Aerospace and Electronic Systems Magazine May 2018 - 45

RF DATA AND ITS CLOUDIFICATION
Currently, RF data are considered to be either time-domain baseband in-phase and quadrature (IQ) data or frequency-domain
(spectrum) data. The RF data are produced by radio receivers that
can cover a wide band but operate one narrowband at a time. IQ
data require very high sampling rates and are generally appropriate for signals of short duration. For a long time, the datafication
of RF was not practical because of the very high sampling rate
required. Nevertheless, this has been very desirable, and over the
years, many platforms that produce "digital IQ" have appeared,
with generally increasing capabilities. For spectrum data, usually
only an absolute value obtained as a result of long-term averaging
is retained, and the data rate is much lower.
At first, considering RF data, just "digital IQ" or spectrum data
may seem adequate. However, because it does not include any
knowledge about the RF signal, such as location and center frequency, this definition is insufficient, particularly for RF data analytics. We define RF data as the time-domain IQ data paired with
all of the metadata (data about data), such as RF center frequency,
bandwidth, and location. Similarly, we define the spectrum values
paired with metadata as RF spectrum data. Note that the "digital
IQ" is not the raw RF signal and cannot be interpreted correctly
without knowing, for example, the RF center frequency and bandwidth. In a private cloud, these parameters are present, but implied,
which is why they have been ignored previously. In the presence
of multiple data streams generated by different types of devices,
this metadata can no longer be left implied, it must be defined explicitly and must be paired, i.e. associated with an IQ stream or
spectrum data. Furthermore, in addition to RF center frequency
and bandwidth, metadata can include many other parameters, such
as location, time stamp, sampling frequency, noise figure, signalto-noise ratio (SNR), antenna beamwidth, and polarization.
In a cloud architecture, the time instant a signal is processed
(actual time) is different from the time instant an RF signal passes
through the antenna (signal time). Therefore, the metadata must
make this distinction when paired with a signal. Furthermore,
the signal time can be obtained from the time stamp attached at
digitization by knowing the group delay of the analog RF circuit.
This shows that the metadata allow all information about the RF
MAY - JUNE 2018

signal impinging on the antenna to be described, i.e., it is an abstraction.
Simple metadata have been defined in [12]. The list of metadata must be extensible to support additional and more sophisticated RF data services. For example, both the metadata and the RF
signal may have attributes such as current value, average, variance,
maximum, minimum, precision, probability, and probability density function. These attributes should also be included. Some metadata parameters can be represented by a single value, while others,
such as frequency response, require a more complex data structure. Furthermore, any additional information that could be useful
can also be metadata, including information about the RF client
or other known RF devices and spectrum policies. In general, the
metadata can be formally represented by using general approaches
to describe any parameters and objects, such as ontologies [12]
or extensible markup language (XML). Ontologies and XML are
powerful formats that are both human readable and machine readable and would support a wide variety of RF cloud services.
We refer to radio receivers that produce RF data defined in
this way as RF clients. Therefore, an RF client includes a collection of resources associated with the frequency band selection, amplification, downconversion and upconversion of RF signals, and
producing data and metadata packets. The physical infrastructure
of an RF client is decoupled from the configuration of this infrastructure. Note that an RF client may transmit metadata packets
without corresponding data packets. RF clients can perform some
signal processing, preprocessing, and simple signal analysis and
can also have some network control functions, i.e., the separation
of functionalities between clients and clouds is flexible. Similar
flexibility has been discussed previously in [11], [12] but not based
on an open interface.
The ongoing trend of datafication of RF will lead to its cloudification, i.e., storing and processing of RF data by using cloud
techniques. While traditional data are generated per day or per
hour and big data more often, RF spectrum data are generated per
nanosecond or even more often. As a result, such huge amounts of
data produced by the datafication will naturally result in its cloudification.
The RF cloud or a RF cloud platform involves RF clients connected to a cloud. The cloud, or more accurately, cloud of clouds,

IEEE A&E SYSTEMS MAGAZINE

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine May 2018