International Telecommunication Union

Radiocommunication Sector

Report ITU-R SM.2542-0

(06/2024)
SM Series: Spectrum management

Next generation spectrum monitoring –

proactive, autonomous and data-driven
ii Rep. ITU-R SM.2542-0

Foreword
The role of the Radiocommunication Sector is to ensure the rational, equitable, efficient and economical use of the radio-
frequency spectrum by all radiocommunication services, including satellite services, and carry out studies without limit
of frequency range on the basis of which Recommendations are adopted.
The regulatory and policy functions of the Radiocommunication Sector are performed by World and Regional
Radiocommunication Conferences and Radiocommunication Assemblies supported by Study Groups.

Policy on Intellectual Property Right (IPR)

ITU-R policy on IPR is described in the Common Patent Policy for ITU-T/ITU-R/ISO/IEC referenced in Resolution
ITU-R 1. Forms to be used for the submission of patent statements and licensing declarations by patent holders are
available from http://www.itu.int/ITU-R/go/patents/en where the Guidelines for Implementation of the Common Patent
Policy for ITU-T/ITU-R/ISO/IEC and the ITU-R patent information database can also be found.

Series of ITU-R Reports

(Also available online at https://www.itu.int/publ/R-REP/en)

Series Title
BO Satellite delivery
BR Recording for production, archival and play-out; film for television
BS Broadcasting service (sound)
BT Broadcasting service (television)
F Fixed service
M Mobile, radiodetermination, amateur and related satellite services
P Radiowave propagation
RA Radio astronomy
RS Remote sensing systems
S Fixed-satellite service
SA Space applications and meteorology
SF Frequency sharing and coordination between fixed-satellite and fixed service systems
SM Spectrum management
TF Time signals and frequency standards emissions

Note: This ITU-R Report was approved in English by the Study Group under the procedure detailed in Resolution
ITU-R 1.

Electronic Publication
Geneva, 2024
© ITU 2024
All rights reserved. No part of this publication may be reproduced, by any means whatsoever, without written permission of ITU.
 Rep. ITU-R SM.2542-0 1

REPORT ITU-R SM.2542-0

Next generation spectrum monitoring –

proactive, autonomous and data-driven
(2024)

TABLE OF CONTENTS
Page

Policy on Intellectual Property Right (IPR) ............................................................................. ii

1 Terms, definitions and abbreviations ............................................................................. 2

1.1 Big data ............................................................................................................... 2
1.2 Artificial intelligence .......................................................................................... 2
1.3 Machine learning ................................................................................................ 2
1.4 Radio frequency machine learning ..................................................................... 2
1.5 Abbreviations ...................................................................................................... 3

2 Introduction .................................................................................................................... 3

3 Distributed spectrum monitoring .................................................................................... 5

3.1 Elements of a distributed spectrum monitoring system ...................................... 5
3.2 Big data challenges of a distributed spectrum monitoring system ..................... 7

4 Big data spectrum monitoring ........................................................................................ 10

4.1 Big data spectrum monitoring benefits ............................................................... 11
4.2 Big data spectrum monitoring solution............................................................... 11
4.3 RF collection layer: example big data spectrum monitoring network ................ 12
4.4 Data storage layer ............................................................................................... 13
4.5 Data management layer ...................................................................................... 15

5 Realtime data-driven spectrum awareness using RFML ................................................ 18

6 Summary......................................................................................................................... 20

7 References ...................................................................................................................... 21

Annex 1 – Solution for data-driven AI and big data spectrum monitoring in Korea
(Republic of) ................................................................................................................... 21

Annex 2 – Mobile public transport based big data acquisition for spectrum mapping ........... 28
2 Rep. ITU-R SM.2542-0

Scope
This Report applies new trends in data sciences including artificial intelligence and big data
technologies to the automation of spectrum monitoring. The Report summarizes the current reactive
and ad hoc spectrum management approaches, and then details next-generation approaches that are
proactive, autonomous and data-driven, using artificial intelligence and big data. The Report gives
several implementation examples that illustrate the power of these techniques.

1 Terms, definitions and abbreviations

1.1 Big data

Recommendation ITU-T Y.3600 – Big data – Cloud computing based requirements and capabilities,
defines big data as “a paradigm for enabling the collection, storage, management, analysis and
visualization, potentially under real-time constraints, of extensive datasets with heterogeneous
characteristics.” Examples of datasets characteristics include high-volume, high-velocity, high-
variety, etc. This Report discusses generation of spectrum big data from densely distributed
monitoring networks that will require both local and central processing to route, store and display the
resulting information. Because of the network and cloud-computing requirements for such a
deployment, ITU publications dealing with information infrastructure are cited here.

1.2 Artificial intelligence

Recommendation ITU-T M.3080 – Framework of artificial intelligence enhanced telecom operation
and management (AITOM), defines artificial intelligence (AI) as a computerized system that uses
cognition to understand information and solve problems. ISO/IEC 2382-28 defines AI as “an
interdisciplinary field, usually regarded as a branch of computer science, dealing with models and
systems for the performance of functions generally associated with human intelligence, such as
reasoning and learning”. In computer science, AI research is defined as the study of “intelligent
agents”: any device that perceives its environment and takes actions to achieve its goals. This includes
pattern recognition, the application of machine learning and related techniques. Artificial-intelligence
is the whole idea and concept of machines being able to carry out tasks in a way that mimics human
intelligence and would be considered “smart”. Further studies in this respect are invited.

1.3 Machine learning

Recommendation ITU-T Y.3172 – Architectural framework for machine learning in future networks
including IMT-2020, defines machine learning (ML) as processes that enable computational systems
to understand data and gain knowledge from it without necessarily being explicitly programmed.
Supervised machine learning and unsupervised machine learning are two examples of machine
learning types.

1.4 Radio frequency machine learning

“… recent research has shown deep machine learning to be an enabling technology for cognitive radio
applications as well as a useful tool for supplementing expertly defined algorithms for spectrum
sensing applications such as signal detection, estimation and classification (termed here as Radio
Frequency Machine Learning, or RFML). A major driver for the usage of deep machine learning in
the context of wireless communications is that little, to no, a priori knowledge of the intended spectral
environment is required, given that there is an abundance of representative data to facilitate training
and evaluation.” [1]
 Rep. ITU-R SM.2542-0 3

1.5 Abbreviations
AI Artificial intelligence
AOA Angle of arrival
API Application programming interface
BI Business intelligence
DB Database
DF Direction finder(ing)
DMR Digital mobile radio
FFT Fast Fourier transform
GUI Graphical user interface
IoT Internet of things
I/Q In-phase/quadrature phase
KPI Key performance indicator
LAN Local area network
LTE Long term evolution
ML Machine learning
PU Primary user
QoE Quality of experience
QoS Quality of service
RBW Resolution bandwidth
RF Radio frequency
RFML Radio frequency machine learning
RSS Received signal strength
SQL Structured query language
TDOA Time-difference-of-arrival
UAV Unmanned aerial vehicle
WCDMA Wideband code division multiple access

2 Introduction
The radio spectrum is a limited resource and the increase in demand for spectrum for emerging new
radio technologies remains unabated. Continuous advancements in wireless communication
technologies have resulted in new standards with:
– Higher carrier frequencies with diverse band assignments;
– Wider modulation bandwidth;
– Lower transmitter power levels to increase system capacity through frequency reuse.
These technology trends result in the deployment of radio communication systems that have a high
number of lower power transmitters to provide coverage in a given geographic area. Modern spectrum
4 Rep. ITU-R SM.2542-0

monitoring strategies need to account for these types of emitter deployments, while still being able to
monitor ‘traditional’ emitters such as FM and TV broadcast or air traffic communications based on
traditional analogue AM transmissions.

FIGURE 1
Trend towards high-density of low-power emitters

Given these conditions, spectrum monitoring has become a complex task, requiring fine grained and
dense spectrum usage data across frequency, time and location. Traditional approaches to spectrum
monitoring are not always able to provide the density of data required to manage today’s spectrum.
Many national authorities maintain specialized services for the systematic monitoring of radio
emissions to create knowledge about spectrum use and to provide feedback to the regulatory process
and if needed, enforcement procedures. These services create occupancy data for the assessment of
utilization [2] to efficiently manage the spectrum that forms the basis for many spectrum management
tasks, from maximizing license revenues to verifying compliance and resolving interference
problems. At the same time, the rapid expansion in wireless communications, both in the number of
transmitters and in standards, make measuring and tracking changes in spectrum utilization more
challenging.
Traditionally, spectrum monitoring relies on the skills of trained operators, using tools such as
spectrograms, power spectral density and demodulation, to identify emitters, locate violators, and
report on spectrum usage characteristics. On the other hand, while spectrum engineering tools such
as coverage maps showing propagation models and effects offer visibility into spectrum usage, they
rely on an individual’s expertise and/or tacit knowledge to analyse spectrum monitoring data and take
the appropriate actions. These skills are hard to replicate or scale and limit the ability of
administrations to face ever growing challenges to spectrum management.
On the other hand, future data-driven spectrum monitoring aims at less human intervention and more
autonomous surveillance. For instance, data-driven spectrum monitoring can detect specific signals
even in the case of a high-density of emitters. It can also collect large volumes of spectrum data
autonomously. The data-driven spectrum monitoring can link between previous and future data due
to accumulated database giving the potential to increase the efficiency of spectrum supervisors.
Advanced spectrum monitoring systems can help administrations to manage abnormal spectrum
events more easily and, depending on the network density, with a high coverage area.
 Rep. ITU-R SM.2542-0 5

Data-driven spectrum monitoring can be used to identify trends and anticipate future events by
analysing the pattern of signals. This enables the monitoring service to prepare accordingly.
Spectrum allocation practices traditionally have embraced “Static Allocations” where a license is
issued to a primary user exclusively on a long-term basis over large geographical regions. With the
exception of some services e.g. the broadcast services, it is rare that licensees make full use of the
frequencies assigned to them. In many cases, some portion of the allocated band(s) may remain
underutilized. Data-driven spectrum monitoring may enable more efficient spectrum management
practices.
To improve spectrum management practices, spectrum utilization needs to be accurately
characterized at the national, regional, urban and local level. This drives the need to identify available
spectrum across time and location in the area of interest. Available spectrum may also be referred to
as spectrum holes – spatial and temporal. The temporal spectrum hole appears in a time interval when
there is no Primary User transmission. The spatial spectrum hole appears in a location or area where
no Primary User transmissions are detected.

3 Distributed spectrum monitoring

Next Generation Spectrum Management takes advantage of and aggregates the capacity of
underutilized frequency bands employing an integrated approach that leverages new advances in
radio frequency (RF) sensing, distributed and mobile sensing that enable geolocation measurements,
centralized databases, analytics and ML to increase capacity, improve coverage and mitigate
interference.
An autonomous approach to spectrum monitoring can adapt the monitoring tasks to changing
conditions and directly answer meaningful queries through dashboards that may provide insight into
radio spectrum KPIs, such as utilization, efficiency, allocation, users, and available spectrum by
location, time and frequency band. An example of next generation distributed spectrum monitoring
is given in Annex 1, describing the experience of Korea (Republic of) with the use of these techniques
for spectrum monitoring.

3.1 Elements of a distributed spectrum monitoring system

A coordinated and concentrated effort to collect large-scale spectrum allocation and usage
information at a micro level requires many geographically distributed spectrum monitoring sensors
that are adaptable to changing operational environments and technical requirements.
Managing a network of distributed sensors, and the voluminous data they generate requires fusing
data from multiple sensors and selecting the optimal dataset to drive decisions. This necessitates
intelligent algorithms that can analyse the data and apply domain knowledge to generate relevant
information for various stakeholders.
A fundamental task in spectrum monitoring is to identify specific usage patterns in the various
dimensions (time, frequency, location, direction, etc.) to reveal available spectrum. Building detailed
spectrum occupancy maps at the national, regional, urban and local level requires continuous
spectrum monitoring, and is challenging and time consuming.
Based on the above, the following elements can form a scalable, dynamic spectrum monitoring
system providing an economical national solution operating down to the local level, and over a large
range of frequencies:
a) Fixed traditional monitoring systems will continue to be inevitable for monitoring, especially
when large antennas are required. Examples for that apply to the monitoring at HF, VHF and
UHF frequency bands or high sensitivity satellite monitoring stations. Such stations may also
be useful to make DF measurements for interference resolution and identification of illegal
6 Rep. ITU-R SM.2542-0

transmitters. It has to be noted that this concept does not address organizational situations in
case of network failure, i.e. the operation of a network independent manned monitoring
station;
b) Lower cost small form factor RF sensor systems may be used for long term measurement,
such as seasonal occupancy evaluation or to evaluate specific interference cases of
intermittent weak signals that may require long term measurements;
c) Mobile and portable monitoring systems can be used to achieve a high spatial resolution for
dynamic spectrum management which fixed stations cannot provide. Annex 2 to this Report
provides examples of mobile and portable monitoring solutions to address localized spectrum
measurements of field strength, coverage and detection of interference.
The large amount of monitoring data reported back to the control centre will add to big data
management issues, making centralized databases and analytical tools with real-time dashboards
critical.
Displays giving generalized locations of multiple emitters based on the license database, as well as
DF and RSS measurements are useful in evaluating emitter density, spectrum occupancy and
spectrum holes at the local level. Determining the precise location of interfering emitters will require
TDOA, AOA and Hybrid geolocation techniques. Due to the low power of many emitters, mobile
and distributed small sensor monitoring/geolocation systems will become more important.
Additionally, these systems will need to detect and locate emitters in the higher frequency ranges.
3.1.1 Local/edge processing at monitoring receivers
The goal of local processing is to transform the raw wideband data created by the monitoring receiver
into an aggregated and transportable form of information and moving spectral intelligence closer to
the edge of the network. Having processing and storage at the local level results in faster analysis,
comparative processing of spectrum snapshots, and real-time machine decision making. The use of
signal processing, enabled by Digital Down Converters (DDCs), increases the visibility and accuracy
of the signalling environment and changes within that environment. The local processing further
reduces the burden on the communication network that ties together the receivers, central processors
and data storage without losing important data and analysis results.
Artificial Intelligence (AI), particularly in the form of Radio Frequency Machine Learning (RFML),
may become a necessity to process the large volume of real-time monitoring data across many
locations. Sensors with local processors and RFML capabilities will be needed to reduce the amount
of network backhaul data (refer to § 5). Less-capable secondary sensors can be used to extend
geographic coverage in the most economical way.
Some I/Q time-series data will need to be recorded locally and transferred to the control centre for
processing of unknown signals where it could be used to train a neural network to recognize the signal
across the entire monitoring network.
3.1.2 Data security
Given the amount of data that may be collected and transferred to a control centre, it may be advisable
to implement appropriate data security measures, which may involve data encryption, or other data
security measures. This topic is addressed in detail in Recommendation ITU-T X.1601 – Security
framework for cloud computing. New technologies such as Blockchain may be relevant to such
networks. Blockchain (e.g. alliance chain mode) is a distributed ledger technology developed through
the integration and innovation of computer technologies such as distributed data storage, peer-to-peer
transmission, consensus mechanisms, encryption algorithms and smart contracts. It has the
characteristics of transparency, traceability and tamper resistance.
 Rep. ITU-R SM.2542-0 7

3.2 Big data challenges of a distributed spectrum monitoring system

Deploying high numbers of distributed RF sensors and receivers to perform continuous monitoring
across wide segments of the spectrum creates network capacity challenges which may result in high
costs associated with maintaining network connections to the processing centre(s). The volume of
monitoring data is directly related to the type of data being sent by the RF sensors – time-series data,
FFT-based spectral data, or RF parametric data (described below). Each type of data is needed for
different measurements typical to monitoring tasks. For example, signal classification and technical
identification often requires I/Q time-series data whereas occupancy measurements can be done with
either spectral data or RF parametric data. I/Q time-series data is the heaviest form of monitoring data
needing the highest level of network capacity to transfer a specific bandwidth as compared with
spectrum or RF parametric data.
3.2.1 Monitoring data types and impact on network bandwidth
As an example, performing continuous streaming of I/Q time-series data at 20 MHz acquisition
bandwidth requires 1.25 (filter shape) * 20 MHz (bandwidth) * 2 (I and Q) * 16 bits (I or Q) =
800 Mbit/s data rate to be sustained, close to the speed of Gigabit LAN. A further constraint is that
I/Q data capture must be continuous, so that the original modulation is not affected which is essential
when classifying the signal or performing technical identification. One hour of continuous I/Q time-
series streaming produces about 360 GB of monitoring data.
By comparison, observation of a continuous spectral streaming of a 20 MHz bandwidth using trace
processing to reduce time resolution and assuming a 16k-point FFT roughly produces a 10 Mbit/s
data rate. One hour of continuous FFT-based spectrum streaming produces considerably less data
than I/Q streaming, roughly 4.5 GB. Reducing the spectrum update rate to 1 scan/s reduces the data
rate to about 24 kB/s. In this case, one hour of 1 scan/s monitoring produces 86 MB. Transfer and
storage of raw spectral data requires significant bandwidth and storage media.
RF parametric data is the result of processing performed on the spectral data from an RF sensor or
receiver. A level or other form of threshold is applied to the spectral data and each energy exceeding
the threshold has basic external parameters extracted and stored and/or accumulated for a period of
time. This form of energy detection can benefit occupancy measurements and does not transfer or
process noise (or signal) energy below the energy threshold (see Fig. 2). Parameters that can be
extracted include:
– centre frequency;
– bandwidth;
– amplitude/field strength;
– time stamp.
If an observation interval is applied to the data collection, additional parameters can be computed and
stored. These are each computed within the specified interval:
– Percent occupancy;
– time of first and last intercept;
– duration (minimum, maximum, average);
– number of detections;
– bandwidth statistics (minimum, maximum, average);
– amplitude/field strength statistics (minimum, maximum, average).
8 Rep. ITU-R SM.2542-0

FIGURE 2
Development of RF parametric data

Using an observation interval allows the spectral sweep to update in real time, while the data reporting
is lessened to one record per detected energy per each observation interval.
This process improves the data efficiency of spectrum monitoring converting raw spectral data into
textual data that can be transferred or stored for further processing, analysis, and provide well
formatted data to ML algorithms.
To create FFT-spectral or RF parametric data, a monitoring system must have the capability to digitize
the receiver output (FFT samples) in real-time and incorporate local processing. Going further, local
intelligence to identify anomalies and access to the license database increases the sophistication and
cost of the monitoring station but lowers the cost of the network traffic. Ideally, an RF monitoring
station has the ability to produce all necessary forms of data to support different monitoring tasks. As
monitoring systems become more densely deployed to detect low-level wireless services and
localized interference, more processing aided by ML algorithms will be needed to reduce the network
data as much as possible.
3.2.2 ML algorithms applied to spectrum monitoring
There is a large volume of technical papers and presentations on ML based spectrum monitoring.
This Report provides a basic summary of the developments and applications. The first step in the
process of development may be to discuss the purpose of effort to develop ML and what benefits will
be obtained. Typical spectrum monitoring tasks that support decisions related to management include
making frequency allocations, detecting and classifying interference, determining license
compliance, and predictive maintenance (possibly for the case of terrestrial or satellite downlink
monitoring).
As the concept of dynamic spectrum sharing develops and is realized in some cognitive networks
(e.g. IoT), making frequency allocations in real-time, autonomously, based on deep learning of the
local spectrum must be enabled by extensive monitoring of the bands – and more specifically –
occupancy data [3]. Granular occupancy data of the frequencies to be shared in the appropriate
locations will be essential to training the ML algorithms to realize intelligent allocations.
Implementing this capability may require a combination of classification and regression test
algorithms.
For the detection and classification of interference, an ML algorithm will need to observe and learn
the local spectrum. Training data will need to represent normal signal energy and then use a form of
anomaly detection to detect interference. Signal classification will also likely be important for this
 Rep. ITU-R SM.2542-0 9

task and need to make use of neural networks – requiring I/Q time-series data from the monitoring
receivers. See § 5.
License compliance and predictive maintenance applications will rely heavily on either spectral or
RF parametric data (centre frequency, bandwidth, field strength) observed and learned over long
periods of time and processed by a regression algorithm.
3.2.3 Common types of ML algorithms
There are four basic forms of ML that may be applied to spectrum monitoring: classification,
regression test, anomaly detection, and clustering. These are described briefly.
– Classification answers the question, “Does this data have the attributes that belong with this
group or that?” The output is a category rather than a value. The decision is based on
comparison to the training data. This may be applied to occupancy data in support of
autonomous frequency selection and allocation.
– Regression attempts to predict the relationship between independent and dependent datasets,
looking for a line fit of the data representing a form of correlation. The output is typically a
value between −1 and +1 where 0 represents no correlation, +1 positive correlation, and −1
negative correlation. Regression testing operating on FFT-spectrum or RF parametric data
may be useful for predictive maintenance applied to a satellite earth station by closely
monitoring RSS on a downlink and taking into consideration local environmental data on
humidity, precipitation, temperature, and wind. It is important to consider that the RF
parameters of an emission measured by a monitoring station may fluctuate due to changes in
the local environment, traffic conditions, propagation changes caused by humidity or
temporary blockage of the signal.
– Anomaly detection looks for outliers in the dataset. Traditional spectrum monitoring software
has relied on rule-based anomaly detection using setpoints on key parameters to trigger an
alert or take some action. An ML-based anomaly detector develops patterns in the training
data and attempts to fit new observations into one of those patterns. Data that does not fit is
flagged as a potential anomaly. Anomaly detection operating on spectral or RF parametric
data may be useful for detection of interference or the emergence of an unknown emission.
– Clustering causes the input, which is usually raw unlabelled data, to be sorted into collections
based on the similarity of parametric qualities describing the data. This may be useful when
applied to detection and classification of novel signals where the frequency plan is unknown.
These algorithms and applications are only examples of what is available today commercially in
spectrum monitoring solutions. Some current software solutions provide high accuracy signal
classification of modern services such as 4G LTE, Wi-Fi, and 5G using I/Q time-series data processed
by a neural network.
As monitoring stations may require more advanced processing, trade-offs between cost,
sophistication, density of sensors, and network backhaul capacity will need to be evaluated prior to
specifying a solution.
3.2.4 Metadata and time synchronization
Distributed monitoring systems likely need to embrace a store-forward architecture to account for
network latencies and disruptions. Store-forward is a data communication technique in which a
message transmitted from a source node is stored at an intermediary device before being forwarded
to the destination node.
In a typical deployment, each monitoring site detects signals (frequency bands) of interest, performs
some processing and generates detailed characteristics on the spectrum. As these monitoring nodes
work independently, comparing data from different sensors without a reference base creates
10 Rep. ITU-R SM.2542-0

challenges in gaining valuable insights. Therefore, it is essential to have metadata on site readings
accompanied with a precision time stamp so that monitoring sensors can be synchronized to undertake
coordinated spectrum measurement surveys.
3.2.5 Sensor data format
There are many approaches to spectrum measurement resulting in many kinds of data, the data needs
to be “graded” or “classified” by its characteristics. Examples of this might include:
a) RF background noise;
b) Energy or power spectral density;
c) I/Q samples or FFT samples;
d) Decimated I/Q samples;
e) Spectrum occupancy data;
f) Geolocation data;
g) Level of metadata supporting the measurement data.
3.2.6 Heterogeneity in spectrum monitoring equipment
A distributed spectrum monitoring system will use spectrum monitoring sensors with distinctive
characteristics, operating parameters often from multiple vendors. A key challenge is to support
different measurement equipment that support different communication APIs, sampling rates, sweep
range, noise figure and frequency ranges.
To make the spectrum monitoring system sustainable and permanent, yet able to evolve over time, it
must be considered a system-of-systems, where the entire system has an indefinite lifetime, while the
individual monitoring elements that comprise it have finite lifetimes. For example, if a receiver in the
system becomes outdated it may be replaced by a current one without requiring the replacement of
the whole system.
3.2.7 Local and central databases
Distributed site databases can contain detailed information on smaller geographic areas and can be
updated quicker to support spectrum sharing. A centralized database can leverage monitoring results
and data from various sites to drive analytics. It is not a question of one or the other but leveraging
both, depending on the situational need. The ability to merge spectrum monitoring data will enable
fine-grained analysis of spectrum occupancy at the national, regional, urban and local level.

4 Big data spectrum monitoring

Spectrum monitoring networks need the ability to collect and integrate large amounts of data. The
influx of data (volume, variety, and velocity) will strain traditional operational procedures. An
information management approach augmented by AI will result in continuous cataloguing of the
spectrum on a real-time basis. This data supports analytics, real-time dashboards and mapping tools
to enable visualization of current and historical spectrum usage, showing coverage areas, interference
areas and unused frequencies:
– Applying AI techniques to learn the spectrum environment;
– Information on unused and under-used frequency bands;
– Interference reports with geolocation and signal recognition;
– Predictive analytics and data mining to assess KPIs on spectrum utilization, spectrum
efficiency, spectrum allocation, spectrum users, available spectrum, and interferences;
– Spectrum/frequency allocation alternatives and improvements based on user requirements.
 Rep. ITU-R SM.2542-0 11

4.1 Big data spectrum monitoring benefits

1 On-demand survey of spectrum utilization and classification (primary, secondary, licensed,
unlicensed, static or dynamic).
2 Quantification of the available spectrum by spectrum bands and location (global, regional
and local).
3 Coverage analysis for QoS and QoE assessment.
4 Identification, geolocation and cataloguing of interference sources and signals of interest.
5 Detection of unauthorized transmissions, jamming devices and rogue services.
6 Enable proactive monitoring to enhance total spectrum usage and spectral efficiency by
exploiting underutilized spectrum with dynamic allocation where possible.
7 Implement ITU-recommended electromagnetic monitoring, technical verification, Direction
Finding (DF), and Geolocation measurements.
8 Ensure a fairly accurate signal detection and DF / geolocation.
9 Enable future expansion/integration of sensors from different manufacturers.

4.2 Big data spectrum monitoring solution

The solution as shown in Fig. 3 is an example of an intelligence-enabled monitoring network,
a central database layer, and an application layer for fine grained monitoring and enforcement of the
electromagnetic spectrum covering U/V/S/EHF frequency spectrum.
12 Rep. ITU-R SM.2542-0

FIGURE 3
Big data spectrum monitoring solution – a conceptual view
Collection Layer

RFML
MONITORING SMART SENSORS
SENSOR RFML
MONITORING SMART SENSORS
SENSOR

Data Storage Layer

License
Monitoring Database
Database

Management Layer

Spectrum Usage Statistics Management Reporting Command Center

4.3 RF collection layer: example big data spectrum monitoring network

A big data spectrum monitoring network necessitates an overall change in the industry from
equipment-oriented systems to data processing-oriented systems. The RF monitoring sensor will
evolve from basic spectrum monitoring functions to one equipped with processing intelligence to
survey spectrum, catalogue signals, detect interference and capture violations.
 Rep. ITU-R SM.2542-0 13

FIGURE 4
National big data spectrum monitoring network

The spectrum big data platform illustrated in Fig. 4 includes a network of RF monitoring sensors
capable of supporting existing monitoring assets along with new sensors that offer enhanced coverage
and intelligent processing. A centralized big data repository collects the spectrum monitoring data,
violations, alarms, geolocation data and signal identification to enable users to make informed
decisions or allow the system to autonomously make decisions.

4.4 Data storage layer

Information in the central database should be stored using industry standard formats and unique
identifiers so that the data from various different monitoring systems can be merged into a common
database.
It is imperative that a big data spectrum monitoring solution enables the operator to undertake
coordinated spectrum measurement campaigns and the ability to task monitoring sensors to perform
one or more functions in tandem or in parallel. This enables operators to easily prepare Monitoring
Task Plans (MTPs) for multiple stations and provision the same on a regional or local level.
All the information stored in the central database is stored using unique identifiers so that the data
from stand-alone and third-party systems can be merged into the database.
14 Rep. ITU-R SM.2542-0

FIGURE 5
Data format for spectrum big data gathering

4.4.1 Collection of big data (definition of collected data)

The collection and processing format, illustrated in Fig. 5, consists of both meta data and spectrum data.
The collected data is processed into accumulated statistical spectrum data such as maximum, average,
and minimum values by processing sequential FFT spectrum data from RF sensors.

FIGURE 6
Example architecture of internet-based spectrum big data gathering process
 Rep. ITU-R SM.2542-0 15

As described in more detail in Fig. 6, the spectrum collection system is configured to repeatedly
process the entire band of the spectrum monitoring receiver (e.g. 20 MHz to 6 GHz) and transmit the
statistically processed spectrum data such as maximum, minimum, and average traces during the
collection time.
4.4.2 Data pre-processing
The parameters necessary for spectrum monitoring tasks are processed into a pre-processing format
for spectrum big data analysis by time series, frequency, and location based on license data (for
example) to improve the speed of analysis. In addition, the local processing in each of the RF sensors
enables detection of illegal transmissions and interference in near real time.
4.4.3 Data analysis and visualization
Spectral raw data and pre-processed data can be developed for data processing and visualization (as in
Fig. 7) by various open source software packages for storage, analysis, and processing.
Spectrum data should be managed for each location and support detailed analysis of monitoring tasks
and other activities through post-processing. Longer-term data can be stored using compression
methods such as compression algorithms and statistical processing of spectrum raw data.

FIGURE 7
Spectrum monitoring with local processing for illegal transmission and interference detection

4.5 Data management layer

System management functions should be easy to use, facilitate collaboration, and enable data
integration and analysis.
The need to automate monitoring tasks, group sensors based on functional capabilities and geographic
proximity will be needed to enable concurrent collection of data from one or more radio monitoring
sensors to gather fine grained data on a national, regional, urban or local level.
Comparative big data spectrum monitoring techniques enable the operators to generate views of
spectrum activity across geography and time. The operators undertake coordinated spectrum
16 Rep. ITU-R SM.2542-0

measurement campaigns with the ability to task monitoring sensors to perform one or more functions
in tandem or in parallel. Cooperative measurements are automated by grouping sensors based on
functional capabilities and geographic proximity.
The use of cooperative groups of sensors working as teams can also provide real time views of activity
and further enable pro-active monitoring and alarms to changes in the environment. These changes
may identify unexpected use, overuse or loss of service including interference events.
The use of intuitive GUIs enables the user to:
– Manage cooperative spectrum monitoring teams;
– Manage tasking;
– Query results (data analytics);
– Set the time period for measurements;
– Set the geographical area and location;
– Integrate multiple data sets and databases to facilitate analysis in support of decision making.
Integration of different systems, through the use of APIs will support commercially available
Business Intelligence (BI) / database reporting tools which provide an interface to build dashboards
to gain access to a data schema. Published data schema make it possible to create new views and
perspectives on the data. This also enables the integration of multiple data sets and associated views.
Through this data schema and the integration of BI tools, users are able to store and provide wide
access to the monitoring data. The data can be accessed by multiple clients, each with their own
customized view of the data of interest for the intended consumer, organization, or department. With
the availability of both spectrum monitoring and geolocation data, dashboards can show the results
of the various datasets in very informative and insightful ways. The datasets allow for spectrum
occupancy data to be analysed across a band combined with geographical information. This data can
be filtered and rearranged using various dashboard tools (see Figs 8 to 11).
Annex 1 to this report provide an example and use case of AI and Big Data, describing the experience
of Korea (Republic of) with the use of AI and big data for spectrum monitoring, including a typical
intuitive GUI.
 Rep. ITU-R SM.2542-0 17

FIGURE 8
Frequency occupancy based on hourly intervals

FIGURE 9
3D visualization of spectrum occupancy over a frequency range and time
18 Rep. ITU-R SM.2542-0

FIGURE 10
Built-in query for evaluating spectrum occupancy in 2D and 3D formats

FIGURE 11
Dashboards for spectrum assessment by frequency, time and location

5 Realtime data-driven spectrum awareness using RFML

With advances in wireless communications and wireless devices, and increasing diversity of new and
adaptive wireless systems, it is important to not only detect and classify existing protocols, but to also
identify, learn and adapt to future protocols and wireless systems as they are deployed – including
 Rep. ITU-R SM.2542-0 19

those in mobile/cellular networks, IoT networks, autonomous systems, and other emerging
technologies.
RFML signal processing can use neural networks to learn RF protocol and emitter characteristics in
order to rapidly detect and classify their type [4]. As new wireless systems, services, and protocols
are introduced, RFML based spectrum sensing systems can identify new and anomalous types of
emissions, help to label, and annotate them in datasets, and train to identify them when seen in the
future. Further, microprocessor hardware has increasingly become optimized for cost and energy
efficient neural network inference and can be deployed to the edge of the monitoring network.
Deploying RFML at the edge provides near real-time sensing, with the ability to efficiently detect
and identify numerous spectrum events, providing an efficient and effective way to audit and analyse
activity over broad frequency bands and long-time intervals. This can be done without the need for
storage or transmission of raw I/Q time-series data files. A data-driven workflow describing the
interactions of edge inference, event analytics, data curation, model retraining and deployment for
new signals and emissions is shown in Fig. 12.

FIGURE 12
Data-driven spectrum sensing workflow

Leveraging AI through RFML-based spectrum sensing enables efficient and scalable signal detection,
identification, analytics, anomaly, and trend detection with near-real-time insight into signal activity
and band occupancy. It provides a smart view into signals activity across a wide range of bands and
technologies and event streams of metadata describing signal types, power levels, properties and
behaviours. This can be leveraged by automation at scale or visualized by users as shown in the
spectrogram illustrating the detection, identification, and annotation of RF emissions in a wide band.
(See Fig. 13).
20 Rep. ITU-R SM.2542-0

FIGURE 13
Sample graphical output of RFML-based sensor showing signal classification and confidence level

AI driven spectrum monitoring has the potential to offer fully autonomous 24/7 signal detection and
recognition and analytics from multiple monitoring sites to support:
– Confirmation that licensed protocols and systems are being used;
– Confirmation that spectrum allocations are being actively used;
– Interference detection, identification, and feedback;
– Available spectrum assessment and feedback for optimization;
– Enablement of air-interface digital twin information;
– Rapid learning of new types of emitters and protocols;
– Automated analytics and monitoring of wide range of trends in spectrum access.

6 Summary
There have been many advancements on the application of ML and AI to spectrum monitoring and
there is a large volume of technical papers outlining the use, the algorithms, and the limitations of the
technology, as well as commercially available software solutions. These developments will likely
lead to more autonomous spectrum monitoring systems that are not only based on classic automation,
but also on intelligent, learning capabilities able to collect and fuse data from multiple sources (e.g.
spectral data, environment, traffic, events, etc.) and make rapid and good decisions at scale.
Dense distributed monitoring networks produce high quality and real-time occupancy data that makes
the discovery of available spectrum across the covered area possible. It also could support accurate
predictions for autonomous allocation of frequencies for specific wireless applications like IoT [2].
In some cases, spectrum sensing capabilities could be co-located and co-deployed with network
infrastructure such as within base station and radio unit hardware, and could where appropriate,
feasible and applicable coordinate with network elements.
Intelligent distributed spectrum monitoring systems hold the potential to improve the efficiency of
regulatory personnel by reducing the effort to obtain the monitoring data required to manage the
spectrum and allowing insight into spectrum activity trends across many locations, bands, types of
emitters, and timescales which were previously unmanageable through other means.
 Rep. ITU-R SM.2542-0 21

In order to build a large-scale spectrum monitoring and automated spectrum allocation system, it is
necessary to integrate spectrum sensing data through common rules and standardization of
measurements. It should be possible to measure the related RF frequencies for each site using different
spectrum sensors, and to distribute them in a standardized integrated format in order to obtain the
trust of spectrum big data analysis. Use of standard protocols for RF datasets can result in
interoperability and efficient deployment, integration, and automation of such systems on a broad
scale.

7 References
[1] Wong, Lauren J., IV William H.Clark, Bryse Flowers, R. Michael Buehrer, Alan J. Michaels and
William C. Headley. “The RFML Ecosystem: A Look at the Unique Challenges of Applying Deep
Learning to Radio Frequency Applications.” arXiv: Signal Processing (2020): n. pag. Online:
https://arxiv.org/pdf/2010.00432
[2] Recommendation ITU-R SM.1046 – Definition of spectrum use and efficiency of a radio system
[3] Baltiiski, P., Iliev, I., Kehaiov, B. et al. Long-Term Spectrum Monitoring with Big Data Analysis and
Machine Learning for Cloud-Based Radio Access Networks. Wireless Pers Commun 87, 815–835
(2016). https://doi.org/10.1007/s11277-015-2631-8
[4] T. J. O’Shea, T. Roy and T. C. Clancy, “Over-the-Air Deep Learning Based Radio Signal
Classification,” in IEEE Journal of Selected Topics in Signal Processing, vol. 12, no. 1, pp. 168-179,
Feb. 2018, doi: 10.1109/JSTSP.2018.2797022

Annex 1

Solution for data-driven AI and big data spectrum monitoring

in Korea (Republic of)

Korea is performing spectrum monitoring through big data analysis, using a dedicated software
program which can realize the data-driven big data analysis. The spectrum is monitored by three
monitoring sites in Korea as a pilot system. Each cloud cluster server uses sensors covering 30 MHz
to 7.5 GHz to collect spectrum data for building a spectrum database.
The government of Korea plans to expand the data-driven big data spectrum monitoring system to
the whole country in the future. There is a clear direction that spectrum management should include
big data and AI (Artificial Intelligence) technology and have a global perspective.
22 Rep. ITU-R SM.2542-0

FIGURE A1-1
Spectrum big data analysis workflow

Figure A1-1 illustrates the spectrum big data analysis workflow including data workspace, analysis
workspace and visual workspace.
The RF sensors autonomously collect designated spectrum data considering the national frequency
allocation table and radio license database. The system produces spectrum data which calculates
through an FFT process the statistical values including maximum, minimum, and average. The
generated spectrum data is periodically transferred to a cloud big data server. The cloud server saves
two forms of the spectrum data including raw data and pre-processing data. When spectrum data is
saved at the cloud server, it is stored as meta data which can distinguish each parameter. The meta
data consists of several variables as listed in the table below. This spectrum data can be accessed at
any regional office and central office. Administrators can easily access the spectrum data due to its
organized meta data form.

TABLE A1-1
Meta Data Format
Class Category Type
Sensor type String
Sensor ID Integer
Operator’s management information String
Antenna model String
Sensor & General Information Spectrum measurement time Float
Spectrum form Integer
Sensor temperature Float
Sensor RF status Integer array
Measurement time zone String
 Rep. ITU-R SM.2542-0 23

TABLE A1-1 (end)

Class Category Type
Measurement date and time String
Measurement longitude Float
Measurement latitude Float
Measurement height Float
GPS lock status Integer
Measurement Information Measurement frequency Double
Measurement bandwidth Float
Reference level Integer
Number of spectrum bins Integer
Interval of spectrum bins Float
Signal existence flag Integer array
Signal Information Average noise level Float array

The raw spectrum data serves as a near real time streaming spectrum form for interference and illegal
frequency detection using AI or other algorithms. The pre-processing data is used for fast big data
analysis of spectrum monitoring tasks such as occupancy, radio noise and radio quality.
Figure A1-2 shows display examples for occupancy, radio noise, radio quality, interference and
illegal frequency analysis.

FIGURE A1-2
Major analysis factors of the data-driven big data spectrum monitoring solution

a. Spectrum occupancy
24 Rep. ITU-R SM.2542-0

b. Radio noise

c. Radio quality
 Rep. ITU-R SM.2542-0 25

d. Illegal frequency

A spectrum occupancy screen such as illustrated in Fig. A1-2 is an indicator of how efficiently each
allocated spectrum is utilized. The spectrum efficiency plays a significant role in spectrum retrieval
and reallocation (spectrum refarming) because the spectrum is a limited resource.
Radio noise environment is another important factor. During spectrum monitoring a combination of
the spectrum of interest and the noise is measured. The radio noise refers to random fluctuations of
the instantaneous spectrum because of artificial and natural noise.
Radio quality is an additional important factor. An analysis result of radio quality indicates whether
or not a signal complies with licensed radio quality parameters such as SNR, channel power,
occupancy bandwidth, etc.
A remaining important monitoring factor within the dedicated software program is an autonomous
illegal frequency search. This is important since an illegal frequency can influence and disrupt the
legal frequency authorized by the administration.
Illegal and interference frequency analysis shows statistical results of time series occurrences and
historical data for each frequency band or space. An illegal or interference frequency is detected by
using streaming spectrum data from the automatically updated licensed database.
The dedicated software program can analyse the spectrum three-dimensionally, with x, y and z
coordinates as illustrated in Fig. A1-3.
Figure A1-3 illustrates spectrum occupancy as measured at three monitoring stations for each axis as
x: selected monitoring station(space), y: frequency channel assigned(frequency) and z :spectrum
occupancy for assigned period for analysis condition(time).
26 Rep. ITU-R SM.2542-0

FIGURE A1-3
Three-dimensional spectrum monitoring analysis

FIGURE A1-4
Streaming data analysis

The software monitors the illegal frequency and interference spectrum in real time. The streaming
data analysis can track previous records which detected illegal or suspicious spectrum.
This analysis technique is illustrated in Fig. A1-4. The suspicious spectrum can be observed as a
sudden change in the streaming data flow. Figure A1-4 illustrates the detection of an illegal frequency
(upper trace) and the analysis results of a historically similar illegal spectrum occurrence at previous
times (lower trace). This analysis can be selected to be compared with a particular region at the current
time to roughly estimate location of transmitter.
 Rep. ITU-R SM.2542-0 27

FIGURE A1-5
Space data analysis by big data spectrum monitoring

FIGURE A1-6
Raw spectrum data analysis by drill-through method based on big data

The big data spectrum monitoring solution can visualize the spatial distribution of the spectrum
occupancy, radio noise, illegal frequency, and intensity of the spectrum. Figure A1-5 shows the
spatial distribution of these four parameters, each plotted on a geographic map.
The big data spectrum monitoring system can analyse various parameters based on raw spectrum
data, including electric field strength, channel power, noise level, etc. Administrations can access the
raw spectrum data by a drill-through method which analyses the raw spectrum data (in the spectrum
DB) to gain additional insights. Figure A1-6 illustrates this function.
The data-driven AI spectrum monitoring solution uses state-of-the-art machine learning technology
such as an anomaly detection AI algorithm. The AI spectrum monitoring solution can observe and
detect unusual spectral images and events by using an AI training method which examines spectrum
images based on training results. The AI training methods include supervised, unsupervised and
reinforcement learning. The AI spectrum monitoring solution typically uses the supervised learning
method for analysing the spectrogram images.
In summary, the data-driven big data spectrum monitoring solution can be realized through a
dedicated platform that measures the important spectrum KPIs of spectrum occupancy, spectrum
quality, spectrum noise, and illegal frequencies. Administrations can investigate important spectrum
incidents based on big data analysis.
28 Rep. ITU-R SM.2542-0

There are clear advantages of the data-driven big data spectrum monitoring solution:
First, a big data spectrum monitoring solution can deal with large amounts of spectrum data. The
spectrum data is collected and analysed 24/7. It is autonomously connected with the national license
database for processing without human intervention. It is very hard to monitor these large amounts
of data by a conventional approach.
Second, it is possible to anticipate future spectrum events by using data-driven big data spectrum
monitoring to compare past spectrum data to the present. Administrations can predict future spectrum
events based on spectrum big data analysis. Using this big data analysis, administrations can supervise
predictable spectrum interference based on previous historical data trends.

Annex 2

Mobile public transport based big data acquisition for spectrum mapping

For spectrum managers implementing dynamic spectrum sharing, it is important to have a realistic
overview of spectrum coverage (downlink) and spatial spectrum usage (uplink). This can be achieved
by collecting spectral data in many different locations across an area of interest (e.g. urban areas) and
plotting the data on a map display. Spectrum mapping creates a continuous heatmap of spectrum
coverage and usage and requires spatial interpolation algorithms. The better the density of the
collected data, the more accurate the interpolation algorithm can predict power levels between the
measurement points. To achieve this data density in urban areas, the use of public transport and
service vehicles like busses, trams, taxis, garbage trucks etc. to carry RF measurement equipment is
ideal. These vehicles are normally operating around the important areas of the city with a route that
supports collection of dense spectral data which could save administrations costs and manpower.
Further, the data density in terms of location and time on public transport cannot be achieved by a
few dedicated monitoring vehicles.
Figure A2-1 illustrates the data density consideration. The area of interest is surrounded by the dashed
red line. The lowest accuracy is achieved if driving only along the purple route adjacent to E11.
Medium accuracy is achieved by driving the red route following E11, D63 and D86, then turning
right up to E11. The highest accuracy and data density is achieved by taking the blue route following
the interior of the area in addition to the red route.
 Rep. ITU-R SM.2542-0 29

FIGURE A2-1
Alternate driving routes to achieve different spectrum measurement densities

The RF monitoring sensor and antenna for public transport applications should have the following
characteristics:
– Compact format;
– Embedded mass storage;
– 12-16 VDC input, to be powered by vehicle;
– Embedded backup battery in case vehicle power is lost;
– Embedded cellular networking router (LTE, 5G, Wi-Fi);
– Compact wideband antenna.
Beside the hardware requirements software automation is key for this application. The data collection
and upload must be fully automated and is described in the following process circle (Fig. A2-2).
30 Rep. ITU-R SM.2542-0

FIGURE A2-2
Spectrum measurement process for mobile monitoring platforms

The process starts with the definition of the measurement job parameters. A measurement job is a set
of parameters and features to be executed by the RF sensor if the defined conditions (e.g. time and
area) are met. As cell sizes and transmitting power is decreasing and the mapping requires a good
spatial resolution to be meaningful, the goal is to collect as many scans per time interval while driving.
It is recommended to complete one complete scan, including re-tuning time with a spatial resolution
of at least 100 m. Expressed in other words, for each 100 m of driving, a band or channel of interest
is re-visited. This requires high-speed monitoring receivers. The minimum tuning speed to fulfil the
100 m requirement depends on driving speed, frequency range to be monitored, resolution bandwidth
(RBW), and trace averaging settings. Table A2-1 provides required re-visit times and scan speeds for
an example frequency range and different driving speeds to fulfil the 100 m spatial resolution
requirement. The scan speed includes time for re-tuning the receiver, as continuous scanning is
required. If a frequency list scan is used, the RF sensor must be able to complete the list within the
re-visit time for the next scan to start.

TABLE A2-1
Scan speed needed for 100 m resolution by drive speed
Drive speed Required re-visit time Required scan speed over
(km/h) example range 80 MHz - 6 GHz
30 12 s 494 MHz/s
50 7.2 s 823 MHz/s
80 4.5 s 1.32 GHz/s
100 3.6 s 1.64 GHz/s
120 3s 2 GHz/s
 Rep. ITU-R SM.2542-0 31

For a standard scan, the RBW needs to be chosen in relation to the smallest signal bandwidth expected
in the frequency range of interest. For channel list scan, an appropriate RBW per channel can be
selected. A list scan saves disc space but can increase the re-visit time in certain scenarios.
While the public transport drives around the city area, the received signal strength at the monitoring
receiver is varying significantly. When driving near strong emitters, the receiver can get into
saturation/overloading. For this reason, gain and attenuation settings are very important. Monitoring
from a public-transport vehicle results in close proximity to many emitters along the route and
therefore, a planned reduction in the sensitivity of the RF sensor (by adding input attenuation)
provides some immunity against overloads caused by strong signals.
The amount of data per RF sensor can range from a few hundred gigabyte up to a terabyte of data per
day. Field trials from 6.5 hours of data collection in a frequency range from 80 MHz up to 6 GHz
produced a data amount of 300 GB. To save disc space, automation of the record start and stop times
is needed. This can be achieved using an area vector-based trigger (which relies on the GPS location
of the vehicle) or vehicle ignition-based trigger.
The collected data is transferred to a data server to process a map of the spectrum together with other
measurements from other vehicles or fixed sensors. The data transfer can be achieved by installing a
Wi-Fi or 5G access point in the depot or garage of the vehicles so that as soon as the router of the RF
sensor connects to the access point, it will start its data transfer. During operation, it is recommended
only to stream the trace of the live spectrum as well as the health and operational status of the RF
sensor to a control centre instead of transferring the raw measurement data via the network as it will
use a lot of cellular bandwidth.
As the embedded storage of the RF sensor is limited, it is suggested to automate data deletion after
upload to the server is completed.
If a large number of public transports are equipped with RF sensors, a good time resolution in urban
busy areas can be achieved. The more often vehicles equipped with RF equipment drive in the same
areas using the same or similar routes, the better spectrum usage trends over time can be identified.
As mentioned above, big data from mobile RF sensors combined (optionally) with fixed stations can
be used to process continuous heatmaps of spectrum coverage and usage for urban areas. After the
data is collected and transferred to a (cloud) server, the data mining and fusion takes place. To make
the measurements comparable, they must be converted to field strength values considering the signal
path losses and antenna factors as well as measurement settings of the individual RF sensors. The
antenna factor and associated uncertainty has to be characterized for each specific combination of
vehicle type and antenna positioning. For each defined channel, a map is produced. To produce a
continuous map, interpolation between the measurement points needs to be calculated. Interpolation
can range from a simple linear interpolation to one supported by wave propagation models. If high
precision maps with terrain and clutter data combined with a wave propagation model are used for
the interpolation, the field strength inside the measurement gaps (where no vehicle was driving) can
be estimated in the most accurate way. Besides geographical interpolation, time aggregation also
needs to be done. Depending on the time accuracy requirement for the evaluation, this can range from
a daily average, which means all collected data of a day is aggregated, up to 10 minutes time
resolution in busy areas. The time resolution depends on the geographical re-visit time of public
transport and fixed stations in the measurement area.
After the data mining and fusion, the data rendering/visualization needs to offer meaningful maps for
spectrum managers to evaluate the real-world spectrum situation. The basic map includes a field
strength distribution with colour coded dBµV/m values. The user can visualize the field strength map
for all pre-defined channels and the band maps that will display the aggregated field strength for an
entire band. See Fig. A2-3 for a process flow for creation of a spectral map.
32 Rep. ITU-R SM.2542-0

If the spectrum mapping tool is connected to the transmitter database, licensed stations can be
displayed together with the heatmap. By correlating the maps with licensed station information, the
operator can identify hotspots which are not assigned to a station.

FIGURE A2-3
Mobile spectrum measurement workflow

The field strength maps (Fig. A2-4) can provide:

– Coverage mapping
• Aggregated view for total band to evaluate total availability;
• Customized views for different providers, which provide best coverage in which area?
Available spectrum?
• Comparison to wave propagation calculations, is the calculated coverage like the
measured coverage?
– Frequency usage:
• Mapping the uplink;
• Evaluate areas of high and low usage, identify overloads;
• Identify and localize interference or illegal spectrum usage;
• Evaluate potential and built up a data foundation for spectrum sharing.
The big data system should offer further maps based on the field strength maps. This can be binary
coverage maps, where the user can define a threshold for which the areas above are coloured as
covered and below as not covered (see Fig. A2-5). A further tool for big data spectrum mapping are
radiation maps in V/m. By this administrations can verify if radiation limits are met in all areas. Such
maps are also well suited to be published to eliminate concerns for health problems from radiation.
Rep. ITU-R SM.2542-0 33

FIGURE A2-4
Field strength maps

FIGURE A2-5
Coverage map