Abstract
Conventional diagnostic platforms for diagnosing complex field failures in cellular networks suffer from several drawbacks, such as garbage collection spikes or pauses, memory fragmentation, memory exhaustion, out-of-memory kernel crashes, and/or LLM context-window overflow. This disclosure describes techniques for diagnosing complex field failures in cellular networks by performing high-throughput, low-latency lazy decoding of raw binary trace files. For example, temporal anchors received from an edge device are used to selectively load and parse only the relevant frames of the binary trace file that correspond to the anomaly, bypassing non-relevant protocol frames without payload expansion. Protocol state trees are instantiated within a contiguous pre-allocated memory arena, which is reclaimed via a bulk pointer reset once root cause analysis is over. An LLM-driven diagnostic engine constrains token generation by a cloud-based LLM using edge-generated semantic summary of the anomaly and deterministic protocol grammar rules.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Zhu, Longlong; Ke, Allan; and Huang, Kuo-Yu, "Cellular Network Failure Diagnosis Using High-Throughput, Low-Latency Lazy Decoding", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12081