti_rx,
0, 0,
0, 0,
...@@ -5114,7 +5114,7 @@ int phy_procedures_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eN ...@@ -5114,7 +5114,7 @@ int phy_procedures_nrUE_RX(PHY_VARS_NR_UE *ue,UE_nr_rxtx_proc_t *proc,uint8_t eN
#if UE_TIMING_TRACE #if UE_TIMING_TRACE
start_meas(&ue->dlsch_channel_estimation_stats); start_meas(&ue->dlsch_channel_estimation_stats);
#endif #endif
nr_pbch_channel_estimation(ue,0,0,ue->symbol_offset+i,i-1,i_ssb,n_hf); nr_pbch_channel_estimation(ue,0,0,ue->symbol_offset+i,i-1,(pbch_config.ssb_index)&7,pbch_config.half_frame_bit);
#if UE_TIMING_TRACE #if UE_TIMING_TRACE
stop_meas(&ue->dlsch_channel_estimation_stats); stop_meas(&ue->dlsch_channel_estimation_stats);
#endif #endif
...@@ -5183,7 +5183,7 @@ start_meas(&ue->generic_stat); ...@@ -5183,7 +5183,7 @@ start_meas(&ue->generic_stat);
if(nr_tti_rx==5 && ue->dlsch[ue->current_thread_id[nr_tti_rx]][eNB_id][0]->harq_processes[ue->dlsch[ue->current_thread_id[nr_tti_rx]][eNB_id][0]->current_harq_pid]->nb_rb > 20){ if(nr_tti_rx==5 && ue->dlsch[ue->current_thread_id[nr_tti_rx]][eNB_id][0]->harq_processes[ue->dlsch[ue->current_thread_id[nr_tti_rx]][eNB_id][0]->current_harq_pid]->nb_rb > 20){
//write_output("decoder_llr.m","decllr",dlsch_llr,G,1,0); //write_output("decoder_llr.m","decllr",dlsch_llr,G,1,0);
//write_output("llr.m","llr", &ue->pdsch_vars[eNB_id]->llr[0][0],(14*nb_rb*12*dlsch1_harq->Qm) - 4*(nb_rb*4*dlsch1_harq->Qm),1,0); //write_output("llr.m","llr", &ue->pdsch_vars[eNB_id]->llr[0][0],(14*nb_rb*12*dlsch1_harq->Qm) - 4*(nb_rb*4*dlsch1_harq->Qm),1,0);
openair1/SCHED_NR/phy_procedures_nr_gNB.c
write_output("rxdataF0_current.m" , "rxdataF0", &ue->common_vars.common_vars_rx_data_per_thread[ue->current_thread_id[nr_tti_rx]].rxdataF[0][0],14*ue->frame_parms.ofdm_symbol_size,1,1); write_output("rxdataF0_current.m" , "rxdataF0", &ue->common_vars.common_vars_rx_data_per_thread[ue->current_thread_id[nr_tti_rx]].rxdataF[0][0],14*ue->frame_parms.ofdm_symbol_size,1,1);
//write_output("rxdataF0_previous.m" , "rxdataF0_prev_sss", &ue->common_vars.common_vars_rx_data_per_thread[next_thread_id].rxdataF[0][0],14*ue->frame_parms.ofdm_symbol_size,1,1); //write_output("rxdataF0_previous.m" , "rxdataF0_prev_sss", &ue->common_vars.common_vars_rx_data_per_thread[next_thread_id].rxdataF[0][0],14*ue->frame_parms.ofdm_symbol_size,1,1);
......
...@@ -638,7 +638,7 @@ static void *UE_thread_synch(void *arg) { ...@@ -638,7 +638,7 @@ static void *UE_thread_synch(void *arg) {
return &UE_thread_synch_retval; return &UE_thread_synch_retval;
} }
void processSubframeRX( PHY_VARS_NR_UE *UE, UE_nr_rxtx_proc_t *proc) { void processSlotRX( PHY_VARS_NR_UE *UE, UE_nr_rxtx_proc_t *proc) {
// Process Rx data for one sub-frame // Process Rx data for one sub-frame
if (slot_select_nr(&UE->frame_parms, proc->frame_tx, proc->nr_tti_tx) & NR_DOWNLINK_SLOT) { if (slot_select_nr(&UE->frame_parms, proc->frame_tx, proc->nr_tti_tx) & NR_DOWNLINK_SLOT) {
//clean previous FAPI MESSAGE //clean previous FAPI MESSAGE
...@@ -661,7 +661,7 @@ void processSubframeRX( PHY_VARS_NR_UE *UE, UE_nr_rxtx_proc_t *proc) { ...@@ -661,7 +661,7 @@ void processSubframeRX( PHY_VARS_NR_UE *UE, UE_nr_rxtx_proc_t *proc) {
#ifdef UE_SLOT_PARALLELISATION #ifdef UE_SLOT_PARALLELISATION
phy_procedures_slot_parallelization_nrUE_RX( UE, proc, 0, 0, 1, UE->mode, no_relay, NULL ); phy_procedures_slot_parallelization_nrUE_RX( UE, proc, 0, 0, 1, UE->mode, no_relay, NULL );
#else #else
phy_procedures_nrUE_RX( UE, proc, 0, 1, UE->mode); phy_procedures_nrUE_RX( UE, proc, 0, 1, UE->mode, UE_mac->phy_config.config_req.pbch_config);
// printf(">>> nr_ue_pdcch_procedures ended\n"); // printf(">>> nr_ue_pdcch_procedures ended\n");
#endif #endif
} }
...@@ -725,7 +725,7 @@ static void *UE_thread_rxn_txnp4(void *arg) { ...@@ -725,7 +725,7 @@ static void *UE_thread_rxn_txnp4(void *arg) {
} }
AssertFatal ( 0== pthread_mutex_unlock(&proc->mutex_rxtx), "[SCHED][UE] error unlocking mutex for UE RXn_TXnp4\n" ); AssertFatal ( 0== pthread_mutex_unlock(&proc->mutex_rxtx), "[SCHED][UE] error unlocking mutex for UE RXn_TXnp4\n" );
processSubframeRX(UE, proc); processSlotRX(UE, proc);
//printf(">>> mac ended\n"); //printf(">>> mac ended\n");
// Prepare the future Tx data // Prepare the future Tx data
#if 0 #if 0
...@@ -1123,7 +1123,7 @@ void *UE_thread(void *arg) { ...@@ -1123,7 +1123,7 @@ void *UE_thread(void *arg) {
// pickStaticTime(lastTime); // pickStaticTime(lastTime);
} //UE->mode != loop_through_memory } //UE->mode != loop_through_memory
else { else {
processSubframeRX(UE,proc); processSlotRX(UE,proc);
getchar(); getchar();
} // else loop_through_memory } // else loop_through_memory
} // while !oai_exit } // while !oai_exit
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment