gNB: fix precoding matrix related compile error
This commit is contained in:
@@ -412,9 +412,9 @@ uint8_t nr_generate_pdsch(PHY_VARS_gNB *gNB,
|
|||||||
|
|
||||||
///Layer Precoding and Antenna port mapping
|
///Layer Precoding and Antenna port mapping
|
||||||
// tx_layers 1-8 are mapped on antenna ports 1000-1007
|
// tx_layers 1-8 are mapped on antenna ports 1000-1007
|
||||||
// The precoding info is supported by nfapi such as numPRGs, prgSize, and PMIdx
|
// The precoding info is supported by nfapi such as num_prgs, prg_size, prgs_list and pm_idx
|
||||||
// The same precoding matrix is applied on prgSize RBs, Thus
|
// The same precoding matrix is applied on prg_size RBs, Thus
|
||||||
// pmi = PMIdx[rbidx/prgSize], rbidx =0,...,rbSize-1
|
// pmi = prgs_list[rbidx/prg_size].pm_idx, rbidx =0,...,rbSize-1
|
||||||
// The Precoding matrix:
|
// The Precoding matrix:
|
||||||
// The Codebook Type I and Type II are not supported yet.
|
// The Codebook Type I and Type II are not supported yet.
|
||||||
// We adopt the precoding matrices of PUSCH for 4 layers.
|
// We adopt the precoding matrices of PUSCH for 4 layers.
|
||||||
@@ -426,8 +426,8 @@ uint8_t nr_generate_pdsch(PHY_VARS_gNB *gNB,
|
|||||||
for (int rb=0; rb<rel15->rbSize; rb++) {
|
for (int rb=0; rb<rel15->rbSize; rb++) {
|
||||||
//get pmi info
|
//get pmi info
|
||||||
uint8_t pmi;
|
uint8_t pmi;
|
||||||
if (rel15->precodingAndBeamforming.prgSize > 0)
|
if (rel15->precodingAndBeamforming.prg_size > 0)
|
||||||
pmi = rel15->precodingAndBeamforming.PMIdx[(int)rb/rel15->precodingAndBeamforming.prgSize];
|
pmi = rel15->precodingAndBeamforming.prgs_list[(int)rb/rel15->precodingAndBeamforming.prg_size].pm_idx;
|
||||||
else
|
else
|
||||||
pmi = 0;//no precoding
|
pmi = 0;//no precoding
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user