Integration: 2026.w30

- #324 ci: fix pull_request_target fork checkout error and Node 20 deprecation warning
- #143 Harmonize PUSCH CDM groups and port allocation
- #182 PMI and MCS table index in phy-test
- #285 Cleanup RU_t for NR and LTE
- #325 Simplify nr_fill_xx_indication()
- #140 Introduce group based processing for 5G uplink MU-MIMO and a MU-MIMO physical simulator
- #292 simplify and comment how the UE generates PSS reference signal
- #328 ci: split PCF CN config out of shared mini_nonrf_config.yaml
- #234 [NR UE] PDSCH Rx: Pre-allocate large buffers on heap
- #319 fix(nr_ue_rrc): calculate shortMAC-I and fill in RRCReestablishmentRequest
- #282 Two-pass min1/min2 cnProc for BG1 and BG2.
- #310 doc(t2): Update T2 related documentation
- #330 Force RedCap UE on the initialBWP

Closes: #278

Force RedCap UE on the initialBWP (#330)

Closes: #278
Reviewed-by: Robert Schmidt <robert.schmidt@openairinterface.org>
Reviewed-By: Luis Pereira <lpereira@allbesmart.pt>

Update timing thresholds using the average of three runs and apply a 20%
allowed margin.

This reflects the decoding time improvements introduced by #282.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>

doc(t2): Update T2 related documentation (#310)

Update T2 documentation and DPDK build in Dockerfile.
Reviewed-by: Shubhika Garg <shubhika.garg@openairinterface.org>
Reviewed-by: Robert Schmidt <robert.schmidt@openairinterface.org>

Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>

Update the DPDK version for fhi72 build with T2 support from 22.11 to
22.11.3 to align with LDPC_OFFLOAD_SETUP.md. Both 22.11 and 22.11.3 are
compatible with the AMD-T2-SDFEC_25-03-1.patch and have been tested with
the T2 card.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>

Update the T2 card build instructions to use the latest DPDK patch,
AMD-T2-SDFEC_25-03-1.patch. The required DPDK version for builds with T2
offload enabled is now 22.11 or 22.11.3, replacing the previous patch
and version requirements.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>

Signed-off-by: Romain Beurdouche <romain.beurdouche@eurecom.fr>

Two-pass min1/min2 cnProc for BG1 and BG2. (#282)

nrLDPC_cnProc_BG1_2pass and nrLDPC_cnProc_BG2_2pass replace the
LUT-exclude-self approach with a two-pass algorithm:

- Pass 1: single sweep over all numBN inputs to collect min1, min2 (two
  smallest |vk|) and the full sign product via sign_epi8.
- Pass 2: re-read each vk, select min2 where |vk|==min1 (tie approx),
  remove self sign with sign_epi8(vsgn_all, vk), and store.

Memory reads per CN group: `2*numBN` vs `numBN*(numBN-1)` for LUT
approach:

    numBN= 4:  8 vs  12  (1.5x  fewer)
    numBN= 7: 14 vs  42  (3.0x  fewer)
    numBN=10: 20 vs  90  (4.5x  fewer)
    numBN=19: 38 vs 342  (9.0x  fewer)

A generic nrLDPC_cnProc_group_2pass(buf, res, numBN, M, off) helper does
the work; the BG1/BG2 wrappers loop over their degree groups and derive
the BN stride from lut_numCnInCnGroups_BG1/2_R13/15 exactly as the
existing functions do.  No new LUT tables are required.

Placed inside the #else (non-AVX512) block alongside the existing
BG1/BG2 functions; AVX512 path is unchanged.

It also removes the cnProc/bnProc/bnProcPC generators and simplifies the
original implementation (several hundreds of lines of repetitive code
replaced by generic functions.)
Reviewed-by: Laurent THOMAS <laurent.thomas@open-cells.com>

fix(nr_ue_rrc): calculate shortMAC-I and fill in RRCReestablishmentRequest (#319)

Currently the shortMAC-I field is hardcoded in RRCReestablishmentRequest
message, fix it by calculating as written in TS 38.331 section 5.3.7.4.

Before this fix, test with OCUDU gNB when triggering re-establishment
will cause gNB verification of shortMAC-I fail, then gNB will send
RRCSetup instead of RRCReestablishment.

After this fix, OCUDU gNB will send RRCReestablishment after UE sent
RRCReestablishmentRequest.
Reviewed-by: Robert Schmidt <robert.schmidt@openairinterface.org>
Reviewed-by: Guido Casati <guido.casati@openairinterface.org>

[NR UE] PDSCH Rx: Pre-allocate large buffers on heap (#234)

Performs memory optimization to speed up OAI UE.

1. Replace per slot memory allocation and free of buffers dl_ch_mag/b/r,
   rho_dl, rxdataF_comp and pdsch_dl_ch_estimates with one-time
   allocation at init and free at teardown by pre-allocating one buffer
   set per actor.
2. Remove unnecessary memset on buffers as they are immediately written
   afterwards before being read.
Reviewed-by: Robert Schmidt <robert.schmidt@openairinterface.org>
Reviewed-by: Sakthivel Velumani <s.velumani@northeastern.edu>

ci: split PCF CN config out of shared mini_nonrf_config.yaml (#328)

The u0_25prb multi-QoS CI scenario needs PCF (use_local_pcc_rules: no),
but in f02c84fc that configuration has been placed in the shared
5g_rfsimulator/mini_nonrf_config.yaml. Every other rfsim pipeline mounts
that file without running oai-pcf.

In some setups different from Linux (i.e. WSL2) this causes issues.

With use_local_pcc_rules: no, SMF attempts associating to oai-pcf when
creating SM Context even when no PCF container exists. If that HTTP call
exceeds the wait timeout (100 ms), AMF gets error and does not complete
the SM context creation, a later Update SM Context then fails ("Could
not find Nsmf_PDUSession URI").

Native Linux is usually fine: when the failed PCF association is fast
enough, context creation completes successfully.

This commit moves PCF-specific configuration into
5g_rfsimulator_u0_25prb only. The shared config is restored to
use_local_pcc_rules: yes and a single oai DNN (pre-f02c84fc status).

Changes:

- Add new file 5g_rfsimulator_u0_25prb/mini_nonrf_config.yaml (PCF,
  dual-DNN, use_local_pcc_rules: no) and mount in
  5g_rfsimulator_u0_25prb/docker-compose.yaml -Revert
  5g_rfsimulator/mini_nonrf_config.yaml
Reviewed-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Reviewed-by: Thomas Schlichter <thomas.schlichter@iis.fraunhofer.de>

simplify and comment how the UE generates PSS reference signal (#292)

a small simplification in UE PSS amplitude
Reviewed-by: Sakthivel Velumani <s.velumani@northeastern.edu>
Reviewed-by: Alex Jiao <alex.jiao@keysight.com>

Taken from three runs and averaged.
Signed-off-by: Robert Schmidt <robert.schmidt@openairinterface.org>

I took three runs, averaged the respective times. For the thresholds,
always apply a 20% margin.
Signed-off-by: Robert Schmidt <robert.schmidt@openairinterface.org>

Currently the shortMAC-I field is hardcoded in RRCReestablishmentRequest message,
fix it by calculating as written in TS 38.331 section 5.3.7.4.
Signed-off-by: Alex Jiao <alex.jiao@keysight.com>

Introduce group based processing for 5G uplink MU-MIMO and a MU-MIMO physical simulator (#140)

This MR introduces the baseband processing at the gNB required to
support Uplink MU-MIMO gNB by updating ULSCH_id based processing to FAPI
group-based joint processing. The PHY layer can now perform spatial
demultiplexing for multiple UEs scheduled on the exact same time and
frequency resources.

Additionally, this MR introduces a dedicated MU-MIMO physical simulator
(nr_ulsim_mu_mimo) to validate the joint receiver's performance.

The simulator can be compiled with the other physical simulators using
./build_oai --phy_simulators

It can be run using the following:

    sudo ./nr_ulsim_mu_mimo -n 300 -s 30 -S 30 -R 273 -W 1 -y 2 -z 2 -L 3 -N 2 -g C,l,10

This is the following output:

    *****************************************
    [UE 0] SNR 30.000000: n_errors (4/300,0/10,0/0,0/0) (negative CRC), false_positive 0/300, errors_scrambling (30357/4140000,335/138000,0/0,0/0)

    SNR 30.000000: Channel BLER (1.333333e-02,0.000000e+00,-nan,-nan Channel BER (7.332609e-03,2.427536e-03,-nan,-nan) Avg round 1.03, Eff Rate 9070.2667 bits/slot, Eff Throughput 98.33, TBS 9224 bits/slot
    DMRS-PUSCH delay estimation: min 0, max 10, average 1.9
    *****************************************

    *****************************************
    [UE 1] SNR 30.000000: n_errors (6/300,0/10,0/0,0/0) (negative CRC), false_positive 0/300, errors_scrambling (32485/4140000,1104/138000,0/0,0/0)

    SNR 30.000000: Channel BLER (2.000000e-02,0.000000e+00,-nan,-nan Channel BER (7.846618e-03,8.000000e-03,-nan,-nan) Avg round 1.03, Eff Rate 9070.2667 bits/slot, Eff Throughput 98.33, TBS 9224 bits/slot
    DMRS-PUSCH delay estimation: min 0, max 10, average 1.9
    *****************************************
Reviewed-by: Roberto Louro Magueta <rmagueta@allbesmart.pt>
Reviewed-by: Robert Schmidt <robert.schmidt@openairinterface.org>

Simplify nr_fill_xx_indication() (#325)

nr_fill_xx_indication() in the UE is hard to understand, this is a small
clarification step.
Reviewed-by: Francesco Mani <email@francescomani.it>

Signed-off-by: Francesco Mani <email@francescomani.it>

UL Indications typically is fast, so the 1-3us have all a guard interval
of 100%. Worse, in phy-test, we typically don't actually forward any RLC
PDUs, so we don't measure much anyway. Remove this from all configs.
Signed-off-by: Robert Schmidt <robert.schmidt@openairinterface.org>

Add llr[2] and llr_buf_max to pdsch_scratch_t, allocated once at init at
worst-case size (12 * 14 * N_RB_DL * 8 * 4). Remove the per-slot
malloc16_clear + free from nr_ue_pdsch_procedures. The zero-init is also
dropped as nr_dlsch_layer_demapping unconditionally overwrites all G
positions before any function reads the buffer.
Signed-off-by: Rupanjali <rupanjali.srivastava@openairinterface.org>

Replace all per-slot heap allocations for rxdataF_comp, dl_ch_mag/magb/magr,
rho_dl and pdsch_dl_ch_estimates with a pre-allocated scratch buffer array
stored in PHY_VARS_NR_UE. One scratch set is allocated per DL actor
(num_dl_actors from UE params, defaulting to 1); the actor index is derived
as proc->nr_slot_rx % pdsch_num_actors, giving each parallel DL actor its
own non-overlapping buffer set and eliminating heap contention between actors.

Also remove redundant memset() calls on dl_ch_estimates_ext and rxdataF_ext
stack VLAs, which are fully overwritten by nr_dlsch_extract_rbs() before use.
Signed-off-by: Rupanjali <rupanjali.srivastava@openairinterface.org>

The u0_25prb multi-QoS CI scenario needs PCF (use_local_pcc_rules: no),
but in f02c84fc that configuration has been placed in the shared
5g_rfsimulator/mini_nonrf_config.yaml. Every other rfsim pipeline
mounts that file without running oai-pcf.

In some setups different from Linux (i.e. WSL2) this causes issues.

With `use_local_pcc_rules: no`, SMF attempts associating to oai-pcf when
creating SM Context even when no PCF container exists. If that HTTP call
exceeds the wait timeout (100 ms), AMF gets error and does not complete
the SM context creation, a later Update SM Context then fails
("Could not find Nsmf_PDUSession URI").

Native Linux is usually fine: when the failed PCF association is fast
enough, context creation completes successfully.

This commit moves PCF-specific configuration into 5g_rfsimulator_u0_25prb
only. The shared config is restored to `use_local_pcc_rules: yes` and a single
oai DNN (pre-f02c84fc status).

Changes:
- Add new file 5g_rfsimulator_u0_25prb/mini_nonrf_config.yaml (PCF, dual-DNN,
  use_local_pcc_rules: no) and mount in 5g_rfsimulator_u0_25prb/docker-compose.yaml
- Revert 5g_rfsimulator/mini_nonrf_config.yaml
Signed-off-by: Guido Casati <guido.casati@openairinterface.org>
Reported-by: Thomas Schlichter <thomas.schlichter@iis.fraunhofer.de>

ci: fix pull_request_target fork checkout error and Node 20 deprecation
warning (#324)

Fixes:
- The Jenkins dispatch workflow failed on fork PRs with
  Error: Refusing to check out fork pull request code from a
  'pull_request_target' workflow. This workflow runs with the base
  repository's GITHUB_TOKEN, secrets, default-branch cache scope, and
  runner access. Fetching and executing a fork's code in that trusted
  context commonly leads to "pwn request" vulnerabilities. To opt in,
  review the risks at https://gh.io/securely-using-pull_request_target
  and set 'allow-unsafe-pr-checkout: true' on the actions/checkout step.
- Node 20 warnings on GitHub Actions

Related commit: 8e30ec80 ("ci: migrate Jenkins dispatch to
pull_request_target")

See the commit message for more details
Reviewed-by: Robert Schmidt <robert.schmidt@openairinterface.org>
Reviewed-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>

Signed-off-by: Robert Schmidt <robert.schmidt@openairinterface.org>

nr_fill_rx_indication() increments itself, remove full memset for the related struct (100 elements, one used)
Signed-off-by: Laurent THOMAS <laurent.thomas@open-cells.com>

Instead of this somewhat opaque function, fill the struct in place to
make it clearer what gets filled. This should make the code clearer in
general.
Co-authored-by: Robert Schmidt <robert.schmidt@openairinterface.org>
Signed-off-by: Laurent THOMAS <laurent.thomas@open-cells.com>
Signed-off-by: Robert Schmidt <robert.schmidt@openairinterface.org>

Signed-off-by: Laurent THOMAS <laurent.thomas@open-cells.com>

Cleanup RU_t for NR and LTE (#285)

For both LTE and NR, removes:

- txfh_in_fep
- fh_north_in
- eth_params_t -> initialized directly via ifdevice

For NR, removes:

- do_prach
- fh_north_out
- fh_south_asynch_in
- fh_north_asynch_in

Removes useless time-domain functions pointers for split 7.2.
Reviewed-by: Robert Schmidt <robert.schmidt@openairinterface.org>
Reviewed-by: Laurent THOMAS <laurent.thomas@open-cells.com>

Signed-off-by: Rakesh Mundlamuri <rakesh.mundlamuri@openairinterface.org>

Signed-off-by: Teodora Vladić <teodora.vladic@openairinterface.org>
Fixes: 7b7afae0 ("feat(oran_fhlib_5g): Changes to run K release")

Move ru_thread create into `start_NR_RU()`
Signed-off-by: Teodora Vladić <teodora.vladic@openairinterface.org>

`synch_to_ext_device` not relevant for NR
Signed-off-by: Teodora Vladić <teodora.vladic@openairinterface.org>

    - bump the detect-changes checkout from v4 to the latest v7
      to clear the "Node 20 is deprecated" warning. It runs on
      Node 24.
    - actions/checkout now refuses to check out fork PR code from a
      pull_request_target workflow, which broke the trigger-jenkins
      job. It was checking out the fork's head SHA via
      'ref: github.event.pull_request.head.sha'.
      However, the job only forwards the github.event payload to
      Jenkins via curl.
    - paths-filter v3 runs on Node 20, which GitHub Actions runners
      now flag as deprecated. Update to v4 which runs on Node 24.

Fixes: 8e30ec80 ("ci: migrate Jenkins dispatch to pull_request_target")
Signed-off-by: Shubhika Garg <shubhika.garg@openairinterface.org>
Tested-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>

Previously, `eth_params_t *` from `openair0_device_t` pointed
to `eth_params_t` from `RU_t`. However, only for transport
methods using ethernet, we shall initialize necessary params
(addresses and ports).
Signed-off-by: Teodora Vladić <teodora.vladic@openairinterface.org>

The MMSE is computed using (H^H*H + N*I)^-1. While computing the MMSE
term H^H*H + N*I before the inverse, the term H^H*H is scaled using
shift (log2_max). However, the term N (noise) is not scaled. The fix scales the
noise term using the shift.
Signed-off-by: Rakesh Mundlamuri <rakesh.mundlamuri@openairinterface.org>
Fixes: 1b77adcb ("Segregate the LLR and channel compensation related common functions into different files")

Signed-off-by: Rakesh Mundlamuri <rakesh.mundlamuri@openairinterface.org>

Signed-off-by: Rakesh Mundlamuri <rakesh.mundlamuri@openairinterface.org>

This commit introduces a dedicated simulation environment to validate
the uplink MU-MIMO joint processing framework. It models a scenario
where a single gNB serves multiple spatially multiplexed UEs sharing
the exact same time and frequency resources.

The simulator executes the complete physical layer baseband pipeline:
- PUSCH data generation and LDPC encoding per UE
- OFDM modulation (TX)
- Signal propagation through a multi-path channel model
- OFDM demodulation (RX) at the gNB
- Joint PUSCH decoding, equalization, and spatial demultiplexing

Performance metrics such as scrambling errors and BLER per UE is
evaluated at the end of the simulation.
Signed-off-by: Rakesh Mundlamuri <rakesh.mundlamuri@openairinterface.org>