Jetson Agx Orin with OAI 5G: E2E with Demapper Issue

[VivOwoH]

Hi,
I'm running OpenAirInterface 5G NR gNB with original demapper using --loader.demapper.shlibversion _capture to log received soft symbols. I'm confused why downlink BLER does not decrease like uplink BLER.

Setup:

graph LR
		Host[Host PC<br/>Ubuntu 24.04]
		
		subgraph J1["Jetson AGX Orin #1<br/>(gNB & Core)"]
		gNB[OAI gNB] --> Core[OAI 5G Core]
		Core --> ExtDN[oai-ext-dn<br/>]
		end
		
		U1[USRP B210<br/>gNB Side]
		
		ATT40[40 dB<br/>Attenuator]
		
		U2[USRP B210<br/>UE Side]
		
		subgraph J2["Jetson AGX Orin #2<br/>(UE)"]
		UE[OAI NR UE<br/>oai-nr-ue] --> TUN[oaitun_ue1<br/>]
		end
		
		Host ---|USB3| J1
		Host ---|USB3| J2
		
		J1 ---|USB3| U1
		
		U1 --> ATT40
		ATT40 --> U1
		
		ATT40 -->|RX2 ← TX/RX<br/>Downlink<br/>Band 78<br/>BW: 24 PRBs<br/>3604.8 MHz | U2
		U2 -->|TX/RX → RX2<br/>Uplink| ATT40
		
		U2 ---|USB3| J2
		
		TUN -.->|5G Tunnel<br/>End-to-End| ExtDN
    
    style J1 fill:#565656,stroke:#333,stroke-width:2px,color:#ddd
		style J2 fill:#565656,stroke:#333,stroke-width:2px,color:#ddd		
		style U1 fill:#ffffcc,stroke:#333,stroke-width:2px,color:#000
		style U2 fill:#ffffcc,stroke:#333,stroke-width:2px,color:#000
		style ATT40 fill:#ffcccc,stroke:#333,stroke-width:2px,color:#000
		style ExtDN fill:#ccffcc,stroke:#333,stroke-width:2px,color:#000
		style TUN fill:#ccf0ff,stroke:#333,stroke-width:2px,color:#000
		style gNB fill:#fff,stroke:#333,stroke-width:1px,color:#000
		style Core fill:#fff,stroke:#333,stroke-width:1px,color:#000
		style UE fill:#fff,stroke:#333,stroke-width:1px,color:#000

Loading

The .env file is as below:

USRP_SERIAL=310FC9A
USRP_SERIAL_UE=310735D

MINI_CONFIG=../common/mini_nonrf_config.yaml
GNB_CONFIG=../common/gnb.sa.band78.24prbs.conf

GNB_IMAGE=oai-gnb-cuda
UE_IMAGE=oai-nr-ue-cuda

GNB_TAG=latest
UE_TAG=latest
AMF_TAG=v2.1.0-1.2
SMF_TAG=v2.1.0-1.2
UPF_TAG=v2.1.0-1.2
TRFGEN_TAG=latest

##############################
#### Additional gNB Options
##############################

GNB_EXTRA_OPTIONS=--thread-pool 6,7,8,9,10,11

GNB_EXTRA_OPTIONS=--loader.demapper.shlibversion _orig
GNB_EXTRA_OPTIONS=--loader.demapper.shlibversion _capture --thread-pool 6

# 24 PRBs config
UE_EXTRA_OPTIONS=-C 3604800000 -r 24 --numerology 1 --ssb 24 --thread-pool 4,5

Problem:

BLER logs show that DLSCH BLER per frame remains high, unlink ULSCH BLER. Why does this happen, and how to resolve it?

Thanks for your help!

[SebastianCa]

Hi @VivOwoH,
packet counts are extremely low with just a single error. Could you run an iperf3 to generate traffic?

Which version of the OAI do you use for the soft-UE?

[VivOwoH]

Hi, thanks for your reply! For the version I am not exactly sure, I'm using the latest oai-nr-ue-cuda docker image as per research kit tutorial in documentation. But the DLSCH issue was indeed too little traffic that is skewing the BLER statistics. Appreciate the help!

[wirelesssun]

Dear VivOwoH, Thanks For YOUR System Architecture, If I Only have one Jetson AGX Orin #1 (gNB & Core), I have not the Jetson AGX Orin #2(UE); Whether I can use a COTS UE to Access and Communicate with the left USRP B210
gNB Side by Air Interface ? Many Thanks!

[SebastianCa]

Hi @wirelesssun,
yes, you can connect a COTS UE to the gNB (see Fig. 19).

[wirelesssun]

Dear SebastianCa, Thanks For the Immediately respomse from YOU ^_^

I have see the Fig. 19 Overview of the deployed setup. See Ettus OAI reference architecture for details.

But the COTS was Quectel RM520N-GL, Has any friend successfully Accessed using a mobile phone? Such as Apple, Samsung, OPPO, HUAWEI and so on.......

[SebastianCa]

I would suggest to open a thread in discussions to get community feedback.

[wirelesssun]

#1063

[wirelesssun]

Dear YOU, How could we know the USRP_SERIAL and the USRP_SERIAL_UE, Command Lines?

The .env file is as below:

USRP_SERIAL=310FC9A
USRP_SERIAL_UE=310735D

MINI_CONFIG=../common/mini_nonrf_config.yaml
GNB_CONFIG=../common/gnb.sa.band78.24prbs.conf

GNB_IMAGE=oai-gnb-cuda
UE_IMAGE=oai-nr-ue-cuda

GNB_TAG=latest
UE_TAG=latest
AMF_TAG=v2.1.0-1.2
SMF_TAG=v2.1.0-1.2
UPF_TAG=v2.1.0-1.2
TRFGEN_TAG=latest

[SebastianCa]

Hi @wirelesssun,
just run uhd_find_devices directly on the host.

[VivOwoH]

Hi Sionna team,

Thanks for your help previously! Following up on my system - I am now running into some practical issues testing with USRP B210 that I would really appreaciate some help.

Issue 1: Cannot set specific MCS
I would like to lock the system to a different modulation order (it currently idles at QPSK but I want to test 16QAM and above). I have tried setting both min and max MCS in the .env file or the gnb configuration file:

--MACRLCs.[0].dl_max_mcs 16 --MACRLCs.[0].dl_min_mcs 16
--MACRLCs.[0].ul_max_mcs 16 --MACRLCs.[0].ul_min_mcs 16

But the MAC scheduler doesn't seem to respect these parameters, and the system still varies MCS dynamically. Is there another way to force scheduler to use a fixed MCS?

Issue 2: USRP B210 buffer overflow during traffic tests

When I run uplink iperf3 tests (docker exec -ti oai-nr-ue iperf3 -u -t 10 -i 1 -b 1M -B 12.1.1.2 -c 192.168.72.135) to stress the system, I get USRP buffer overflows ('L' and 'U' in logs). This eventually crashes the link:

ERROR: CODE_GENERATION_ERROR (trx_usrp_read:763)
rx_rf: Asked for 7680 samples, got 5123 from USRP
[NR_MAC] W UE 46fd: Detected UL Failure on PUSCH after 1 PUSCH DTX
[NR_RRC] Timer T310 expired - Radio link failure!

The buffer issue happens consistently with uplink traffic and also when using higher PRB configuration. Is this bandwidth limitation issue with B210 or otherwise? What parameters could I adjust to help?

Summary of questions:
Setup: 2x USRP B210, 2x Jetson AGX Orin, Band 78 @ 3604.8 MHz, 24 PRBs

1. How to overwrite to a fixed MCS?
2. Has anyone dealt with B210 buffer overflow issues when testing with uplink traffic?

I realize these could be OAI or USRP issues rather than Sionna-related, but I haven't had much luck on other forums and thought someone here might have encountered similar challenges.

Thanks in advance!

Screenshot log for reference:

[SebastianCa]

Hi @VivOwoH,
as far as I know there is no min_mcs parameter in OAI, thus, you see no effect. Setting the max_mcs should work (see neural receiver as example).

Regarding the buffer overflows: did you activate any specific tutorial (e.g., neural receiver)? Did you set the correct power mode? The thread assignment can also help (reserve a specific core for the ru-thread as defined in the gnb config).