Part of: OCUDU on COSMOS · 4G / 5G & O-RAN · Step 3 of the OCUDU learning path · Previous: ocudu-ota · Next: ocudu-mplane
Split the OCUDU base station at the O-RAN 7.2x fronthaul: OCUDU runs as the O-DU on a COSMOS server and sends frequency-domain IQ samples over Ethernet to a commercial Benetel RAN650 radio unit, which radiates a 100 MHz n78 cell. A Quectel modem registers with the Open5GS core over that cell, opens a PDU session and measures throughput.
This is the production shape of an open radio access network, and the first step at which the radio is not software you can instrument. A commercial O-RU is a sealed box with its own processor, firmware and stored configuration. It keeps that configuration across runs, and other users drive the same radio with settings of their own. So before every bring-up the tutorial puts the radio into the state OCUDU needs and proves it is there. Skipping that check fails silently: the fronthaul counters stay clean and the cell is useless.
Concepts
- O-DU / O-RU — the distributed unit and the radio unit as the O-RAN Alliance defines them. The 3GPP CU and DU stay together in OCUDU's
gnb; the O-RU holds the lowest physical layer and the RF.- 7.2x fronthaul — the O-RAN split inside the physical layer. The O-DU sends frequency-domain IQ as eCPRI frames on VLAN 4; the control, user and synchronization planes (C/U/S-Plane) share that link.
- PTP (G.8275.1) — the time protocol the S-Plane uses. The fronthaul switch is the grandmaster; the O-DU server and the O-RU are its clients.
- Standalone O-RU — an O-RU that runs from configuration stored on itself. OCUDU's
gnbhas no M-Plane (NETCONF) client, so the O-RU must run standalone here; the M-Plane is the subject of ocudu-mplane.- eAxC — the identifier of one antenna-carrier stream on the fronthaul. The DU's
dl_port_id,ul_port_idandprach_port_idmust name the streams the O-RU actually uses.- CPU isolation — the O-DU busy-polls its fronthaul threads on cores reserved for it with
isolcpus, derived per server.
| Link | Between | Carried over |
|---|---|---|
| Fronthaul (C/U-Plane) | O-DU ↔ O-RU | eCPRI on VLAN 4 over the server's E810 fronthaul port DATA1a, through the fronthaul switch; OCUDU opens a raw socket on the physical function, so no DPDK and no virtual functions are needed |
| Synchronization (S-Plane) | switch → O-DU and O-RU | PTP G.8275.1, domain 24; the O-DU server runs ptp4l on DATA1a |
| O-RU state | O-DU server → O-RU | SSH from COSMOS tooling on the O-DU server, before the DU starts; OCUDU itself never talks to the O-RU's management interface |
| O-RU power | console → O-RU | omf tell from the console of the domain your O-RU reservation targets |
| Air | O-RU ↔ modem | band n78, 100 MHz, 30 kHz subcarrier spacing, TDD DDDDDDDSUU |
| N2, N3, N6 | gNB ↔ Open5GS ↔ internet | on the O-DU server, as in the earlier tutorials; UE pool 192.168.100.0/22 |
Where the O-RU's address and channel come from. You never type the O-RU's address or its power-control name. The parameter ocudu_oru names the radio (benetel); with ocudu_ru_ip: auto and ocudu_ru_name: auto every mode asks the COSMOS O-RU tools on the O-DU server for the radio's current management address and its omf tell name when it starts. The channel works the same way: ocudu_center_freq_mhz: reservation and ocudu_bw_mhz: reservation take the band, centre and bandwidth of your O-RU reservation, so the cell is always on the spectrum you booked. auto and reservation mean "ask the testbed, now". To pin a value, set it in your --params file and stay inside your reservation.
| Domains | rrail, osc, sb1 |
| Nodes | 2, named in OMF_NODES |
| On the console | python3, ansible-playbook, ssh, scp |
cosmos-tutorial check ocudu-72 verifies, on each node:
Not in the package - you provide:
You will also need:
Danger: The Benetel and the LiteOn share one RF space. Every O-RU reservation carries its band, centre and bandwidth, and the scheduler refuses overlapping bookings, so the two may be powered only on non-overlapping channels. A DU left streaming at an O-RU breaks every other user's cell: always run the teardown.
On the console of the domain your O-RU reservation targets, name the O-DU server first and the UE node second. The check is read-only.
export OMF_NODES="<du-node>,<ue-node>"
cosmos-tutorial check ocudu-72
It ends with RUNNABLE or NOT RUNNABLE and the reasons: the OCUDU build and an Intel E810 on the O-DU server, the O-RU management key in its image, the COSMOS O-RU tools listing the Benetel, a 5G modem on the UE node, and no other DU running. If the O-DU server lacks the image, load it first:
omf load -i ocudu.ndz -t <du-node> -r 0 -o 2400 -b 600
omf tell -a on -t <du-node>
The image carries no CPU isolation, because the right cores differ per server. The first run derives and applies it for the fronthaul NIC and reboots the O-DU server once; later runs on the same image load reuse it.
export OMF_NODES=<node>,<node>
cosmos-tutorial check ocudu-72
cosmos-tutorial run ocudu-72 # ansible
cosmos-tutorial run ocudu-72 --mode manual # manual
cosmos-tutorial jupyter ocudu-72 # jupyter
cosmos-tutorial run ocudu-72 --mode parity # parity
cosmos-tutorial run ocudu-72 --mode teardown # teardown
All three modes read one parameter file, use one log parser, and write the same JSON result into ~/cosmos-runs/ocudu-72/<UTC>-<mode>/ on the console. OMF_NODES is required and ordered: first the O-DU server, then the UE node. The O-RU is not in OMF_NODES: it is an instrument, named in the parameters and powered from the console under your O-RU reservation.
The playbook runs a preflight check, resolves the O-RU's address and power name, stops a previous DU, cold-cycles the O-RU and waits until it is really down and back up, applies CPU isolation if it is missing (one reboot), reads the channel from your reservation, renders the gNB configuration for this server's isolated cores, and runs the bring-up on the O-DU server. It then configures and registers the modem, brings up the data call, measures ping and throughput, and writes the result.
The manual mode runs the commands below for you, from the console. To follow them yourself, start on the console:
D=$(cosmos-tutorial which ocudu-72)
RU_IP=$(python3 $D/cosmos_addr.py --value auto oru benetel --du <du-node>)
RU_NAME=$(python3 $D/cosmos_addr.py --value auto oru benetel --du <du-node> --field omf)
eval "$(python3 $D/ocudu-72_lib.py rf $D/ocudu-72_params.yml)"; echo "$OCUDU_RF_CENTER_MHZ MHz / $OCUDU_RF_BW_MHZ MHz"
Step 1 — Isolate the fronthaul cores (O-DU server, once per image load). cosmos-ofh-isolation derives the isolated cores from the fronthaul NIC's NUMA node and writes them into the kernel command line. A reboot makes them take effect.
ssh root@<du-node> "cosmos-ofh-isolation --nic DATA1a --apply && systemctl reboot"
ssh root@<du-node> "grep -o 'isolcpus=[^ ]*' /proc/cmdline" # after the reboot: a non-empty per-host set
Step 2 — Cold-cycle the O-RU (console). The Benetel starts transmitting only on a fresh boot against a correctly timed fronthaul; restarting the DU under a running O-RU does not revive it. Power it off, wait until it no longer answers ping from the O-DU server, hold it off for 30 s, and power it on. In standalone mode it is up when its SSH port (22) answers; its NETCONF port stays closed on purpose.
omf tell -a offh -t $RU_NAME
ssh root@<du-node> "ping -c1 -W1 $RU_IP" # repeat until it fails
omf tell -a on -t $RU_NAME
ssh root@<du-node> "timeout 2 bash -c '</dev/tcp/$RU_IP/22' && echo RU-UP" # repeat until RU-UP
omf tell refuses if the O-RU reservation is not yours, is not active now, or targets another domain, and names the console to use. The first omf tell to an instrument is sometimes dropped: confirm by ping, not by the reply.
Step 3 — Render the gNB configuration (console). The template takes the cell from the parameters and the channel step, and the CPU affinities from this server's isolated cores, its core count and its hyperthread siblings, so the main worker pool never shares a physical core with the fronthaul threads.
ISO=$(ssh root@<du-node> "grep -oE 'isolcpus=[^ ]+' /proc/cmdline | sed -E 's/isolcpus=//; s/(managed_irq|domain|nohz),?//g'")
NCPU=$(ssh root@<du-node> "nproc --all")
SIB=$(ssh root@<du-node> "cat /sys/devices/system/cpu/cpu*/topology/thread_siblings_list | sort -u | tr '\n' ' '")
python3 $D/ocudu-72_lib.py render-gnb $D/ocudu-72_params.yml $D/ocudu-72_gnb.yml.tmpl gnb-ocudu-72.yml "$ISO" "$NCPU" "$SIB"
ssh root@<du-node> "mkdir -p /root/ocudu /root/oru-profiles"
scp gnb-ocudu-72.yml root@<du-node>:/root/ocudu/gnb-ocudu-72.yml
scp $D/ocudu-72_bringup.sh root@<du-node>:/root/
scp $D/host/oru-profiles/oru-apply.py $D/host/oru-profiles/benetel-ran650.yml root@<du-node>:/root/oru-profiles/
Step 4 — Bring up the O-DU (O-DU server). The bring-up script does, in order: the performance CPU governor; the O-RU's standalone mode (/root/benetel-ru-mode.sh nomplane, which reboots the O-RU only if the mode changes); a wait until the O-RU's own radio bring-up has finished for this boot; the O-RU state check against its profile, writing your channel into the O-RU and rebooting it once if anything had to change; the fronthaul NIC with jumbo frames, promiscuous mode and its interrupts off the isolated cores; PTP, stepped once and then slewing; Open5GS; and the gNB as the named unit ocudu-72, launched under taskset over all cores.
RU_CENTER_HZ=$OCUDU_RF_CENTER_HZ RU_BW_HZ=$OCUDU_RF_BW_HZ /root/ocudu-72_bringup.sh /opt/ocudu/build/apps/gnb/gnb /root/ocudu/gnb-ocudu-72.yml /tmp/ocudu-72.log /tmp/ocudu-72-stdout.log DATA1a /etc/linuxptp/ptp4l-g8275.conf /root/benetel-ru-mode.sh nomplane $RU_IP /root/oru/ssh_keys/id_ed25519
systemd-run --unit=ocudu-iperf --collect /usr/bin/iperf3 -s -B 192.168.100.1 -p 5201
Set RU_CENTER_HZ and RU_BW_HZ to the values the channel step printed on the console. The O-RU management key ships in the image and is never copied into the tutorial.
Step 5 — The modem (UE node). The modem steps are those of ocudu-ota with this tutorial's values: band mask 77:78 and the OCUDU APN COSMOS-O. The helper's call starts the data call and, if the modem holds a stale session from an earlier run, deregisters, reboots the modem and retries.
scp $D/ocudu-72_modem.py $D/ocudu-72_iperf.sh root@<ue-node>:/root/ # from the console
systemctl stop ModemManager
read -r QMI WWAN <<<"$(python3 /root/ocudu-72_modem.py resolve auto auto)"
python3 /root/ocudu-72_modem.py configure /dev/ttyUSB2 00101 77:78 COSMOS-O
python3 /root/ocudu-72_modem.py wait-reg /dev/ttyUSB2 240 00101
python3 /root/ocudu-72_modem.py call /dev/ttyUSB2 $QMI $WWAN COSMOS-O 240 # note the IPv4 address ($UE) and gateway ($GW)
ip addr add $UE/22 dev $WWAN; ip link set $WWAN mtu 1400 up
ip route replace 192.168.100.0/22 dev $WWAN
for t in 192.168.100.1 8.8.8.8; do ip route replace $t/32 via $GW dev $WWAN; done
ping -I $WWAN -c 10 -i 0.3 -W 2 192.168.100.1
ping -I $WWAN -c 10 -i 0.3 -W 2 8.8.8.8
/root/ocudu-72_iperf.sh 192.168.100.1 5201 $UE 10 -R # DL
/root/ocudu-72_iperf.sh 192.168.100.1 5201 $UE 10 # UL
The notebook runs on the console and drives both nodes over SSH. It performs the same steps and plots downlink and uplink throughput. It can also be run non-interactively with papermill, which writes an executed copy to a new file.
The parity mode runs Ansible, manual and Jupyter one after the other on the same O-DU server, O-RU and modem, then compares their results. It passes when every mode brings the cell up and registers the modem, local ping loss is at most 20 %, there are no fronthaul drops, downlink is in 20–1500 and uplink in 1–600 Mbit/s, and the modes agree within 60 % (downlink) and 80 % (uplink). The harness runs the teardown on every exit, including a failure or Ctrl-C.
The bring-up prints one block per stage. A healthy run looks like this:
=== [1b] O-RU: THIS boot's radio bring-up finished, per-stack state asserted, Tx count set ===
RU bring-up finished this boot: Mon Sep 14 02:21:19 UTC 2026 (RU up 152s)
C0300 (number of Tx) = 0x1010101
ptp4l locked: ptp4l[2139.061]: rms 3149 max 4739 freq -6742 +/- 390 delay
stepped: |offset| = 148 ns
slewing: |offset| = 1 ns (UTC timescale)
AMF OK
GNB-ALIVE, cell up
ether_rx rx_bytes: 892800
Read it line by line: the O-RU finished its radio bring-up this boot, so the registers read back are real; its transmitter count is set; PTP stepped once and now slews within a few nanoseconds; the AMF runs; the gNB is alive with its cell up; and ether_rx rx_bytes is non-zero, so the DU's raw socket really receives fronthaul frames. A zero there means the NIC is not promiscuous.
The modem then shows +C5GREG: 0,1 and an IPv4 address in 192.168.100.0/22, and both pings succeed. The last verified parity run (2026-10-30, rrail, Benetel RAN650) measured a downlink of 157–200 Mbit/s and an uplink of 75–77 Mbit/s. A parity run ends with PARITY: PASS, the two spread lines inside their tolerances, and rig released.
Change any of these with cosmos-tutorial run ocudu-72 --params my.yml. An unknown key or a wrong type is refused. cosmos-tutorial params ocudu-72 prints the file.
| Key | Default | Meaning |
|---|---|---|
ocudu_oru |
benetel |
Which O-RU drives the cell: a cosmos-oru name, benetel | liteon | foxconn. |
ocudu_ru_name |
auto |
The O-RU's omf tell name. auto = from cosmos-oru-tools (orus.json) on the O-DU host; a value pins it. |
ocudu_ru_console |
`` | "" = run omf tell on THIS console. |
ocudu_ru_ip |
auto |
The O-RU's management address. auto = the address that answers from the O-DU host, from cosmos-oru-tools (orus.json; the Benetel moves between two); an O-RU listed there without an address (the Foxconn) is looked up by its omf name in DNS on the O-DU host. |
ocudu_ru_mode |
nomplane |
⚠️ THE decisive step: in M-Plane mode the RU radiates only while a NETCONF client supervises it, so with srsRAN you get a perfectly clean fronthaul (0 drops) and NO RF at all. |
ocudu_ru_mode_script |
auto |
auto = the image's benetel-ru-mode.sh, else cosmos-oru-tools' copy; a path pins it |
ocudu_ru_key |
auto |
the Benetel's SSH key on the O-DU host. auto = BENETEL_KEY in the O-RU logins the run pushes from |
ocudu_gnb_bin |
/opt/ocudu/build/apps/gnb/gnb |
Per-run files on the node live in the run's own directory $COSMOS_NODE_RUN_DIR (/var/tmp/cosmos/, set by cosmos-tutorial; /var/tmp/cosmos/ocudu-72.current points at the latest run). |
ocudu_gnb_cfg |
gnb-ocudu-72.yml |
RENDERED from ocudu-72_gnb.yml.tmpl |
ocudu_gnb_log |
ocudu-72.log |
srsRAN log.filename: "gNB started" + OFH drops live HERE |
ocudu_gnb_stdout |
ocudu-72-stdout.log |
|
ocudu_core_start |
/root/open5gs/start.sh |
|
ocudu_fh_iface |
DATA1a |
the E810 PF carrying VLAN 4 (srsRAN OFH uses a raw socket) |
ocudu_ptp_iface |
DATA1a |
|
ocudu_ptp_cfg |
/etc/linuxptp/ptp4l-g8275.conf |
|
ocudu_band |
78 |
Verified: n78 3750 MHz / 100 MHz (273 PRB @ 30 kHz, dl_arfcn 650000). |
ocudu_tac |
7 |
|
ocudu_plmn |
00101 |
|
ocudu_imsi |
auto |
auto = read from the modem's SIM (AT+CIMI); a number here is only a cross-check |
ocudu_dnn |
COSMOS-O |
STANDING CONVENTION: one APN per STACK, so which stack a modem is on is obvious at a glance even with several cells live. |
ocudu_ue_pool |
192.168.100.0/22 |
|
ocudu_ue_gw |
192.168.100.1 |
|
ocudu_center_freq_mhz |
reservation |
Benetel RAN650 (ocudu_oru: benetel) - its complete radio parameter set Channel (2026-10-16): taken from the O-RU RESERVATION (omf-shared-scheduler: band, centre, bandwidth), so the cell is always on the spectrum you booked. |
ocudu_bw_mhz |
reservation |
|
ocudu_rf_fallback_center_mhz |
3750 |
|
ocudu_rf_fallback_bw_mhz |
100 |
|
ocudu_ru_bandwidth_mhz |
channel |
ru_ofh.ru_bandwidth_MHz: channel = the channel bandwidth above |
ocudu_ru_mac |
8c:1f:64:d1:11:f0 |
|
ocudu_du_mac |
00:11:22:33:44:66 |
the MAC the RU is configured to talk to |
ocudu_pci |
2 |
STANDING CONVENTION: unique physCellId AND cell identity PER O-RU, so a UE moving between radios registers fresh instead of trying to re-establish on what looks like the same cell. |
ocudu_sector_id |
2 |
srsRAN builds the NR Cell Identity from gnb_id + sector_id |
ocudu_compr_method_prach |
bfp |
Fronthaul timing + compression: the values proven against THIS RAN650 (2026-09-02). |
ocudu_compr_bitwidth_prach |
9 |
|
ocudu_nof_antennas_dl |
2 |
4T4R leaves this RU's transmit branches badly imbalanced (measured 33.7 / 1.1 / 24.1 / 22.8 dBm at equal attenuation), which keeps the single-port SSB strong while wrecking the multi-port PDSCH: the UE reports a fine RSRP, reports CQI 2, and DL HARQ exhausts its retransmissions. 2T2R on the two STRONG branches is what the published working RAN650 configuration does, and what took DL from 8.0 to 62.5 Mbit/s here. |
ocudu_dl_port_id |
[0, 2] |
|
ocudu_ignore_prach_start_symbol |
True |
The O-RU stamps its PRACH U-plane packets with symbolId 0..11 for the occasion. srsRAN validates that field against the start symbol it expects and silently drops the packets when they disagree, so a preamble that is physically present is never handed to the detector. |
ocudu_prach_cp_enabled |
True |
this RU has prach_freq_offset_dynamic=true in its ru_config.cfg, i.e. it FOLLOWS the DU PRACH C-plane. |
ocudu_prach_config_index |
159 |
VERIFIED 2026-09-09: this is also what srsRAN derives for this cell when the field is omitted, so it is pinned rather than copied from another vendor. |
ocudu_nof_cb_preambles_per_ssb |
omit |
omit = srsRAN's default (the LiteOn needs 4, see its set) |
ocudu_tdd_period_slots |
10 |
ru_bandwidth_MHz in the gNB yml follows the channel bandwidth above (2026-10-16) TDD pattern. |
ocudu_tdd_dl_slots |
7 |
|
ocudu_tdd_dl_symbols |
6 |
|
ocudu_tdd_ul_slots |
2 |
|
ocudu_tdd_ul_symbols |
4 |
|
ocudu_t1a_max_cp_dl |
470 |
|
ocudu_t1a_min_cp_dl |
419 |
|
ocudu_t1a_max_cp_ul |
336 |
|
ocudu_t1a_min_cp_ul |
285 |
|
ocudu_t1a_max_up |
345 |
|
ocudu_t1a_min_up |
294 |
|
ocudu_ta4_max |
1960 |
from the OCUDU config that DID work end-to-end (Foxconn, 2026-08-19). ta4 is the DU UL RECEPTION WINDOW: if it closes before the O-RU delivery arrives, the closed-window handler pops the PRACH context as "incomplete" with no drops logged, which is exactly our symptom. 200 and 500 both failed. |
ocudu_ta4_min |
0 |
|
ocudu_ignore_ecpri_payload_size |
True |
|
ocudu_ignore_ecpri_seq_id |
omit |
omit = srsRAN's default |
ocudu_max_proc_delay |
omit |
No expert_phy max_proc_delay: with 15 slots of processing budget OCUDU 26.10 misses every DL transmit window on the Benetel (2026-10-23). omit = srsRAN's default. |
ocudu_iq_scaling |
5.5 |
⭐ SHORT RANGE => LOW GAINS. |
ocudu_ru_ul_gain_correction |
0 |
RU-side uplink gain, written to /etc/ru_config.cfg by the bring-up. |
ocudu_preamble_rx_target_pw |
-100 |
Open-loop Msg1 power: the UE sets its preamble power to preamble_rx_target_pw + pathloss. |
ocudu_ssb_block_power |
-4 |
Advertised SSB power, used by the UE to estimate pathloss. -4 dBm is what ran end to end. |
ocudu_pucch_resource_common |
11 |
|
ocudu_zero_correlation_zone |
0 |
Zero-correlation zone: 0, as in the Foxconn and published RAN650 configurations. |
ocudu_prach_freq_start |
0 |
MUST be 0 on this RU. |
ocudu_prach_port_id |
[4, 5] |
srsRAN only notifies a PRACH occasion when EVERY port of its PRACH buffer has been filled (prach_context::have_all_res_been_written), so this list must not name a port the RU is silent on. |
ocudu_nof_antennas_ul |
2 |
|
ocudu_ul_port_id |
[0, 1] |
|
ocudu_mcs_table |
qam64 |
Link adaptation. |
ocudu_olla_target_bler |
0.1 |
|
ocudu_fh_vlan |
4 |
⚠️ These two MUST equal what the O-RU has in its own /etc/ru_config.cfg — in standalone mode the RU is the authority and silently ignores anything else. |
ocudu_static_compr_hdr |
False |
MUST match the RU's own compression= in /etc/ru_config.cfg -- in standalone mode the RU is the authority and a mismatch corrupts every U-plane section while all counters look healthy: static_compressed -> true dynamic_compressed -> false <-- what THIS RAN650 carries, and what the proven config uses Read it before every run: ssh -i root@ 'grep ^compression /etc/ru_config.cfg' |
ocudu_liteon_center_freq_mhz |
reservation |
LiteOn FlexFi (ocudu_oru: liteon) - its complete radio parameter set Source: configs/ocudu-liteon-72/gnb-ocudu-liteon.yml (srsRAN's LiteOn reference design + the COSMOS fixes), WORKING end to end 2026-09-14 and 2026-10-05 (DL 210 / UL 42.6 Mbit/s). |
ocudu_liteon_bw_mhz |
reservation |
|
ocudu_liteon_rf_fallback_center_mhz |
3750 |
|
ocudu_liteon_rf_fallback_bw_mhz |
100 |
|
ocudu_liteon_ru_bandwidth_mhz |
omit |
|
ocudu_liteon_ru_mac |
e8:c7:4f:25:89:39 |
this RU's eth1 (fronthaul) MAC |
ocudu_liteon_du_mac |
00:11:22:33:44:66 |
the RU's du-mac-address |
ocudu_liteon_pci |
1 |
|
ocudu_liteon_sector_id |
1 |
|
ocudu_liteon_compr_method_prach |
bfp |
|
ocudu_liteon_compr_bitwidth_prach |
9 |
|
ocudu_liteon_nof_antennas_dl |
4 |
4T4R, NOT the reference's 2T1R: the RU runs dl-ul-layer 4T4R, and a DU sending 2 DL streams never gives it a complete set (DU reception went from 439,200 to 1,756,800 bytes, exactly 4x, at 4 ports). |
ocudu_liteon_dl_port_id |
[0, 1, 2, 3] |
|
ocudu_liteon_nof_antennas_ul |
4 |
|
ocudu_liteon_ul_port_id |
[0, 1, 2, 3] |
|
ocudu_liteon_prach_port_id |
[4, 5, 6, 7] |
The RU carries eAXC_id 4 5 6 7; srsRAN refuses fewer PRACH ports than reception antennas. |
ocudu_liteon_ignore_prach_start_symbol |
omit |
|
ocudu_liteon_prach_cp_enabled |
True |
|
ocudu_liteon_prach_config_index |
159 |
|
ocudu_liteon_prach_freq_start |
2 |
|
ocudu_liteon_zero_correlation_zone |
0 |
|
ocudu_liteon_nof_cb_preambles_per_ssb |
4 |
PRACH TA is only right for preamble 0 on this RU (2026-09-20: the reported TA is a lookup of the preamble index, so Msg3 lands outside the FFT window). 4 = srsRAN's minimum; effectively single-UE. |
ocudu_liteon_preamble_rx_target_pw |
omit |
|
ocudu_liteon_ssb_block_power |
omit |
|
ocudu_liteon_pucch_resource_common |
omit |
|
ocudu_liteon_mcs_table |
omit |
the reference design: no qam64 cap (that is a Benetel fix) |
ocudu_liteon_olla_target_bler |
omit |
|
ocudu_liteon_tdd_period_slots |
omit |
omit = srsRAN's default pattern, as the proven config |
ocudu_liteon_tdd_dl_slots |
omit |
|
ocudu_liteon_tdd_dl_symbols |
omit |
|
ocudu_liteon_tdd_ul_slots |
omit |
|
ocudu_liteon_tdd_ul_symbols |
omit |
|
ocudu_liteon_t1a_max_cp_dl |
350 |
|
ocudu_liteon_t1a_min_cp_dl |
200 |
|
ocudu_liteon_t1a_max_cp_ul |
350 |
|
ocudu_liteon_t1a_min_cp_ul |
200 |
|
ocudu_liteon_t1a_max_up |
450 |
2026-09-28: 300/0 left the RU marking 64 % of the U-plane RX_EARLY; swept on the RU's own counters, 450/350 is the centre of the 93 %-on-time plateau. |
ocudu_liteon_t1a_min_up |
350 |
|
ocudu_liteon_ta4_max |
1960 |
At 500 every PRACH message from this RU arrived just after the window closed; 1960 = 0 misses. |
ocudu_liteon_ta4_min |
0 |
|
ocudu_liteon_ignore_ecpri_payload_size |
False |
|
ocudu_liteon_ignore_ecpri_seq_id |
True |
the LiteOn does not maintain the eCPRI sequence id |
ocudu_liteon_max_proc_delay |
10 |
At 15 the MAC ran so far ahead that this RU's PRACH reached the scheduler after the RAR window ("Could not transmit RAR within the window"); 10 -> RAR sent, Msg3 crc=OK, registered (2026-09-14). |
ocudu_liteon_iq_scaling |
5.0 |
|
ocudu_liteon_fh_vlan |
4 |
|
ocudu_liteon_static_compr_hdr |
True |
the LiteOn uses STATIC compression headers (the Benetel dynamic) |
ocudu_foxconn_center_freq_mhz |
3750 |
Foxconn RPQN-7801 (ocudu_oru: foxconn) - its complete radio parameter set Source: configs/ocudu-foxconn-72/gnb-ocudu-foxconn.yml, WORKING end to end 2026-10-26 (srv18.osc, TCP DL 176-181 / UL 136-144 Mbit/s). |
ocudu_foxconn_bw_mhz |
100 |
|
ocudu_foxconn_rf_fallback_center_mhz |
3750 |
|
ocudu_foxconn_rf_fallback_bw_mhz |
100 |
|
ocudu_foxconn_ru_bandwidth_mhz |
omit |
|
ocudu_foxconn_ru_mac |
6c:ad:ad:00:0d:60 |
RRH_SRC_MAC_ADDR in the RU's RRHconfig_xran.xml |
ocudu_foxconn_du_mac |
00:11:22:33:44:66 |
RRH_DST_MAC_ADDR |
ocudu_foxconn_pci |
3 |
|
ocudu_foxconn_sector_id |
3 |
|
ocudu_foxconn_compr_method_prach |
bfp |
|
ocudu_foxconn_compr_bitwidth_prach |
9 |
|
ocudu_foxconn_nof_antennas_dl |
2 |
TX branches 2 and 4 read ~6 dB below 1 and 3: 4 DL ports gave 40-80 % BLER, ports 0+2 give 2-6 %. |
ocudu_foxconn_dl_port_id |
[0, 2] |
|
ocudu_foxconn_nof_antennas_ul |
4 |
|
ocudu_foxconn_ul_port_id |
[0, 1, 2, 3] |
|
ocudu_foxconn_prach_port_id |
[4, 5, 6, 7] |
RRH_EAXC_ID_TYPE3 = 0x4..0x7 |
ocudu_foxconn_ignore_prach_start_symbol |
omit |
|
ocudu_foxconn_prach_cp_enabled |
True |
|
ocudu_foxconn_prach_config_index |
159 |
|
ocudu_foxconn_prach_freq_start |
2 |
|
ocudu_foxconn_zero_correlation_zone |
0 |
|
ocudu_foxconn_nof_cb_preambles_per_ssb |
omit |
|
ocudu_foxconn_preamble_rx_target_pw |
omit |
|
ocudu_foxconn_ssb_block_power |
omit |
|
ocudu_foxconn_pucch_resource_common |
omit |
|
ocudu_foxconn_mcs_table |
omit |
|
ocudu_foxconn_olla_target_bler |
omit |
|
ocudu_foxconn_tdd_period_slots |
omit |
|
ocudu_foxconn_tdd_dl_slots |
omit |
|
ocudu_foxconn_tdd_dl_symbols |
omit |
|
ocudu_foxconn_tdd_ul_slots |
omit |
|
ocudu_foxconn_tdd_ul_symbols |
omit |
|
ocudu_foxconn_t1a_max_cp_dl |
470 |
C-plane 285-470: with srsRAN's reference 250-420 the RU counted ~28 % of C-plane late (xRN c_late). |
ocudu_foxconn_t1a_min_cp_dl |
285 |
|
ocudu_foxconn_t1a_max_cp_ul |
470 |
|
ocudu_foxconn_t1a_min_cp_ul |
285 |
|
ocudu_foxconn_t1a_max_up |
196 |
|
ocudu_foxconn_t1a_min_up |
80 |
|
ocudu_foxconn_ta4_max |
1960 |
|
ocudu_foxconn_ta4_min |
25 |
|
ocudu_foxconn_ignore_ecpri_payload_size |
True |
|
ocudu_foxconn_ignore_ecpri_seq_id |
False |
|
ocudu_foxconn_max_proc_delay |
omit |
No max_proc_delay: at 10 the DU dropped ~6000 DL U-plane messages/s as late (2026-10-26). |
ocudu_foxconn_iq_scaling |
1.0 |
|
ocudu_foxconn_fh_vlan |
4 |
|
ocudu_foxconn_static_compr_hdr |
True |
RRH_CMPR_HDR_PRESENT = 0 on the RU applies to BOTH directions: with a DL udCompHdr the RU reads the IQ one byte off, trips "AD9025 PA protection" and no UE ever sees the cell (the whole 09-14 state). |
ocudu_gnb_overrides |
0 keys | A YAML mapping deep-merged into the rendered gNB config (after the O-RU set), e.g. for emulated UEs: ocudu_gnb_overrides: {cu_cp: {inactivity_timer: 7200}, cell_cfg: {ta: {ta_cmd_offset_threshold: -1}}} |
ocudu_ue_mode |
modem |
modem = steps 4-6 register the Quectel on the second node and measure; none = the cell only (no second node needed; the contract records ue_mode none). |
ocudu_at_port |
/dev/ttyUSB2 |
|
ocudu_qmi_dev |
auto |
auto = the Quectel's own QMI device and netdev, found from /dev/quectel-qmi or /dev/quectel-at (a UE node can carry two modems: node1.rrail has a SIMCom on cdc-wdm0/wwan0 and the Quectel on cdc-wdm1/wwan1). |
ocudu_wwan |
auto |
|
ocudu_nr5g_bands |
77:78 |
|
ocudu_remote_target |
8.8.8.8 |
|
ocudu_ping_count |
10 |
|
ocudu_iperf_secs |
10 |
|
ocudu_iperf_port |
5201 |
|
ocudu_ru_cycle |
True |
The O-RU only starts transmitting on a FRESH boot that comes up against a correctly-timed fronthaul: restarting the DU under a running RU does not revive it (every counter stays healthy and TXMeanPower reports "No signal"). |
ocudu_ru_off_wait_s |
30 |
|
ocudu_ru_boot_timeout_s |
420 |
|
ocudu_gnb_timeout_s |
120 |
|
ocudu_reg_timeout_s |
240 |
To change a parameter, put only the keys you change in a file of your own and pass it to any mode. A key the tutorial does not have, or a value of the wrong type, is refused before anything starts:
cat > my.yml <<'EOF'
ocudu_iperf_secs: 20
EOF
cosmos-tutorial run ocudu-72 --params my.yml
Most of the fronthaul parameters were measured against this radio and must match what the O-RU itself carries: the VLAN, the compression-header mode, the TDD pattern and the PRACH ports. Change them only together with the O-RU. To change the tutorial itself, cosmos-tutorial copy ocudu-72 puts an editable copy in the current directory, which then takes precedence over the installed one.
The O-RU's state decides the result. A commercial O-RU keeps what the last user wrote into it, and a DU that manages the Benetel over the M-Plane leaves different values from the ones OCUDU needs:
| Left by an M-Plane DU | OCUDU | |
|---|---|---|
mimo_mode |
1_2_3_4_4x4 |
1_2_3_4_4x2 |
tdd_pattern |
DDDSU |
DDDDDDDSUU |
compression |
static_compressed |
dynamic_compressed |
| M-Plane | enabled | disabled (nomplane) |
Each mismatch alone gives a clean fronthaul and a useless radio. The bring-up therefore checks the O-RU against its profile (host/oru-profiles/benetel-ran650.yml) and refuses to start the DU until it matches. The profile lists which registers the O-RU sets for itself at boot; those are verified, never written.
The management mode moves the PRACH streams. On the Benetel, M-Plane mode sends PRACH on eAxC 8,9 and standalone mode on 4,5,…; OCUDU's prach_port_id asks for the standalone ones. Run OCUDU against a radio left in M-Plane mode and no UE ever attaches, while every downlink and uplink counter reads healthy. Establish the O-RU's mode first; never change the DU's port IDs to follow an eAxC you saw in a capture.
ether_rx rx_bytes is the honest counter. Fronthaul drop counters stay at 0 when nothing arrives at all. TX_TOTAL on the O-RU counts uplink frames and is 0 with no UE scheduled; it says nothing about whether the O-RU radiates. Judge radiation by a modem scan and the DU's PRACH counters.
Throughput depends on the O-RU, the CPU platform and the gains. The modem sits metres from the O-RU, so both the DU's iq_scaling and the O-RU's uplink gain are kept low; too high clips the downlink or saturates an uplink branch without any error. Every OCUDU build's measurements, by O-RU and CPU platform, are on OCUDU performance by build.
Other O-RUs. The packaged bring-up configures and verifies the Benetel. OCUDU also works with the LiteOn FlexFi (standalone by design, so its state is the same for every DU) and the Foxconn RPQN-7801, each with its own complete parameter set, never one edited from another radio's. The LiteOn needs two timing values that differ from its reference configuration, ru_ofh.ta4_max: 1960 and expert_phy.max_proc_delay: 10, and its jumboframe setting re-applied after every boot. The Foxconn procedure is on OCUDU with the Foxconn O-RU.
Many failures on this rig share one signature: every fronthaul counter looks healthy and still no UE attaches, or throughput is a fraction of the expected value. Check the O-RU's state before tuning anything else. To place a failure in the attach sequence, compare the last milestone reached with the flow on 5G SA fundamentals.
| Symptom | Cause | Fix |
|---|---|---|
gNB started, 0 fronthaul drops, and no UE sees a cell |
the O-RU still carries another DU's compression, mimo_mode or TDD pattern |
let the bring-up apply the OCUDU state and reboot the O-RU from its own OS; a power cycle alone does not pick up a changed configuration on this radio |
| The DU detects no PRACH although the downlink and uplink counters are clean | the O-RU was left in M-Plane mode, so its PRACH moved to eAxC 8,9 | bash /root/benetel-ru-mode.sh nomplane (the bring-up does it); do not change prach_port_id to follow an observed eAxC |
| RSRP looks fine but CQI 2, downlink about 3 Mbit/s, about 75 % downlink errors | the O-RU carries mimo_mode 1_2_3_4_4x4 and tdd_pattern DDDSU from an M-Plane DU |
restore the OCUDU state (the bring-up's O-RU check) |
ether_rx rx_bytes 0, no PRACH ever |
the fronthaul NIC is not promiscuous; OCUDU's raw socket only sees frames the NIC accepts | ip link set DATA1a promisc on (the bring-up does it) |
Invalid CPU core selected 'N' |
isolcpus hides the isolated cores from OCUDU's affinity check |
launch the gNB inside taskset -c 0-<ncpus-1> (the bring-up does it) |
Throughput low with every fronthaul counter clean (rx_late=0, no skipped symbols) |
main_pool_cpus includes the hyperthread siblings of the isolated fronthaul cores; no counter reports the contention |
compare the affinities with /sys/devices/system/cpu/cpu*/topology/thread_siblings_list; the rendered configuration keeps the main pool off them |
The O-RU reports No signal; the DU counts nof_late_up_dl in the thousands per second and ether_tx tx_bytes stays 0 |
with OCUDU 26.10, expert_phy.max_proc_delay: 15 makes the DU drop every downlink packet as late on the Benetel |
leave max_proc_delay at its default for the Benetel, as the template does |
Every uplink symbol dropped: no section was decoded correctly, or the O-RU transmits garbage |
a compression-header flag differs from what the O-RU sends or expects | match enable_ul_static_compr_hdr (and enable_dl_static_compr_hdr) to the O-RU's compression= setting; this Benetel needs false (ocudu_static_compr_hdr) |
| Downlink BLER stuck at 20–30 % while link adaptation has already backed off | iq_scaling too high clips the downlink at this short range |
keep ocudu_iq_scaling at 5.5 |
| Uplink collapses to a few kbit/s; one receive port reads exactly 0 dBFS on the O-RU | the O-RU's uplink gain saturates a branch at this range | keep ocudu_ru_ul_gain_correction: 0 |
The modem camps (+QENG: "servingcell","NOCONN"…) but stays at +C5GREG: 0,0; the AMF logs Registration reject [23] |
after a cold start the modem still holds an identity from a core that has since restarted, is rejected, and backs off for 12 min | reboot the modem (AT+CFUN=1,1); the helper does it when registration stalls |
LiteOn: PRACH missed incoming User-Plane PRACH messages, then no uplink PRACH context |
the uplink receive window closes just before the LiteOn's PRACH arrives | ru_ofh.ta4_max: 1960 |
LiteOn: preambles detected, then Could not transmit RAR within the window |
the MAC runs so far ahead that the PRACH reaches the scheduler after the window's last downlink slot | expert_phy.max_proc_delay: 10 |
LiteOn: PRACH detected but Msg3 crc=KO intermittently, while the downlink is good |
the reported timing advance is correct only for preamble index 0, so Msg3 lands outside the cyclic prefix | retry (AT+CFUN=1,1); for a single UE, prach: nof_cb_preambles_per_ssb: 4; do not raise gains |
| LiteOn: downlink BLER around 70 % on a 2× Xeon Gold 6126 server while another platform is clean | not explained; platform-dependent | compare on another CPU platform before tuning the radio |
Registers, gets an address, 100 % ping loss; ip -s link show $WWAN shows TX errors rising by one per packet |
the modem's USB data endpoint is stuck | USB device reset of the modem (the helper does it when it sees TX errors) |
omf tell … No such node for the O-RU |
the O-RU was named by a short name | use the name the tutorial resolves (cosmos_addr.py … --field omf), without -d |
omf tell refuses: not reserved right now, reserved by …, or run omf tell from console.… |
O-RU power needs your own active O-RU reservation, used from the console of its target domain | reserve in the O-RU scheduler, or log in to the console the message names |
Preflight: OMF_NODES selected 1 node(s) |
a node is Unknown (UNREACHABLE) in omf stat and does not answer SSH |
check that it is on and booted, power-cycle it with omf tell, and report the chassis manager to the testbed operators |
After any mode, run the teardown from the console. On the UE node it stops the data call and hands the modem back to ModemManager; on the O-DU server it stops the ocudu-72 and ocudu-iperf units and the Open5GS network functions. The O-RU is left powered, because it takes about five minutes to come back.
OMF_NODES=<du-node>,<ue-node> cosmos-tutorial run ocudu-72 --mode teardown
Never stop the gNB with pkill: pkill -f apps/gnb/gnb matches your own shell's command line. Power the O-RU off only if nobody else needs it, from the console of your O-RU reservation's domain:
omf tell -a offh -t $(python3 $(cosmos-tutorial which ocudu-72)/cosmos_addr.py --value auto oru benetel --du <du-node> --field omf)
omf tell -a offh -t <du-node>,<ue-node>
If others use the O-DU server for work without a fronthaul, restore its kernel command line as in Removing isolation, or reload ocudu.ndz.
Package cosmos-experiments-ocudu 0.1.0-31 - on every console it is released to (cosmos-tutorial list shows the installed version). The code is installed on the console in /usr/share/cosmos-experiments/ocudu-72/. To change anything, cosmos-tutorial copy ocudu-72 puts an editable copy in the current directory, and cosmos-tutorial run ocudu-72 then runs that copy instead of the installed one (cosmos-tutorial which ocudu-72 shows which will run). Repository (GitLab account needed, not needed to run or change it): packages/cosmos-experiments-ocudu/usr/share/cosmos-experiments/ocudu-72.
Last verified 2026-11-19: ansible mode PASS from the 0.1.0-31 code, on sb1 (2× Intel Xeon Gold 6226, 24C/24T, Benetel RAN650 at 3350 MHz / 100 MHz, Quectel modem): 0 fronthaul drops, DL 175 / UL 74.1 Mbit/s, ping 27 ms. Before that, 2026-10-30: three-mode parity PASS from the installed cosmos-experiments-ocudu package, cold start (all devices off), on rrail (1× AMD EPYC 9355P, Benetel RAN650 and Quectel modem): DL 157–200 / UL 75–77 Mbit/s. Results by build, O-RU and CPU platform are on OCUDU performance by build.
References: srsRAN Project documentation (the ru_ofh reference) · OCUDU documentation · O-RAN Alliance specifications (Open Fronthaul CUS-Plane and M-Plane) · linuxptp · Open5GS documentation · iperf3 · on this wiki: OCUDU on COSMOS · O-RAN and the 7.2x fronthaul · O-RU pool · Preparing a real-time RAN host · The commercial UE.
Author(s): COSMOS team · Package: cosmos-experiments-ocudu (cosmos-tutorial run ocudu-72) · Tags: ocudu, srsran, oran, 7.2, fronthaul, o-du, benetel, liteon, quectel