Pre-Emptive Conflict Detection Architecture for O-RAN Service Management and Orchestration
Before changing a network parameter, inspect what comparable cells suggest about its side effects. A pre-emptive warning gives the proposing application evidence for its decision.
Look for side effects before applying a change
An rApp can improve the metric it manages while degrading a metric managed by another application. Pre-emptive conflict detection asks what the network’s existing observations can tell us about those side effects before a configuration change is applied.
We compare the target cell with meaningfully similar cells already using the current and proposed parameter values. Their KPI distributions provide evidence about possible degradation. The proposing rApp receives that evidence and decides how to act on it.
Compare peers under two settings
Follow a proposed change from setting A to setting B. Switch between two constructed sets of proposed-setting peers to see how the warning changes. The example illustrates the comparison behind the paper’s Figures 1, 3 and 4.
What happened in comparable cells?
Target cell: setting A, access success rate 97%. Proposal: change to B. Higher access success is better.
Existing setting A
Match usage patterns and remaining configuration.
Proposed setting B
Match usage patterns and remaining configuration.
Other settings
Excluded from this two-setting comparison.
Similar peers selected within each eligible group
Each dot is one illustrative peer. Lines span the observed examples, not confidence intervals. The dashed marker is the target’s 97%. Both rows use the same 85–100% axis.
Target: 97%. Existing-setting peers average 97.0%; proposed-setting peers average 92.0%. Higher access success is better.
The proposed-setting peers have worse access success than the target cell. Report possible degradation to the proposing rApp; it decides whether the tradeoff is acceptable.
The numerical values and three-peer groups are constructed for this explanation. The paper’s method compares KPI evidence from similar cells; this small mean comparison is not a fitted predictor or a reproduced experiment.
What we implemented and observed
The prototype used simulated rApps to generate configuration-change proposals and a conflict-detection rApp to assess them. It consumed performance, configuration and KPI data from a telecommunications network segment of approximately 800 cells. The functions were implemented in Python and developed in Jupyter notebooks.
For each proposal, we first group cells by the value of the parameter being changed. Within the current-value and proposed-value groups, similarity uses time-series performance behaviour and the remaining configuration parameters. Leaving the changed parameter out of that second comparison allows us to find otherwise comparable cells across the two groups.
The paper shows KPI-distribution comparisons with and without predicted degradation, and a table reporting expected effects across several KPI categories. This supplies evidence for a decision; the method does not assess the importance of one KPI relative to another.
A warning is evidence for a decision
These comparisons depend on observed associations and on the quality of the peer matches. They do not establish that changing the parameter causes the difference, or guarantee the outcome in a new cell. A lack of sufficiently comparable observations limits what can be inferred.
An rApp may still apply a proposal after a warning if the predicted tradeoff serves the operator’s goals. Confidence estimates for the predictions and integration with operator intent are identified as further work. The reported examples do not establish a production-wide detection accuracy or a universal safety guarantee.
From comparison to contextual decisions
The cell-similarity paper explains how to inspect the evidence for choosing comparable cells. CAMINO develops the use of operator intent and external context when assessing a proposed change.
TACIT addresses arbitration through prediction reputation, while ORACLE addresses a verifiable workflow. These papers contribute different mechanisms and evaluations within my autonomous network-management research.
Paper and citation
J. Armstrong, E. Fallon, S. Fallon, “Pre-Emptive Conflict Detection Architecture for O-RAN Service Management and Orchestration,” 2024 IEEE International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT), pp. 335–340, 2024. DOI: 10.1109/IAICT62357.2024.10617647.