gNB stops working on Raspberry Pi 5

[gruyaume]

I am running the srsRAN gnb on a Raspberry Pi 5 / 8 GB. While the gnb works for a few hours, it then suddenly stops working. It doesn't crash though, it just stops functioning and I need to go manually restart it. I am no expert in RF or srsRAN and I would appreciate some help in identifying the root cause and the steps I can take to remediate it.

Note: I tried lowering the bandwidth to 10MHz, but the phones would not connect. I don't know if my config was wrong or if the phones I have don't support 10MHz in n77.

Setup

• srsRAN: 3ed363d
• RF: n77, 20MHz bandwidth
• Radio: USRP B205 mini
• UEs: COTS UEs (iPhone 16, Crosscall, Moto G5)
• OS: Ubuntu 24.04 (6.8.0-1047-raspi)

Config

Here's my config file:

cu_cp:
  amf:
    addr: 10.202.0.3
    port: 38412
    bind_addr: 10.202.0.5
    supported_tracking_areas:
      - tac: 1
        plmn_list:
          - plmn: "99901"
            tai_slice_support_list:
              - sst: 1
  inactivity_timer: 300
  security:
    nea_pref_list: nea2,nea1
    nia_pref_list: nia2,nia1
cu_up:
  ngu:
    socket:
      - bind_addr: 10.202.0.5
ru_sdr:
  device_driver: uhd
  device_args: type=b200
  clock: internal
  srate: 23.04
  tx_gain: 80
  rx_gain: 40

cell_cfg:
  dl_arfcn: 665000
  band: 77
  channel_bandwidth_MHz: 20
  common_scs: 30
  plmn: "99901"
  tac: 1
  pdcch:
    dedicated:
      ss2_type: common
      dci_format_0_1_and_1_1: false
  prach:
    prach_config_index: 160
  pdsch:
    mcs_table: qam64
  pusch:
    mcs_table: qam64

log:
  filename: stdout
  all_level: warning

remote_control:
  enabled: true
  bind_addr: 0.0.0.0
  port: 8001

metrics:
  autostart_stdout_metrics: true
  enable_json: true 

pcap:
  mac_enable: disable
  ngap_enable: disable

Observability

Logs

I do see these logs

2026-02-13 23:38:58.7192026-02-14T04:38:58.702323 [RF      ] [W] Real-time failure in RF: late
2026-02-13 23:38:58.7192026-02-14T04:38:58.702345 [RF      ] [W] Real-time failure in RF: underflow

I have attached more logs here:

Logs-2026-02-14 06_54_41.txt

Metrics

I have a dashboard for my srsRAN deployment, but I don't see a metric that points to a clear root cause.

[s5uishida]

@gruyaume

Did the srsRAN_Project gNB work on the RP5 with 4 cores?
I ran an older revision of srsRAN_Project in ZMQ mode on a VM (host is x64 PC), but it required 5 CPU cores. (No real devices used)

[gruyaume]

Yes, the srsRAN gnb starts fine on the Pi5 with 4 cores, and it integrates with the 5G core, phones connect, etc. But after a couple of hours it stops working. I will try with Ubuntu Real Time Kernel and see if that helps.

[andrepuschmann]

Hey @gruyaume - thanks for reporting this. Just to be sure, the behavior you're describing is happening after ~3h of running the gNB? You restart and it runs again fine for 3h?

[gruyaume]

The gnb worked fine for ~12h, then I had to restart it. Then I made changes to the setup so I don't know if it would have worked fine for that same amount of time.

Since then I made the following adjustments to improve the gNB stability: 

• Use Ubuntu Real-Time
• Use a high real-time scheduling priority (with chrt)
• Set scaling governor to Performance (with srsRAN performance script)
• Disable DRM KMS polling (with srsRAN performance script)

I will be running this environment for a couple of days to see if the situation is improved.

[andrepuschmann]

Ok, thanks for confirming.
Anyway, in case you're not aware we've moved the development of srsRAN Project to OCUDU, see the announcement here #1470

It would be good if you could update your setup and use OCUDU instead. If the issue still persists with your new setup and OCUDU, please open an issue in the new repo.

[gruyaume]

Ok I updated the setup to OCUDU and I'll open the issue over there if it comes back.

For your information, we also updated the Ella Core docs to use OCUDU in co-hosted deployments: docs.ellanetworks.com/how_to/co_host_with_ocudu