![Implementation of a Wideband Spectrum Sensing Algorithm Using a
Software-Defined Radio (SDR)
Max Robertson and Dr. Mario Bkassiny
State University of New York at Oswego
Department of Electrical & Computer Engineering
{mrobert2, mario.bkassiny}@oswego.edu
Project Description
This project consisted of implementing an autonomous algorithm that would be able to sense the elec-
tromagnetic radio frequency (RF) spectrum using a Universal Software Radio Peripheral (USRP) and to
detect the center frequencies and bandwidths of the local active signals in the SUNY Oswego area. This
information, once captured, is used to evaluate and analyze the spectrum activity in the surrounding RF
environment. By comparing the actual spectral activity to the spectrum allocations of the Federal Com-
munications Commission (FCC), we could give accurate conclusions on to the efficiency and utilization
of the spectral resources.
The proposed signal detection algorithm is based on a smoothed spectral estimation method. By ap-
plying hypothesis testing to the received signal, we could identify the center frequencies and bandwidths
of the active signals, subject to a certain false alarm rate. Some signals that were detected include cell
phone LTE, aeronautical radio-navigation, and Earth-Space communication signals. The project incor-
porates signal processing using MATLAB to create functions and scripts that can be used at different
locations for reproduction of the project simulations. This project demonstrates the feasibility of wide-
band spectrum sensing for Cognitive Radio (CR) applications. It also showed the potential of using
software-defined radio (SDR) platforms for various signal processing applications, including wideband
signal detection.
Materials and Methods
Very few materials were used in this project, the USRP and MATLAB software were all that was
needed to complete the proposed project. The following simulations correspond to a Neyman-Pearson
threshold approach as in [1], with an incorporated smoothing operation, to help assess the data once
taken. The block diagram, in Figure 1, shows the process in which raw data is taken and interpreted.
Note that, the MATLAB code has been programmed such that repeatability is easy.
Note: Fast Fourier Transform (FFT), Smoothing Window Length (SWL), Shifted and Scaled Received
signals (T’(n)), Center Frequency (CF), Bandwidth (BW), Spectrum Utilization (%), and Neyman-
Pearson Threshold η.
Figure 1: Block diagram of the detection algorithm
The decision threshold η is given in (1), where γ−1 denotes the lower inverse incomplete gamma
function, L represents the desired SWL and α represents the desired False-Alarm probability [1]:
η = 2 ∗ γ−1
(L; (1 − α) ∗ Γ(L)) (1)
The signals that were generated were distorted by additive white Gaussian noise (AWGN). We applied
a smoothing operation method to obtain a more accurate spectral estimation, which also decreased the
number of intersection points with the threshold line. The concept of this sliding window[1] allows the
individual points to become ”grouped” and gives a smoother plot. When programming we needed to
also implement a truncation of values, the summation formulas used can be seen in Figure 2, where L
represents the desired SWL value. We note that, as the Sliding Window Length increases, the smoother
the curve becomes, as seen in Figures 3 & 4.
Figure 2: Visual representation of smoothing operation with length L = 3
The signal being processed is a simple modulated sinusoidal signal with a carrier frequency of 20KHz.
As can be seen in Figures 3 & 4, the corresponding peaks are at +/- 20KHz in frequency domain, with
AWGN simulating the behaviour of a ”real” signal. You can see how the increased Sliding Window
Length effects the curve’s smoothed appearance and increases stability.
Simulation and Results
The hardware used in this project consisted of a Universal Software Radio Peripheral (USRP) model:
National Instruments (NI) - NI USRP - 2920, 50MHz to 2200MHz (Figure 5), controlled by MATLAB
software. The received signal strength depends on the antenna characteristics; that is, if we were to
replicate the sensing measurements with a higher gain antenna the results would be more accurate and
more low-power local signals detected. There is a large amount of MATLAB code associated with this
project, some functions were created for easy replication of this project for future work.
Figures 5 & 6: NI USRP - 2920 Hardware
The algorithm has been designed to scan a desired portion of the RF spectrum and acquire a collection
of sub-bands, this process can then be repeated. When creating the algorithm, the desired outcome
could be achieved with known input signals; it was after much fine tuning that the algorithm produces
a quality of work that is more than satisfactory:
Figure 7: Neyman-Pearson (NP) threshold testing using simulated signals
The square-shaped waveform below represents the detection outcome of the proposed algorithm, which
can be used to calculate the Center Frequencies and Bandwidths of the detected active signals. Some
results of frequency utilization and the corresponding FCC allocations are shown below .
8.75 8.8 8.85 8.9 8.95 9 9.05
x 10
7
10
2
10
3
10
4
f(MHz)−|R(f)|.
2
f−vec
Smoothed Periodogram of Recieved Signal, SWL=901, Threshold=1e−06, Utilization=22.618%
8.75 8.8 8.85 8.9 8.95 9 9.05
x 10
7
−1
−0.5
0
0.5
1
1.5
2
Thresholdvalue
f(MHz)
Impulse plot of Scaled/Shifted Recieved Signal, SWL=901, Threshold=1e−06, Utilization=22.618%
Figure 8: Actual result from USRP from 88-90MHz of the spectrum, which corresponds to local FM
radio signals
There are more plots like these that cover different parts of the spectrum, but this process is very time-
consuming for one USRP. Here is another observation of a different sub-band:
Sub-band Start (MHz) Sub-band End (MHz) Utilization % Signal Allocation (FCC)
88 90 22.6180% FM Radio
155 158 4.7179% Maritime Mobile
403 407 5.7111% Meteorological Satellite
700 705 7.4545% Cell Phone LTE
849 852 2.7943% GSM
999 1002 7.7136% Aeronautical Radionavigation
2025 2030 2.6709% Earth/Space Exploration
Table 1: Utilization of locally observed sub-bands
Conclusion
The algorithm has clearly shown that it can achieve the goals of spectrum sensing in the local RF
spectrum, relaying this information graphically, and then analyzing the data to conclude the amount of
active signals in the SUNY Oswego area and their corresponding bandwidths. The expected outcome
is very similar to that of the actual USRP raw data output, which allows for further fine-tuning and
potential for future work.
Future Work
I would love to continue the work achieved here, this is only the foundations; with the algorithm com-
plete it allows us to venture into the idea of a communication network of USRP’s to give constant
coverage and real-time analysis, and with better quality antennas - the results would be impressive.
Acknowledgments
I would like to express my gratitude towards the constant amount of support Professor Bkassiny has
provided for me throughout this research project. I also thank the Electrical and Computer Engineering
Department for this opportunity, and thank the advanced wireless systems research (ADWISR) center
for providing the equipment. Lastly but most importantly, I thank SUNY Oswego for providing support
for my research with the SCAC grant.
References
[1] ”M. Bkassiny, S. K. Jayaweera, Y. Li, and K. A. Avery”. Blind cyclostationary feature detection
based spectrum sensing for autonomous self-learning cognitive radios. In IEEE International Con-
ference on Communications (ICC ’12), Ottawa, Canada, June 2012.](https://image.slidesharecdn.com/a076d2a1-86ee-455a-8500-4874906f9ced-151013230643-lva1-app6892/85/Max_Poster_FINAL-1-320.jpg)
Max_Poster_FINAL
- 1. Implementation of a Wideband Spectrum Sensing Algorithm Using a Software-Defined Radio (SDR) Max Robertson and Dr. Mario Bkassiny State University of New York at Oswego Department of Electrical & Computer Engineering {mrobert2, mario.bkassiny}@oswego.edu Project Description This project consisted of implementing an autonomous algorithm that would be able to sense the elec- tromagnetic radio frequency (RF) spectrum using a Universal Software Radio Peripheral (USRP) and to detect the center frequencies and bandwidths of the local active signals in the SUNY Oswego area. This information, once captured, is used to evaluate and analyze the spectrum activity in the surrounding RF environment. By comparing the actual spectral activity to the spectrum allocations of the Federal Com- munications Commission (FCC), we could give accurate conclusions on to the efficiency and utilization of the spectral resources. The proposed signal detection algorithm is based on a smoothed spectral estimation method. By ap- plying hypothesis testing to the received signal, we could identify the center frequencies and bandwidths of the active signals, subject to a certain false alarm rate. Some signals that were detected include cell phone LTE, aeronautical radio-navigation, and Earth-Space communication signals. The project incor- porates signal processing using MATLAB to create functions and scripts that can be used at different locations for reproduction of the project simulations. This project demonstrates the feasibility of wide- band spectrum sensing for Cognitive Radio (CR) applications. It also showed the potential of using software-defined radio (SDR) platforms for various signal processing applications, including wideband signal detection. Materials and Methods Very few materials were used in this project, the USRP and MATLAB software were all that was needed to complete the proposed project. The following simulations correspond to a Neyman-Pearson threshold approach as in [1], with an incorporated smoothing operation, to help assess the data once taken. The block diagram, in Figure 1, shows the process in which raw data is taken and interpreted. Note that, the MATLAB code has been programmed such that repeatability is easy. Note: Fast Fourier Transform (FFT), Smoothing Window Length (SWL), Shifted and Scaled Received signals (T’(n)), Center Frequency (CF), Bandwidth (BW), Spectrum Utilization (%), and Neyman- Pearson Threshold η. Figure 1: Block diagram of the detection algorithm The decision threshold η is given in (1), where γ−1 denotes the lower inverse incomplete gamma function, L represents the desired SWL and α represents the desired False-Alarm probability [1]: η = 2 ∗ γ−1 (L; (1 − α) ∗ Γ(L)) (1) The signals that were generated were distorted by additive white Gaussian noise (AWGN). We applied a smoothing operation method to obtain a more accurate spectral estimation, which also decreased the number of intersection points with the threshold line. The concept of this sliding window[1] allows the individual points to become ”grouped” and gives a smoother plot. When programming we needed to also implement a truncation of values, the summation formulas used can be seen in Figure 2, where L represents the desired SWL value. We note that, as the Sliding Window Length increases, the smoother the curve becomes, as seen in Figures 3 & 4. Figure 2: Visual representation of smoothing operation with length L = 3 The signal being processed is a simple modulated sinusoidal signal with a carrier frequency of 20KHz. As can be seen in Figures 3 & 4, the corresponding peaks are at +/- 20KHz in frequency domain, with AWGN simulating the behaviour of a ”real” signal. You can see how the increased Sliding Window Length effects the curve’s smoothed appearance and increases stability. Simulation and Results The hardware used in this project consisted of a Universal Software Radio Peripheral (USRP) model: National Instruments (NI) - NI USRP - 2920, 50MHz to 2200MHz (Figure 5), controlled by MATLAB software. The received signal strength depends on the antenna characteristics; that is, if we were to replicate the sensing measurements with a higher gain antenna the results would be more accurate and more low-power local signals detected. There is a large amount of MATLAB code associated with this project, some functions were created for easy replication of this project for future work. Figures 5 & 6: NI USRP - 2920 Hardware The algorithm has been designed to scan a desired portion of the RF spectrum and acquire a collection of sub-bands, this process can then be repeated. When creating the algorithm, the desired outcome could be achieved with known input signals; it was after much fine tuning that the algorithm produces a quality of work that is more than satisfactory: Figure 7: Neyman-Pearson (NP) threshold testing using simulated signals The square-shaped waveform below represents the detection outcome of the proposed algorithm, which can be used to calculate the Center Frequencies and Bandwidths of the detected active signals. Some results of frequency utilization and the corresponding FCC allocations are shown below . 8.75 8.8 8.85 8.9 8.95 9 9.05 x 10 7 10 2 10 3 10 4 f(MHz)−|R(f)|. 2 f−vec Smoothed Periodogram of Recieved Signal, SWL=901, Threshold=1e−06, Utilization=22.618% 8.75 8.8 8.85 8.9 8.95 9 9.05 x 10 7 −1 −0.5 0 0.5 1 1.5 2 Thresholdvalue f(MHz) Impulse plot of Scaled/Shifted Recieved Signal, SWL=901, Threshold=1e−06, Utilization=22.618% Figure 8: Actual result from USRP from 88-90MHz of the spectrum, which corresponds to local FM radio signals There are more plots like these that cover different parts of the spectrum, but this process is very time- consuming for one USRP. Here is another observation of a different sub-band: Sub-band Start (MHz) Sub-band End (MHz) Utilization % Signal Allocation (FCC) 88 90 22.6180% FM Radio 155 158 4.7179% Maritime Mobile 403 407 5.7111% Meteorological Satellite 700 705 7.4545% Cell Phone LTE 849 852 2.7943% GSM 999 1002 7.7136% Aeronautical Radionavigation 2025 2030 2.6709% Earth/Space Exploration Table 1: Utilization of locally observed sub-bands Conclusion The algorithm has clearly shown that it can achieve the goals of spectrum sensing in the local RF spectrum, relaying this information graphically, and then analyzing the data to conclude the amount of active signals in the SUNY Oswego area and their corresponding bandwidths. The expected outcome is very similar to that of the actual USRP raw data output, which allows for further fine-tuning and potential for future work. Future Work I would love to continue the work achieved here, this is only the foundations; with the algorithm com- plete it allows us to venture into the idea of a communication network of USRP’s to give constant coverage and real-time analysis, and with better quality antennas - the results would be impressive. Acknowledgments I would like to express my gratitude towards the constant amount of support Professor Bkassiny has provided for me throughout this research project. I also thank the Electrical and Computer Engineering Department for this opportunity, and thank the advanced wireless systems research (ADWISR) center for providing the equipment. Lastly but most importantly, I thank SUNY Oswego for providing support for my research with the SCAC grant. References [1] ”M. Bkassiny, S. K. Jayaweera, Y. Li, and K. A. Avery”. Blind cyclostationary feature detection based spectrum sensing for autonomous self-learning cognitive radios. In IEEE International Con- ference on Communications (ICC ’12), Ottawa, Canada, June 2012.