diff --git a/openair2/LAYER2/MAC/eNB_scheduler_bch.c b/openair2/LAYER2/MAC/eNB_scheduler_bch.c index becbc15288..f6566ce354 100644 --- a/openair2/LAYER2/MAC/eNB_scheduler_bch.c +++ b/openair2/LAYER2/MAC/eNB_scheduler_bch.c @@ -733,6 +733,7 @@ schedule_SI(module_id_t module_idP, frame_t frameP, sub_frame_t subframeP) dl_config_pdu->dlsch_pdu.dlsch_pdu_rel8.num_bf_vector = 1; // dl_config_pdu->dlsch_pdu.dlsch_pdu_rel8.bf_vector = ; dl_req->number_pdu++; + dl_req->tl.tag = NFAPI_DL_CONFIG_REQUEST_BODY_TAG; // Rel10 fields dl_config_pdu->dlsch_pdu.dlsch_pdu_rel10.tl.tag = NFAPI_DL_CONFIG_REQUEST_DLSCH_PDU_REL10_TAG; diff --git a/openair2/LAYER2/MAC/eNB_scheduler_fairRR.c b/openair2/LAYER2/MAC/eNB_scheduler_fairRR.c index d0e3ca30cb..99c337b528 100644 --- a/openair2/LAYER2/MAC/eNB_scheduler_fairRR.c +++ b/openair2/LAYER2/MAC/eNB_scheduler_fairRR.c @@ -600,6 +600,7 @@ void dlsch_scheduler_pre_processor_fairRR (module_id_t Mod_id, #endif memset(rballoc_sub[0],0,(MAX_NUM_CCs)*(N_RBG_MAX)*sizeof(unsigned char)); memset(min_rb_unit,0,sizeof(min_rb_unit)); + memset(MIMO_mode_indicator[0], 0, MAX_NUM_CCs*N_RBG_MAX*sizeof(unsigned char)); for (CC_id=0; CC_id 0) // If this CC is allocated for MBSFN skip it here @@ -1763,6 +1764,7 @@ schedule_ue_spec_fairRR(module_id_t module_idP, dl_config_pdu->pdu_size = (uint8_t)(2+sizeof(nfapi_dl_config_dci_dl_pdu)); dl_config_pdu->dci_dl_pdu.dci_dl_pdu_rel8.dci_format = NFAPI_DL_DCI_FORMAT_1; dl_config_pdu->dci_dl_pdu.dci_dl_pdu_rel8.aggregation_level = get_aggregation(get_bw_index(module_idP,CC_id),ue_sched_ctl->dl_cqi[CC_id],format1); + dl_config_pdu->dci_dl_pdu.dci_dl_pdu_rel8.tl.tag = NFAPI_DL_CONFIG_REQUEST_DCI_DL_PDU_REL8_TAG; dl_config_pdu->dci_dl_pdu.dci_dl_pdu_rel8.rnti = rnti; dl_config_pdu->dci_dl_pdu.dci_dl_pdu_rel8.rnti_type = 1; // CRNTI : see Table 4-10 from SCF082 - nFAPI specifications dl_config_pdu->dci_dl_pdu.dci_dl_pdu_rel8.transmission_power = 6000; // equal to RS power