Spectrum Allocation for IoT Applications using

Sub-Nyquist Wideband Spectrum Sensing with

Geolocation Database over TV White Space
Vijaya Rajeshwarkar1[0000-0002-2078-758X] , Department of Electronics and
Telecommunicatio, Sinhgad Institute of Technology, Lonavala, Pune, India
1
v.rajeshwarkar@gmail.com
Tanuja S. Dhope4[0000-0003-2907-8509] , Department of Electronics and Communi-
cation, Bharati Vidyapeeth (Deemed to be University) College of Engineering,
Pune, India
2
tanuja_dhope@yahoo.com
Sulakshana S. Patil5[0000-0002-7729-644X] , Department of Electronics and Telecom-
munication, Sinhgad Institute of Technology, Lonavala, Pune, India
3
sulakshanaspatil@gmail.com
Corresponding auther:Tanuja S.Dhope,email:tanuja_dhope@yahoo.com
Abstract—Combining spectrum sensing with a geolocation database
is an effective strategy for increasing the utilization of white space
devices and increasing spectrum access opportunities. This integra-
tion improves detection accuracy while also reducing computational
complexity. However, due to the need for wideband signal capture
and processing, applying advanced spectrum sensing techniques in
tiny devices with limited resources offers considerable hurdles. This
study presents a hybrid approach that combines sub-Nyquist wideband
spectrum sensing and geolocation database utilization to precisely
detect the spectrum situation while minimizing measurement count
and computational intricacy. By leveraging prior channel occupancy
status from geolocation database, our approach refines spectrum sens-
ing to target a limited number of potentially unoccupied channels
within TV white space (TVWS).
Keywords- Geolocation database; Sub-Nyquist rate sampling, wideband spectrum
sensing; TV White Space, multicoset sampling;

Introduction
The Internet of Things (IoT) is driving an increase in linked devices, with 29.3
billion [1] smart gadgets expected by 2023. However, due to the restricted
availability of RF spectrum, this rapid expansion is generating a spectrum crisis.
Traditional fixed frequency band allocation has led in considerable inefficiencies,
with less than 20% of available spectrum utilized [2]. To address this, governments
began implementing measures such as dynamic spectrum sharing, including the
usage of TV White Space (TVWS) [3, 4], 3.5 GHz band deployment of the Citizens

1
Broadband Radio Service (CBRS) [5], and the use of unauthorized frequency
spectrum in 5 GHz range.[6]. TVWS is the unused spectrum allotted to licensed
users, particularly within ultra-high frequency (UHF) range typically utilized for
Television broadcasting (54 to 806 MHz) [7] [8]. These underutilized or unutilized
TV channels offer a chance to overcome spectrum scarcity in IoT applications and
extend wireless connection. TVWS, in particular, provides higher penetration
and propagation characteristics, making it a promising solution for increasing
wireless connection, particularly in locations with limited Wi-Fi access [9].
Unlicensed users (Secondary Users (SUs)) can use these white spaces (WS)
opportunistically while not interfering with licensed users (Primary Users (PUs))
[10]. In [11-15] author proposed new spectrum sensing technique and defined
use cases to empower the rural India.
In [16], the researchers conducted a quantitative evaluation of the unutilized
television spectrum within India. Their findings indicate that, in contrast to
Japan's scenario where an average of 16.67 channels (41.67%) out of a total of
40 are available in 84.3% of the areas [17], in India, a minimum of 12 out of 15
channels (80%) are accessible as TV white space in all the examined areas. In
other regions, like Germany, the accessible TVWS is 19.2 channels (48%), in
the US it's 23.1 channels (58%), Switzerland holds 25.3 channels (63%), and
Denmark comprises 24.4 channels (61%) out of 40 channels [18].
These TVWSs, however, are difficult to determine since they vary with location
and time. As a result, it becomes obligatory for an SU to check for the presence
of PU on a channel. Various methods for recognizing these TVWSs and avoiding
interference are presented. Geolocation database and spectrum sensing are
two of these technologies. Accurate and rapid spectrum sensing is required for
dynamically accessing white spaces while preventing harmful interference to
licensed services like Digital Terrestrial Television (DTT) and wireless microphone
systems. White Space Devices (WSDs) or SUs can either perform spectrum
sensing to detect primary signals or rely on geolocation databases to locate unused
spectrum in the vicinity. WSDs or SUs can either do spectrum sensing to detect
primary signals or use geolocation databases to locate unused spectrum nearby.
Because of its simplicity, the geolocation approach is widely used, however it can
be cautious in returning available channels and only protects registered systems.
Because of its lower cost, energy usage, and hardware complexity, this technique
is preferred over spectrum sensing.
To tackle these limitations and ensure optimal utilization of white space, a hybrid
approach incorporating spectrum sensing and a geolocation database (GLDB) is
suggested. This approach aims to update databases and facilitate harmonious
coexistence with dynamic incumbent systems. Spectrum sensing is classified
into two types: narrowband and wideband. Narrowband techniques are focused
on locating vacant frequency slots within a narrow frequency range, whereas
wideband techniques span a wider frequency range. Narrowband techniques
include energy detection, cyclostationary feature identification, and matched
filter detection [11][19]. However, with low value SNRs, energy detection performs

2
poorly, cyclostationary feature identification is confined to periodic signals, and
matched filter detection necessitates foreknowledge of PU signals. Moreover,
these methods are unable to identify particular opportunities across a wide
spectrum. Consequently, the adoption of wideband spectrum sensing (WSS)
has grown. Traditional WSS approaches demand high-speed ADCs, introducing
unnecessary complexities and expenses. Sub-Nyquist [19] or compressive sensing
approach, on the other hand, use sampling rates below the Nyquist rate, allowing
the use of easily accessible off-the-shelf components while maintaining efficiency.

RELATED WORK
The authors of [20] proposed a consolidated system that brings together the
strengths of spectrum sensing and geolocation database to enhance the utilization
of white spaces. The researchers suggested an approach in [21] that combines a
geolocation database with spectrum sensing to access TVWS. They developed a
mathematical measure of spectrum efficiency for SUs to enable dynamic access
to TVWS. The impact of Programme Making and Special Events (PMSE) on the
reliable utilization of channels by SUs was investigated. The elevated sampling
rate poses a challenge for energy-constrained white space devices engaged in
spectrum sensing. To address this, in [22], the authors proposed a technique
that integrates Geolocation Database and compressive sensing (CS) to achieve
sub-Nyquist sampling rates. Theoretical findings demonstrated the superiority
of their method over the conventional approach. All methodologies mentioned
in [20]-[22] involve SUs directly connecting to the distant Geolocation Database,
leading to escalated load and transmission issues. Consequently, WSS is utilized,
involving the sampling of signals at sub-Nyquist rates, with the aid of high-speed
ADCs to gather these samples.
In [23], the author employed a combined technique that capitalizes on the benefits
of both spectrum sensing and Geolocation Database. The process involves first
acquiring channel occupancy details from the geolocation database, followed
by sequential spectrum sensing. However, this approach prolongs the sensing
duration and could potentially lead to interference or instances of overlooked
detection. In order to ensure trustworthy and dependable communication, specific
APIs are implemented. Authenticated utilizers use IETF's Protocol to Access
White Space (PAWS) to access the Geolocation Database. To accommodate
the presence of wireless microphones, the utilization of spectrum sensing is
implemented.
In [24], the authors introduced a combined method that merges geolocation and
spectrum sensing through compressive sensing. They suggested an algorithm
called data-assisted non-iteratively reweighted least squares (DNRLS) to improve
the precision of detection. This algorithm utilizes the channel power from the
geolocation database for weight calculations, eliminating the need for repetitive
processes. However, adopting this non-iterative approach leads to decreased
possibility of reconstructing signal accurately. Additionally, the presence of an

3
illegal primary user or modifications to the geolocation database can disrupt the
reconstruction.
This study presents a combined strategy that merges sub-Nyquist wideband
spectrum sensing and geolocation databases. The objective is of this study is:
• To efficiently exploit available unused frequencies, especially when dealing
with changing PUs like wireless microphones which are not registered.
• To utilize the previous knowledge of compressive sensing support for
decreasing measurement count and computing efforts in signal recovery.
• To utilize information from the GLDB regarding the occupied TV channels
to reduce the sensor node's sensitivity during sensing. To check whether
this approach transfers complexity from local sensing to the main network,
resulting in lower computational intricacy and energy usage for spectrum
sensing.

Proposed Methodology
3.1 Proposed Model
The suggested hybrid model utilizes TVWSs through a sensor network and
geolocation database (see fig.1). Multiple nodes perform WSS while updating
the database, aiming to enable shared use of TV bands by WSDs while minimizing
interference to other users, such as digital TV broadcast signals and wireless
microphones that are not registered.

Fig. 1: Proposed hybrid scheme

In this scheme, the GLDB plays a crucial role in managing the operation of
WSDs within TV White Spaces. WSDs determine their location and query the
GLDB to obtain information about available frequencies at that specific location.
The geolocation database serves as a control mechanism, ensuring that WSDs
operate within the designated frequencies without causing interference. In this

4
operational scenario, a master device communicates with slave devices. The
master WSD directly connects to the GLDB to obtain operational parameters
for its coverage area's slave WSDs. Using received spectral occupancy status, the
master WSD assigns available frequency spectrum to the slave WSDs. Despite
heavy usage by incumbents like local TV and radio stations, the geolocation
database maintains stability in certain TV spectral bands due to long-term
broadcasting policies. This prior GLDB information enhances spectrum sensing
reliability, compensating for potential errors due to dynamic spectrum occupancy
changes.

3.2 Signal Model

Let the received signal x(t) be represented as:
x(t) = sg (t) + n0 (t) (1)
where sg (t)= Incumbent user signal
n0 (t) = Additive White Gaussian Noise with mean = 0 and variance = σ 2 .
Let x(t) be a continuous time wideband signal which exhibits sparsity in the
frequency domain. This means that the energy of the signal is concentrated
in fragmented bands, resulting in a spectrum measurement that is significantly
smaller compared to the entire frequency spectrum of the signal.
The frequency band in WSS is segmented into N narrowbands with bandwidth B.
The N channels are indexed from 0 to N-1, and during sensing, there can be up to
c active channels, where c is a fraction of N defined by the spectrum utilization
ratio u such that c = (u * N) → u = (c/N) < 1. The channel occupancy values
are used to calculate busy channels, and the spectrum usage ratio determines
channel utilization. The objective here is to detect licensed transmissions and
identify the active channel set
P = P1 , P2 , P3 , . . . . . . .Pc representing the occupied narrowbands. Therefore,
the goal is to determine which narrowbands are currently in use by licensed
users/PUs and which are not being used.
Due to the limited resources and size of SUs, a sub-Nyquist sampling approach
is used in our WSS scheme for minimizing sampling and computation costs. To
enhance energy efficiency, the approach utilizes compressed multicoset sampling,
effectively minimizing costs associated with signal sampling and acquisition.
Practical implementation of multicoset sampling can be realized through a
time-interleaved ADC, where a subset of channels is utilized, simplifying the
process.
The complete process of the proposed approach is outlined through sequential
steps, as illustrated in fig. 2 of the accompanying flowchart:

5
Step 1: Sensors gather signal samples using a multicoset sampler, employing a
sampling rate below the Nyquist threshold.
Step 2: A reduced-dimension measurement matrix is generated utilizing Sub-
Nyquist samples.
Step 3: Occupied channels are detected using a joint sparse recovery method
that exploits the sparsity of the multiband signals. This algorithm integrates
previous information about channel occupancy status acquired from a GLDB,
thereby mitigating computational complexity.
Step 4: Once the white spaces are identified, it can be allocated to SUs.

Fig. 2: Flow diagram

Figure.2 :Flow Chart
The underutilized wideband spectrum's inherent sparsity leads to the adoption
of sub-Nyquist sampling, which collects wide-band frequencies at a lower rate
than Nyquist sampling. Integration of GLDB's prior occupancy information in
spectrum sensing simplifies calculations. The subsequent mathematical expres-
sion elucidates the accumulation of compressive measurements accompanied by
random noise at each sensor device.
y0 = Φx + ξ (2)
Equation (2) is defined as follows:
y0 is the vector containing compressive measurements, with a size of (m × 1).
x represents the original signal vector, also sized (n × 1).

6
Φ, a matrix with dimensions (m × n), serves as the sensing matrix to gather
compressive measurements y0 from the original signal x, where m < n.
Here, m stands for the number of measurements (c < m < n), and n denotes
the number of samples.
ξ is the vector representing random noise.
After identifying the solution within the reduced replacement set, the estimation
of minimum and maximum channel occupancy across all channels takes place.
In the realm of compressed sensing, where m is considerably smaller than n and
the magnitude of ξ ϵ is bounded by an upper limit ϵ (with ∥ξ∥2 < ϵ), the process
unfolds. Under specific conditions like the restricted isometry property (RIP)
pertaining to Φ, the equation’s signal (as in equation 2) can be reconstructed
robustly. As a result, the application of this RIP leads to a resilient signal
recovery, wherein the sparse reconstruction framework selects the solution with
the least sparsity.
2
minx∈Rn ||x||o subject to | |Φx − yo | |2 ≤ ε (3)
Enhancements can be made to the conventional sparse recovery framework by
incorporating supplementary prior data about the signal. When there exists
either partial or complete prior knowledge about the sparse domain's support,
adjustments can be introduced to the sparse recovery framework. This adaptation
aims to identify a signal that not only accounts for the measurements but also
contains the fewest additional elements in its support when compared to the
existing known support. This modified framework ensures that the sparsest
solution is given by selecting only those indices that are present in both the prior
information and the measurements. This approach can significantly improve the
performance of the sparse reconstruction process. Hence, it is feasible to adapt
the sparse recovery framework when there are supplementary entries within the
recognized support set L. The sparse result is then written as:
2
min ||(x)Lc ||o subjectto | |Φx − yo | |2 ≤ ε (4)
The minimization problem in equation (3) aims to find the solution that has
the smallest number of non-zero elements, while still being able to explain the
measurements. Equation (4) modifies this problem to focus only on minimizing
the count of non-zero components that are not part of the known support L,
which denotes the positions where non-zero components are already known to
exist. By reducing the count of non-zero components that are not already known,
the resulting solution will be the sparsest possible solution. This modification
allows for additional preliminary details about the sparse domain's support
within the sparse recovery framework, potentially enhancing the performance of
the recovery structure. The optimization problem described in equation (4) goes
beyond exclusively identifying non-zero values solely in the positions indicated
by L. Instead, it aims to minimize the count of non-zero values in other places
which are outside the known support set L, while still satisfying the measurement
constraints. In the event of errors within the known support set L, where few

7
locations within L aren't truly part of the support, equation (4)'s optimization
task remains capable of accurately reconstructing the signal. Nevertheless, this
might necessitate a larger number of measurements than in scenarios without
errors within the recognized support set L. The least number of measurements
required for exact reconstruction through l0 -minimization is determined by the
restricted orthogonality constant θ, represented as:

θk+2u < 1

In this context:
k =signifies the extent of the identified support segment
(k: =|L|),
u =denotes the scope of the unidentified support segment
(u: = |U|),
Err =represents the magnitude of the discrepancy in the recognized support
section (err: = |L\P|),
so that c= k + u - err
where P is support of x.
In compressive sensing, exact reconstruction of a sparse signal relies on having
a sufficient number of measurements. The restricted orthogonality constant θ
quantifies the minimum required measurements for exact reconstruction. By
incorporating prior knowledge about signal’s support, the measurement count
needed for accurate recovery can be reduced. This reduction in measurements
and sampling rate is beneficial for power-limited devices, as it decreases the data
processing load, making compressive sensing more efficient and practical for such
devices.
Compressive multicoset sampling is used to lower overall expenses of data
gathering and sampling while reducing the amount of energy consumed. The
multicoset sampling is executed by uniformly capturing samples at specific
periods(mN + gi)T , employing a reduced sampling rate of 1/NT = fs /N. This
is carried out with a time offset of(g i T ), where i takes on values from 1,2. . . ..
onwards s, and m is an integer.
Here, T = 1/fs stands for the base sampling period, fs signifies the Nyquist
sampling rate, N denotes the channel count, s represents the number of parallel
s
cosets, and G indicates the multicoset sampling pattern. The set G = {gi }i=1 is
a collection of s different numbers, arbitrarily chosen from the range [0, N − 1],
forming a (N, s) sampling arrangement.
The measurement order of the ith coset, denoted as xgi (b), assumes the value
x(bT) when b = (mN+gi), under the condition that m is part of the set of
integers; otherwise, it is zero. The process of computing this measurement

8
sequence involves utilizing the Fourier transform (FT), resulting in segments
that bear a resemblance to portions of the original spectrum.


Xgi ej2πf T =
PN −1 2πg m
1 j Ni
NT m=0 Xm (f )e ∀f ∈ [0, B] (5)
For 1 ≤ i ≤ s, Xm (f ) is defined as Xm (f + NmT ) where X(f ). is the original
spectrum of the signal and m is an integer between 0 and N-1. Xm (f) represents
a portion of the original spectrum that is moved left by m/NT units and
corresponds to the mth channel.
As
x(t) = sg (t) + no (t) (pl see equation 1)
The FT of above equation will be
X(f ) = Sg (f ) + No (f ) (6)
Rewriting equation (5) in matrix form as:

Xg1 (ej2πf T )

 
Xg2 ej2πf T 
 
 .. 
 . 
Xgs (ej2πf T )
Yo (f )

2π
 2π 2π

ej N g1 0 ej N g1 1 ... ej N g1 (N −1)
 j 2π 2π
e N g2 0 ej N g2 (N −1)

2π 
 . e N g2 1 ···
 . .. 
 . .



 .. 

 . ... 
2π 2π 2π
1 ej N gs 0 ej N gs 1 · · · ej N gs (N −1)
=
NT Q

 
x1 (f )
 x2 (f ) 
 
 .. 
 . 
× xn (f ) = Q [Sg (f ) + N0 (f )] , ∀f ∈ [0, B]
X(f )

(7)
Y0 (f ) = QX(f ) = Q [Sg (f ) + N0 (f )] , ∀f ∈ [0, B] (8)
where Y0 (f ) = vector of length s whose ith row is Xgi (ej2πf T ), X(f) = unidentified
frequency range vector of x(t) with N frequency bands, and Q ∈ Cs×N matrix
and its (i, j)th component value is given as:

9
 1 j 2π
N gi (j−1)
Qi,j = NT e (9)
The support represented by supp (Sg(f)) in equation (8) corresponds to the
occupied channel index set P . This equivalence arises because the value of N in
the multicoset sampler is determined according to the count of channels within
the spectrum under consideration.
In equation (8), the process of reconstructing the unknown matrix X(f) can be
called as the multiple measurement vectors (MMV) problem. X(f) is said to
be row-sparse because it only has non-zero entries in a few rows. This means
that most of the rows in the matrix contain only zeros. The goal of the MMV
problem is to estimate the entire matrix X(f) from a limited set of measurements,
which makes it a challenging task.
3.3 Joint Spectrum Support Recovery Utilizing Prior Geolocation Database Knowl-
edge:
In compressed sensing, the reconstruction of a signal is achieved by solving
an optimization problem. The goal is to determine a sparse or compressible
depiction of signal that best matches with measurements obtained from the
compressed sensing process.
The MMV technique is employed to reconstruct the row-sparse matrix X(f ).
An occupied channel is indicated by a non-zero value within a sparse vector. To
enhance recovery performance under sub-Nyquist sampling, unused spectrum
information from Geolocation Database is incorporated into corresponding rows
with significant magnitude. This leads to enhanced recovery performance with a
reduced number of measurements. The MMV approach aims to decrease the
number of rows containing non-zero values based on the measurement restriction
outlined in equation (10). In line with the MMV problem we write:
min Dlq (X) 0
subject to Y0 = QX (10)

T
Dlq (X)is a vector in RN = [w1 , w2 , . . . wi , . . . wN ]

here

wi = ith row of X which is lq norm

1
PN q
= ∥X[i]∥q = ( j=1 |Xi,j | ) q (11)
Although Equation (10) poses an NP-hard challenge, a solution is derived by
addressing the l1 minimization problem as follows:
min Dlq (X) 1
subject to Y0 = QX (12)
Within multicoset sampling nonzero row locations align with the occupied channel
index set P. Integrating prior information about occupied channels from the
GLDB into the recovery process can enhance outcome and minimize the necessary

10
measurement count. The sensor node can get information on available channels
through GLDB, aiding in reducing energy consumed during high-frequency signal
processing. The correlation between the occupied channel status obtained from
GLDB and the actual channel occupancy set P can be expressed as:
P = T ∪ ∆\∆e (13)
where T = prior information of the occupied channel
Indices from the GLDB (T ⊂ [0, N − 1]),
P is actual channel occupancy set,
P
∆ is the new channel occupancy set: = T ,
∆e is the set of channels declared occupied by the GLDB but currently vacant.
= T / P.
Incorporating previous channel occupancy status from a GLDB in the reconstruc-
tion procedure reduces the count of required measurements and computational
intricacy, leading to more precise local sensing. This also facilitates the iden-
tification of currently occupied channels, making the implementation process
easier. Information about the accessible channels within the sensing area can be
extracted from the Geolocation Database.
3.4 Enhancing Sparse Signal Recovery Using Prior Information and Subspace
Knowledge
To address this issue, two reconstruction methods are employed: a greedy
algorithm and lv norm minimization approach. To improve the performance of
these algorithms, channel occupancy information from geolocation databases is
integrated into the reconstruction procedure.
Modifications made to these algorithms involve constructing a signal approxima-
tion during every repetition. This allows for the incorporation of T , which is a
parameter related to the multicoset sampler used in the sub-Nyquist sampling
process, as an initial condition in the recursive process. By using this informa-
tion, the reconstruction algorithms can make more informed decisions about
which rows of the matrix X(f ) should be non-zero and which ones should be
zero, resulting in better overall performance.The multicoset sampler receives s
sets of sample sequences. In order to simplify the computational intricacy of
the reconstruction process, the covariance matrix of these sample sequences is
calculated as:
   
R = E Y0 (f )Y0H (f ) = Q Rs + σ 2 I QH (14)
here Rs = E[Sg (f )SgH (f )] - N×N major correlation matrix and H represents
Hermitian transpose.
The recognized covariance matrix R is decomposed into non-increasing principal
eigenvalues by means of the rank revealing eigenvalue decomposition as:

11
 σ2
R = EΛE H = Es Λs EsH + H
N T 2 En En (15)
where E = [Es , En ], Λs = diag{λ1 , λ2 , . . . , λc } which has c non-increasing
principal eigenvalues, Es = corresponding eigenvectors and En = eigenvectors √
corresponding to eigenvalue of (s-c). The measurement matrix χs = Es Λs
can be reconstructed from the c largest eigenvalues Λs and the corresponding
eigenvectors Es and hence the equation for linear system can be written as:
χs = Qυs (16)
Here, the support of the sparsest solution in equation (16), denoted as supp(υs )
= supp (Sg(f)), reaches convergence with the original primary signal's support.
As compared to the original sub-Nyquist matrix Y0 (f), use of this χs for sup-
port recovery reduces the computation complexity of white space devices. To
decrease computation complexity and enhance detection performance against
noise permutation, we use Simultaneous Orthogonal Matching Pursuit (SOMP)
on the created reduced-dimension measurement matrix χs . The modified ver-
sion of SOMP is called Subspace Based SOMP (SB-SOMP). By using χs the
computational price needed on the sensor nodes is reduced.
The SB-SOMP algorithm selects the column of matrix Q that has the highest
correlation with the leftover signal segments in every repetition. It adds the
index of the selected column to the estimated support set and removes the
contribution of the selected column from the measurements matrix χs . Then it
repeats the process on the remaining part of the signal. It is worth highlighting
that the chosen column indices from matrix Q correspond to the positions of
non-zero rows in the sparse signal υs . The parameter N for the multicoset
sampler is established according to the quantity of channels in the received
signal. This parameter determines the dimensionality of the measurement
matrix. Utilizing the geolocation database, which contains information about
registered license systems and their usage of specific channels, allows for the
recognition of associated columns within the measurement matrix that could be
engaged or held in reserve by these systems. Therefore, before the repetition
commences, the residual is set up as:
r0 = χs − Qτ +1 (Q†τ +1 χs ) (17)
The set τ , which is a subset of the integers from 0 to N − 1, is incremented by 1
to match respective column indices of Q, which initializes with 1.
The stopping criterion for recovery employed in accordance with sparsity places a
constraint on the count of maximum repetitions permitted in the reconstruction
procedure. Since the channel occupancy status of Television band obtained from
the GLDB provides prior knowledge about the active frequency bands in the
signal, it can be used to estimate the support set more accurately. This means
that the estimated sparsity level is closer to the true sparsity level of the signal.
As a result, the count of iterations needed to recover the data can be reduced,
which speeds up the sensing process.

12
Equation (10) is reformulated as a problem of weighted least squares, represented
in Lagrangian form as:
2 PN 2
min ∥Y0 − QX∥F + λ i=1 ωi (∥X[i]∥2 ) (18)
Here ||.||F = Frobenius norm, ωi = weight, and λ = Lagrange multiplier. In
each iteration, the weighting parameter ωi in equation (18) undergoes an update
depending upon outcome of the preceding iteration. The revised ωi is computed
as follows:
(t)
v−2
ωi = X (t−1) [i] 2 (19)
(t)
where xi is the weight vector with i = 1, . . . ., N, X(t-1) is the (t − 1)th
iteration,v=XXX. To incorporate prior information about channel occupancy
from the geolocation database, the weighting strategy in equation (18) is adjusted
as in equation (19). These new weighting parameters are updated at each iteration
of the recovery process, allowing for the integration of prior knowledge to improve
the accuracy of the reconstruction.
(
v−2
(t) µ X (t−1) [i] 2 , iϵτ
ωi = (t−1)

v−2 (20)
X [i] 2 , otherwise
where µ = constant.
Thus, at t-th iteration the solution is given as:
−1
X (t+1) = W (t) QT (QW (t) QT + λI) Y0 (21)
 
1 1
where W (t) = diag (t) , . . . , (t)
w1 wN

The algorithm stops running when the convergence criterion is met i.e.,
∥X (t+1) −X (t) ∥F
≤ ρ (22)
∥X (t) ∥F
where ρ = value selected by user.
Overall iterations decrease as the initial residue is changed due to inclusion of
geolocation database's prior information. Use of SOMP lowers the expense of
computing in sensor nodes with limited power and resources. c-iterations are
used by conventional OMP for recovering c-sparse signals. Only c − τ iterations
are necessary for SB-SOMP because τ -indices are obtained from GLDB. The
channel occupancy status from the GLDB offers pre-existing awareness about
certain occupied channels, which could potentially be either held in reserve or
utilized by authorized users. Notably, only a small fraction of potentially free
bands requires local spectrum sensing to be carried out.

13
Results and Discussions
The performance of the proposed system is evaluated through simulations,
considering 8 active channels out of a total of 40 available channels, resulting in
a channel utilization ratio of 20%. The TV frequency utilized is 470 MHz.
The suggested method is implemented in NS2 by creating source data and noise
data, followed by the evaluation of performance metrics such as throughput,
Packet Delivery Ratio (PDR), delay, and residual energy. The integration of prior
information regarding channel usage from the geolocation database contributes
to enhanced energy efficiency and reduced delay within the suggested method.
TABLE 1: THROUGHPUT Versus SIMULATION TIME

Simulate time Throughput (bits per second)

Sr. No. Proposed
(sec)
method CPDA
1. 50 149914 128367
2. 75 150384 133289
3. 100 150577 133239
4. 125 151702 133940
5. 150 150630 134262
6. 175 151295 135785
7. 200 151556 135733

Throughput can be defined as the successful transmission of packets across

a communication channel. Fig. 3 illustrates the graphical representation of
throughput plotted against simulation time.

Fig. 3: Throughput

14
Within compressed sensing, the probability of false alarm is influenced by both
Signal-to-Noise Ratio (SNR) and compression ratio. While it rises with higher
SNR, it remains fairly consistent when the compression ratio decreases from
100% to 50%. However, as the compression ratio further drops to 25%, the
likelihood of false alarm increases, signifying that accurate signal recovery is no
longer achievable. This situation subsequently leads to a decline in the network's
throughput. The sparsity values determine the least compression ratio required
for recovering signal efficiently. Initial frequency spectrum occupancy information
from GLDB reduces the spectrum sensing time, and hence the energy required
for sensing will also be reduced. As a consequence, time for data transmission is
also reduced. Thus, there is an improvement in throughput as compared to the
current system that solely relies on the geolocation database for accessing white
spaces. The suggested approach yields a 13% increase in throughput.
Packet delivery ratio (PDR) is crucial for power-efficient nodes with limited
resources. Spectrum situation can cause packet losses, and PDR is affected by
propagation losses and spectrum handoffs during sensing. The Packet Delivery
Ratio (PDR) experiences a decline when primary users (PUs) occupy the channel
for a long time. PDR decreases when the channel is occupied by primary users
(PUs) for a long time. The geolocation database provides prior information,
reducing packet drops and significantly improving PDR by an average of 12% in
comparison with CPDA technique, as shown in Figure 4.
TABLE 2: PDR Versus SIMULATION TIME

Sr. No. Simulate time Packet Delivery Ratio (%)

(sec)
Proposed Method CPDA
1. 50 93.4286 80
2. 75 93.8333 83.1667
3. 100 94 83.1765
4. 125 94.7273 83.6364
5. 150 94.0741 83.8519
6. 175 94.5 84.8125
7. 200 94.6696 84.7862

15
 Fig. 4: Packet Delivery Ratio
Due to joint spectrum sensing with geolocation database, the probability of
accurate channel detection will be high. Thus, data is sent to the BS using that
vacant channel and hence the number of packets dropped will be less.
The suggested system employs a reconstruction procedure based on sparsity-
based stopping criteria. By utilizing TV channel occupancy information from the
geolocation database, the number of iterations needed for recovery is reduced,
leading to minimized computation time and delay. This method enhances
the detection performance, provides smaller sampling rates, and improves the
system's lifetime due to reduced delay. Figure 5 compares the delay between the
proposed and existing methods. Pl write in % improvement.
TABLE 3: DELAY Versus SIMULATION TIME

Simulate time Delay (sec)

Sr. No. Proposed
(sec)
Method CPDA
1. 50 0.006249 0.014715
2. 75 0.006084 0.0135842
3. 100 0.005804 0.0131332
4. 125 0.005744 0.0128727
5. 150 0.005717 0.0121264
6. 175 0.005473 0.0114562
7. 200 0.005335 0.0110763

16
 Fig. 5: Delay
Accessing TV white space spectrum via a combination of Geolocation
Database and spectrum sensing approach minimizes computing intricacy,
which therefore lowers energy consumed. The amount of energy needed for
spectrum sensing is decreased with information about frequency band utilization
from a geolocation database. As shown in Fig. 6, the lowered computational
intricacy and duration results in decreased energy consumption by a node during
communications.
Fig.6: Residual energy

CONCLUSION
This work presents a low complexity hybrid method that combines sub-Nyquist
wideband spectrum sensing and geolocation database utilization. In scenarios
involving dynamic incumbent systems, this method efficiently utilizes available
white spaces. By incorporating prior channel occupancy details from the GLDB,
the required sub-Nyquist sampling ratio is reduced, leading to improved detection
performance as demonstrated by simulations. Moreover, the approach streamlines
the convergence process by reducing the count of repetitions and calculation
time necessary. This suggested approach changes the channel occupancy state by
using the geolocation information, minimizing interference hazards to existing
incumbent users. This allows for accurate spectrum occupancy evaluation while
requiring less computing power, making it suitable for tiny devices. With the
proposed technique, throughput improves by 13% overall. The PDR is improved
by an average of 12% in comparison with CPDA technique,The delay is reduced
by ***%.This approach improves detection performance and allows for reduced
sample. This approach improves detection performance, allows for lower sample
rates, and increases system lifetime due to reduced delay.

17
References
1. Cisco annual internet report 2018-2023..
2. J. Bradley, et al., “Embracing the Internet of Everything To Capture Your
Share of $14.4 Trillion”, White Paper, Cisco 318, 2013.
3. Ofcom, Implementing TV White Spaces, Ofcom, London, U.K.,Feb.
2015.[Online].Available:http://stakeholders.ofcom.org.uk/binaries/consultations/white-
space-coexistence/statement/tvws-statement.pdf.
4. P. Kolodzy, “Spectrum policy task force report”, Federal Communications
Commission, Washington, DC, USA, ET Docket no. 02-135, Nov. 2002.
5. “Amendment of the commission’s rules with regard to commercial opera-
tions in the 3550–3650 MHz band,” Federal Communications Commission,
Washington, DC, USA, ET Docket no. 12-354, June. 2017.
6. “In the matter of revision of part 15 of the commission’s rules to permit
unlicensed national information infrastructure (U-NII) Devices in the 5 GHz
band: Notice of proposed rulemaking,” Federal Commun. Commission,
Washington, DC, USA, ET Docket no. 13-49, May 2015.
7. Tanuja Dhope, Dina Simunic and A.Kerner, “Analyzing the Performance of
Spectrum Sensing Algorithms for IEEE 802.11af Standard in CR Network*”,
Studies in informatics and Control, Vol.21, Issue 1, pp.93-100, March 2012.
ISSN 1220-1766. (SCI and Web of science).
8. Tanuja Dhope, Dina Simunic,“Spectrum Sensing Algorithm for CR Net-
works for Dynamic Spectrum Access for IEEE 802.11af standard”, In-
ter.Journal of Research and Reviews in Wireless Sensor Networks,UK,
Vol.2., No.1,pp.77-84 , March 2012.ISSN: 2047-0037.(Web of science).
9. M. Fitch, M. Nekovee, S. Kawade, K. Briggs, and R. MacKenzie, “Wireless
service provision in TV white space with cognitive radio technology: A
telecom operator’s perspective and experience”, IEEE Commun. Mag., vol.
49, no. 3, pp. 64–73, Mar. 2011.
10. Vijaya G. Rajeshwarkar, Shekhar M. Jagade, P. Chandrasekhar Reddy
and Tanuja S. Dhope, “Regulations and standards for TV white space”,
Helix Journal (ESCI) , Thomson Reuters, Accepted in Publication Process
June 2019(ISSN) E-ISSN: 2319-5592; P-ISSN: 2277-3495.
11. Tanuja Dhope, Dina Simunic and R.Prasad, “Hybrid Detection Method
for CR”,13th International conferenace on SoftCOM, SoftCOM2011,Split-
Hvar-Dubrovnik, SS1-78741-1609,pp.1-5,Sept.2011.Print ISBN: 978-1-4577-
1439-9,INSPEC Accession Number: 12343806.
12. Tanuja Dhope, Dina Simunic and R.Prasad, “TVWS Opportunities and
Regulation:-Empowering Rural India”, 14th International Symposium

18
 on WPMC-11, Brest, France, pp.201-205, Oct. 2011.ISSN: 1347-6890,E-
ISBN: 978- 2-908849-26-4,PrintISBN: 978-1-4577-1786-4,INSPEC Accession
Number: 12386930.
13. Tanuja Dhope, Dina Simunic and R.Prasad, “TVWS opportunities and
regulatory aspects in India”, 14th International Symposium on WPMC
’11, Brest, France, pp. 566-570, Oct’2011. ISSN: 1347-6890,E-ISBN: 978-2-
908849-26-4, PrintISBN: 978-1-4577-1786-4, INSPEC Accession Number:
12386930.
14. T. Dhope, D. Simunic, “Performance Analysis of Covariance Based Detec-
tion in Cognitive Radio”, 35th Jubilee International Convention MIPRO
2012under Green ICT world, Opatija, Croatia, pp.737-742,May.2012. Print-
ISBN: 978-1-4673-2577-6, INSPEC Accession Number: 12865443.
15. T. Dhope, D. Simunic, “Hybrid Detection Method for Spectrum Sensing
in Cognitive Radio”, 35th Jubilee International Convention MIPRO 2012
under Green ICT world. Opatija, Croatia, pp. 765-770, May 2012. Print-
ISBN: 978-1-4673-2577-6, INSPEC Accession Number: 12865443]
16. G. Naik, S. Singhal, A. Kumar and A. Karandikar, “Quantitative assess-
ment of TV white space in India”, 2014 Twentieth National Conference on
Communications (NCC), pp. 1-6, 2014, doi: 10.1109/NCC.2014.6811306.
17. T. Shimomura, T. Oyama and H. Seki, “Analysis of TV White Space Avail-
ability in Japan”, IEEE Vehicular Technology Conference (VTC Fall), Que-
bec City, QC, Canada, 2012, pp. 1-5, doi: 10.1109/VTCFall.2012.6399270.
18. J. van de Beek, J. Riihijarvi, A. Achtzehn and P. Mahonen, “TV White
Space in Europe”, in IEEE Transactions on Mobile Computing, vol. 11,
no. 2, pp. 178-188, Feb. 2012, doi: 10.1109/TMC.2011.203.
19. G.P. Aswathy, K. Gopakumar, “Sub-Nyquist wideband spectrum sensing
techniques for cognitive radio: A review and proposed techniques,”AEU
- International Journal of Electronics and Communications, 104, 44-57,
(2019).
20. Wang, J., Ding, G., Wu, Q. et al., “Spatial-temporal spectrum hole dis-
covery: a hybrid spectrum sensing and geolocation database framework”,
Chin. Sci. Bull. 59, 1896–1902 (2014). doi.org/10.1007/s11434-014-0287-5.
21. N. Wang, Y. Gao and B. Evans, “Database-augmented spectrum sensing
algorithm for cognitive radio”, 2015 IEEE International Conference on Com-
munications (ICC), pp. 7468-7473, 2015, doi: 10.1109/ICC.2015.7249520.
22. Z. Qin, L. Wei, Y. Gao and C. G. Parini, “Compressive spectrum sensing
augmented by geo-location database”, 2015 IEEE Wireless Communications
and Networking Conference Workshops (WCNCW), pp. 170-175, 2015.
doi: 10.1109/WCNCW.2015.7122549.

19
23. Dionisio, R., Ribeiro, J., Marques, P. et al., “Combination of a ge-
olocation database access with infrastructure sensing in TV band”,
Journal of Wireless Communication Network 2014, 210 (2014).
https://doi.org/10.1186/1687-1499-2014-210.
24. Z. Qin, Y. Gao and C. G. Parini, “Data-Assisted Low Complexity Com-
pressive Spectrum Sensing on Real-Time Signals Under Sub-Nyquist Rate”,
IEEE Transactions on Wireless Communications, vol. 15, no. 2, pp. 1174-
1185, Feb. 2016, doi: 10.1109/TWC.2015.2485992.

20