maki/openairinterface5g
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix array sizes and aligment issues in 5G UE dci decoder

Browse Source
This commit is contained in:
Laurent THOMAS
2025-07-25 15:49:35 +02:00
parent 343b983aa9
commit f835705dd1
1 changed files with 35 additions and 77 deletions
Download Patch File Download Diff File Expand all files Collapse all files
openair1/PHY/NR_UE_TRANSPORT/dci_nr.c
+35 -77
View File
@@ -184,26 +184,9 @@ static void nr_pdcch_demapping_deinterleaving(c16_t *llr,
}
}
static void nr_pdcch_llr(NR_DL_FRAME_PARMS *frame_parms,
uint32_t coreset_nbr_rb,
c16_t rxdataF_comp[coreset_nbr_rb * RE_PER_RB],
c16_t *pdcch_llr,
uint8_t symbol)
static void nr_pdcch_llr(uint32_t sz, c16_t *rxF, c16_t *llr)
{
c16_t *rxF = rxdataF_comp;
c16_t *llr = pdcch_llr + symbol * coreset_nbr_rb * RE_PER_RB_OUT_DMRS;
if (!llr) {
LOG_E(NR_PHY_DCI, "pdcch_qpsk_llr: llr is null, symbol %d\n", symbol);
return;
}
LOG_DDD("llr logs: pdcch qpsk llr for symbol %d (pos %d), llr offset %ld\n",
symbol,
symbol * frame_parms->N_RB_DL * RE_PER_RB,
llr - pdcch_llr);
for (int i = 0; i < coreset_nbr_rb * RE_PER_RB_OUT_DMRS; i++) {
for (int i = 0; i < sz; i++) {
// We clip the signal
c16_t res;
res.r = min(rxF->r, 31);
@@ -217,16 +200,14 @@ static void nr_pdcch_llr(NR_DL_FRAME_PARMS *frame_parms,
}
//compute average channel_level on each (TX,RX) antenna pair
static int32_t nr_pdcch_channel_level(int nb_rb,
c16_t dl_ch_estimates_ext[][nb_rb * RE_PER_RB_OUT_DMRS],
int nb_antennas_rx,
int symbol)
static int32_t nr_pdcch_channel_level(int nb_re, int arraySz, c16_t dl_ch_estimates_ext[][arraySz], int nb_antennas_rx)
{
int32_t avgs = 0;
for (int aarx = 0; aarx < nb_antennas_rx; aarx++) {
simde__m128i *dl_ch128 = (simde__m128i *)&dl_ch_estimates_ext[aarx];
//compute average level
int32_t tmp = simde_mm_average(dl_ch128, nb_rb * RE_PER_RB_OUT_DMRS, 0, nb_rb * RE_PER_RB_OUT_DMRS);
// compute average level
// if nb_re is not %4, some samples are missed inside simde_mm_average, so let's discard it
int32_t tmp = simde_mm_average(dl_ch128, nb_re & ~3, 0, nb_re & ~3);
avgs = cmax(avgs, tmp);
}
return avgs;
@@ -238,8 +219,9 @@ static void nr_pdcch_extract_rbs_single(uint32_t rxdataF_sz,
c16_t rxdataF[][rxdataF_sz],
int32_t est_size,
c16_t dl_ch_estimates[][est_size],
c16_t rxdataF_ext[][coreset_nbr_rb * RE_PER_RB_OUT_DMRS],
c16_t dl_ch_estimates_ext[][coreset_nbr_rb * RE_PER_RB_OUT_DMRS],
int arraySz,
c16_t rxdataF_ext[][arraySz],
c16_t dl_ch_estimates_ext[][arraySz],
int symbol,
NR_DL_FRAME_PARMS *frame_parms,
uint8_t *coreset_freq_dom,
@@ -296,25 +278,23 @@ static void nr_pdcch_extract_rbs_single(uint32_t rxdataF_sz,
if ((c_rb + n_BWP_start) < frame_parms->N_RB_DL / 2)
// if RB to be treated is lower than middle system bandwidth then rxdataF pointed
// at (offset + c_br + symbol * ofdm_symbol_size): even case
rxF = rxFbase + (frame_parms->first_carrier_offset + RE_PER_RB * c_rb) + n_BWP_start * RE_PER_RB;
rxF = rxFbase + frame_parms->first_carrier_offset + RE_PER_RB * (c_rb + n_BWP_start);
else
// number of RBs is even and c_rb is higher than half system bandwidth (we don't skip DC)
// if these conditions are true the pointer has to be situated at the 1st part of the rxdataF
// we point at the 1st part of the rxdataF in symbol
rxF = rxFbase + RE_PER_RB * (c_rb + n_BWP_start - frame_parms->N_RB_DL / 2);
} else {
if ((c_rb + n_BWP_start) < frame_parms->N_RB_DL / 2)
if ((c_rb + n_BWP_start) <= frame_parms->N_RB_DL / 2)
// if RB to be treated is lower than middle system bandwidth then rxdataF pointed
// at (offset + c_br + symbol * ofdm_symbol_size): odd case
rxF = rxFbase + frame_parms->first_carrier_offset + RE_PER_RB * (c_rb + n_BWP_start);
else if ((c_rb + n_BWP_start) > frame_parms->N_RB_DL / 2)
// number of RBs is odd and c_rb is higher than half system bandwidth + 1
else
// number of RBs is odd and c_rb is higher than half system bandwidth + 1
// if these conditions are true the pointer has to be situated at the 1st part of
// the rxdataF just after the first IQ symbols of the RB containing DC
// we point at the 1st part of the rxdataF in symbol
rxF = rxFbase + RE_PER_RB * (c_rb + n_BWP_start - frame_parms->N_RB_DL / 2) - 6;
else
rxF = rxFbase + frame_parms->first_carrier_offset + RE_PER_RB * (c_rb + n_BWP_start);
}
const int valid_re[RE_PER_RB_OUT_DMRS] = {0, 2, 3, 4, 6, 7, 8, 10, 11};
for (int i = 0; i < sizeofArray(valid_re); i++) {
@@ -326,50 +306,31 @@ static void nr_pdcch_extract_rbs_single(uint32_t rxdataF_sz,
}
}
static void nr_pdcch_channel_compensation(uint32_t coreset_nbr_rb,
c16_t rxdataF_ext[][coreset_nbr_rb * RE_PER_RB_OUT_DMRS],
c16_t dl_ch_estimates_ext[][coreset_nbr_rb * RE_PER_RB_OUT_DMRS],
c16_t rxdataF_comp[][coreset_nbr_rb * RE_PER_RB],
static void nr_pdcch_channel_compensation(int arraySz,
c16_t rxdataF_ext[][arraySz],
c16_t dl_ch_estimates_ext[][arraySz],
c16_t rxdataF_comp[][arraySz],
int antRx,
uint8_t output_shift)
{
for (int aarx = 0; aarx < antRx; aarx++) {
const int sz = coreset_nbr_rb * RE_PER_RB_OUT_DMRS;
c16_t *dl_ch = dl_ch_estimates_ext[aarx];
c16_t *rxdataF = rxdataF_ext[aarx];
c16_t *rxdataF_c = rxdataF_comp[aarx];
// multiply by conjugated channel
mult_cpx_conj_vector(dl_ch, rxdataF, rxdataF_c, sz, output_shift);
for (int rb = 0; rb < sz; rb++)
LOG_DDD("rxdataF[%d]=(%d,%d) X dlch[%d]=(%d,%d) rxdataF_comp[%d]=(%d,%d)\n",
rb,
rxdataF[rb].r,
rxdataF[rb].i,
rb,
dl_ch[rb].r,
dl_ch[rb].i,
rb,
rxdataF_c[rb].r,
rxdataF_c[rb].i);
// multiply by conjugated channel, this function require size in _m128i, else it doesn't process all samples
mult_cpx_conj_vector(dl_ch_estimates_ext[aarx], rxdataF_ext[aarx], rxdataF_comp[aarx], arraySz, output_shift);
}
}
static void nr_pdcch_detection_mrc(int n_rb, c16_t rxdataF_comp[][n_rb * RE_PER_RB_OUT_DMRS])
static void nr_pdcch_detection_mrc(int sz, c16_t rxdataF_comp[][sz])
{
LOG_D(NR_PHY_DCI, "we enter nr_pdcch_detection_mrc (hard coded 2 antennas)\n");
const int sz = n_rb * RE_PER_RB_OUT_DMRS;
c16_t *rx0 = rxdataF_comp[0];
c16_t *rx1 = rxdataF_comp[1];
// MRC on each re of rb
int i = 0;
for (; i < (sz & ~3); i += 4) {
// input always aligned and accepting tail padding to process all actual samples
for (int i = 0; i < sz; i += 4) {
*(simde__m128i *)(rx0 + i) =
simde_mm_adds_epi16(simde_mm_srai_epi16(*(simde__m128i *)(rx0 + i), 1), simde_mm_srai_epi16(*(simde__m128i *)(rx1 + i), 1));
}
for (; i < sz; i++)
rx0[i] = (c16_t){(rx0[i].r + rx1[i].r) / 2, (rx0[i].i + rx1[i].i) / 2};
}
void nr_rx_pdcch(PHY_VARS_NR_UE *ue,
@@ -387,25 +348,25 @@ void nr_rx_pdcch(PHY_VARS_NR_UE *ue,
const int antRx = frame_parms->nb_antennas_rx;
// Pointer to llrs, 4-bit resolution.
int32_t llr_size = 4 * n_rb * RE_PER_RB_OUT_DMRS;
int32_t llr_size = (rel15->coreset.duration-rel15->coreset.StartSymbolIndex) * n_rb * RE_PER_RB_OUT_DMRS;
c16_t llr[llr_size];
memset(llr, 0, sizeof(llr));
LOG_D(NR_PHY_DCI,
"pdcch coreset: freq %x, n_rb %d, rb_offset %d\n",
rel15->coreset.frequency_domain_resource[0],
n_rb,
rb_offset);
for (int s=rel15->coreset.StartSymbolIndex; s<(rel15->coreset.StartSymbolIndex+rel15->coreset.duration); s++) {
for (int s = rel15->coreset.StartSymbolIndex; s < rel15->coreset.StartSymbolIndex + rel15->coreset.duration; s++) {
LOG_D(NR_PHY_DCI, "in nr_pdcch_extract_rbs_single(rxdataF -> rxdataF_ext || dl_ch_estimates -> dl_ch_estimates_ext)\n");
__attribute__((aligned(32))) c16_t rxdataF_ext[antRx][n_rb * RE_PER_RB_OUT_DMRS];
__attribute__((aligned(32))) c16_t pdcch_dl_ch_estimates_ext[antRx][n_rb * RE_PER_RB_OUT_DMRS];
const int arraySz = ceil_mod(n_rb * RE_PER_RB_OUT_DMRS, 32);
__attribute__((aligned(32))) c16_t rxdataF_ext[antRx][arraySz];
__attribute__((aligned(32))) c16_t pdcch_dl_ch_estimates_ext[antRx][arraySz];
nr_pdcch_extract_rbs_single(ue->frame_parms.samples_per_slot_wCP,
n_rb,
rxdataF,
pdcch_est_size,
pdcch_dl_ch_estimates,
arraySz,
rxdataF_ext,
pdcch_dl_ch_estimates_ext,
s,
@@ -415,7 +376,7 @@ void nr_rx_pdcch(PHY_VARS_NR_UE *ue,
LOG_D(NR_PHY_DCI, "in nr_pdcch_channel_level(dl_ch_estimates_ext -> dl_ch_estimates_ext)\n");
// compute channel level based on ofdm symbol 0
int32_t avgs = nr_pdcch_channel_level(n_rb, pdcch_dl_ch_estimates_ext, antRx, s);
int32_t avgs = nr_pdcch_channel_level(n_rb, arraySz, pdcch_dl_ch_estimates_ext, antRx);
uint8_t log2_maxh = (log2_approx(avgs) / 2) + 5; //+antRx;
#ifdef UE_DEBUG_TRACE
@@ -427,25 +388,22 @@ void nr_rx_pdcch(PHY_VARS_NR_UE *ue,
LOG_D(NR_PHY_DCI, "we enter nr_pdcch_channel_compensation(log2_maxh=%d)\n", log2_maxh);
LOG_D(NR_PHY_DCI, "in nr_pdcch_channel_compensation(rxdataF_ext x dl_ch_estimates_ext -> rxdataF_comp)\n");
// compute LLRs for ofdm symbol 0 only
__attribute__((aligned(32))) c16_t rxdataF_comp[antRx][n_rb * RE_PER_RB];
nr_pdcch_channel_compensation(n_rb,
__attribute__((aligned(32))) c16_t rxdataF_comp[antRx][arraySz];
nr_pdcch_channel_compensation(arraySz,
rxdataF_ext,
pdcch_dl_ch_estimates_ext,
rxdataF_comp,
antRx,
log2_maxh); // log2_maxh+I0_shift
UEscopeCopy(ue, pdcchRxdataF_comp, rxdataF_comp, sizeof(struct complex16), antRx, n_rb * RE_PER_RB, 0);
UEscopeCopy(ue, pdcchRxdataF_comp, rxdataF_comp, sizeof(struct complex16), antRx, n_rb * RE_PER_RB_OUT_DMRS, 0);
if (antRx > 1)
nr_pdcch_detection_mrc(n_rb * RE_PER_RB, rxdataF_comp);
nr_pdcch_detection_mrc(arraySz, rxdataF_comp);
LOG_D(NR_PHY_DCI, "we enter nr_pdcch_llr(for symbol %d), pdcch_vars[eNB_id]->rxdataF_comp ---> pdcch_vars[eNB_id]->llr \n", s);
LOG_D(NR_PHY_DCI, "in nr_pdcch_llr(rxdataF_comp -> llr)\n");
nr_pdcch_llr(frame_parms, n_rb, rxdataF_comp[0], llr, s);
UEscopeCopy(ue, pdcchLlr, llr, sizeof(c16_t), 1, llr_size, 0);
nr_pdcch_llr(n_rb * RE_PER_RB_OUT_DMRS, rxdataF_comp[0], llr + s * n_rb * RE_PER_RB_OUT_DMRS);
}
UEscopeCopy(ue, pdcchLlr, llr, sizeof(c16_t), 1, llr_size, 0);
nr_pdcch_demapping_deinterleaving(llr,
pdcch_e_rx,