Support for Baicells 452 base station

[mozartatplay]

Has anybody tested Baicells 452 (https://www.baicells.com/product/details?id=3e287ca0-1703-4c05-b242-14bb260bf969) 3GPP LTE TDD base station. Given that It talks standard 3GPP interfaces (S1 for LTE, N2/N3 for 5G) - I assume it will work - want to order some but hoping I can have the risk of it not working eliminated before buying

Thanks
David

[MrBrownieCGN]

Hi @mozartatplay — we don't have a Baicells 452 specifically, but we run Baicells NOVA 227 (LTE FDD eNB) and Neutrino 220 (5G NR gNB) with Open5GS in production. The 452 is the same Baicells stack family (TDD LTE), so most of the operational quirks we hit are likely going to apply. Sharing what to expect so you can decide on the order with eyes open:

The good news: S1AP works cleanly out of the box. NGAP on the 5G-capable Baicells units works fine too. SNMP+TR-069 management integrates well — once you configure the right ACS URL for initial provisioning, the rest is straightforward.

Three things to be aware of, all fixable but not obvious:

1. MTU = 1500 is mandatory on the route to the eNB. The Mindspeed T2200 SoC inside the Baicells units hangs on SCTP DATA chunks larger than 1500 bytes. We hit this as recurring 84-second SCTP flap loops with our NOVA 227 — looked like a transport bug for weeks until we traced it to path-MTU. Fix: per-route MTU 1500 in netplan/iproute2 for every Baicells IP. Once enforced at the operator side, completely silent. Worth setting from day one.

2. LGW / SIPTO must stay disabled. Baicells lets you enable Local Gateway / Selective IP Traffic Offload, which sounds attractive (lower-latency intra-eNB traffic). Don't. Enabling it bypasses the Open5GS UPF for traffic between two UEs on the same cell, and the Baicells implementation has issues — session churn, PGW_S5U_TEID=0 errors. Make sure LGW is disabled in your master config and verify on every unit.

3. SCTP_STATUS getsockopt destabilises the units. Don't poll SCTP socket status via getsockopt(SCTP_STATUS) from the core side — for some reason it makes the Baicells SCTP stack flap. We had to switch to ioctl(SIOCOUTQ) instead. If you're not writing custom monitoring code that pokes SCTP socket internals, you won't hit this; just flagging.

A few smaller quirks: SNMPv2c needs an explicit snmpd enable command via CLI on the unit; some timer values (T3396 etc.) are in different units than the menu labels suggest; mibcli settings may not persist across reboots without a commit. None of these are blockers — just be ready for a couple of hours of "why is this not working as documented" the first time.

If you're planning to use 5G SA later and need IMS-Voice over PS, also check that the Aurora/Neutrino-class Baicells radios advertise EPS Network Feature Support correctly — Open5GS got better at the per-UE handling recently (#4504 in flight, #4425 already in main).

Net-net: the 452 will almost certainly work for your S1 use case. The pitfalls above are operational, not architectural.