diff --git a/matlab/tests/viterbi_bler.m b/matlab/tests/viterbi_bler.m
index 2c27d9ee5..c42877bc6 100644
--- a/matlab/tests/viterbi_bler.m
+++ b/matlab/tests/viterbi_bler.m
@@ -1,10 +1,10 @@
 
 clear
 blen=40;
-SNR_values_db=linspace(-6,4,8);
-Nrealizations=5000;
+SNR_values_db=linspace(-3,4,8);
+Nrealizations=6000;
 
-addpath('../../debug/lte/phy/lib/fec/test')
+addpath('../../build/srslte/lib/fec/test')
 
 errors1=zeros(1,length(SNR_values_db));
 errors2=zeros(1,length(SNR_values_db));
@@ -28,7 +28,7 @@ for snr_idx=1:length(SNR_values_db)
         
         decodedData = lteConvolutionalDecode(noisysymbols);
         interleavedSymbols = reshape(reshape(noisysymbols,[],3)',1,[]);
-        [decodedData2, quant] = srslte_viterbi(interleavedSymbols);
+        [decodedData2, quant] = srslte_viterbi(interleavedSymbols, 32);
         
         errors1(snr_idx) = errors1(snr_idx) + any(decodedData ~= Data);
         errors2(snr_idx) = errors2(snr_idx) + any(decodedData2 ~= Data);
diff --git a/srslte/lib/fec/src/viterbi.c b/srslte/lib/fec/src/viterbi.c
index fef653df2..b42400547 100644
--- a/srslte/lib/fec/src/viterbi.c
+++ b/srslte/lib/fec/src/viterbi.c
@@ -80,7 +80,6 @@ int decode37(void *o, uint8_t *symbols, uint8_t *data, uint32_t frame_length) {
 #ifdef LV_HAVE_SSE
 int decode37_sse(void *o, uint8_t *symbols, uint8_t *data, uint32_t frame_length) {
   srslte_viterbi_t *q = o;
-  uint32_t i;
 
   uint32_t best_state;
 
@@ -91,26 +90,22 @@ int decode37_sse(void *o, uint8_t *symbols, uint8_t *data, uint32_t frame_length
   }
 
   /* Initialize Viterbi decoder */
-  init_viterbi37_sse(q->ptr, q->tail_biting ? -1 : 0);
+  init_viterbi37_sse(q->ptr, -1);
 
   /* Decode block */
-  uint8_t *tmp = q->tmp;
   if (q->tail_biting) {
-    memcpy(tmp, symbols, 3 * frame_length * sizeof(uint8_t));
-    for (i = 0; i < 3 * (q->K - 1); i++) {
-      q->tmp[i + 3 * frame_length] = q->tmp[i];
-    }
+    memcpy(q->tmp,                  symbols, 3*frame_length*sizeof(uint8_t));
+    memcpy(&q->tmp[3*frame_length], symbols, 3*frame_length*sizeof(uint8_t));    
+    memcpy(&q->tmp[6*frame_length], symbols, 3*frame_length*sizeof(uint8_t));    
+    update_viterbi37_blk_sse(q->ptr, q->tmp, 3*frame_length, &best_state);
+    chainback_viterbi37_sse(q->ptr,  q->tmp, 3*frame_length, best_state);
+    memcpy(data, &q->tmp[frame_length], frame_length*sizeof(uint8_t));
   } else {
-    tmp = symbols;
+    update_viterbi37_blk_sse(q->ptr, symbols, frame_length+q->K-1, &best_state);
+    chainback_viterbi37_sse(q->ptr, data, frame_length, best_state);
   }
 
-  update_viterbi37_blk_sse(q->ptr, tmp, frame_length + q->K - 1,
-      q->tail_biting ? &best_state : NULL);
-
-  /* Do Viterbi chainback */
-  chainback_viterbi37_sse(q->ptr, data, frame_length,
-      q->tail_biting ? best_state : 0);
-
+  
   return q->framebits;
 }
 
@@ -154,6 +149,7 @@ int init37(srslte_viterbi_t *q, uint32_t poly[3], uint32_t framebits, bool tail_
   }
   if (q->tail_biting) {
     q->tmp = srslte_vec_malloc(3 * (q->framebits + q->K - 1) * sizeof(uint8_t));
+    bzero(q->tmp, 3 * (q->framebits + q->K - 1) * sizeof(uint8_t));
     if (!q->tmp) {
       perror("malloc");
       free37(q);
@@ -188,7 +184,7 @@ int init37_sse(srslte_viterbi_t *q, uint32_t poly[3], uint32_t framebits, bool t
     return -1;
   }
   if (q->tail_biting) {
-    q->tmp = srslte_vec_malloc(3 * (q->framebits + q->K - 1) * sizeof(uint8_t));
+    q->tmp = srslte_vec_malloc(10 * (q->framebits + q->K - 1) * sizeof(uint8_t));
     if (!q->tmp) {
       perror("malloc");
       free37(q);
@@ -198,7 +194,7 @@ int init37_sse(srslte_viterbi_t *q, uint32_t poly[3], uint32_t framebits, bool t
     q->tmp = NULL;
   }
   
-  if ((q->ptr = create_viterbi37_sse(poly, framebits)) == NULL) {
+  if ((q->ptr = create_viterbi37_sse(poly, 3*framebits)) == NULL) {
     fprintf(stderr, "create_viterbi37 failed\n");
     free37(q);
     return -1;
diff --git a/srslte/lib/fec/src/viterbi37_sse.c b/srslte/lib/fec/src/viterbi37_sse.c
index 39c4bf5e2..ffafd693d 100644
--- a/srslte/lib/fec/src/viterbi37_sse.c
+++ b/srslte/lib/fec/src/viterbi37_sse.c
@@ -111,21 +111,19 @@ int chainback_viterbi37_sse(
    */
   endstate %= 64;
   endstate <<= 2;
-
+  
   /* The store into data[] only needs to be done every 8 bits.
    * But this avoids a conditional branch, and the writes will
    * combine in the cache anyway
    */
   d += 6; /* Look past tail */
-  while(nbits-- != 0){
+  while(nbits--) {
     int k;
 
     k = (d[nbits].c[(endstate>>2)/8] >> ((endstate>>2)%8)) & 1;
     endstate = (endstate >> 1) | (k << 7);
     data[nbits] = k;
-#ifdef DEBUG
-//    printf("endstate=%3d, k=%d, w[0]=%d, w[1]=%d\n", endstate, k, d[nbits].s[1]&1, d[nbits].s[2]&1);
-#endif
+    //printf("nbits=%d, endstate=%3d, k=%d, w[0]=%d, w[1]=%d, c=%d\n", nbits, endstate, k, d[nbits].s[1]&1, d[nbits].s[2]&1, d[nbits].c[(endstate>>2)/8]&1);
   }
   return 0;
 }
@@ -163,17 +161,24 @@ void update_viterbi37_blk_sse(void *p,unsigned char *syms,int nbits, uint32_t *b
 #endif
   
   d = (decision_t *) vp->dp;
+  
+  for (int s=0;s<nbits;s++) {
+    memset(d+s,0,sizeof(decision_t));
+  }
+  
   while(nbits--) {
     __m128i sym0v,sym1v,sym2v;
     void *tmp;
     int i;
+
+   // printf("nbits=%d, syms=%d,%d,%d\n", nbits, syms[0], syms[1], syms[2]);fflush(stdout);
     
     /* Splat the 0th symbol across sym0v, the 1st symbol across sym1v, etc */
     sym0v = _mm_set1_epi8(syms[0]);
     sym1v = _mm_set1_epi8(syms[1]);
     sym2v = _mm_set1_epi8(syms[2]);
     syms += 3;
-
+    
     for(i=0;i<2;i++){
       __m128i decision0,decision1,metric,m_metric,m0,m1,m2,m3,survivor0,survivor1;
 
@@ -185,7 +190,6 @@ void update_viterbi37_blk_sse(void *p,unsigned char *syms,int nbits, uint32_t *b
       print_128i("metric_initial", metric);
 #endif
       /* There's no packed bytes right shift in SSE2, so we use the word version and mask
-       * (I'm *really* starting to like Altivec...)
        */
       metric = _mm_srli_epi16(metric,3);
       metric = _mm_and_si128(metric,_mm_set1_epi8(31));
@@ -209,7 +213,7 @@ void update_viterbi37_blk_sse(void *p,unsigned char *syms,int nbits, uint32_t *b
       survivor1 = _mm_or_si128(_mm_and_si128(decision1,m3),_mm_andnot_si128(decision1,m2));
  
       /* Pack each set of decisions into 16 bits */
-      d->s[2*i] = _mm_movemask_epi8(_mm_unpacklo_epi8(decision0,decision1));
+      d->s[2*i]   = _mm_movemask_epi8(_mm_unpacklo_epi8(decision0,decision1));
       d->s[2*i+1] = _mm_movemask_epi8(_mm_unpackhi_epi8(decision0,decision1));
 
       /* Store surviving metrics */
@@ -218,22 +222,6 @@ void update_viterbi37_blk_sse(void *p,unsigned char *syms,int nbits, uint32_t *b
         
     }
 
-
-#ifdef DEBUG      
-      uint8_t wmin=UINT8_MAX;
-      int minstate = 0; 
-      printf("[%d]: ", nbits);
-      for (int j=0;j<64;j++) {
-        printf("%d, ", vp->new_metrics->c[j]);
-        if (vp->new_metrics->c[j] <= wmin) {
-          wmin = vp->new_metrics->c[j];
-          minstate = j; 
-        }
-      }
-      printf("\n");
-      printf("%3d, ",minstate);
-#endif
-
     // See if we need to normalize
     if (vp->new_metrics->c[0] > 100) {
       int i;
diff --git a/srslte/lib/fec/test/viterbi_test.c b/srslte/lib/fec/test/viterbi_test.c
index 463de43bc..f619b50b9 100644
--- a/srslte/lib/fec/test/viterbi_test.c
+++ b/srslte/lib/fec/test/viterbi_test.c
@@ -212,10 +212,13 @@ int main(int argc, char **argv) {
       struct timeval t[3];
       gettimeofday(&t[1], NULL);
       int M = 1; 
+      
+      //srslte_vec_fprint_b(stdout, data_tx, frame_length);
+      
       for (int i=0;i<M;i++) {
         srslte_viterbi_decode_uc(&dec, llr_c, data_rx, frame_length);      
       }
-      
+            
 #ifdef TEST_SSE
       gettimeofday(&t[2], NULL);
       get_time_interval(t);
diff --git a/srslte/lib/phch/src/pdcch.c b/srslte/lib/phch/src/pdcch.c
index 40f817c9e..19256eb14 100644
--- a/srslte/lib/phch/src/pdcch.c
+++ b/srslte/lib/phch/src/pdcch.c
@@ -276,9 +276,6 @@ static int dci_decode(srslte_pdcch_t *q, float *e, uint8_t *data, uint32_t E, ui
     /* unrate matching */
     srslte_rm_conv_rx(e, E, q->rm_f, 3 * (nof_bits + 16));
     
-    /* Normalize LLR */
-    srslte_vec_sc_prod_fff(q->rm_f, (float) 3 * nof_bits/E, q->rm_f, 3*(nof_bits+16));
-  
     /* viterbi decoder */
     srslte_viterbi_decode_f(&q->decoder, q->rm_f, data, nof_bits + 16);
 
diff --git a/srslte/lib/ue/src/ue_dl.c b/srslte/lib/ue/src/ue_dl.c
index 8ed19799e..265344d61 100644
--- a/srslte/lib/ue/src/ue_dl.c
+++ b/srslte/lib/ue/src/ue_dl.c
@@ -406,8 +406,7 @@ int srslte_ue_dl_decode_rnti_rv(srslte_ue_dl_t *q, cf_t *input, uint8_t *data, u
     return SRSLTE_ERROR;
   }
 
-  int found_dci = srslte_ue_dl_find_dl_dci(q, &dci_msg, q->cfi, sf_idx, rnti); 
-  
+  int found_dci = srslte_ue_dl_find_dl_dci(q, &dci_msg, q->cfi, sf_idx, rnti);   
   if (found_dci == 1) {
     
     if (srslte_dci_msg_to_dl_grant(&dci_msg, rnti, q->cell.nof_prb, &dci_unpacked, &grant)) {