Merge remote-tracking branch 'origin/dmrs_channel_estimation_parallelization' into integration_2024_w47 (!2972)

Parallelize PUSCH channel estimation

This MR adds PUSCH channel estimation parallelization which reduces the
processing time. It also adds time measurements for each antenna and makes
num_antennas_per_thread configurable in nr_ulsim and the gNB config file.

Some results:

- **273 PRBs 4 antennas** `./nr_ulsim -n200 -m25 -s40 -S40 -q1 -R273 -r273 -z4 -P`

| Number of threads | ULSCH channel estimation time (us) develop| ULSCH channel estimation time (us)| Antenna Processing time (us) |
| ------            | ------                                    | ------                            | ------ |
| 0                 | 194.06                                    |193.17                             |  134.82|
| 2                 | 194.88                                    |154.02                             |  93.43 |
| 4                 | 195.23                                    |134.27                             |   72.71|
- **273 PRBs 8 antennas** `./nr_ulsim -n200 -m25 -s40 -S40 -q1 -R273 -r273 -z8 -P`

| Number of threads | ULSCH channel estimation time (us) develop| ULSCH channel estimation time (us)| Antenna Processing time (us) |
| ------            | ------                                    | ------                            | ------ |
| 0                 | 387.93                                    |384.99                             |  269.39|
| 2                 | 388.10                                    |279.47                             |  161.32|
| 4                 | 387.47                                    |225.37                             |  106.52|
| 8                 | 390.53                                    |204.48                             |   83.74|

- **106 PRBs MIMO 4 layers** `./nr_ulsim -n100 -m9 -r106 -s10 -W4 -y4 -z4 -P`

| Number of threads | ULSCH channel estimation time (us) develop| ULSCH channel estimation time (us)| Antenna Processing time (us) |
| ------            | ------                                    | ------                            | ------ |
| 0                 | 263.20                                    |268.97                             |  57.49 |
| 2                 | 263.90                                    |252.61                             |  51.22 |
| 4                 | 264.17                                    |192.49                             |  36.29 |

openair1/PHY/defs_gNB.h

@@ -543,6 +543,7 @@ typedef struct PHY_VARS_gNB_s {
   time_stats_t ulsch_decoding_stats;
   time_stats_t ulsch_deinterleaving_stats;
   time_stats_t ulsch_channel_estimation_stats;
+  time_stats_t pusch_channel_estimation_antenna_processing_stats;
   time_stats_t ulsch_llr_stats;
   time_stats_t rx_srs_stats;
   time_stats_t generate_srs_stats;
@@ -562,7 +563,10 @@ typedef struct PHY_VARS_gNB_s {
   notifiedFIFO_t L1_rx_out;
   notifiedFIFO_t resp_RU_tx;
   tpool_t threadPool;
+  int nbSymb;
+  int nbAarx;
   int num_pusch_symbols_per_thread;
+  int dmrs_num_antennas_per_thread;
   pthread_t L1_rx_thread;
   int L1_rx_thread_core;
   pthread_t L1_tx_thread;
@@ -585,6 +589,37 @@ union puschSymbolReqUnion {
   uint64_t p;
 };
 
+typedef struct puschAntennaProc_s {
+  unsigned char Ns;
+  int nl;
+  unsigned short p;
+  unsigned char symbol;
+  unsigned short bwp_start_subcarrier;
+  int aarx;
+  int beam_nb;
+  int numAntennas;
+  nfapi_nr_pusch_pdu_t *pusch_pdu;
+  int *max_ch;
+  c16_t *pilot;
+  int *nest_count;
+  uint64_t *noise_amp2;
+  delay_t *delay;
+  int chest_freq;
+  NR_gNB_PUSCH *pusch_vars;
+  NR_DL_FRAME_PARMS *frame_parms;
+  c16_t ***rxdataF;
+} puschAntennaProc_t;
+
+struct puschAntennaReqId {
+  uint16_t ul_id;
+  uint16_t spare;
+} __attribute__((packed));
+
+union puschAntennaReqUnion {
+  struct puschAntennaReqId s;
+  uint64_t p;
+};
+
 typedef struct LDPCDecode_s {
   PHY_VARS_gNB *gNB;
   NR_UL_gNB_HARQ_t *ulsch_harq;

openair1/SIMULATION/NR_PHY/ulsim.c

@@ -213,6 +213,7 @@ int main(int argc, char *argv[])
   int params_from_file = 0;
   int threadCnt=0;
   int max_ldpc_iterations = 5;
+  int num_antennas_per_thread = 1;
   if ((uniqCfg = load_configmodule(argc, argv, CONFIG_ENABLECMDLINEONLY)) == 0) {
     exit_fun("[NR_ULSIM] Error, configuration module init failed\n");
   }
@@ -224,7 +225,7 @@ int main(int argc, char *argv[])
   InitSinLUT();
 
   int c;
-  while ((c = getopt(argc, argv, "--:O:a:b:c:d:ef:g:h:i:k:m:n:op:q:r:s:t:u:v:w:y:z:C:F:G:H:I:M:N:PR:S:T:U:L:ZW:E:X:")) != -1) {
+  while ((c = getopt(argc, argv, "--:O:a:b:c:d:ef:g:h:i:k:m:n:op:q:r:s:t:u:v:w:y:z:A:C:F:G:H:I:M:N:PR:S:T:U:L:ZW:E:X:")) != -1) {
 
     /* ignore long options starting with '--', option '-O' and their arguments that are handled by configmodule */
     /* with this opstring getopt returns 1 for non-option arguments, refer to 'man 3 getopt' */
@@ -395,6 +396,10 @@ int main(int argc, char *argv[])
       }
       break;
 
+    case 'A':
+      num_antennas_per_thread = atoi(optarg);
+      break;
+
     case 'F':
       input_fd = fopen(optarg, "r");
       if (input_fd == NULL) {
@@ -512,6 +517,7 @@ int main(int argc, char *argv[])
       printf("-w Start PRB for PUSCH\n");
       printf("-y Number of TX antennas used at UE\n");
       printf("-z Number of RX antennas used at gNB\n");
+      printf("-A Number of antennas per thread for PUSCH channel estimation\n");
       printf("-C Specify the number of threads for the simulation\n");
       printf("-E {SRS: [0] Disabled, [1] Enabled} e.g. -E 1\n");
       printf("-F Input filename (.txt format) for RX conformance testing\n");
@@ -559,6 +565,7 @@ int main(int argc, char *argv[])
   gNB = RC.gNB[0];
   gNB->ofdm_offset_divisor = UINT_MAX;
   gNB->num_pusch_symbols_per_thread = 1;
+  gNB->dmrs_num_antennas_per_thread = num_antennas_per_thread;
   gNB->RU_list[0] = calloc(1, sizeof(**gNB->RU_list));
   gNB->RU_list[0]->rfdevice.openair0_cfg = openair0_cfg;
 
@@ -952,6 +959,7 @@ int main(int argc, char *argv[])
     reset_meas(&gNB->rx_pusch_symbol_processing_stats);
     reset_meas(&gNB->ulsch_decoding_stats);
     reset_meas(&gNB->ulsch_channel_estimation_stats);
+    reset_meas(&gNB->pusch_channel_estimation_antenna_processing_stats);
     reset_meas(&gNB->rx_srs_stats);
     reset_meas(&gNB->generate_srs_stats);
     reset_meas(&gNB->get_srs_signal_stats);
@@ -1512,6 +1520,7 @@ int main(int argc, char *argv[])
       printDistribution(&gNB->phy_proc_rx,table_rx, "Total PHY proc rx");
       printStatIndent(&gNB->rx_pusch_stats, "RX PUSCH time");
       printStatIndent2(&gNB->ulsch_channel_estimation_stats, "ULSCH channel estimation time");
+      printStatIndent3(&gNB->pusch_channel_estimation_antenna_processing_stats, "Antenna Processing time");
       printStatIndent2(&gNB->rx_pusch_init_stats, "RX PUSCH Initialization time");
       printStatIndent2(&gNB->rx_pusch_symbol_processing_stats, "RX PUSCH Symbol Processing time");
       printStatIndent(&gNB->ulsch_decoding_stats,"ULSCH total decoding time");

openair2/GNB_APP/L1_nr_paramdef.h

@@ -59,6 +59,8 @@
 #define HLP_L1TX_BO "Backoff from full-scale output at the L1 entity(frequency domain), ex. 12 would corresponding to 14-bit input level (6 dB/bit). Default 36 dBFS for OAI RU entity"
 #define CONFIG_STRING_L1_PHASE_COMP                        "phase_compensation"
 #define HLP_L1_PHASE_COMP "Apply NR symbolwise phase rotation"
+#define CONFIG_STRING_NUM_ANTENNAS_PER_THREAD              "dmrs_num_antennas_per_thread"
+#define HLP_NUM_ARX "Number of antennas per thread for PUSCH channel estimation"
 /*----------------------------------------------------------------------------------------------------------------------------------------------------*/
 /*                                            L1 configuration parameters                                                                             */
 /*   optname                                         helpstr   paramflags    XXXptr              defXXXval                  type           numelt     */
@@ -84,6 +86,7 @@
   {CONFIG_STRING_L1_TX_THREAD_CORE,                    NULL,       0,         .uptr=NULL,           .defintval=-1,              TYPE_UINT,     0},          \
   {CONFIG_STRING_L1_TX_AMP_BACKOFF_dB,                 HLP_L1TX_BO,0,         .uptr=NULL,           .defintval=36,              TYPE_UINT,     0},         \
   {CONFIG_STRING_L1_PHASE_COMP,                        HLP_L1_PHASE_COMP,PARAMFLAG_BOOL, .uptr=NULL,.defintval=1,               TYPE_UINT,     0},         \
+  {CONFIG_STRING_NUM_ANTENNAS_PER_THREAD,              HLP_NUM_ARX,0,         .uptr=NULL,           .defintval=1,               TYPE_UINT,     0},         \
 }
 // clang-format on
 #define L1_CC_IDX                                          0
@@ -105,6 +108,7 @@
 #define L1_TX_THREAD_CORE                                  16
 #define L1_TX_AMP_BACKOFF_dB                               17
 #define L1_PHASE_COMP                                      18
+#define NUM_ANTENNAS_PER_THREAD                            19
 
 /*----------------------------------------------------------------------------------------------------------------------------------------------------*/
 #endif

openair2/GNB_APP/gnb_config.c

@@ -914,6 +914,7 @@ void RCconfig_NR_L1(void)
       LOG_I(NR_PHY, "L1_RX_THREAD_CORE %d (%d)\n", *(L1_ParamList.paramarray[j][L1_RX_THREAD_CORE].iptr), L1_RX_THREAD_CORE);
       gNB->TX_AMP = (int16_t)(32767.0 / pow(10.0, .05 * (double)(*L1_ParamList.paramarray[j][L1_TX_AMP_BACKOFF_dB].uptr)));
       gNB->phase_comp = *L1_ParamList.paramarray[j][L1_PHASE_COMP].uptr;
+      gNB->dmrs_num_antennas_per_thread = *(L1_ParamList.paramarray[j][NUM_ANTENNAS_PER_THREAD].uptr);
       LOG_I(NR_PHY, "TX_AMP = %d (-%d dBFS)\n", gNB->TX_AMP, *L1_ParamList.paramarray[j][L1_TX_AMP_BACKOFF_dB].uptr);
       AssertFatal(gNB->TX_AMP > 300, "TX_AMP is too small, must be larger than 300 (is %d)\n", gNB->TX_AMP);
       // Midhaul configuration