Design of A Passive Radar Based On RTL-SDR Technol
Design of A Passive Radar Based On RTL-SDR Technol
Design of A Passive Radar Based On RTL-SDR Technol
Keywords: Antenna, directional channel, Reference channel, Doppler, GNU Radio, Oscillator,
RTL-SDR.
1. Introduction
The incessant technological development of UAV unmanned aerial vehicles is increasing; this is how several
models are currently available to anyone. In this sense, there is no restriction for its use, and there are UAV mod-
els that are manufactured for autonomous navigation and support the transfer of significant loads, reaching a few
kilograms, thus becoming potential means for the transfer of weapons and sensors. In this way, they represent a
potential threat to privacy and invasion of airspace in sensitive infrastructures [1].
The Armed Forces Research Center has raised the need to develop new perimeter surveillance solutions, espe-
cially in critical infrastructures. In this way, it is proposed to face the malicious or negligent use of manned and
unmanned aircraft, where it is important to detect their approach to military infrastructures or restricted areas. In
this sense, the most efficient solution is the implementation of a radar. However, there are two types of active
and passive radars; For presence evasion based on electronic measurements, passive radars consolidate a robust
and effective solution [2].
It is important to consider the size of these vehicles since they are getting smaller and smaller and are made of
non-metallic materials, which makes their detection difficult. Likewise, they are capable of rising to small
heights and navigating at low speeds, thus avoiding being captured by conventional radars [3].
For this reason, the present proposal focuses on the design of a passive radar with great functionalities such as
RTL-SDR technology. In this way, low frequencies are used for active use of radar, thus allowing wavelengths
the size of the UAV to be obtained, thus expanding the radar cross-section. In addition, the integration times used
are longer, thus offering a higher resolution of the Doppler effect and thus improving the processing of these
signals [4].
In the present investigation, 4 sections have been structured that allow describing the work carried out. Thus, in
the first section, the use of a passive two-channel reception radar with a reflected signal comparison algorithm is
proposed. Then, in the second section, signal processing is described, which describes how false detections are
discriminated and how the Doppler effect works in passive radar. The third part describes the necessary hardware
and the RTL-SDR technology to be used, as well as the compatible software. Finally, section 4 describes obtain-
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
ing data and handling the signal from a command line interface. To end the article, the conclusions and recom-
mendations reached are written.
Air vehicles are distributed for both the military and civil fields. This is how the development of a passive radar
for surveillance and monitoring of the airspace is proposed. However, it is a task with a high degree of difficulty
since the objects to be detected are increasingly smaller and operate at low altitude. In this way, it has aroused
interest in venturing into UAV detection and tracking techniques, but there is a problem in suppressing clutter
since these vehicles are influenced by false alerts. Likewise, the tracking of trajectories is interrupted by objects
or vehicles that are at the same height[5], [6].
Unlike active radar, passive radar uses transmitters found on FM, TV, or mobile phone frequencies. In this way,
the signals that exist in the environment are taken advantage of, this is how the signals are reflected in small
objects such as cars and light aircraft. The use of two antennas makes it possible to have two reception channels
that, through an algorithm, the reflected signal can be compared with a copy of the real signal, thus achieving a
bistatic radar display.
A passive radar can have separate receivers where the reference channel directly receives the light signal from
the transmitter while the second channel detects the echoes of the signal that has been bounced in the middle (see
Figure 1).
2
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
For the different tests, a frequency within the range of FM communications has been chosen, in this sense the
equations and measurements of the elements of the yagi antenna are described in the Table 1.
3
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
Next, the antenna simulation was carried out using the MMANA-GAL software, which is a free program for the
analysis of antennas. [7], [8]. This computer tool does not require many resources and is easy to use.
One aspect to consider in the simulation is the diameter of the rod that is used in the conductors whose
measurement is 3/8 of an inch. Likewise, it is important to enter the correct frequency, which in this case is 102.1
MHz. With these data, the software proceeds to process the information, delivering the radiation diagram as a
result, as can be seen in Figure 3.
The SWR standing wave ratio reaches an approximate value of 9.6% with reflected power, which allows
working without impedance coupling. The main building blocks of the antenna are 3/8-inch aluminum tubes for
the reflector, feeder, and directors. For the base and the mast, a one-inch square aluminum tube was used, in this
way the antenna is quite light since it has a large size. In the Figure 4, you can see the antenna implemented and
located at a height of 5 meters from the floor.
4
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
For the reference channel, it is recommended not to use a directive antenna since in most cases the topology of
Ecuador does not allow the FM antennas to be close. This is the case, so an RTL-SDR communications kit is
included whose main element is a monopole antenna as you can see in the Figure 5.
3. Signal Processing
One of the main encounters that passive radar faces is extracting noise from puny echoes. On some occasions,
this noise is caused by the leakage of the route signal towards the observation channel, as well as the signals
bounced off by objects such as mountains, buildings and constructions. The set of these signals is called clutter,
and to suppress it an algorithm has been developed that discriminates noise and makes it easier to observe the
echoes of the target. In equation (1), it is presented to the signal s0 n the signal received by the observation
channel which is the sum of the echo signal from the target star n , the clutter signal scl n and an auditory
noise v n .
s0 n star n scl n v n (1)
The signal received by the reference channel is sr n likewise, the reference signal is assumed to be a faultless
duplicate of the transmitted signal. On the other hand, the clutter signal is a combination of a direct path copy of
the transmitted signal and several reflections on multiple objects. Thus, the clutter signal can be modeled as a
sum M of scaled and delayed copies of the reference signal as shown in equation (2).
5
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
M 1
scl n bk sr n k (2)
k 0
In the disorder signal the coefficients bk they are unknown because they depend on the objects around the
radar. Appropriate estimates can be used bk to find the coefficients of the clutter signal. After subtracting the
clutter signal from the observation channel signal, the echo signal may be estimated.
M 1
star n s0 n scl n s0 n bk sr n k (3)
k 0
Then in the clutter removal algorithm the adaptive filtering can find the appropriate estimate for the noise
coefficients. The coefficients are obtained from the input data, see Figure 6.
To explain how an adaptive filter can determine the optimal x's, we start by building a signal model for the target
echoes. For N different targets, the echo signal can be represented as a sum of N copies scaled, delayed and with
Doppler shift with respect to the reference signal [9], [10]. In the following formula wk corresponds to the
magnitude, d k to delay and f k the Doppler shift of the target's k-th echo. While Fs is the sample rate.
M 1
j 2 nf k
star n wk sr n d k exp (4)
k 0 Fs
Different frequencies of complex exponential functions are orthogonal to each other. The target echoes are
linearly independent of the reference signal due to the Doppler shift, provided the Doppler shift is non-zero and
the signals are monitored over a long enough time period [9], [10].
The clutter signal has been considered to be a perfect linear combination of lagged copies of the reference signal.
The components of s0 n (observation channel) linearly connected to delayed copies of sr n (reference
channel). If these components are subtracted from the observation channel, only the echo signal of the target will
be obtained star n .
In terms of vector spaces, it can be said that the disorder signal scl n it occupies a vector space whose base
vectors are lagged copies of the reference signal. This space can be called "messy space." The disorder
coefficients bk are the coordinates of the disorder signal in the space mentioned above. The echo pattern is
orthogonal to "messy space" because it is linearly independent of each delayed duplicate of the location signal.
Calculating the projection of the observation channel signal in the "cluttered space" allows us to discover the
clutter signal. After removing the clutter signal, just the echo signal remains.
6
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
To remove passive radar clutter, a number of adaptive filtering methods may be utilized. The least-squares filter
(also known as the Wiener filter) will be utilized as the algorithm. This filter is a block algorithm because it
divides the input data into pieces and calculates a separate set of filter coefficients for each chunk. For the vector
space formulation of the clutter reduction process discussed above, the least-squares technique is straightforward
to derive.
Let xk be a vector having L samples of the reference signal which is delayed by k samples, xk n k . The
vector set xk for k between 0 and M are the basic vectors of the disorder space. Then we can define a matrix X,
which is an L × M matrix with columns that are the base vectors xk for k between 0 and M.
Let β be an arbitrary vector of length M, the product of X with β is a vector of length L. X transforms any point
in the disorder space (a vector of length M) into its representation in the signal space (a vector of length L). Also,
let it be a vector containing L samples of the observation signal s0 n . Finally, for any disorder vector β letϵ
the remainder between and Xβ.
y X (5)
Or equivalently:
y X (6)
To find the optimal value of β (called ) it is sought that the term Xβ captures the largest possible fraction of ,
so that it leaves the smallest component behind after subtracting it. The optimal disorder vector can be
obtained by minimizing the square magnitude of ϵ. The corresponding vector of minimum magnitude it then
becomes the best estimate of the target signal.
Since the noise term it is not correlated with the reference signal, it cannot be removed, but as long as it is
small compared to the target signal, it can be ignored. So the problem has been reduced to an ordinary least
squares regression. The solution to this type of problem is well known and is given by equation 7.
X HX X H y
1
(7)
H
In the above equation X represents the Hermitian transpose of X, since the elements of X can be complex
H
numbers. In practice, the matrix X X sometimes it is ill-conditioned, which means that the solution to the least
squares problem is not uniquely determined. One way to deal with this is to use the Tikhonov regularization
technique. The modified equation for the disorder coefficient vector is shown below using Tikhonov
regularization, where Г is an M × M matrix called the Tikhonov matrix.
X H X H X H y
1
(8)
Different options for the Tikhonov matrix result in different types of regularization. A common option is
I , where I is the identity matrix and α is a constant. This choice (which is equivalent to regularization
L2 ) adds a penalty on the square magnitude of the solution vector to select solutions with smaller magnitudes.
Once the clutter coefficient vector has been obtained, it is easy to recover the target echo signal:
y X (9)
and with all this the clutter removal process is complete. The echo signals are then processed to reveal the char-
acteristics of the target.
When the clutter has been removed, the properties of interest are the time delay and the Doppler shift of each
target echo. Which are related to the distance and speed of the target, respectively. The exact relationship be-
tween time and Doppler shift depends on the geometry of the receiver and transmitter, as well as the location and
heading of the target.
Considering the echo signal from the target star n from equation 4, where wk corresponds to the magnitude,
d k to the delay, f k the Doppler shift of the target's k-th echo and Fs the sampling frequency. Target detection
is equivalent to discovering what delay values and Doppler changes are present in the echo signal.
This can be achieved by calculating the cross-ambiguity function X s star , f . Cross-ambiguity may be
thought of as an extension of cross-correlation. By calculating the cross-correlation of two comparable signals,
we can infer their delay. The cross-ambiguity operation is similar, but in addition to the delay, it provides for the
determination of the signal's frequency difference. This is achieved by taking advantage of the fact that sinusoids
of different frequencies are orthogonal to each other.
7
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
N 1
j 2 nf
X s1, s 2 , f s1 n s2* n exp (10)
k 0 Fs
The outcome of calculating the cross-ambiguity function over frequency shift values and a range of delay is a
two-dimensional surface known as a Doppler range map. The targets will show as peaks on this surface at values
of and F, which correspond to the target echoes' delay and Doppler shift.
The Coherent Processing Interval (CPI) is the value of N that must be carefully determined since it restricts the
system's Doppler resolution. In general, adopting a higher CPI allows signals with more minor frequency dis-
crepancies to be distinguished (this implies that Doppler resolution and CPI are inversely proportional). For in-
stance, a CPI of one second relates to a minimum Doppler resolution of 1Hz.
The cross-ambiguity function is difficult to calculate directly, particularly when a large number of Doppler and
range intervals are required. However, there is a rough approach that considerably reduces the computation time.
The core concept is to employ a frequency domain method that enables us to apply the FFT. Because the range
of Doppler shifts we're interested in is narrow relative to the input signal's bandwidth (<500 Hz vs. 200 kHz), we
compress the incoming signals before conducting the FFT.
The sum can re-indexed to divide it into two components in order to derive the fast cross-ambiguity algorithm,
where N = ML:
N 1 L 1
j 2 lf j 2 kMf
X , f l kM s2* l kM exp exp (11)
k 0 l 0 Fs Fs
L and M are chosen such that f max L Fs , where f max is the maximum Doppler shift of interest. Under
these conditions, the term exp j 2 lf Fs 1 so it can be disregarded. Next, we define the function
g k , :
L 1
g k , s1 l kM s2* l kM (12)
l 0
N 1
j 2 kMf
X , f g k , exp (13)
k 0 Fs
8
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
To achieve a coherent dual channel receiver, it is necessary for the RTL-SDRs to use the same clock.
The way to join two SDRs using common crystal oscillators, although the way in which it is done is
controversial, it is shown that in practice it works (Bernales et al., Sf; VonEhr et al., 2016). The SDRs that were
purchased feature TCXO oscillators. The TCXO is a quartz crystal oscillator that reduces the amount of change
in oscillation frequency when ambient temperature changes occur. The TCXO outperforms a standard crystal
oscillator in several ways, most notably in terms of frequency stability across a wide temperature range. In the
Figure 7, presents the SDR modules with TCXO oscillator.
To use the same clock reference, the TCXO oscillator must be removed from a device. Then the yellow wire
is soldered to a pin for the conventional oscillator, which has continuity to a removed TCXO pin. The black wire
is then soldered to a pin of a capacitor located to one side of the oscillator site. In the, you can see the result of
the procedure described in this paragraph. See Figure 8.
9
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
In the other RTL-SDR device the yellow wire goes to the same pin as in the previous case, while the black wire
is soldered to the negative pin of the electrolytic capacitor (Bernales et al., Nd).
To protect the devices and avoid movements that could disorder the cables, the RTL-SDRs have been placed in a
plastic case. It should be mentioned that direct glue has not been used between the case and the devices, rather
only screws have been used indirectly for fastening. This allows the devices to be easily removed if required.
The observation antenna is the Yagi-Uda and the reference antenna is the monopole antenna. Each antenna has a
coaxial cable and a 50 Ω MCX connector. The RTL-SDRs have been placed in the Yagi antenna support and in
addition two 1.80m long USB type B male-female cables are being used.
A set of tools and software programs is necessary to be able to perform data collection and processing. GNU
Radio is a free and open-source software development toolkit for creating software radios. It may be used to
develop software-defined radios (SDR) using affordable and easily accessible RF hardware, or it can be utilized
without hardware in a simulation-like environment. It is extensively used to support both wireless
communications research and real-world radio systems in research, university, government, industry, and
hobbyist domains.
HDF (Hierarchical Data Format) is a collection of file formats (HDF4, HDF5) for storing and organizing
massive datasets.
HDF5 is a software package that enables the management of exceedingly big and complicated data sets. The
HDF5 technology package contains the following items:
• A flexible data model capable of representing very complex data items as well as an extensive range
of information.
• A file format that is totally portable and has no restriction on the quantity or amount of data items in
the collection.
A software library that runs on a variety of computing platforms, from laptops to massively parallel
systems, and implements a high-level API with C, C ++, Fortran 90, and Java interfaces.
A comprehensive set of integrated performance enhancements designed to improve access time and
storage space.
Tools and applications for managing, manipulating, visualizing, and analyzing the collection's data.
The HDF5 data model, file format, application programming interface (API), library, and tools are all
open-source and publicly accessible..
10
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
Anaconda is an environment that makes it easy to manage data tools and machine learning in a single area.
Anaconda allows you to manage environments, that is, it allows you to have separate dependencies for different
projects.
FFmpeg is a free, open-source toolkit capable of decoding, encoding, transcoding, mux, demux, stream, filter,
and play virtually any format. FFmpeg can be compiled on various operating systems, including Windows. A
command line tool for converting media files between formats.
Inside FFmpeg is the ffmpeg command line tool for converting audio or video from one format to another. You
can also capture and encode in real time from DirectShow, a TV card, or other compatible device.
5. Data Collection
Passive radar data recording is done through GNU Radio, for which it is first necessary to build a flow chart that
can handle two RTL-SDR devices. In GNU Radio all blocks have their name located at the top. The following
diagram shows the connection of the equipment to perform data acquisition with the GNU Radio flow diagram.
In the Figure 9, it presents the equipment already connected and in the correct location. It is preferable that the
monopole antenna is located behind the reflector of the Yagi-Uda 5 element antenna.
After a few seconds of clicking on execute, the graph in Figure 10 will open. The graph at the bottom shows the
data in frequency, both for the observation channel and the reference channel. In this case, the data in blue
correspond to the observation channel and the data in red to the reference channel.
11
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
To obtain a video, use the command: python main.py --config prconFigureyaml through the Anaconda Prompt
command terminal. the time it takes to run the command is 38 minutes and depending on the size of the. hdf5 file
it may take more or less than the mentioned time. Then to obtain a video of Doppler frequency vs distance in
kilometers of the data acquired with GNU Radio, the command is used:
python range_doppler_plot.py --config prconFigure yaml --mode video
In the video it is expected to obtain data similar to those of the Figure , when an object is encountered in the
direction of the antenna.
The localized objects have a speed estimate with an estimated error of 1.89% on the approach path towards the
reference antenna and about 3.03% when moving away from the location of the antennas. Ref. Figure 11.
6. Conclusions
Passive radars are very important to establish detection and monitoring of unmanned vehicles since they do not
emit any signal that reveals their position. In addition, the establishment of coherent channels allows to extend its
functionality allowing to detect objects that are cm from the floor as well as km from it.
It is important to consider the type of oscillator that the RTL-SDR modules have and establish their timing for
coherent channels. In this way the passive radar considers the location of UAVs in the order of 1320 pounds.
The use of Python and anaconda allows to manage the data in a faster way than in conjunction with the hdf5
software, it facilitates the generation of videos and multimedia files that facilitate detection and surveillance
operations.
Acknowledgments
The authors wish to thank the Armed Forces Research Center, CIDFAE, for its special support in the
development of this proposal.
References
[1] C. Hu, Y. Wang, R. Wang, T. Zhang, J. Cai, y M. Liu, «An improved radar detection and tracking method
for small UAV under clutter environment», Sci. China Inf. Sci., vol. 62, n.o 2, p. 29306, feb. 2019, doi:
10.1007/s11432-018-9598-x.
[2] B. García y C. Antón, «Nuevo concepto de radares pasivos multiestáticos basados en LTE-R para vigilan-
cia de líneas ferroviarias de alta velocidad», p. 8, 2018.
[3] C. Liang, N. Cao, X. Lu, y Y. Ye, «UAV Detection Using Continuous Wave Radar», en 2018 IEEE Inter-
national Conference on Information Communication and Signal Processing (ICICSP), Singapore, sep.
2018, pp. 1-5. doi: 10.1109/ICICSP.2018.8549736.
12
ICORTAR 2021 IOP Publishing
Journal of Physics: Conference Series 2199 (2022) 012023 doi:10.1088/1742-6596/2199/1/012023
[4] J. Klare, O. Biallawons, y D. Cerutti-Maori, «UAV detection with MIMO radar», en 2017 18th Interna-
tional Radar Symposium (IRS), Prague, Czech Republic, jun. 2017, pp. 1-8. doi:
10.23919/IRS.2017.8008140.
[5] M. Ezuma, O. Ozdemir, C. K. Anjinappa, W. A. Gulzar, y I. Guvenc, «Micro-UAV Detection with a Low-
Grazing Angle Millimeter Wave Radar», en 2019 IEEE Radio and Wireless Symposium (RWS), Orlando,
FL, USA, ene. 2019, pp. 1-4. doi: 10.1109/RWS.2019.8714203.
[6] C. Schupbach, C. Patry, F. Maasdorp, U. Boniger, y P. Wellig, «Micro-UAV detection using DAB-based
passive radar», en 2017 IEEE Radar Conference (RadarConf), Seattle, WA, USA, may 2017, pp. 1037-
1040. doi: 10.1109/RADAR.2017.7944357.
[7] G. Fang, J. Yi, X. Wan, Y. Liu, y H. Ke, «Experimental Research of Multistatic Passive Radar With a
Single Antenna for Drone Detection», IEEE Access, vol. 6, pp. 33542-33551, 2018, doi:
10.1109/ACCESS.2018.2844556.
[8] J. Zhao, X. Fu, Z. Yang, y F. Xu, «Radar-Assisted UAV Detection and Identification Based on 5G in the
Internet of Things», Wirel. Commun. Mob. Comput., vol. 2019, pp. 1-12, jul. 2019, doi:
10.1155/2019/2850263.
[9] S. Li, Y. Chai, M. Guo, y Y. Liu, «Research on Detection Method of UAV Based on micro-Doppler Ef-
fect», en 2020 39th Chinese Control Conference (CCC), Shenyang, China, jul. 2020, pp. 3118-3122. doi:
10.23919/CCC50068.2020.9189414.
[10] M. Pieraccini, L. Miccinesi, y N. Rojhani, «A Doppler Range Compensation for Step-Frequency Continu-
ous-Wave Radar for Detecting Small UAV», Sensors, vol. 19, n.o 6, p. 1331, mar. 2019, doi:
10.3390/s19061331.
[11] E. A. C. Bernales, J. Morales, y F. Cortés, «Desarrollo de Radar HF Marino basado en Radio Definida por
Software», p. 100.
[12] K. VonEhr, S. Hilaski, B. E. Dunne, y J. Ward, «Software Defined Radio for direction-finding in UAV
wildlife tracking», en 2016 IEEE International Conference on Electro Information Technology (EIT),
Grand Forks, ND, USA, may 2016, pp. 0464-0469. doi: 10.1109/EIT.2016.7535285.
13