Release Notes#
This page details new features, bug fixes, performance improvements, limitations, and documentation updates by release.
Note
With the Aerial Testbed 1.0 release, this project has been renamed from “ARC-OTA” to “Aerial Testbed”. The “ARC-OTA” name is still used to refer to legacy releases below.
Aerial Testbed Release 1.0 (April 2026)#
New Features#
Aerial Testbed (ATB) is now supported on DGX Spark. An over-the-air testbed can now be implemented on DGX Spark. In this release single-cell operation is supported.
dApp (distributed App) capabilities have been expanded. Users can now write a dApp in Python, C, CUDA or PyTorch and easily deploy on ATB. The user dApp can access various signals, for example front-haul time-frequency resource block I/Q samples, CSI, and other signals, in the PUSCH pipeline using the dApp subscription manager. The dApp framework enables real-time machine learning to support applications such as ISAC (Integrated Sensing and Communications) and spectrum management. Refer to the Data Lake and dApp sections of this documentation for full details.
Data Lake has new capabilities to save the downlink transport block.
The OAI GPU-accelerated Soft-UE is supported on Jetson-ORIN and DGX Spark with higher data rates and stability.
ATB installation is now simplified with a new installation script. For step-by-step commands (for DGX Spark and Ubuntu Server), refer to the Automated Installation of Aerial CUDA-Accelerated RAN section in the installation guide.
Known Issues and Workarounds#
On occasion the L2 can stop scheduling, the UE will repeatedly perform RACH/random access and the following error codes 0x03 and 0x06 will be logged in the cubb logs. The errors are an indication that there is late L2 response. The gNB will need to be restarted.
In a RF cabled or OTA 2 cell configuration, there are occasional failures in which UE traffic drops to zero in cell 2. The traffic in cell 1 is not impacted. The gNB will need to be restarted.
In a 2-cell OTA configuration the DL and UL throughput is observed to be reduced in each cell when compared to a single cell configuration.
When using 2-layers in the uplink with SRS enabled, high BLER can be observed. The gNB needs to be restarted.
Notes on SSB block power:
If using Foxconn RU or WNC RU with digital-power-scaling legacy, the L1
cuPHY-CP/cuphycontroller/config/cuphycontroller_P5G_FXN_GH.yamlshould be configured as follows:fs_offset_dl: 7 fs_offset_ul: -5
And the L2
targets/PROJECTS/GENERIC-NR-5GC/CONF/gnb-vnf.sa.band78.273prb.aerial.confshould be configured as follows:ssPBCH_BlockPower = -34
If using digital-power-scaling o-ran, the L1 should be configured as follows:
fs_offset_dl: 0 fs_offset_ul: 0
And the L2 should be configured as follows:
ssPBCH_BlockPower = -11
There is a bug in the WNC RU where, if multiple DUs send fronthaul (FH) packets, UEs will see MIB and SIB but will not receive Msg2; at the L1 you will only see CRC errors.
ARC-OTA Release 1.7 (January 2026)#
New Features and Bug Fixes#
GPU (Jetson AGX) accelerated soft-UE: Support for 30 MHz RF BW. Limited data rates (30/30 Mbps DL/UL).
Multi-UE support: DL and UL traffic to 100 UEs (using UE simulator over eCPRI)
UL Heavy TDD Pattern (DSUUU): OTA validation with OAI L2+
Data Lake:
Real-time gNB I/Q sample capture from 4 cells
Real-time soft UE data collection enables correlation across gNB and UE.
Data is accessible to E3 agent and in RAM database for use by dApps Framework.
The system can be configured to not capture data in the RAM database if the data is only to be consumed by the E3 agent.
MAC PDUs are now captured even when PUSCH CRC fails.
Aerial CUDA-accelerated RAN software 25-3
OAI gNB SW:
ARC1.7_integrationbranch (2025.w46baseline)
Performance Improvements#
TDD patterns: DDDDD DSUUU, DSUUU
MIMO layers
DL: 4 layers
UL: 2 layers
Single cell peak throughput
1 UE:
DL: ~1.2 Gbps
UL: ~240 Mbps (DSUUU)
4 UEs:
DL: ~1.1 Gbps
UL: ~135 Mbps
Deprecations#
Foxconn O-RUs are no longer supported.
Known Issues and Workarounds#
Multiple UE support: Aggregated throughput may be low when scheduling multiple UEs simultaneously. Throughput occasionally drops to zero after running traffic with multiple UEs for some time. The current workaround is to restart the test.
SRS is only supported when PUSCH is simultaneously transmitted by the UE. SRS operation with 2L UL or operation with many UEs may have stability and performance issues after some time.
ARC-OTA Release 1.6 (August, 2025)#
New Features and Bug Fixes#
4T4R O-RUs
WNC 1220
WNC 3210
Multi-UE support
55 UEs attach and ping (UE simulator over eCPRI)
2 UEs traffic
Process SRS samples received from RU for use by L2 (4T4R; cabled)
Aerial Data Lake: Real-time I/Q sample capture from 2 cells
This enables real-time capture of samples to DRAM for use in real-time processing.
Previously, real-time samples were captured to SSD and only enabled offline processing, such as for model training or visualization.
Real-time processed samples can now be used by algorithms to perform closed-loop actions on L1/L2 with low latency.
Aerial CUDA-accelerated RAN software 25-2
OAI gNB SW: ARC1.6_integration branch (2025.w29 baseline)
Performance Improvements#
TDD pattern: DDDDD DSUUU
MIMO layers
DL: 4 layers
UL: 1 layer
Single cell peak throughput
1 UE:
DL: ~1.2 Gbps
UL: ~126 Mbps
2 UEs:
DL: ~1.4 Gbps
UL: ~75 Mbps
2 cell aggregated peak throughput (1 UE per cell)
DL: ~1.4 Gbps
UL: ~180 Mbps
Deprecations#
Dell PowerEdge R750 Server + A100X HW is no longer supported.
Gigabyte Edge E251-U70 Server is no longer supported.
Known Issues and Workarounds#
There are connection drops after running iPerf traffic for some time. The UE will re-initiate the connection autonomously.
Data stalls with traffic to multiple UEs: Throughput occasionally drops to zero after running traffic with multiple UEs for some time. The current workaround is to restart the test.
ARC-OTA Release 1.5 (July, 2024)#
New Features and Bug Fixes#
Grace Hopper integration
OAI ARM CPU support
Hopper GPU cuBB support
CBRS O-RU
CBRS RU Interop
Multi-UE support
Multi-UE CSI dataset blueprint (using Aerial Data Lake and PyAerial)
Updated OAI branch 2024.w21+ARC1.5
Multi-L2 Support (2 Cells)
Tip
For additional context, you can also review the artifacts in the 2024.w21+ARC1.5 branch.
Note
The Grace Hopper platform requires OAI CN version 2024-June.
Performance Improvements#
MIMO layers
DL: 2 layers -> 4 layers
Peak throughput
SMC-GH
DL: ~460Mbps -> ~1.03Gbps
UL: ~112Mbps -> ~125Mbps
Dell R750 / Gigabyte Edge E251-U70
Multi-L2 (2 Cell) throughput
SMC-GH
DL: ~700Mbps (per Cell)
UL: ~105Mbps (per Cell)
ARC-OTA Release 1.3 (May, 2024)#
New Features and Bug Fixes#
Full support for master gitlab repo gitlab.eurecom.fr/oai/ making it available for use, modification and distribution. Refer to the Licensing page for the OAI license.
OAI_Aerial private branch is deprecated and will no longer be maintained.
Developer extension - Sterling k8 service management and monitoring (refer to this document for more details)
Updated OAI branch
2024.w15with a standalone patch (w15.4).
Performance Improvements#
Frame structure and slot format
DDDSU -> DDDSU + DDDDDDSUUU
Bi-directional UDP Traffic
> 3.5 hours exercised -> > 4.0 hours exercised
ARC-OTA Release 1.2 (March, 2024)#
New Features and Bug Fixes#
Support for the Dell R750 platform to host gNB
Support for NVIDIA Converged Accelerators A100X
Continued support for Gigabyte Edge E251-U70 with NVIDIA discrete cards A100 GPU, CX6-DX SmartNIC
OAI_Aerial_v2.2.1 updated to OAI_Aerial_v2.2.2
ARC-OTA Release A1.1 (January, 2024)#
New Features and Bug Fixes#
Kernel cmdline configuration updated.
Updates to the core assignment in the Aerial configuration.
Updates to PTP and phc2sys core assignment.
Changes to phc2sys cmdline.
Changes to the L2 Docker run cmd to use all non-isolated cores.
Changes to the L2 configuration, max DL MCS defaults to 25.
Removal of unnecessary ORU firmware installation step because of Foxconn firmware default updates.
OAI_Aerial_v2.0 updated to OAI_Aerial_v2.2.1 throughout.
Known Issues and Workarounds#
256 QAM is not supported on ARC-OTA. You must disable 256 QAM support by issuing the following command at the gNB command license:
--gNBs.[0].force_256qam_off