Merge branch 'next' of github.com:softwareradiosystems/srsLTE into next

CHANGELOG

@@ -1,6 +1,10 @@
 Change Log for Releases
 ==============================
 
+## 18.03.1
+  * Fixed compilation for NEON
+  * Fixed logging and RLC AM issue
+
 ## 18.03
   * Many bug-fixes and improved stability and performance in all parts
 

cmake/modules/SRSLTEVersion.cmake

@@ -20,5 +20,5 @@
 
 SET(SRSLTE_VERSION_MAJOR 18)
 SET(SRSLTE_VERSION_MINOR 3)
-SET(SRSLTE_VERSION_PATCH 0)
+SET(SRSLTE_VERSION_PATCH 1)
 SET(SRSLTE_VERSION_STRING "${SRSLTE_VERSION_MAJOR}.${SRSLTE_VERSION_MINOR}.${SRSLTE_VERSION_PATCH}")

lib/include/srslte/interfaces/ue_interfaces.h

@@ -487,9 +487,11 @@ typedef struct {
   uint32_t cfo_loop_pss_conv;
   uint32_t cfo_ref_mask;
   bool average_subframe_enabled;
-  int time_correct_period; 
+  bool estimator_fil_auto;
+  float estimator_fil_stddev;
+  uint32_t estimator_fil_order;
+  int time_correct_period;
   std::string sss_algorithm;
-  float estimator_fil_w;   
   bool rssi_sensor_enabled;
   bool sic_pss_enabled;
   float rx_gain_offset;

lib/include/srslte/phy/ch_estimation/chest_dl.h

@@ -74,6 +74,7 @@ typedef struct {
   float snr_vector[12000];
   float pilot_power[12000];
 #endif
+  bool smooth_filter_auto;
   uint32_t smooth_filter_len; 
   float smooth_filter[SRSLTE_CHEST_MAX_SMOOTH_FIL_LEN];
 
@@ -123,6 +124,13 @@ SRSLTE_API void srslte_chest_dl_set_smooth_filter(srslte_chest_dl_t *q,
 SRSLTE_API void srslte_chest_dl_set_smooth_filter3_coeff(srslte_chest_dl_t* q, 
                                                          float w); 
 
+SRSLTE_API void srslte_chest_dl_set_smooth_filter_gauss(srslte_chest_dl_t* q,
+                                                        uint32_t order,
+                                                        float std_dev);
+
+SRSLTE_API void srslte_chest_dl_set_smooth_filter_auto(srslte_chest_dl_t* q,
+                                            bool enable);
+
 SRSLTE_API void srslte_chest_dl_set_noise_alg(srslte_chest_dl_t *q, 
                                               srslte_chest_dl_noise_alg_t noise_estimation_alg); 
 

lib/src/phy/ch_estimation/chest_dl.c

@@ -155,6 +155,7 @@ int srslte_chest_dl_init(srslte_chest_dl_t *q, uint32_t max_prb)
 
     q->rsrp_neighbour = false;
 
+    q->smooth_filter_auto = false;
     q->smooth_filter_len = 3; 
     srslte_chest_dl_set_smooth_filter3_coeff(q, 0.1);
     
@@ -263,41 +264,66 @@ int srslte_chest_dl_set_cell(srslte_chest_dl_t *q, srslte_cell_t cell)
 /* Uses the difference between the averaged and non-averaged pilot estimates */
 static float estimate_noise_pilots(srslte_chest_dl_t *q, uint32_t port_id, srslte_sf_t ch_mode)
 {
-  int nref=SRSLTE_REFSIGNAL_NUM_SF(q->cell.nof_prb, port_id);
+  const float weight = 1.0f;
+  float sum_power = 0.0f;
+  uint32_t count = 0;
+  uint32_t npilots = SRSLTE_REFSIGNAL_NUM_SF(q->cell.nof_prb, port_id);
+  uint32_t nsymbols =
+      (ch_mode == SRSLTE_SF_MBSFN) ? srslte_refsignal_mbsfn_nof_symbols() : srslte_refsignal_cs_nof_symbols(port_id);
+  uint32_t nref = npilots / nsymbols;
+  uint32_t fidx = srslte_refsignal_cs_fidx(q->cell, 0, port_id, 0);
 
-  if (q->average_subframe) {
-    if (ch_mode == SRSLTE_SF_MBSFN) {
-      nref /= 4;
-    } else {
-      nref /= 2;
-    }
+  cf_t *input2d[nsymbols + 2];
+  cf_t *tmp_noise = q->tmp_noise;
+
+  for (int i = 0; i < nsymbols; i++) {
+    input2d[i + 1] = &q->pilot_estimates[i * nref];
   }
 
-  /* Substract noisy pilot estimates */
-  srslte_vec_sub_ccc(q->pilot_estimates_average, q->pilot_estimates, q->tmp_noise, nref);  
-  
-#ifdef FREQ_SEL_SNR
-  /* Compute frequency-selective SNR */
-  srslte_vec_abs_square_cf(q->tmp_noise, q->snr_vector, nref);
-  srslte_vec_abs_square_cf(q->pilot_estimates, q->pilot_power, nref);
-  srslte_vec_div_fff(q->pilot_power, q->snr_vector, q->snr_vector, nref);
-  
-  srslte_vec_fprint_f(stdout, q->snr_vector, nref);
-#endif
-  
-  /* Compute average power. Normalized for filter len 3 using matlab */
-  float norm  = 1;
-  if (q->average_subframe) {
-    norm = 32;
+  input2d[0] = &q->tmp_noise[0];
+  if (nsymbols > 3) {
+    srslte_vec_sc_prod_cfc(input2d[2], 2.0f, input2d[0], nref);
+    srslte_vec_sub_ccc(input2d[0], input2d[4], input2d[0], nref);
   } else {
-    if (q->smooth_filter_len == 3) {
-      float a = q->smooth_filter[0];
-      float norm3 = 6.143*a*a+0.04859*a-0.002774;
-      norm /= norm3;
+    srslte_vec_sc_prod_cfc(input2d[2], 1.0f, input2d[0], nref);
+  }
+
+  input2d[nsymbols + 1] = &q->tmp_noise[nref];
+  if (nsymbols > 3) {
+    srslte_vec_sc_prod_cfc(input2d[nsymbols - 1], 2.0f, input2d[nsymbols + 1], nref);
+    srslte_vec_sub_ccc(input2d[nsymbols + 1], input2d[nsymbols - 3], input2d[nsymbols + 1], nref);
+  } else {
+    srslte_vec_sc_prod_cfc(input2d[nsymbols - 1], 1.0f, input2d[nsymbols + 1], nref);
+  }
+
+  for (int i = 1; i < nsymbols + 1; i++) {
+    uint32_t offset = ((fidx < 3) ^ (i & 1)) ? 0 : 1;
+    srslte_vec_sc_prod_cfc(input2d[i], weight, tmp_noise, nref);
+
+    srslte_vec_sum_ccc(&input2d[i - 1][0], &tmp_noise[offset], &tmp_noise[offset], nref - offset);
+    srslte_vec_sum_ccc(&input2d[i - 1][1 - offset], &tmp_noise[0], &tmp_noise[0], nref + offset - 1);
+    if (offset) {
+      tmp_noise[0] += 2.0f * input2d[i - 1][0] - input2d[i - 1][1];
+    } else {
+      tmp_noise[nref - 1] += 2.0f * input2d[i - 1][nref - 2] - input2d[i - 1][nref - 1];
     }
+
+    srslte_vec_sum_ccc(&input2d[i + 1][0], &tmp_noise[offset], &tmp_noise[offset], nref - offset);
+    srslte_vec_sum_ccc(&input2d[i + 1][1 - offset], &tmp_noise[0], &tmp_noise[0], nref + offset - 1);
+    if (offset) {
+      tmp_noise[0] += 2.0f * input2d[i + 1][0] - input2d[i + 1][1];
+    } else {
+      tmp_noise[nref - 1] += 2.0f * input2d[i + 1][nref - 2] - input2d[i + 1][nref - 1];
+    }
+
+    srslte_vec_sc_prod_cfc(tmp_noise, 1.0f / (weight + 4.0f), tmp_noise, nref);
+
+    srslte_vec_sub_ccc(input2d[i], tmp_noise, tmp_noise, nref);
+    sum_power = srslte_vec_avg_power_cf(tmp_noise, nref);
+    count++;
   }
-  float power = norm*srslte_vec_avg_power_cf(q->tmp_noise, nref);
-  return power; 
+
+  return sum_power / (float) count * sqrtf(weight + 4.0f);
 }
 
 static float estimate_noise_pss(srslte_chest_dl_t *q, cf_t *input, cf_t *ce) 
@@ -443,6 +469,53 @@ void srslte_chest_dl_set_smooth_filter3_coeff(srslte_chest_dl_t* q, float w)
   q->smooth_filter[1] = 1-2*w; 
 }
 
+void srslte_chest_dl_set_smooth_filter_gauss(srslte_chest_dl_t* q, uint32_t order, float std_dev)
+{
+  const uint32_t filterlen = order + 1;
+  const int center = (filterlen - 1) / 2;
+  float *filter = q->smooth_filter;
+  float norm_p = 0.0f;
+
+  if (filterlen) {
+
+    for (int i = 0; i < filterlen; i++) {
+      filter[i] = expf(-powf(i - center, 2) / (2.0f * powf(std_dev, 2)));
+      norm_p += powf(filter[i], 2);
+    }
+
+    const float norm = srslte_vec_acc_ff(filter, filterlen);
+
+    srslte_vec_sc_prod_fff(filter, 1.0f / norm, filter, filterlen);
+    q->smooth_filter_len = filterlen;
+  }
+}
+
+void srslte_chest_dl_set_smooth_filter_auto(srslte_chest_dl_t *q, bool enable) {
+  q->smooth_filter_auto = enable;
+}
+
+uint32_t srslte_chest_dl_interleave_pilots(srslte_chest_dl_t *q, cf_t *input, cf_t *tmp, cf_t *output, uint32_t port_id, srslte_sf_t ch_mode) {
+  uint32_t nsymbols = (ch_mode == SRSLTE_SF_MBSFN)?srslte_refsignal_mbsfn_nof_symbols(port_id):srslte_refsignal_cs_nof_symbols(port_id);
+  uint32_t nref = (ch_mode == SRSLTE_SF_MBSFN)?6*q->cell.nof_prb:2*q->cell.nof_prb;
+
+  if (srslte_refsignal_cs_fidx(q->cell, 0, port_id, 0) < 3) {
+    srslte_vec_interleave(input, &input[nref], tmp, nref);
+    for (int l = 2; l < nsymbols - 1; l += 2) {
+      srslte_vec_interleave_add(&input[l * nref], &input[(l + 1) * nref], tmp, nref);
+    }
+  } else {
+    srslte_vec_interleave(&input[nref], input, tmp, nref);
+    for (int l = 2; l < nsymbols - 1; l += 2) {
+      srslte_vec_interleave_add(&input[(l + 1) * nref], &input[l * nref], tmp, nref);
+    }
+  }
+
+  nref *= 2;
+  srslte_vec_sc_prod_cfc(tmp, 2.0f / nsymbols, output, nref);
+
+  return nref;
+}
+
 static void average_pilots(srslte_chest_dl_t *q, cf_t *input, cf_t *output, uint32_t port_id, srslte_sf_t ch_mode)  {
   uint32_t nsymbols = (ch_mode == SRSLTE_SF_MBSFN)?srslte_refsignal_mbsfn_nof_symbols(port_id):srslte_refsignal_cs_nof_symbols(port_id);
   uint32_t nref = (ch_mode == SRSLTE_SF_MBSFN)?6*q->cell.nof_prb:2*q->cell.nof_prb;
@@ -523,7 +596,16 @@ void chest_interpolate_noise_est(srslte_chest_dl_t *q, cf_t *input, cf_t *ce, ui
     q->cfo = chest_estimate_cfo(q);
   }
 
+  /* Estimate noise */
+  if (q->noise_alg == SRSLTE_NOISE_ALG_REFS) {
+    q->noise_estimate[rxant_id][port_id] = estimate_noise_pilots(q, port_id, ch_mode);
+  }
+
   if (ce != NULL) {
+    if (q->smooth_filter_auto) {
+      srslte_chest_dl_set_smooth_filter_gauss(q, 4, q->noise_estimate[rxant_id][port_id] * 200.0f);
+    }
+
     /* Smooth estimates (if applicable) and interpolate */
     if (q->smooth_filter_len == 0 || (q->smooth_filter_len == 3 && q->smooth_filter[0] == 0)) {
       interpolate_pilots(q, q->pilot_estimates, ce, port_id, ch_mode);
@@ -533,13 +615,11 @@ void chest_interpolate_noise_est(srslte_chest_dl_t *q, cf_t *input, cf_t *ce, ui
     }
   
     /* Estimate noise power */
-    if (q->noise_alg == SRSLTE_NOISE_ALG_REFS && q->smooth_filter_len > 0) {
-      q->noise_estimate[rxant_id][port_id] = estimate_noise_pilots(q, port_id, ch_mode);
-    } else if (q->noise_alg == SRSLTE_NOISE_ALG_PSS) {
+    if (q->noise_alg == SRSLTE_NOISE_ALG_PSS) {
       if (sf_idx == 0 || sf_idx == 5) {
         q->noise_estimate[rxant_id][port_id] = estimate_noise_pss(q, input, ce);
       }
-    } else {
+    } else if (q->noise_alg != SRSLTE_NOISE_ALG_REFS) {
       if (sf_idx == 0 || sf_idx == 5) {
         q->noise_estimate[rxant_id][port_id] = estimate_noise_empty_sc(q, input);        
       }

lib/src/phy/ue/ue_sync.c

@@ -41,7 +41,7 @@
 
 #define MAX_TIME_OFFSET 128
 
-#define TRACK_MAX_LOST          4
+#define TRACK_MAX_LOST          100
 #define TRACK_FRAME_SIZE        32
 #define FIND_NOF_AVG_FRAMES     4
 #define DEFAULT_SAMPLE_OFFSET_CORRECT_PERIOD  0

srsue/src/main.cc

@@ -261,6 +261,18 @@ void parse_args(all_args_t *args, int argc, char *argv[]) {
      bpo::value<bool>(&args->expert.phy.average_subframe_enabled)->default_value(true),
      "Averages in the time domain the channel estimates within 1 subframe. Needs accurate CFO correction.")
 
+    ("expert.estimator_fil_auto",
+     bpo::value<bool>(&args->expert.phy.estimator_fil_auto)->default_value(false),
+     "The channel estimator smooths the channel estimate with an adaptative filter.")
+
+    ("expert.estimator_fil_stddev",
+     bpo::value<float>(&args->expert.phy.estimator_fil_stddev)->default_value(1.0f),
+     "Sets the channel estimator smooth gaussian filter standard deviation.")
+
+    ("expert.estimator_fil_order",
+     bpo::value<uint32_t>(&args->expert.phy.estimator_fil_order)->default_value(4),
+     "Sets the channel estimator smooth gaussian filter order (even values perform better).")
+
     ("expert.time_correct_period",
      bpo::value<int>(&args->expert.phy.time_correct_period)->default_value(5),
      "Period for sampling time offset correction.")
@@ -269,10 +281,6 @@ void parse_args(all_args_t *args, int argc, char *argv[]) {
      bpo::value<string>(&args->expert.phy.sss_algorithm)->default_value("full"),
      "Selects the SSS estimation algorithm.")
 
-    ("expert.estimator_fil_w",
-     bpo::value<float>(&args->expert.phy.estimator_fil_w)->default_value(0.1),
-     "Chooses the coefficients for the 3-tap channel estimator centered filter.")
-
     ("expert.pdsch_csi_enabled",
      bpo::value<bool>(&args->expert.phy.pdsch_csi_enabled)->default_value(false),
      "Stores the Channel State Information and uses it for weightening the softbits. It is only compatible with TM1.")

srsue/src/phy/phch_worker.cc

@@ -441,13 +441,11 @@ bool phch_worker::extract_fft_and_pdcch_llr() {
   if (phy->get_pending_ack(tti) || decode_pdcch) {
     
     // Setup estimator filter 
-    float w_coeff = phy->args->estimator_fil_w; 
-    if (w_coeff > 0.0) {
-      srslte_chest_dl_set_smooth_filter3_coeff(&ue_dl.chest, w_coeff); 
-    } else if (w_coeff == 0.0) {
-      srslte_chest_dl_set_smooth_filter(&ue_dl.chest, NULL, 0); 
-    }
-    
+    srslte_chest_dl_set_smooth_filter_gauss(&ue_dl.chest,
+                                            phy->args->estimator_fil_order,
+                                            phy->args->estimator_fil_stddev);
+    srslte_chest_dl_set_smooth_filter_auto(&ue_dl.chest, phy->args->estimator_fil_auto);
+
     if (!phy->args->snr_estim_alg.compare("refs")) {
       srslte_chest_dl_set_noise_alg(&ue_dl.chest, SRSLTE_NOISE_ALG_REFS);
     } else if (!phy->args->snr_estim_alg.compare("empty")) {

srsue/src/phy/phy.cc

@@ -98,7 +98,9 @@ void phy::set_default_args(phy_args_t *args)
   args->cfo_correct_tol_hz  = 50; 
   args->time_correct_period = 5; 
   args->sss_algorithm       = "full";
-  args->estimator_fil_w     = 0.1; 
+  args->estimator_fil_auto  = false;
+  args->estimator_fil_stddev = 1.0f;
+  args->estimator_fil_order  = 4;
 }
 
 bool phy::check_args(phy_args_t *args) 
@@ -107,10 +109,6 @@ bool phy::check_args(phy_args_t *args)
     log_h->console("Error in PHY args: nof_phy_threads must be 1, 2 or 3\n");
     return false; 
   }
-  if (args->estimator_fil_w > 1.0) {
-    log_h->console("Error in PHY args: estimator_fil_w must be 0<=w<=1\n");
-    return false; 
-  }
   if (args->snr_ema_coeff > 1.0) {
     log_h->console("Error in PHY args: snr_ema_coeff must be 0<=w<=1\n");
     return false; 

srsue/ue.conf.example

@@ -150,7 +150,9 @@ enable = false
 #                       sampling frequency offset. Default is enabled. 
 # sss_algorithm:        Selects the SSS estimation algorithm. Can choose between 
 #                       {full, partial, diff}. 
-# estimator_fil_w:      Chooses the coefficients for the 3-tap channel estimator centered filter. 
+# estimator_fil_auto:   The channel estimator smooths the channel estimate with an adaptative filter.
+# estimator_fil_stddev: Sets the channel estimator smooth gaussian filter standard deviation.
+# estimator_fil_order:  Sets the channel estimator smooth gaussian filter order (even values perform better).
 #                       The taps are [w, 1-2w, w]
 # metrics_period_secs:  Sets the period at which metrics are requested from the UE. 
 #
@@ -204,7 +206,9 @@ enable = false
 #time_correct_period = 5
 #sfo_correct_disable = false
 #sss_algorithm       = full
-#estimator_fil_w     = 0.1
+#estimator_fil_auto  = false
+#estimator_fil_stddev  = 1.0
+#estimator_fil_order  = 4
 #average_subframe_enabled = true
 #sic_pss_enabled     = true
 #pregenerate_signals = false