Autonomous information processing at the tactical level�Paper 026

Rune Langleite, Emil P. Andersen, Anette F. Opland, and Frank T. Johnsen

Foto: AdobeStock

1. Background and motivation
2. Limitations
3. Architecture/implementation
4. Conclusions & future work

​

​

​

​

​

• Incremental work, based on our paper «Big Data Solutions on Tactical Infrastructure» presented at ICCRTS 2022.

Agenda

2

Motivation

• Today's land-based combat platforms use little or no autonomous systems for the acquisition, analysis and sharing of data from sensors

​

• Existing systems are generally stand-alone and/or isolated systems with a relatively high degree of human interaction, tightly coupled with local ICT infrastructure and sensors

​

• Sharing sensor data often requires human interaction

Focus area

• Big data at the tactical edge.
• The motivating factor is that with IoT/IoBT there is a significant increase in available sensor data
• Too much to handle with old, manual approaches
• Need for digitalization and automation
• As recent trend analyses indicate, AI/ML is a necessary approach in future systems

​

IoT/IoBT value chain

​

​

• In civilian IoT systems, the Platform is typically in the cloud
• For military applications, we can’t expect continuous network availability
• DIL environment
• Must operate without reach back to cloud
• Platform = Tactical platform
• We consider the edge computing paradigm, where things/connectivity/data/analytics are in physical proximity of each other.

​

Premises and hypothesis

​

​

Hypothesis

​

• We can realize a technical architecture using FOSS supporting both basic systems of thinking, which may be used in a stand-alone tactical node.

​

​

Premises

​

• AI is needed in the tactical domain to achieve information superiority through digitalization and automated analysis.

​

• Digitalization involves speeding up manual processes, and so the AI system should support the two basic systems of thinking (i.e., fast and slow).

Limitations in this study

• Big Data-stack in one tactical node
• FOSS (Free Open Source Software)

​

• The aim is to study how to build the next-generation, AI-enabled tactical platform
• Analysis of sensor data in near real time
• Automated alerts

​

• For future work, we will also study collaboration between multiple nodes

Software architecture�Literature studies helped us identify Lambda architecture as the best approach

Figure source: T. Matyashovsky, "DZone - Big Data Zone - Lambda Architecture with Apache Spark", https://dzone.com/articles/lambda-architecture-with-apache-spark

Approach

• Software components to process sensor data as part of the tactical ICT platform

​

• Sensor examples
• RF
• Radar
• Video
• Audio

​

• Local state of continous processing and updating

​

• Collaboration between nodes (future work) for even further possibilities

sensor input

sensor mount

information output

Software�Lambda-architecture with FOSS

• Realizing the lambda-arkitekture¹⁾ with FOSS
• Message queue: Kafka²⁾ (+ Connector)
• Processing layers: Spark³⁾ (+ SANSA)
• Storage: Kafka log + GraphDB⁴⁾

1. https://en.wikipedia.org/wiki/Lambda_architecture -
2. https://kafka.apache.org/ -
3. https://spark.apache.org/ -
4. https://graphdb.ontotext.com/ -

Use case: RF

• One node
• Detection
• Classification
• Collaborating nodes
• Geolocation

​

• Generate target data through reasoning

S1: (x₁, y₁)

S2: (x₂, y₂)

S3: (x₃, y₃)

S1:

• Frq: 45.000MHz
• Amplitude: 8dBm
• TOA: 2023-02-02T23:00:05.122Z

​

S2:

• Frq: 45.000MHz
• Amplitude: 3dBm
• TOA: 2023-02-02T23:00:05.123Z

​

S3:

• Frq: 45.000MHz
• Amplitude: 12dBm
• TOA: 2023-02-02T23:00:05.121Z

S1:

• Emitter_1:
• CL: 8-QAM
• Confidence: 80%

​

S2:

• Emitter_1:
• CL: 8-QAM
• Confidence: 74%

​

S3:

• Emitter_1:
• CL: 8-QAM
• Confidence: 95%

S1:

• Emitter_1:
• HB: d1

​

S2:

• Emitter_1:
• HB: d2

​

S3:

• Emitter_1:
• HB: d3

​

Emitter_1:

• Grid: 34WCB12345678

Conclusions

• What we have learned
• Lambda architecture
• Realized the architecture using FOSS
• This is a solution for big data processing in tactical nodes
• Tested with simple sensors (temperature, air pressure in the previous paper)
• We are currently working on the RF case

​

• AI application is domain specific
• Need for domain experts to construct models
• What we have now is a general infrastructure, on which different models can run

​

• The way forward
• Put in place model(s) that can handle different data (e.g., RF, audio, …)
• Orchestration of services, evaluating resource use
• Collaborating nodes

Future work

Tactical platform node

BMS wireframe

Thank you for your attention!

​

​

​

​

​

Point of contact:�Rune.Langleite@ffi.no

​

​

​

​

Questions?

​