This topic describes how the Uhana by VMware platform uses artificial intelligence (AI) to detect poor subscriber experiences and provide root causes to help fix problems in the radio access network (RAN).

The following topics describe how alerts are treated in the Uhana by VMware platform.

Data sources

The Uhana by VMware platform ingests raw signaling and traffic traces and processes them into subscriber session records and cell records. These processed records serve as inputs for building AI-based applications on top of the real-time stream processing engine (RSPE).

For more information, see Understand data sources and events.

The following topics list some critical key performance indicators (KPIs) in these records.

Subscriber session record KPIs

Subscriber session records might include the following RAN KPIs for an individual subscriber session.

Subscriber experience KPIs

  • Downlink throughput per radio bearer
  • Uplink throughput
  • Session setup cause per radio bearer
  • Session release cause per radio bearer
  • Voice quality for VoLTE

Packet loss KPIs

  • Downlink MAC BLER
  • Uplink MAC BLER
  • Downlink RLC loss rate
  • Uplink RLC loss rate
  • Downlink PDCP loss rate
  • Uplink PDCP loss rate

Radio KPIs

Traffic KPIs

  • MAC, PDCP volumes on downlink and uplink
  • Modulation and coding scheme usage
  • Downlink and uplink resource usage

Handover KPIs

  • Handover type
  • Handover target cell
  • Handover result

Cell record KPIs

Cell records contain KPIs that are periodically reported by cells.

Traffic KPIs

  • Downlink and uplink volumes
  • Active users on downlink and uplink
  • Resource usage on control and data channels on uplink and downlink

Radio KPIs

  • Path loss distribution
  • Uplink interference per PRB per antenna branch
  • Uplink interference on PUCCH and PUSCH

Subscriber experience modeling

The Uhana by VMware platform uses offline subscriber session and cell records to train models about subscriber experience KPIs.

The following are several examples of subscriber experience metrics which illustrate how Uhana by VMware models might be trained.

  • Downlink throughput for default radio bearer per frequency band - supervised learning regression model
  • Uplink voice quality for VoLTE - supervised learning classification model indicating whether voice quality is bad
  • Session release cause - supervised learning classification model indicating whether session release cause is abnormal

With these models, Uhana by VMware can access the impact of a RAN feature on a subscriber's experience.

RAN root-cause analysis

Using the models trained offline, the Uhana by VMware platform infers in real-time, the root-cause of degraded quality subscriber experiences in the RAN and provides the impact of resolving the problem.

For every subscriber session whose experience does not satisfy operator-defined expectations, the Uhana by VMware platform identifies anomalies in RAN features that can degrade subscriber experience. These anomalies are identified by learning these features' nominal values for the network and inferring deviations from these values. Each anomaly is then associated with a root-cause label. The impact on subscriber experience is deduced by adjusting the anomalous features to their appropriate nominal values and applying the offline model on the new feature set to derive the new subscriber experience.

The following RAN root cause descriptions are supported.

  • Uplink/Downlink interference - poor signal quality for a given signal strength on the uplink or downlink
  • Load imbalance - imbalance in distribution of resources among cells covering the same geographical region
  • Coverage - large path losses for users in a specific cell relative to all the users in a given frequency band

Alert generation

The Uhana by VMware platform performs a spatiotemporal clustering of sessions that are impacted by the same problem and forms alerts. The spatiotemporal clustering gets a complete view of the impact caused by a specific RAN problem and also reduces the number of duplicate alerts generated.

The following properties are included in each alert description.

  • Alert ID - a unique ID for the alert
  • Duration - the time period for which the alert was raised
  • Symptom - the subscriber experience metric that was impacted
  • Affected entities - one or more cells where a RAN problem is impacting subscribers
  • Impacted sessions - the number of subscriber sessions impacted by the alert
  • Percentage impact - the percentage of subscriber sessions in the Affected entities impacted by the alert
  • Root cause - the root-cause label associated with the alert

Each alert has a Visibility page that provides KPIs relevant to the alert and an Insights page that describes potential fix recommendations for the alert.

check-circle-line exclamation-circle-line close-line
Scroll to top icon