One-line positioning
For AI-RAN base stations, simulated terminals quantitatively generate signaling- and user-plane inputs while real terminals quantify service experience, turning “does AI optimization actually work?” into measurable, reproducible and comparable test conclusions.
Solution overview
AI for RAN is becoming a major direction in base station evolution: AI/ML functions for energy saving, load balancing, mobility optimization, beam management and CSI feedback enhancement are entering base stations. Semantic and token communications are technology reserves for 6G. These areas are broadly at a forward-looking technology stage, and this solution targets their R&D and validation scenarios.
AI outcomes depend strongly on scenarios: the same model can perform very differently under heavy/light traffic or high-speed rail/fixed-location conditions. Drive tests and live-network statistics cannot reproduce scenarios under control or independently attribute before/after differences, leaving conclusions unreviewable. Validating AI-RAN requires reproducible inputs followed by output measurement on the same basis.
The solution addresses this with two sets of sensing capabilities:
- Signaling-plane sensing: XproUESimulator simulates large numbers of terminals. An X86 server, GPU physical-layer acceleration cards and SDR RF fronthaul modules form the instrument, with OTA air-interface and SMA RF feeder connections. UE behavior and traffic models are programmable, supplying controlled, reproducible inputs to base station AI functions.
- User-plane sensing: XproVideo measures video quality and service experience from real Android terminals, quantifying what users perceive as stalling, buffering, latency and bit rate.
Together with the application and security simulation tester (real live-network service traffic), IMS voice core network tester (voice service loop) and core network simulation, these create a fully surrounding test environment. Data-plane traffic travels from the generator through simulated terminals, the base station and N3 / N6 interfaces back to the generator for statistics; voice services loop from simulated terminals through the base station and core network to IMS.
Core capabilities
① Scenario-based AI for RAN testing
| AI for RAN application scenario | Test focus |
|---|---|
| Network energy saving | Validate energy savings from AI real-time deactivation/activation of cells and NG-RAN nodes, together with dropped-call and handover-success costs |
| Load balancing | Validate UE performance feedback after AI/ML load prediction and offloading, assessing load balancing improvement |
| Mobility optimization | Validate AI/ML mobility optimization, focusing on handover success improvement and optimization objective determination |
| Beam management / downlink beam prediction | Validate spatial-domain downlink beam prediction using measurements from beam set B |
| CSI feedback enhancement | Validate spatial-frequency CSI compression using two-sided AI models; study AI/ML CSI compression; time-domain CSI prediction using terminal-side models |
| Semantic communication (joint source-channel coding) | End-to-end quality assessment for semantic communication test platforms (6G evolution direction) |
| Token communication | Token communication testing platform: TCE token communication efficiency, TER token error rate (two-point measurement of raw link-layer values and residual values after completion), TTFT time to first token, TPOT time per output token, semantic fidelity (CLIP / LPIPS), and cross-domain degradation rate (6G evolution direction) |
② Signaling-plane sensing and optimization-effect testing
- UE behavior simulation: programmatically construct near/mid/far user distributions, handovers, calls / SMS and other signaling behavior
- Traffic model simulation: high-volume access, high-volume calls, high-speed rail models and highway models
- Quantified signaling-plane optimization effects: accurately measure base station energy savings, dropped-call rates and handover success rates
- Scenario comparisons: compare the same AI model’s optimization effects across traffic models to validate generalization
- Programmable and controllable protocol stacks: programmable NAS / RRC and controllable PDCP / RLC / MAC for abnormal and boundary signaling
③ User-plane service experience and poor-quality testing (XproVideo)
- Real-terminal perspective: Android App collecting end-to-end service experience from the terminal side, supporting Android 15
- Video quality quantification: stall count/duration, initial buffering duration, total buffering duration, bit rate, VMOS score, frame rate and resolution, with automated testing of mainstream video applications
- Service experience metrics: web/App opening latency, HTTP download rate (up to 20 threads), IPv4 / IPv6 traffic share, TCP / UDP session statistics, packet capture (customizable size) and DNS filtering
- Quantitative generation of poor-quality characteristics: quantitatively generate poor user experience to validate reduced stalling and web opening times after base station optimization
- Long-duration automation: server-dispatched scripts support unattended 12 / 24-hour stability tests; packet captures and screen recordings are automatically returned for metric analysis
- Terminal diversity: simulate traffic behavior of phones, modules, robots and other terminal types
④ Basic base station function and performance testing
- Functions and protocols: 3GPP and domestic specification features, protocol conformance, interface testing, end-to-end service procedures, version iteration and release regression
- Performance and capacity: large-scale concurrent terminal access, single-/multi-user throughput, end-to-end latency, extended coverage, mixed services and quantified QoS metrics
- Base station energy consumption testing: Layer 4–7 traffic generator + simulated core network + terminal simulator (near/mid/far users) form the topology; UEs and user-plane traffic are automatically mapped, with out-of-band signaling synchronizing UE IP mappings to the generator
- Traffic excitation testing: replay application traffic to validate base station DPI and intelligent scheduling; supports L3 / L7 replay, multiple flows from one PCAP, packet content editing, large PCAP import, SNAT mode, multiple UE simulators and different services per UE
Key technical specifications
| Item | Parameters / description |
|---|---|
| Communication technologies | NR (FDD / TDD) Release 18; NTN (FDD) 3GPP Release 18 and versions 1.6 / 2.1 of a well-known Chinese satellite operator’s specifications; 6G is an evolution direction |
| Terminal simulation capacity | Up to 16,000 RRC-connected users (beam-bound mode); 4,000 active users (whole system) |
| Schedulable UEs per slot | 40 (uplink and downlink combined) |
| Cell bandwidth | All bandwidths up to 100MHz (8 beams), 200MHz (4 beams) |
| MIMO | 1T1R / 1T2R / 2T4R |
| Multi-system cascading | 4 or more cascaded systems |
| Programmability | Programmable NAS / RRC; controllable PDCP / RLC / MAC |
| Ephemeris and uplink synchronization | Connect to mainstream channel emulators for LEO / MEO / GEO ephemeris parsing and uplink synchronization |
| RF connections | OTA air-interface coupling or direct SMA RF cable connection |
| Video and experience measurement | Stalling / buffering / bit rate / VMOS / frame rate / resolution; web/App opening latency; IPv4 / IPv6 dual-stack statistics |
| Terminal environment | Android App, supporting Android 15 |
Target customers
- Base station equipment manufacturers (AI-RAN functional R&D, integration and validation testing)
- Terrestrial telecom operators and their research institutes (AI for RAN introduction assessment)
- Research institutes and universities (AI-RAN, semantic communication and token communication research)
Verifiable facts
- The X86 + GPU + SDR multi-terminal simulator has been delivered as a product, with OTA and direct SMA RF connection options
- Testing support is available for five deployed AI for RAN scenarios (energy saving, load balancing, mobility optimization, beam management and CSI feedback enhancement)
- XproVideo is a mature tool already delivered and continuously operating on networks, producing video quality and service experience metrics from real terminals
- Semantic and token communication testing platforms target 6G evolution; this page describes capability directions
- Complete topologies and methods are available for base station energy consumption and traffic excitation testing
