Update tutorial

Optimization pass for channel modelling on transmission in vrtsim. This
comes from a requirement of O-RU to perform channel modelling with minimum
latency.

- Reduced MAX_CHANNEL_LENGTH to 200 and introduced SAVED_SAMPLES_LEN
- Refactored channel modelling actor allocation to parallelize work better
- Improved batching and task creation for channel modelling
- Replaced static global saved_samples with local static buffer in vrtsim_write_with_chanmod

CMakeLists.txt

@@ -282,7 +282,7 @@ add_boolean_option(TRACY_ENABLE OFF "Enable tracy instrumentation" ON)
 if (TRACY_ENABLE)
   # the tracy version here should match the tracy server version
   # below is latest release as of this commit
-  CPMAddPackage("gh:wolfpld/tracy#0.12.2")
+  CPMAddPackage("gh:wolfpld/tracy#v0.12.2")
 endif()
 
 set (OCP_ITTI ${OPENAIR_DIR}/common/utils/ocp_itti)

doc/tutorial_resources/vrtsim/channel_modelling_basics.md

@@ -0,0 +1,100 @@
+# About
+
+In this step usage of channel modelling with `vrtsim` will be explained
+
+# Configure 
+
+Modify the gNB config file at `targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb.sa.band78.fr1.106PRB.usrpb210.conf`. Add the following at the end:
+
+```
+channelmod = { 
+  max_chan=10;
+  modellist="modellist_rfsimu_1";
+  modellist_rfsimu_1 = (
+    {
+        model_name                       = "server_tx_channel_model"
+      	type                             = "AWGN";			  
+      	ploss_dB                         = 0;
+        noise_power_dB                   = 0; 
+        forgetfact                       = 0;  
+        offset                           = 0;      
+        ds_tdl                           = 0;      
+    }    
+  );
+};
+```
+
+# (Optional) Run UE with scope to visualize the impact
+
+```
+cmake ../../ -DENABLE_IMSCOPE=ON
+```
+
+```
+cmake --build . --target imscope
+```
+
+Add `--imscope` to UE command line
+
+# Run
+
+```
+sudo ./nr-softmodem -O ../../targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb.sa.band78.fr1.106PRB.usrpb210.conf --gNBs.[0].min_rxtxtime 6 --device.name vrtsim --vrtsim.role server --vrtsim.chanmod 1
+```
+
+```
+sudo ./nr-uesoftmodem -C 3619200000 -r 106 --numerology 1 --ssb 516 --band 78 --device.name vrtsim
+```
+
+# Modifying the channel model
+
+Try the following settings:
+
+```
+  modellist_rfsimu_1 = (
+    {
+        model_name                       = "server_tx_channel_model"
+      	type                             = "Rayleigh8";			  
+      	ploss_dB                         = 0;
+        noise_power_dB                   = 0; 
+        forgetfact                       = 0;  
+        offset                           = 0;      
+        ds_tdl                           = 0;      
+    }    
+  );
+```
+```
+  modellist_rfsimu_1 = (
+    {
+        model_name                       = "server_tx_channel_model"
+      	type                             = "AWGN";			  
+      	ploss_dB                         = -5;
+        noise_power_dB                   = 0; 
+        forgetfact                       = 0;  
+        offset                           = 0;      
+        ds_tdl                           = 0;      
+    }    
+  );
+```
+```
+  modellist_rfsimu_1 = (
+    {
+        model_name                       = "server_tx_channel_model"
+      	type                             = "AWGN";			  
+      	ploss_dB                         = -10;
+        noise_power_dB                   = 0; 
+        forgetfact                       = 0;  
+        offset                           = 0;      
+        ds_tdl                           = 0;      
+    }    
+  );
+```
+
+Add noise:
+
+```
+channelmod = { 
+  max_chan=10;
+  modellist="modellist_rfsimu_1";
+  noise_power_dBFS = -42;
+```

doc/tutorial_resources/vrtsim/channel_modelling_external_channel.md

@@ -0,0 +1,92 @@
+## Option 1: Work via Emitter
+
+### About
+
+Precomputed 3GPP 38.901 TDL profiles (A to E) with statistical MIMO channels, plus a Python emitter that publishes taps to VRTSIM over a nanomsg PUB socket using FlatBuffers. Supports multiple antenna shapes (up to 64x64), speed presets (walking about 1.5 m/s, car about 30 m/s), spatial correlation, mutual coupling, LOS steering for D and E, and time-varying lognormal shadowing.
+
+**Key Features:**
+* Precomputed 3GPP 38.901 TDL taps (A to E) stored in a CIR database, published by a Python emitter over a PUB socket using FlatBuffers.
+* Avoids on-the-fly channel synthesis during runs, which reduces CPU load and improves timing stability.
+* Database contains profiles, delay spreads, antenna shapes, and speed presets, so scenarios are reproducible and easy to share.
+* VRTSim consumes taps from the socket and applies them per slot, which decouples tap generation from application.
+* Channel model is editable in the external repo, so new profiles or motion models can be added without touching VRTSim.
+
+### Clone Channel Emulator
+```bash
+git clone https://gitlab.eurecom.fr/oai/raytracing-channel-emulator.git
+```
+
+### Setup Schema
+```bash
+python3 -m pip install --user flatbuffers nanomsg numpy
+cd ~/raytracing-channel-emulator/server/api
+flatc --python -o . taps.fbs    # generates ./Phy/Taps.py
+```
+
+### Generate CIR DB
+```bash
+cd ~/raytracing-channel-emulator/server/external_taps
+python3 CIR_generator.py --out ./cir_db.bin --demo
+```
+
+### Run CIR Emitter
+```bash
+python3 emit_from_db.py --bind tcp://127.0.0.1:5555 --model TDL-A --ds-ns 10 --nrx 1 --ntx 1 --interval 0.5
+```
+
+### Clone OAI
+```bash
+git clone https://gitlab.eurecom.fr/oai/openairinterface5g
+git fetch origin vrtsim_cirdb_read:vrtsim_cirdb_read
+git checkout vrtsim_cirdb_read
+```
+
+### Build
+```bash
+cd openairinterface5g/cmake_targets
+cd build 
+cmake ../.. -GNinja   -DCMAKE_BUILD_TYPE=RelWithDebInfo   -DOAI_VRTSIM_TAPS_CLIENT=ON   -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
+cmake --build . --target taps_client vrtsim rfsimulator nr-softmodem nr-uesoftmodem ldpc params_libconfig -j"$(nproc)"
+```
+
+### Run gNB
+```bash
+sudo ./nr-softmodem -O ../../targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb.sa.band78.fr1.106PRB.usrpb210.conf --device.name vrtsim --vrtsim.role server --vrtsim.taps-socket tcp://127.0.0.1:5555 --gNBs.[0].min_rxtxtime 6
+```
+
+### Run UE
+```bash
+sudo ./nr-uesoftmodem -C 3619200000 -r 106 --numerology 1 --ssb 516 --band 78 --device.name vrtsim --vrtsim.role client
+```
+
+---
+
+## Option 2: Read DB Directly (In-process CIRDB)
+
+### About
+
+**Key Features:**
+* Adds an in-process CIR database path in VRTSim that applies taps internally, so no external emitter is required and no socket exchange occurs.
+* The dataset is produced offline using the CIR DB generator in the channel-emulator repo, then consumed by VRTSim for deterministic replay.
+* Avoids on-the-fly channel synthesis during runs, which reduces CPU load and improves timing stability.
+* Reduces transport jitter and context switches compared to a socket publisher, which helps at higher PRB, MIMO, and UE counts.
+
+### Run gNB
+```bash
+sudo ./nr-softmodem \
+  -O ./../../targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb.sa.band78.fr1.106PRB.usrpb210.conf \
+  --device.name vrtsim \
+  --vrtsim.role server \
+  --gNBs.[0].min_rxtxtime 3 \
+  --vrtsim.cirdb 1 \
+  --vrtsim.cirdb_yaml /openairinterface5g/radio/vrtsim/cir_db.yaml \
+  --vrtsim.cirdb_file /openairinterface5g/radio/vrtsim/cir_db.bin \
+  --vrtsim.cirdb_model_id 0 \
+  --vrtsim.cirdb_ds_ns 10 \
+  --vrtsim.cirdb_speed_mps 1.5
+```
+
+### Run UE
+```bash
+sudo ./nr-uesoftmodem -C 3619200000 -r 106 --numerology 1 --ssb 516 --band 78 --device.name vrtsim --vrtsim.role client
+```

doc/tutorial_resources/vrtsim/first_steps.md

@@ -0,0 +1,145 @@
+# About
+
+In this step the basics of compiling and running `vrtsim` will be explained.
+
+# About vrtsim
+
+vrtsim is a realtime/fixed timescale split8 device emulator. The main difference between
+rfsim and vrtsim is the timing constraints vrtsim forces on the applications that use it.
+
+# Compiling
+
+Start by making a new directory under `cmake_targets`
+
+```bash
+mkdir -p cmake_targets/build/
+```
+
+Compile vrtsim + RAN elements
+
+```bash
+cd /cmake_targets/build
+cmake ../../ -GNinja
+cmake --build . --target vrtsim nr-softmodem nr-uesoftmodem ldpc params_libconfig params_yaml
+```
+
+# Running
+
+Run a basic testcase to verify connectivity, no core network present
+
+```bash
+sudo ./nr-softmodem -O ../../targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb.sa.band78.fr1.106PRB.usrpb210.conf --gNBs.[0].min_rxtxtime 6 --device.name vrtsim --vrtsim.role server
+```
+
+```bash
+sudo ./nr-uesoftmodem -C 3619200000 -r 106 --numerology 1 --ssb 516 --band 78 --device.name vrtsim
+```
+
+# Verification
+
+Goal is to have the UE connect for now (i.e. enter RRC Connected mode). The success depends on your
+CPU speed.
+
+## Example NR UE output when UE is connected
+
+```
+[PHY]   Initial sync: pbch decoded sucessfully, ssb index 0
+[PHY]   pbch rx ok. rsrp:51 dB/RE, adjust_rxgain:-1 dB
+[NR_PHY]   Cell Detected with GSCN: 0, SSB SC offset: 516, SSB Ref: 0.000000, PSS Corr peak: 99 dB, PSS Corr Average: 61
+[PHY]   [UE0] In synch, rx_offset 444704 samples
+[PHY]   [UE 0] Measured Carrier Frequency offset 6 Hz
+[PHY]   Initial sync successful, PCI: 0
+[PHY]   HW: Configuring channel 0 (rf_chain 0): setting tx_freq 3619200006 Hz, rx_freq 3619200006 Hz, tune_offset 0
+[PHY]   Got synch: hw_slot_offset 29, carrier off 6 Hz, rxgain 110.000000 (DL 3619200006.000000 Hz, UL 3619200006.000000 Hz)
+[PHY]   UE synchronized! decoded_frame_rx=404 UE->init_sync_frame=0 trashed_frames=14
+[PHY]   Resynchronizing RX by 444704 samples
+[HW]   received write reorder clear context
+[NR_RRC]   SIB1 decoded
+[NR_MAC]   TDD period index = 6, based on the sum of dl_UL_TransmissionPeriodicity from Pattern1 (5.000000 ms) and Pattern2 (0.000000 ms): Total = 5.000000 ms
+[NR_MAC]   Set TDD configuration period to: 8 DL slots, 3 UL slots, 10 slots per period (NR_TDD_UL_DL_Pattern is 7 DL slots, 2 UL slots, 6 DL symbols, 4 UL symbols)
+[NR_MAC]   Configured 1 TDD patterns (total slots: pattern1 = 10, pattern2 = 0)
+[PHY]   N_TA_offset changed from 0 to 800
+[MAC]   Initialization of 4-Step CBRA procedure
+[NR_MAC]   PRACH scheduler: Selected RO Frame 421, Slot 19, Symbol 0, Fdm 0
+[PHY]   PRACH [UE 0] in frame.slot 421.19, placing PRACH in position 2828, Msg1/MsgA-Preamble frequency start 0 (k1 0), preamble_offset 6, first_nonzero_root_idx 0, preambleIndex = 27
+[PHY]   [UE 0] RAR-Msg2 decoded
+[NR_MAC]   [UE 0][RAPROC][RA-RNTI 010b] Got BI RAR subPDU 5 ms
+[NR_MAC]   [UE 0][RAPROC][RA-RNTI 010b] Got RAPID RAR subPDU
+[NR_MAC]   [UE 0][RAPROC][422.10] Found RAR with the intended RAPID 27
+[MAC]   received TA command 31
+[NR_MAC]   [RAPROC][422.19] RA-Msg3 transmitted
+[MAC]   [UE 0][423.10][RAPROC] 4-Step RA procedure succeeded. CBRA: Contention Resolution is successful.
+[NR_RRC]   [UE0][RAPROC] Logical Channel DL-CCCH (SRB0), Received NR_RRCSetup
+[RLC]   Added srb 1 to UE 0
+[NR_RRC]   State = NR_RRC_CONNECTED
+```
+
+## Example gNB output when connected
+
+```
+[NR_PHY]   [RAPROC] 421.19 Initiating RA procedure with preamble 27, energy 31.7 dB (I0 0, thres 120), delay 0 start symbol 0 freq index 0
+[NR_MAC]   421.19 UE RA-RNTI 010b TC-RNTI 2c32: initiating RA procedure
+[NR_MAC]   UE 2c32: Msg3 scheduled at 422.19 (422.10 TDA 0) start 0 RBs 8
+[NR_MAC]   UE 2c32: 422.10 Generating RA-Msg2 DCI, RA RNTI 0x10b, state 1, preamble_index(RAPID) 27, timing_offset = 0 (estimated distance 0.0 [m])
+[NR_MAC]   422.10 Send RAR to RA-RNTI 010b
+[NR_MAC]    422.19 PUSCH with TC_RNTI 0x2c32 received correctly
+[MAC]   [RAPROC] Received SDU for CCCH length 6 for UE 2c32
+[RLC]   Activated srb0 for UE 11314
+[RLC]   Added srb 1 to UE 11314
+[NR_MAC]   Activating scheduling Msg4 for TC_RNTI 0x2c32 (state WAIT_Msg3)
+[NR_RRC]   Decoding CCCH: RNTI 2c32, payload_size 6
+[NR_RRC]   [--] (cellID 0, UE ID 1 RNTI 2c32) Create UE context: CU UE ID 1 DU UE ID 11314 (rnti: 2c32, random ue id 1c14de2d3f000000)
+[RRC]   activate SRB 1 of UE 1
+[NR_RRC]   [DL] (cellID bc614e, UE ID 1 RNTI 2c32) Send RRC Setup
+[NR_MAC]   UE 2c32 Generate Msg4: feedback at  423.17, payload 225 bytes, next state nrRA_WAIT_Msg4_MsgB_ACK
+[NR_MAC]    423.17 UE 2c32: Received Ack of Msg4. CBRA procedure succeeded (UE Connected)
+[NR_MAC]   Adding new UE context with RNTI 0x2c32
+[NR_RRC]   [UL] (cellID bc614e, UE ID 1 RNTI 2c32) Received RRCSetupComplete (RRC_CONNECTED reached)
+```
+
+## Troubleshooting:
+In case the UE does not connect:
+
+Interrupt the process using `CTRL + C` and analyze `vrtsim` stdout at the end. You should see the histogram
+print from `vrtsim`
+
+Example from a working configuration:
+```
+[HW]   VRTSIM: Realtime issues: TX 0.00%, RX 0.00%
+[HW]   VRTSIM: Read/write too early (suspected radio implementaton error) TX: 0, RX: 0
+[HW]   VRTSIM: Average TX budget 913.111 uS (more is better)
+[HW]   VRTSIM: TX budget histogram: 6468 samples
+[HW]   Bin 0    [0.0 - 100.0uS]:                0
+[HW]   Bin 1    [100.0 - 200.0uS]:              0
+[HW]   Bin 2    [200.0 - 300.0uS]:              0
+[HW]   Bin 3    [300.0 - 400.0uS]:              0
+[HW]   Bin 4    [400.0 - 500.0uS]:              1
+[HW]   Bin 5    [500.0 - 600.0uS]:              0
+[HW]   Bin 6    [600.0 - 700.0uS]:              0
+[HW]   Bin 7    [700.0 - 800.0uS]:              1
+[HW]   Bin 8    [800.0 - 900.0uS]:              125
+[HW]   Bin 9    [900.0 - 1000.0uS]:             6242
+[HW]   Bin 10   [1000.0 - 1100.0uS]:            0
+[HW]   Bin 11   [1100.0 - 1200.0uS]:            0
+```
+
+The histogram above indicates if the application encountered any timing related issues. If the average TX budget
+is low (near ~0) or bin 0 is being filled that could mean that your CPU is too slow.
+
+Possible fixes: 
+ - Try disabling idle states
+ ```bash
+ sudo cpupower idle-set -D 0
+ ```
+ - Change cpu frequency scaling governor to performance
+ ```bash
+ sudo cpupower frequency-set --governor performance
+ ```
+ - Modify this line in the code of the UE and rebuild:
+```diff
+-#define NR_UE_CAPABILITY_SLOT_RX_TO_TX (3)
++#define NR_UE_CAPABILITY_SLOT_RX_TO_TX (6)
+```
+
+If that doesnt work, adjust `--vrtsim.timescale <timescale>` argument on `nr-softmodem`.
+Start lowering the timescale (e.g. to 0.2) to slow down time until the UE connects.

doc/tutorial_resources/vrtsim/ip_traffic.md

@@ -0,0 +1,95 @@
+# About
+
+In this step using `iperf` to test the modems performance is explained.
+
+# Running
+
+## Option 1: --noS1 mode
+
+Select this option on slower machines. This doesn't require core network.
+
+1. Run gNB
+
+```bash
+sudo ./nr-softmodem -O ../../targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb.sa.band78.fr1.106PRB.usrpb210.conf --gNBs.[0].min_rxtxtime 6 --device.name vrtsim --vrtsim.role server --noS1 --do-ra
+```
+
+2. Create UE namespace
+
+```bash
+sudo ip netns add ue_ns
+```
+
+2. Run UE in the new namespace
+
+```bash
+sudo ip netns exec bash
+sudo ./nr-uesoftmodem -C 3619200000 -r 106 --numerology 1 --ssb 516 --band 78 --device.name vrtsim --noS1 --do-ra
+```
+
+3. Run iperf server in the UE namespace
+
+```bash
+sudo ip netns exec bash
+iperf -s -B 10.0.1.2
+```
+
+4. Run iperf client in host namespace
+
+```bash
+iperf -c 10.0.1.2 -B 10.0.1.1
+```
+
+## Option 2: with core network
+
+1. Start `cn` in `doc/tutorial_resources/oai-cn5g` with `docker compose up`
+
+2. Run gNB
+
+```bash
+sudo ./nr-softmodem -O ../../targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb.sa.band78.fr1.106PRB.usrpb210.conf --gNBs.[0].min_rxtxtime 6 --device.name vrtsim --vrtsim.role server
+```
+
+3. Verify connection to core network in gNB stdout
+
+```
+[NGAP]   Send NGSetupRequest to AMF
+[NGAP]   3584 -> 0000e000
+[NGAP]   Served GUAMIs for AMF OAI-AMF (assoc_id=3):
+[NGAP]    GUAMI:
+[NGAP]      PLMN: MCC=001, MNC=01
+[NGAP]      AMF Region ID: 1
+[NGAP]      AMF Set ID: 1
+[NGAP]      AMF Pointer: 1
+[NGAP]   Supported PLMN 0: MCC=001 MNC=01
+[NGAP]   Supported slice (PLMN 0): SST=0x01 SD=000
+[NGAP]   Received NGSetupResponse from AMF
+```
+
+4. Connect the UE. 
+
+```bash
+sudo ./nr-uesoftmodem -C 3619200000 -r 106 --numerology 1 --ssb 516 --band 78 --device.name vrtsim  --uicc0.imsi 001010000000001
+```
+
+5. Verify interface is created
+
+```
+[OIP]   Interface oaitun_ue1 successfully configured, IPv4 10.0.0.2, IPv6 (null)
+```
+
+```bash
+ifconfig | grep tun   
+```
+
+6. Run iperf in oai-ext-dn container
+
+```bash
+docker exec -it oai-ext-dn bash
+iperf -c 10.0.0.2 -B 192.168.70.135
+```
+
+## Troubleshooting
+
+Use the same steps as from `first_steps.md`. This test adds additional requirements
+on the CPU which might cause the UE to fail.

openair1/PHY/NR_UE_ESTIMATION/nr_ue_measurements.c

@@ -219,7 +219,7 @@ void nr_ue_ssb_rsrp_measurements(PHY_VARS_NR_UE *ue,
   int SNRtimes10 = dB_fixed_x10(signal_pwr) - dB_fixed_x10(ue->measurements.n0_power_avg);
   ue->measurements.ssb_sinr_dB[ssb_index] = SNRtimes10 / 10.0;
 
-  LOG_D(PHY,
+  LOG_I(PHY,
         "[UE %d] ssb %d SS-RSRP: %d dBm/RE (%f dB/RE), SS-SINR: %f dB\n",
         ue->Mod_id,
         ssb_index,

radio/vrtsim/vrtsim.c

@@ -20,6 +20,7 @@
  */
 
 #include "PHY/TOOLS/tools_defs.h"
+#include "notified_fifo.h"
 #include <sys/socket.h>
 #include <sys/types.h>
 #include <netinet/in.h>
@@ -54,7 +55,10 @@
 typedef enum { ROLE_SERVER = 1, ROLE_CLIENT } role;
 
 #define MAX_NUM_ANTENNAS_TX 4
-#define MAX_CHANNEL_LENGTH (1 << 20)
+#define MAX_CHANNEL_LENGTH 200
+#define SAVED_SAMPLES_LEN (MAX_CHANNEL_LENGTH - 1)
+#define BATCH_SIZE 4096
+#define FIRST_BATCH_SIZE (BATCH_SIZE / 2)
 
 #define ROLE_CLIENT_STRING "client"
 #define ROLE_SERVER_STRING "server"
@@ -112,6 +116,7 @@ typedef struct {
   uint64_t rx_samples_late;
   uint64_t rx_early;
   uint64_t rx_samples_total;
+  int num_tx_timing;
   tx_timing_t *tx_timing;
   peer_info_t peer_info;
   int chanmod;
@@ -120,14 +125,12 @@ typedef struct {
   int tx_num_channels;
   int rx_num_channels;
   channel_desc_t *channel_desc;
+  int num_chanmod_actors;
   Actor_t *channel_modelling_actors;
   char *taps_socket;
   int client_num_rx_antennas;
 } vrtsim_state_t;
 
-// Sample history for channel impulse response
-static c16_t saved_samples[MAX_NUM_ANTENNAS_TX][MAX_CHANNEL_LENGTH] __attribute__((aligned(32))) = {0};
-
 static void histogram_add(histogram_t *histogram, double diff)
 {
   histogram->num_samples++;
@@ -312,10 +315,11 @@ static int vrtsim_connect(openair0_device *device)
   }
 
   // Handle channel modelling after number of RX antennas are known
-  int num_tx_stats = 1;
+  vrtsim_state->num_tx_timing = 1;
   if (vrtsim_state->chanmod || vrtsim_state->taps_socket) {
-    vrtsim_state->channel_modelling_actors = calloc_or_fail(vrtsim_state->peer_info.num_rx_antennas, sizeof(Actor_t));
-    for (int i = 0; i < vrtsim_state->peer_info.num_rx_antennas; i++) {
+    vrtsim_state->num_chanmod_actors = vrtsim_state->peer_info.num_rx_antennas * vrtsim_state->tx_num_channels;
+    vrtsim_state->channel_modelling_actors = calloc_or_fail(vrtsim_state->num_chanmod_actors, sizeof(Actor_t));
+    for (int i = 0; i < vrtsim_state->num_chanmod_actors; i++) {
       init_actor(&vrtsim_state->channel_modelling_actors[i], "chanmod", -1);
     }
     if (vrtsim_state->taps_socket) {
@@ -327,10 +331,10 @@ static int vrtsim_connect(openair0_device *device)
     } else {
       load_channel_model(vrtsim_state);
     }
-    num_tx_stats = vrtsim_state->peer_info.num_rx_antennas;
+    vrtsim_state->num_tx_timing = vrtsim_state->num_chanmod_actors;
   }
-  vrtsim_state->tx_timing = calloc_or_fail(num_tx_stats, sizeof(tx_timing_t));
-  for (int i = 0; i < num_tx_stats; i++) {
+  vrtsim_state->tx_timing = calloc_or_fail(vrtsim_state->num_tx_timing, sizeof(tx_timing_t));
+  for (int i = 0; i < vrtsim_state->num_tx_timing; i++) {
     vrtsim_state->tx_timing[i].tx_histogram.min_samples = 100;
     // Set the histogram range to 3000uS. Anything above that is not interesting
     vrtsim_state->tx_timing[i].tx_histogram.range = 3000.0;
@@ -371,25 +375,24 @@ typedef struct {
   vrtsim_state_t *vrtsim_state;
   openair0_timestamp timestamp;
   c16_t *samples[MAX_NUM_ANTENNAS_TX];
+  c16_t *first_samples[MAX_NUM_ANTENNAS_TX];
   int nsamps;
   int nbAnt;
   int flags;
   int aarx;
+  int task_index;
+  int num_tasks;
+  int tx_timing_index;
+  c16_t samples_copy[1];
 } channel_modelling_args_t;
 
 static void perform_channel_modelling(void *arg)
 {
-  channel_modelling_args_t *channel_modelling_args = arg;
-  vrtsim_state_t *vrtsim_state = channel_modelling_args->vrtsim_state;
-  int nsamps = channel_modelling_args->nsamps;
-  int aarx = channel_modelling_args->aarx;
-  int nb_tx_ant = channel_modelling_args->nbAnt;
-  c16_t **input_samples = (c16_t **)channel_modelling_args->samples;
-
-  int aligned_nsamps = ceil_mod(nsamps, (512 / 8) / sizeof(cf_t));
-  cf_t samples[aligned_nsamps] __attribute__((aligned(64)));
-  // Apply noise from global settings
-  get_noise_vector((float *)samples, nsamps * 2);
+  channel_modelling_args_t *args = arg;
+  vrtsim_state_t *vrtsim_state = args->vrtsim_state;
+  int nsamps = args->nsamps;
+  int aarx = args->aarx;
+  int nb_tx_ant = args->nbAnt;
 
   channel_desc_t *channel_desc = vrtsim_state->channel_desc;
 
@@ -417,90 +420,149 @@ static void perform_channel_modelling(void *arg)
     }
   }
 
-  for (int aatx = 0; aatx < nb_tx_ant; aatx++) {
-    c16_t *previous_samples = saved_samples[aatx];
-    for (int i = 0; i < nsamps; i++) {
-      cf_t *impulse_response = channel_impulse_response_p[aatx];
-      for (int l = 0; l < channel_desc->channel_length; l++) {
-        int idx = i - l;
-        // TODO: Use AVX2 for this
-        c16_t tx_input = idx >= 0 ? input_samples[aatx][idx]
-                                  : previous_samples[(channel_modelling_args->timestamp + i + idx) % MAX_CHANNEL_LENGTH];
-        samples[i].r += tx_input.r * impulse_response[l].r - tx_input.i * impulse_response[l].i;
-        samples[i].i += tx_input.i * impulse_response[l].r + tx_input.r * impulse_response[l].i;
+  // Calculate number of batches to process per thread.
+  int num_batches_to_process = 0;
+  if (nsamps > args->task_index * FIRST_BATCH_SIZE) {
+    num_batches_to_process = 1;
+    int nsamps_after_first_batch = max(0, nsamps - args->num_tasks * FIRST_BATCH_SIZE);
+    int num_full_size_batches = nsamps_after_first_batch / (args->num_tasks * BATCH_SIZE);
+
+    int nsamps_left = nsamps_after_first_batch - (num_full_size_batches * BATCH_SIZE);
+    if (nsamps_left > args->task_index * BATCH_SIZE)
+      num_batches_to_process++;
+    num_batches_to_process += num_full_size_batches;
+  }
+
+  for (int batch_index = 0; batch_index < num_batches_to_process; batch_index++) {
+    int batch_start_offset;
+    int batch_size;
+    if (batch_index == 0) {
+      batch_start_offset = args->task_index * FIRST_BATCH_SIZE;
+      batch_size = FIRST_BATCH_SIZE;
+    } else {
+      batch_start_offset = args->num_tasks * (FIRST_BATCH_SIZE + (batch_index - 1) * BATCH_SIZE) + args->task_index * BATCH_SIZE;
+      batch_size = BATCH_SIZE;
+    }
+    int batch_nsamps = min(batch_size, nsamps - (batch_start_offset - args->timestamp));
+    int aligned_nsamps = ceil_mod(batch_nsamps, (512 / 8) / sizeof(cf_t));
+    cf_t samples[aligned_nsamps] __attribute__((aligned(64)));
+    get_noise_vector((float *)samples, batch_nsamps * 2);
+
+    c16_t **input_samples = batch_start_offset == 0 ? args->first_samples : (c16_t **)args->samples;
+    for (int aatx = 0; aatx < nb_tx_ant; aatx++) {
+      for (int i = 0; i < batch_nsamps; i++) {
+        cf_t *impulse_response = channel_impulse_response_p[aatx];
+        cf_t sample = {0, 0};
+        for (int l = 0; l < channel_desc->channel_length; l++) {
+          int idx = i - l;
+          // TODO: Use AVX2 for this
+          c16_t tx_input = input_samples[aatx][batch_start_offset + idx];
+          sample.r += tx_input.r * impulse_response[l].r - tx_input.i * impulse_response[l].i;
+          sample.i += tx_input.i * impulse_response[l].r + tx_input.r * impulse_response[l].i;
+        }
+        samples[i].r += sample.r;
+        samples[i].i += sample.i;
       }
     }
-  }
 
-  // Convert to c16_t
-  c16_t samples_out[aligned_nsamps] __attribute__((aligned(64)));
+    // Convert to c16_t
+    c16_t samples_out[aligned_nsamps] __attribute__((aligned(64)));
 #if defined(__AVX512F__)
-  for (int i = 0; i < aligned_nsamps / 8; i++) {
-    simde__m512 *in = (simde__m512 *)&samples[i * 8];
-    simde__m256i *out = (simde__m256i *)&samples_out[i * 8];
-    *out = simde_mm512_cvtsepi32_epi16(simde_mm512_cvtps_epi32(*in));
-  }
+    for (int i = 0; i < aligned_nsamps / 8; i++) {
+      simde__m512 *in = (simde__m512 *)&samples[i * 8];
+      simde__m256i *out = (simde__m256i *)&samples_out[i * 8];
+      *out = simde_mm512_cvtsepi32_epi16(simde_mm512_cvtps_epi32(*in));
+    }
 #elif defined(__AVX2__)
-  for (int i = 0; i < aligned_nsamps / 4; i++) {
-    simde__m256 *in = (simde__m256 *)&samples[i * 4];
-    simde__m128i *out = (simde__m128i *)&samples_out[i * 4];
-    *out = simde_mm256_cvtsepi32_epi16(simde_mm256_cvtps_epi32(*in));
-  }
+    for (int i = 0; i < aligned_nsamps / 4; i++) {
+      simde__m256 *in = (simde__m256 *)&samples[i * 4];
+      simde__m128i *out = (simde__m128i *)&samples_out[i * 4];
+      *out = simde_mm256_cvtsepi32_epi16(simde_mm256_cvtps_epi32(*in));
+    }
 #else
-  for (int i = 0; i < nsamps; i++) {
-    samples_out[i].r = lroundf(samples[i].r);
-    samples_out[i].i = lroundf(samples[i].i);
-  }
+    for (int i = 0; i < nsamps; i++) {
+      samples_out[i].r = lroundf(samples[i].r);
+      samples_out[i].i = lroundf(samples[i].i);
+    }
 #endif
 
-  vrtsim_write_internal(channel_modelling_args->vrtsim_state,
-                        channel_modelling_args->timestamp,
-                        samples_out,
-                        channel_modelling_args->nsamps,
-                        aarx,
-                        channel_modelling_args->flags,
-                        aarx);
+    vrtsim_write_internal(args->vrtsim_state,
+                          args->timestamp + batch_start_offset,
+                          samples_out,
+                          batch_nsamps,
+                          aarx,
+                          args->flags,
+                          args->tx_timing_index);
+  }
 }
 
 static int vrtsim_write_with_chanmod(vrtsim_state_t *vrtsim_state,
                                      openair0_timestamp timestamp,
-                                     void **samplesVoid,
+                                     c16_t **samplesVoid,
                                      int nsamps,
                                      int nbAnt,
                                      int flags)
 {
-  AssertFatal(nbAnt < MAX_NUM_ANTENNAS_TX, "Number of antennas %d exceeds maximum %d\n", nbAnt, MAX_NUM_ANTENNAS_TX);
+  AssertFatal(nbAnt <= MAX_NUM_ANTENNAS_TX, "Number of antennas %d exceeds maximum %d\n", nbAnt, MAX_NUM_ANTENNAS_TX);
+
+  static int actor = 0;
+  // Sample history for channel impulse response
+  static c16_t saved_samples[MAX_NUM_ANTENNAS_TX][SAVED_SAMPLES_LEN] __attribute__((aligned(32))) = {0};
+  static openair0_timestamp last_timestamp = 0;
+
   for (int aarx = 0; aarx < vrtsim_state->peer_info.num_rx_antennas; aarx++) {
-    notifiedFIFO_elt_t *task = newNotifiedFIFO_elt(sizeof(channel_modelling_args_t), 0, NULL, perform_channel_modelling);
-    channel_modelling_args_t *args = (channel_modelling_args_t *)NotifiedFifoData(task);
-    args->vrtsim_state = vrtsim_state;
-    args->timestamp = timestamp;
-    args->nsamps = nsamps;
-    args->nbAnt = nbAnt;
-    args->flags = flags;
-    args->aarx = aarx;
-    for (int i = 0; i < nbAnt; i++) {
-      args->samples[i] = samplesVoid[i];
+    for (int task_index = 0; task_index < vrtsim_state->tx_num_channels; task_index++) {
+      // Allocate extra space at the end of the message to copy historical samples
+      size_t extra_size = sizeof(c16_t) * (BATCH_SIZE + SAVED_SAMPLES_LEN) * nbAnt;
+      notifiedFIFO_elt_t *task = newNotifiedFIFO_elt(sizeof(channel_modelling_args_t) + extra_size, 0, NULL, perform_channel_modelling);
+      channel_modelling_args_t *args = NotifiedFifoData(task);
+      if (task_index == 0) {
+        for (int aatx = 0; aatx < nbAnt; aatx++) {
+          c16_t *buffer = (c16_t *)(args + 1) + aatx * (FIRST_BATCH_SIZE + SAVED_SAMPLES_LEN);
+          size_t gap_samples = timestamp - last_timestamp;
+          if (gap_samples > 0) {
+            size_t gap_samples_needed = min(SAVED_SAMPLES_LEN, gap_samples);
+            memset(&buffer[SAVED_SAMPLES_LEN - gap_samples_needed], 0, sizeof(c16_t) * gap_samples_needed);
+            if (gap_samples < SAVED_SAMPLES_LEN) {
+              size_t samples_from_saved = SAVED_SAMPLES_LEN - gap_samples_needed;
+              memcpy(&buffer[0], &saved_samples[aatx][SAVED_SAMPLES_LEN - samples_from_saved], sizeof(c16_t) * samples_from_saved);
+            }
+          } else {
+            memcpy(buffer, saved_samples[aatx], sizeof(c16_t) * SAVED_SAMPLES_LEN);
+          }
+          memcpy(&buffer[SAVED_SAMPLES_LEN], &samplesVoid[aatx][0], sizeof(c16_t) * FIRST_BATCH_SIZE);
+          args->first_samples[aatx] = buffer + SAVED_SAMPLES_LEN;
+        }
+      }
+      for (int i = 0; i < nbAnt; i++) {
+        args->samples[i] = samplesVoid[i];
+      }
+      args->vrtsim_state = vrtsim_state;
+      args->timestamp = timestamp;
+      args->nsamps = nsamps;
+      args->nbAnt = nbAnt;
+      args->flags = flags;
+      args->aarx = aarx;
+      args->task_index = task_index;
+      args->num_tasks = vrtsim_state->tx_num_channels;
+      actor = (actor + 1) % vrtsim_state->num_chanmod_actors;
+      args->tx_timing_index = actor;
+      pushNotifiedFIFO(&vrtsim_state->channel_modelling_actors[actor].fifo, task);
     }
-    pushNotifiedFIFO(&vrtsim_state->channel_modelling_actors[aarx].fifo, task);
-  }
-  int start_index = timestamp % MAX_CHANNEL_LENGTH;
-  int end_index = min(start_index + nsamps, MAX_CHANNEL_LENGTH);
-  int cp_nsamps = end_index - start_index;
-  for (int aatx = 0; aatx < nbAnt; aatx++) {
-    c16_t *samples = (c16_t *)samplesVoid[aatx];
-    memcpy(&saved_samples[aatx][start_index], &samples[0], sizeof(c16_t) * cp_nsamps);
   }
 
-  if (end_index < start_index + nsamps) {
-    // wrap around condition, write at beginning of buffer
-    cp_nsamps = nsamps - cp_nsamps; // remaining samples
-    start_index = 0;
+  // Save samples for next round
+  if (nsamps < MAX_CHANNEL_LENGTH) {
     for (int aatx = 0; aatx < nbAnt; aatx++) {
-      c16_t *samples = (c16_t *)samplesVoid[aatx];
-      memcpy(&saved_samples[aatx][start_index], &samples[0], sizeof(c16_t) * cp_nsamps);
+      memmove(&saved_samples[aatx][0], &saved_samples[aatx][nsamps], sizeof(c16_t) * (SAVED_SAMPLES_LEN - nsamps));
+      memcpy(&saved_samples[aatx][SAVED_SAMPLES_LEN - nsamps], &samplesVoid[aatx][0], sizeof(c16_t) * nsamps);
+    }
+  } else {
+    for (int aatx = 0; aatx < nbAnt; aatx++) {
+      memcpy(saved_samples[aatx], &samplesVoid[aatx][nsamps - SAVED_SAMPLES_LEN], sizeof(c16_t) * (SAVED_SAMPLES_LEN));
     }
   }
+  last_timestamp = timestamp + nsamps;
   return nsamps;
 }
 
@@ -515,7 +577,7 @@ static int vrtsim_write(openair0_device *device, openair0_timestamp timestamp, v
   timestamp -= device->openair0_cfg->command_line_sample_advance;
   vrtsim_state_t *vrtsim_state = (vrtsim_state_t *)device->priv;
   bool channel_modelling = vrtsim_state->chanmod || vrtsim_state->taps_socket;
-  return channel_modelling ? vrtsim_write_with_chanmod(vrtsim_state, timestamp, samplesVoid, nsamps, nbAnt, flags)
+  return channel_modelling ? vrtsim_write_with_chanmod(vrtsim_state, timestamp, (c16_t **)samplesVoid, nsamps, nbAnt, flags)
                            : vrtsim_write_internal(vrtsim_state, timestamp, (c16_t *)samplesVoid[0], nsamps, 0, flags, 0);
 }
 
@@ -569,18 +631,18 @@ static void vrtsim_end(openair0_device *device)
 
   tx_timing_t *tx_timing = vrtsim_state->tx_timing;
   if (vrtsim_state->chanmod || vrtsim_state->taps_socket) {
-    for (int i = 0; i < vrtsim_state->peer_info.num_rx_antennas; i++) {
+    for (int i = 0; i < vrtsim_state->num_chanmod_actors; i++) {
       shutdown_actor(&vrtsim_state->channel_modelling_actors[i]);
     }
     free(vrtsim_state->channel_modelling_actors);
-    for (int i = 1; i < vrtsim_state->peer_info.num_rx_antennas; i++) {
+    for (int i = 1; i < vrtsim_state->num_tx_timing; i++) {
       histogram_merge(&tx_timing->tx_histogram, &tx_timing[i].tx_histogram);
       tx_timing->tx_early += tx_timing[i].tx_early;
       tx_timing->tx_samples_late += tx_timing[i].tx_samples_late;
       tx_timing->average_tx_budget += tx_timing[i].average_tx_budget;
       tx_timing->tx_samples_total += tx_timing[i].tx_samples_total;
     }
-    tx_timing->average_tx_budget /= vrtsim_state->peer_info.num_rx_antennas;
+    tx_timing->average_tx_budget /= vrtsim_state->num_tx_timing;
     free_noise_device();
     if (vrtsim_state->taps_socket) {
       taps_client_stop();

targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb.sa.band78.fr1.106PRB.usrpb210.conf

@@ -248,3 +248,21 @@ e2_agent = {
   #sm_dir = "/path/where/the/SMs/are/located/"
   sm_dir = "/usr/local/lib/flexric/"
 };
+
+
+channelmod = { 
+  max_chan=10;
+  modellist="modellist_rfsimu_1";
+  modellist_rfsimu_1 = (
+    {
+        model_name                       = "server_tx_channel_model"
+      	type                             = "AWGN";			  
+      	ploss_dB                         = 0;
+        noise_power_dB                   = 0; 
+        forgetfact                       = 0;  
+        offset                           = 0;      
+        ds_tdl                           = 0;      
+    }    
+  );
+};
+