View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by University of Johannesburg Institutional Repository

Implementation of a GNU Radio and Python

FMCW Radar Toolkit
Themba W. Mathumo∗† , Theo G. Swart∗ and Richard W. Focke†
∗ Depart. of Electrical and Electronic Engineering Science, University of Johannesburg, South Africa
† Defence, Peace, Safety and Security, Council for Scientific and Industrial Research, South Africa
tmathumo@csir.co.za, tgswart@uj.ac.za, rfocke@csir.co.za

Abstract—The use of GNU Radio in order to explore FMCW purposes. Both studies have a similar implementation: USRP
radar is growing rapidly, where it is used for radar signal pro- N210 and GNU Radio are used in their conceptual designs
cessing. In this paper we implement FMCW surveillance radar using FMCW waveform.
for drone detection on SDR using GNU Radio and the USRP
B210. The requirement was to design and implement an FMCW Research and development on Reconfigurable Hardware
surveillance radar to detect a drone with a radar cross-section of Interface for Computation and Radio (RHINO) is being done
0.1 m2 and a maximum range of 150 m for the purpose of point at the University of Cape Town [6]. RHINO is an open source
detection. The signal processing takes the beat frequency which development consisting of hardware and software which is
has pulse compression gain and performs coherent integration. It also designed for radar application. Such hardware opens the
windows and displays the results on a range Doppler map using
two-dimensional fast Fourier transform. The pulse repetition opportunity for the practical realization of radar.
frequency of the waveform is selected as 2200 Hz with a chirp The application area of drone detection is of interest as
bandwidth of 28 MHz, allowing for resolving a maximum velocity drones are gaining popularity in transportation and are also
of 30 m/s unambiguously, which is the typical maximum speed of increasingly being used for criminal purposes, such as spying
a small drone. Experiments were conducted using a human being and transportation of drugs into prison [7]. Thus, it is of
and a car target as a drone was not available. The B210 induced a
phase drift in the results which causes a Doppler shift. The phase interest to determine whether SDR is capable of being used
drift was resolved by creating a phase equalization matrix, which for radar surveillance applications, as it could lead to radar
was used to correct the phase drift in real-time. Verification of systems deployable against the increasingly prevalent drones.
the designed radar system yielded a 63% probability of detection
and probability of false alarm of 1.7 × 10−6 . II. FMCW WAVEFORMS
Index Terms—Doppler radar, frequency modulation, radar Frequency-modulated continuous-wave (FMCW) radar has
signal processing, software defined radar been the center of interest when designing radar in GNU Radio
because of its low power requirements and low cost [5]. An
I. I NTRODUCTION FMCW waveform, also known as a chirp, is modeled as [8]:
There is a growing demand for cheap portable radar sys- fdev
tems which are configurable and can be used for different f (t) = fT + t, (1)
T
applications. Typical examples are for medical imaging [1] where f (t) is the transmitted waveform of a center transmit
and synthetic aperture radar (SAR) [2]. frequency fT plus the slope. The slope is the relationship
Software defined radio (SDR) is a radio transceiver in between fdev and T , where fdev is the stop frequency of the
which most of the key parameters of radio operations, that chirp in hertz. Fig. 1 shows the reflected signal from a moving
are typically implemented in hardware, are implemented in target, where the signal shifts up or down because of the
software. SDR has allowed radar to be explored in software Doppler effect. The problem is that the beat frequency consists
[3]. Radar is primarily used to measure range and radial of the Doppler and/or range component, and it is impossible
velocity using transmission and reception of radio frequency to determine them using only one beat frequency.
(RF) and signal processing. GNU Radio is one of the platforms
that is used for designing software defined radar together with A. Two-dimensional Signal Processing for FMCW Radar
the Universal Software Radio Peripheral (USRP) as front end. Since it is not possible to determine both range and velocity
Ralston and Hargrave [3] introduce ground penetrating radar using a single up/down chirp, and since the triangular chirp
(GPR) and the concept of using SDR for radar ranging. The does not work for multiple targets, the double fast Fourier
requirements for GPR radar are usually high bandwidth, timing transform (FFT) technique for FMCW signal processing is
accuracy and high transmit power. The prototype was built used for multiple targets, as shown in Fig. 2.
using GNU Radio and USRP2 since these are open source The chirp sequence is multiple pulses within a coherent
and generic. processing interval (CPI), and TCPI is the dwell time at
Sundaresan et al. [4] discuss SDR radar as a potential tech- which a number of pulses are transmitted and received. The
nology to be used for target detection and tracking. Prabaswara transmitted signal is mixed with the Doppler-shifted and time-
et al. [5] discuss the use of SDR radar for weather surveillance delayed reflected signal. The M pulses are arranged in a
 f (t) The performance of the low-cost USRP B210 transceiver is
investigated for its performance for radar use in the C band.
fdev τ
At the time of these investigations there was no previous radar
work done using the B210.

fB A. Radar Waveform Design

Transmitted signal The radar is designed based on user requirements which
Reflected signal from are then interpreted to be the design parameters using signal
moving target
design rules [10]. The given user parameters are that the
fT TCPI t
surveillance radar should be able to detect a drone as far as
away as 150 m. The parameters are given in Table I. The
Fig. 1. Up chirp transmitted waveform with time-delayed reflected waveform
from a stationary target [9]. design parameters were calculated using the above-mentioned
signal design rules. The user parameters were adjusted depend-
Transmitted signal
ing on the constraints imposed by the B210 and the result
f (t)
Reflected signal from a target of waveform design. Instead of having range resolution of
fdev
0.1 m, the achieved range resolution is 5.35 m because of
the maximum bandwidth achieved by the B210; the velocity
fT t resolution that was expected by the user is 1 m/s but only
TCPI
FFT FFT FFT FFT
1.2 m/s was achieved based on the chosen pulse repetition
frequency (PRF).

fR
B. Radar Model
FFT
Table II gives the specifications of the USRP B210 and the
antenna available that can be used for the experiments. Only
the important parameters which are used to determine the radar
fD performance are listed.
Fig. 2. Chirp processing by the method of pulse Doppler processing. The radar characteristics are used to predict the radar
performance by evaluating the SNR vs. range, as shown in
Fig. 3. The SNR vs. range will be improved in order to meet
two-dimensional matrix in which range and Doppler FFT the radar performance requirements.
processing is performed in order to obtain a range Doppler The signal return of interest is buried under the noise floor
matrix. Then the double FFT can be performed on the range since the SNR is negative. This means that during detection
Doppler matrix in order to obtain the range and radial velocity we will not be able to detect the targets.
of targets. The receiver operating characteristic (ROC) curve in Fig. 4
shows that in order to meet the user’s performance requirement
III. S YSTEM D ESIGN the SNR has to be 12 dB. Various techniques such as intro-
The user requirement is to design and implement an FMCW ducing amplifiers, the use of coherent integration and Doppler
ground-based surveillance radar. It should be able to detect
TABLE I
small drones for the purpose of point defence, which is a low- T HE USER PARAMETERS AND THE CALCULATED DESIGN PARAMETERS
range application, and have the following properties: User Design
Value Value
• A range coverage of up to 150 m—the maximum range Parameter Parameter
allows time for detection and decision making after the Center
5.5 GHz PRF 2200
target has been identified. frequency
2 Range Required: 150 m Number of pulses 50
• Probability of detection of 80% is required for a 0.1 m
−6
Range Required: 0.1 m,
target, with probability of false alarm of 10 , meaning Range of samples 227
resolution achievable: 5.35 m.
1 false alarm per 1 million observations. Velocity 30 m/s Bandwidth 28 MHz
• The sampling rate that we could achieve using the B210 Velocity Required: 1 m/s, #pulses
CPI= PRF 22.73 m/s
is 28 MHz, which results in the range resolution to be resolution achievable: 1.2 m/s.
5.35 m, and not 0.1 m as expected using the range
TABLE II
resolution equation [10]. S PECIFICATIONS OF EQUIPMENT USED
◦
• A desired azimuth coverage of 45 . The advantage of
Parameter description Value
using a directional antenna is that the antenna can achieve C-band 5.5 GHz
higher gain, the narrower the antenna the higher the gain. Maximum transmit power 6.6 dBm
A drone is relatively small, thus the 45◦ azimuth coverage Bandwidth 28 MHz
is enough for the radar to cover multiple drones in the Noise figure 3.8 dB
Antenna 10 dB
direction that the radar is facing.
 waveform design. Fig. 7 shows the flow graphs of the im-
plemented GNU Radio block. The B210 handles the transmit
and receive waveform and GNU Radio applies the process
of dechirping. The controller is embedded in the FMCW
transmitter block. The data from the FMCW transmitter is
written in real time and is passed as meta data via JSON
(Java Script Object Notation) so that further signal processing
steps that require the data can proceed. Further processing
was done in Python, and the data packets were sent using
the user datagram protocol (UDP). Detection and windowing
Fig. 3. SNR vs. range were performed and the results displayed in a two-dimensional
range and Doppler plot.
The UDP has a 65535 byte maximum size that can be
transferred as a payload, 8 bytes is the UDP header and 20
bytes is the IP header. It affects the design since there is a
bottleneck for data that can be processed at the same time. The
range samples M was calculated to be 227 per pulse, when the
pulse has PRF of 2200 Hz and the decimated sampling rate is
500 kHz. Since the data over UDP could not be complex, it
was interleaved into float data which resulted in 454 samples
being transferred over UDP. The waveform block was used to
Fig. 4. ROC curve for various SNR values. The SNR of the detector design the waveform, then the waveform was stored in a file
performance requirement is 12 dB. using the GNU Radio file sink. The file source was used to

processing gain will be employed to improve the current SNR

to equal the performance requirement.
1) Effect of Low-noise Amplifier (LNA): The HMC460
LNA is introduced to improve the noise figure of the system,
thereby improving the SNR. The LNA has a noise figure of
2.5 dB and a gain of 14 dB. The effective noise figure is
determined by cascading the noise factor of the receive chain
using Frii’s equation [9]:
 
F2 − 1
N F (dB) = 10 log F1 + , (2)
G1
where F1 is the noise factor of the LNA, G1 the gain of the
LNA and F2 is the noise factor of the USRP B210. Fig. 5. SNR vs. range after implementing the LNA, power amplifier and
2) Effect of Power Amplifier: The TWAL 0412-20 amplifier pulse integration.
was available, which is a 20 W amplifier with 43 dBm of
power. The output power of the radar system is improved by
36 dB from the initial output power of 7 dBm.
3) Signal Processing Gain: The application of coherent
integration can help to improve the SNR, since multiple pulses
will be transmitted. The returned pulses np can be integrated
coherently by adding the I and Q data, which is equivalent to
increasing the SNR by a gain of 10 log np .
Fig. 5 shows the improvement obtained in SNR vs. range
when implementing these measures.
IV. I MPLEMENTATION
Fig. 6 is the design for the radar system. The SDR ap-
proach allows less to be done in firmware but still builds
on this technology by allowing the development to be done
in software. The first step was to develop the independent
GNU Radio block using C++ and Python that was responsible
for generating the FMCW waveform using parameters of the Fig. 6. Design for the radar showing how the user interacts with the system.
 Fig. 7. GNU Radio implementation using the chirp signal which is displayed on the spectrum analyzer.

generate the waveform in real time and send the samples to Receive 2M samples and Apply windowing on
Create N × M matrix
create M complex pairs every pulse length
the USRP.
The signal processing is done sequentially, as shown in Get waveform
Windowing in Doppler
parameters from meta
Fig. 8. After receiving the meta data it is read from the JSON data
dimension

file that is shared over the network using Samba. The meta Store the
data sets the PRF, sampling rate, bandwidth, center frequency complex data in
a file
2D FFT and FFT shift

and the modulation waveform. The data is received from pulse

to pulse and Hanning windowing is performed from pulse Plot 2D range and
Doppler map
to pulse. According to the design, 50 pulses are received on
which the two-dimensional FFT is performed. After the pulses Store the Perform CFAR
detections in a detection on envelope
have been coherently integrated, the result is displayed on an file data

FFT plot. The pulses are arranged in a two-dimensional matrix Fig. 8. The signal processing algorithm.
and the windowing is done on the Doppler dimension together
with the second FFT. The results are displayed on the two-
dimensional Doppler range map.
V. R ESULTS AND D ISCUSSION
In order to validate the transmit waveform according to
the design parameters, we used a computer with GNU Radio
installed. This computer was responsible for generating the
waveform, and was connected to the USRP transmitting the
signal, which was measured on the spectrum analyzer. The
measured signal on the spectrum analyzer showed that the
bandwidth of the signal was 28 MHz, shown in Fig. 9 by
∆Mkr, but the measured output power of the USRP was
1.39 dBm, and not the 6.6 dBm given in the specifications. Fig. 9. Measured signal on the spectrum analyzer.
The power of the signal is increased by the use of a power
amplifier.
A human being and a car were used as targets in the should be at 0 m/s, but in Fig. 11 it is not centered at the
experiments although the radar was designed to detect a drone. 0 Hz Doppler bin due to the phase drift effect caused by the
The experiments were done with reduced power: instead of the B210, the black line shows that the velocity is increasing with
43 dBm amplifier, a 13 dBm amplifier was used because of distance. The B210 contributes an unknown phase shift that
health implications and to account for using targets which have varied with range on the range Doppler map.
a bigger radar cross section (RCS) than a drone. Fig. 10 shows To correct the phase drift, the phase correction algorithm
the setup and where the experiment was conducted. shown in Fig. 12 was implemented in Python. The process was
The target is approximately 60 m from the radar. The target started by estimating the phase drift by calculating the slope
and the clutter are not moving, so it is expected that they of the phase drift. The Doppler shift is the rate of change of
 Estimate the needed
phase drift (−ϕ)

Populate 1D matrix with

phase angle ϕ increments

Phase increment with Perform 1D FFT and

pulse N in the Doppler FFT shift in the
dimension (2D matrix) Doppler dimension

Read 2D range Doppler 1D IFFT on the

matrix from file Doppler dimension

Element-wise
1D FFT in the
multiplication with
Doppler dimension
cos(ϕ) and sin(ϕ)
Fig. 10. Setup used for the experiments.
1D FFT on the Doppler
dimension and FFT shift

Phase corrected range

and Doppler map

Fig. 12. Phase correction algorithm flow graph.

Fig. 11. Range Doppler map without phase correction with clutter and
stationary target that was placed at approximately 60 m.

phase [9], which is calculated using:

∇ϕ 2v
= . (3)
2π λP RF
The main aim of creating the equalization matrix was to
create a negative slope with the same phase drift as the range Fig. 13. Range Doppler map showing a car moving away from the radar.
This is to test the radar for high speed.
Doppler matrix that had to be corrected. The phase drift
was assumed to increase linearly, thus the equalization matrix
was implemented in such a way that its phase drift increases After performing the element-wise multiplication in the time
linearly. The process of creating the 2D equalization matrix domain, the results are transformed back into the frequency
starts by creating a 1D array, with the data in increments of domain and viewed on the range Doppler map.
phase correction angle ϕ. The 1D array is transformed to 2D In Fig. 13 we can see the car at the velocity of approxi-
by incrementing the phase by the pulse number on the Doppler mately 20 m/s when the car was moving away from the radar.
dimension—in the design 50 pulses were used for Doppler The phase drift has been corrected—there is no longer a slope
processing, thus the equalization matrix should be populated or clutter, and the target has 0 Hz Doppler shift.
with 50 pulses so that the sizes are the same.
The data from the experiment was saved in a 2D FFT A. Probability of Detection Performance and Verification
matrix, thus when it is read it undergoes inverse fast Fourier In order to confirm the probability of detection a human was
transform (IFFT) to get back to the time domain. Element-wise placed at a range of approximately 150 m, as it was specified
multiplication is performed between the experimental data and from the user requirement for probability of detection of 80%.
the equalization matrix, which has complex cosine and sine A number of trials should be repeated and saved to file to
pairs. This will only cause phase change on the experimental determine how many times the radar is able to detect the target;
data without affecting the amplitude, as can be shown using 1000 trials were performed. The human target was going back
the Fourier transform pair [9]: and forth in a circular motion within a distance of 5 m so that
on the detector it will appear on the 28-th range bin. This
ej2πf0 t x[n] ←→ X[f − f0 ]. (4) will cause the RCS of the target to fluctuate since the range
resolution of the radar is 5.35 m and the target is moving in radar using SDR (i.e. GNU Radio and B210) was done since
the range bin. it can be a portable and cost efficient solution.
The target was detected 639 times out of the 1000 trials. A radar was designed and modeled using the given user
The achieved Pd is 63.9%, but the expected probability of requirements and performance specifications. The radar was
detection is 80%: expected to detect a drone up to 150 m away with 80%
Ndetections probability of detection and 10−6 probability of false alarm.
Pd = . (5)
Ntrials The USRP B210 is part of the user requirements, and it had
We use the Gaussian approximation to determine the con- to be used as transceiver. It was studied in order to understand
fidence interval [11], where P is the mean of Pd , N is the how it performs and what its limitations are. The waveform
number of trials and z is the critical value given for various designed has PRF of 2200 Hz, at a center frequency of
values of probability: 5.5 GHz and bandwidth of 28 MHz; this results in resolution
r of 5.35 m, maximum unambiguous velocity of 30 m/s and
P (1 − P ) velocity resolution of 1.2 m/s when 50 pulses are used for
xL , xu = P ± z ,
N Doppler processing. The system was only capable of reaching
xL , xu = 0.639 ± 0.013. 63.9% probability of detection and 1.7 × 10−6 probability of
false alarm.
The reason why the detector could not perform optimally
The bandwidth was still a limiting factor in terms of range
is that the strong signal return from the close-by objects may
resolution; however, reading the waveform from the file made
have masked the weak signal return from the target of interest
it possible to increase the resolution. According to the spec-
as the RCS changes. From the design we could not meet the
ifications of the B210, 56 MHz of instantaneous bandwidth
range resolution specification of 0.1 m for the RCS of 0.1 m2
can be achieved, but in the implementation only 28 MHz was
target, the range resolution that was achieved is 5.35 m. This
achieved when both transmitter and receiver channels were
results in more clutter energy being integrated into a single
used. The phase drift caused by the local oscillator of the B210
cell along with the target, thus affecting the achieved SNR.
is one of the limiting factors in terms of getting an accurate
B. Probability of False Alarm Performance Verification Doppler shift. During the experiments a Doppler shift was
The other performance specification that needs to be met is induced, so a phase correction algorithm was implemented to
the probability of false alarm. To do that the radar is directed solve the issue by creating an equalization matrix.
in the vertical direction so that the radar only measures the R EFERENCES
leakage, thermal noise and opportunistic signals within the [1] J. Marimuthu, K. S. Bialkowski and A. M. Abbosh, “Reconfigurable
vicinity of the radar. A 100 trials of measurements are to Software Defined Radar for medical imaging,” in Australian Microwave
be recorded, then we will use the constant false alarm rate Symp., Melbourne, Australia, Jun. 26–27, 2014, pp. 15–16.
[2] D. M. Chinnam, J. Madhusudhan, C. Nandhini, S. N. Prathyusha, S.
(CFAR) detector to estimate the rate of false detections. The Sowmiya, R. Ramanathan and K. P. Soman, “Implementation of a low
CFAR detector dynamically adjusts the threshold and, thus, cost synthetic aperture radar using software defined radio,” in Int. Conf.
it is not expected that noise will cause any false alarms. Computing Commun. Netw. Technol., Karur, India, Jul. 29–31, 2010, pp.
1–7.
However, the system can generate spurious signal or receive [3] J. Ralston and C. Hargrave, “Software defined radar: An open source
opportunistic signals, thus once in a while a false target will platform for prototype GPR development,” in Int. Conf. Ground Pene-
be detected. trating Radar, Shanghai, China, Jun. 4–8, 2012, pp. 172–177.
[4] S. Sundaresan, C. Anjana, T. Zacharia and R. Gandhiraj, “Real time
As a rule of thumb the value N should be at least N ≥ implementation of FMCW radar for target detection using GNU radio
10/Pf a , resulting in 100000000 trials [11]. Lower values of and USRP,” in Proc. Int. Conf. Commun. Signal Proc., Melmaruvathur,
N can still be used and relate it to time [9]: India, Apr. 2–4, 2015, pp. 1530–1534.
[5] A. Prabaswara, A. Munir and A. B. Suksmono, “GNU Radio based
Tm F AR software-defined FMCW radar for weather surveillance application,” in
Pf a = , (6) Int. Conf. Telecommun. Syst. Services Applic., Bali, Indonesia, Oct. 20–
M
21, 2011, pp. 227–230.
where F AR is the false alarm rate (i.e. how many false alarms [6] M. Inggs, G. Inggs, A. Langman and S. Scott, “Growing horns: Applying
occur in a second) and Tm is the time interval at which the Rhino software defined radio system to radar,” in IEEE Radar Conf.,
Kansas City, MO, May 23–27, 2011, pp. 951–955.
the number of resolution cells M are collected. Each trial [7] (2016, June 26). Accurately detect and locate drones and their operators.
consists of 226 samples of data, which represents a pulse at Available: https://www.echoradarsystems.com/drone-detection-system/
a PRF of 2200 Hz. The data capture was repeated for 100 [8] H. Rohling and M. Kronauge, “New radar waveform based on a chirp
sequence,” Int. Radar Conf., Lille, France, Oct. 13–17, 2014, pp. 1–4.
trials. The calculated Pf a is 1.7 × 10−6 , which is a good [9] M. A. Richards, J. A. Scheer and W. A. Holm, Principles of Modern
enough approximation for 1 false detection for 1 million radar Radar: Basic Principles, SciTech Publishing, 2010.
observations. [10] D. E. Barrick, “FM/CW radar signals and digital processing,” NOAA
Technical Report ERL 283-WPL 26, 1973.
[11] J. D. Echard, “Estimation of radar detection and false alarm probability,”
VI. C ONCLUSION IEEE Trans. Aerosp. Electron. Syst., vol. 27, no. 2, pp. 255–260, Mar.
Drones are becoming popular for varying applications, but 1991.
they are also being used for illegal activities such as spying
and flying into restricted areas, thus a study of surveillance