Sequential Sensing Scheme for Cognitive Radio Based on a Block of Received Samples

Abstract

A method and system for determining whether a given electromagnetic frequency is in use includes applying a transformation to an amplitude of received samples, adjusting the transformed samples by a constant based on a minimum detection signal-to-noise ratio; combining the adjusted samples to produce a test statistic; and using a processor to make a determination regarding if the frequency is in use based on the test statistic exceeded or falling below a threshold, said test statistic being based on Ξq=Σi=1q×L(|ri|2−Δ), where q is the block index, ri is the ith received sample, and Δ is the constant.

Description

BACKGROUND

1. Technical Field

The present invention relates to cognitive radio and, more particularly, to systems and methods for determining whether a given spectrum band is unoccupied based on a block of received samples.

2. Description of the Related Art

Cognitive radio (CR) that supports secondary (unlicensed) users to access licensed spectrum bands not being currently occupied can dramatically improve spectrum utilization. Since the licensed (primary) users are prior to the secondary users (SUs) in utilizing the spectrum, the secondary and opportunistic access to licensed spectrum bands is only allowed to have negligible probability of deteriorating the quality of service of the primary users (PUs). Spectrum sensing performed by the secondary users to detect the unoccupied spectrum bands, is an important step in meeting this requirement.

Several spectrum sensing schemes, such as matched-filter detection, energy detection, and cyclostationary detection, have been proposed and investigated. Among these sensing schemes, energy detection does not rely on any deterministic knowledge about the primary signals and has low complexity. However, energy detection entails considerable amount of sensing time at the low detection signal-to-noise ratio (SNR) level, e.g., the sensing time is inversely proportional to SNR². To overcome this shortcoming, another sensing scheme, the sequential probability ratio test (SPRT), has been proposed for CR.

The SPRT has been widely used in many scientific and engineering fields since it was introduced in the 1940s. For given detection error probabilities, the SPRT requires a small average sample number for testing simple hypotheses. However, the SPRT-based sensing schemes proposed to date have several potential drawbacks: First, SPRT needs deterministic information or the statistical distribution of certain parameters of the primary signals. Acquiring such deterministic information or statistical distributions is practically difficult. Secondly, when the primary signals are taken from a finite alphabet, the test statistic of the SPRT based sensing scheme involves a special function, which incurs high implementation complexity. Thirdly, SPRT adopts the Wald's choice on the thresholds. However, the Wald's choice, which works well for the non-truncated SPRT, increases error probabilities when applied for the truncated SPRT.

SUMMARY

A method for determining whether a given electromagnetic frequency is in use includes applying a transformation to an amplitude of received samples, adjusting the transformed samples by a constant based on a minimum detection signal-to-noise ratio, combining the adjusted samples to produce a test statistic; and using a processor to make a determination regarding if the frequency is in use based on the test statistic exceeded or falling below a threshold, the test statistic being based on Ξ_q=Σ_i=1^q×L(|r_i|²−Δ), where q is the block index, r₁ is the i^th received sample, and Δ is the constant.

A system for determining whether a given electromagnetic frequency is in use, includes a transformation module configured to transform an amplitude of received samples, an adjustment module configured to adjust the transformed samples by a constant based on a minimum detection signal-to-noise ratio and to combine the adjusted samples to produce a test statistic, and a test module configured for making a determination using a processor as to whether the frequency is in use based on the test statistic exceeding or falling below a threshold, the test statistic being based on Ξ_q=Σ_i=1^q×L(|r_i|²−Δ), where q is the block index, r_i is the i^th received sample, and Δ is the constant.

These and other features and advantages will become apparent from the following detailed description of illustrative embodiments thereof, which is to be read in connection with the accompanying drawings.

BRIEF DESCRIPTION OF DRAWINGS

The disclosure will provide details in the following description of preferred embodiments with reference to the following figures wherein:

FIG. 1 shows a block/flow diagram that illustrates an illustrative embodiment of the present principles.

FIG. 2 shows a block/flow diagram illustrating an exemplary method for determining whether a frequency is in use according to the present principles.

FIG. 3 shows a block/flow diagram illustrating an exemplary system for determining whether a frequency is in use according to the present principles.

FIG. 4 shows a block/flow diagram illustrating an exemplary method for adjusting thresholds according to design specifications.

FIG. 5 shows a graph that illustrates how the test statistic for received samples is used to determine whether a frequency is in use.

FIG. 6 shows a graph that illustrates how the test statistic for received sample blocks is used to determine whether a frequency is use.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

Cognitive radio (CR) supports secondary and opportunistic access to licensed spectrum to improve spectrum utilization. The present principles are directed to a truncated, sequential sensing scheme having a simple test statistic. The present principles deliver a considerable reduction in average sensing time needed to determine whether a given band is unoccupied, while maintaining detection performance that is comparable to prior art techniques. Referring to FIG. 1, a general outline of the present principles is shown. First, appropriate thresholds are determined 102 that produce suitable probabilities of false alarm 104 and of misdetection 106 according to design specifications. Next, channel occupancy is determined 108. Determining whether a given band is occupied involves first sampling the channel 110. A test statistic is then calculated based on said samples 112, and is compared to the above-described thresholds 114 to produce an occupancy determination.

Embodiments described herein may be entirely hardware, entirely software or including both hardware and software elements. In a preferred embodiment, the present invention is implemented in software, which includes but is not limited to firmware, resident software, microcode, etc.

Embodiments may include a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system. A computer-usable or computer readable medium may include any apparatus that stores, communicates, propagates, or transports the program for use by or in connection with the instruction execution system, apparatus, or device. The medium can be magnetic, optical, electronic, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium. The medium may include a computer-readable medium such as a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk, etc.

The application makes use of the following notation. Upper boldface and low boldface letters are used to denote matrices and vectors, respectively; I_M denotes an M×M identity matrix; E[•] denotes the expectation operator, (•)^T denotes the transpose operation; N_p^q denote a set of consecutive integers from p to q, N_p^q:={p, p+1, . . . , q}, where p is a non-negative integer and q is a positive integer or infinity; I_{x≧0} denotes an indicator function defined as I_{x≧0}=1 if x≧0 and I_{x≧0}=0 if x<0.

Consider a narrow-band CR communication system having a single secondary user (SU). The SU shares the same spectrum with a single primary user (PU) and needs to detect the presence/absence of the PU to determine whether it is permissible to use the spectrum. This is shown as block 108 in FIG. 1. Detecting for the primary signals is formulated as a binary hypothesis testing problem as

H ₀ :r _i =w _i , i=1, 2, . . . , ad infinitum  (1)

H ₁ :r _i =s _i +w _i , i=1, 2, . . . , ad infinitum  (2)

where r is the signal received by the SU at time instant i, w_i is additive white Gaussian noise, s_i is the transmitted signal of the PU, and H₀ and H₁ denote the null and alternative hypotheses, respectively. It may be further assumed that1) w_is are modeled as independent and identically distributed (i.i.d.) complex Gaussian random variables (RVs) with means zero and variances σ_w², i.e., w_i˜CN (0,σ_w²);2) the primary signal samples s_i are i.i.d.;3) w_i and s_i are statistically independent; and4) the perfect knowledge on the noise variance σ_w² is available at the SU.

The same assumptions have been made in energy detection. In practice, the noise variance σ_w² can be known a priori by an appropriate measurement. In energy detection, the energy of the received signal samples is first computed and then is compared to a predetermined threshold. The test procedure of energy detection is given as

$T\left(r\right)=\frac{1}{M}\sum _{i=1}^M{r_i}^2\stackrel{\stackrel{n_1}{>}}{\underset{\underset{n_0}{<}}{=}\kappa }$

where r:=[r₁, r₂, . . . , r_M], T(r) denotes the test statistic, M represents the number of samples available for making a decision, and κ denotes a threshold for energy detection.

One principal advantage for energy detection is that, in its sensing process, energy detection requires no deterministic knowledge of the primary signals and thus is known as a form of non-coherent detection. On the other hand, one major drawback of energy detection is that, at the low detection SNR level, it requires a large sensing period. For energy detection, the number of sensing samples increases on the order of SNR⁻² as SNR decreases.

To solve this problem, a simple sequential detection scheme may be used, having the following statistic,

$\begin{array}{cc}{\Lambda }_N=\sum _{i=1}^N\left({r_i}^2-\Delta \right)& \left(3\right)\end{array}$

where Δ is a predetermined constant. Calculation of the test statistic is shown as block 112 in FIG. 1. The parameter Δ satisfies σ_w²<Δ<σ_w²(1+SNR_m) with SNR_m denoting the minimum detection SNR. Assuming that the detector needs to make a decision with M samples, the following testing procedure applies, shown as block 114 in FIG. 1:

Reject H₀:

if Λ_N≧b and N≦M−1 or if Λ_M≧γ;  (4)

Accept H₀:

if Λ_N≦a and N≦M−1 or if Λ_M<γ;  (5)

Continue Sensing:

if Λ_N∈(a,b) and N≦M−1  (6)

where a, b, and γ are three predetermined thresholds with a<0, b>0, and a<γ<b, and M is the truncated size of the test. Since each term in the cumulative sum Λ_N is a shifted squared random variable (RV), the test procedure (4)-(6) is termed the sequential shift chi-square test (SSCT). In statistical terms, the SSCT is a truncated sequential test. Referring now to FIG. 5, an exemplary test region of the SSCT is shown which includes two stopping boundaries: the lower- and upper-boundary labeled “a” and “b” respectively. The vertical axis represents the value of the test statistic after a given sample, while the horizontal axis represents the index of a given sample. The γ threshold is employed when the number of received samples reaches M and no decision has been made at or before the M^th sample.

It is evident from (4)-(6) that the test statistic depends only on the amplitudes of the received samples and the constant Δ. As will be shown below, the choice of the constant Δ depends on the minimum detection SNR instead of the exact operating SNR value, which is typically difficult to obtain in practice. To distinguish these two different SNRs, we denote the operating SNR as SNR_o.

Normalizing Λ_N by ρ_w²/2, equation (3) can be rewritten as

$\begin{array}{cc}{\Lambda }_N=\sum _{i=1}^N\left(v_i-2\Delta /{\sigma }_w^2\right)& \left(7\right)\end{array}$

where Λ_N:=2Λ_N/ρ_w² and v_i:=2|r_i|²/ρ_w². Let ξ_N denote the sum of v_i for i=1, . . . , N, i.e., ξ_N=Σ_i=1^Nv_i and let Δ denote 2Δ/ρ_w². With this notation, Λ_N can be rewritten as

Λ _N=ξ_N− Δ.  (8)

For notional convenience, Λ₀ and ξ₀ are defined as zero. Let a_i and b_i be two parameters defined as follows: a_i=0 for N₀^P, a_i=ā+i Δ for i∈N_P+1^+∞, and b_i= b+i Δ for b∈N₀^+∞, where ā:=2a/ρ_w²b:=2b/ρ_w² and P denotes the largest integer less than or equal to −a/Δ, i.e., P:=└−a/Δ┘. Applying the preceding transformation (8), (4)-(6) can be rewritten as

Reject H₀:

if ξ_N≧b_N and N≦M−1 or if ξ_M≧ γ_M;  (9)

Accept H_o:

if ξ_N≧a_N and N≦M−1 or if ξ_M≧ γ_M;  (10)

Continue Sensing:

if ξ_N∈(a_N,b_N) and N≦M−1  (11)

where γ_M= γ+M Δ with γ=2γ/ρ_w². Clearly then, a_M< γ_M<b_M. P_FA,M and P_MD,M are defined as false-alarm and misdetection probabilities, respectively.

It should be noted that the SSCT is not merely an SPRT. In the non-truncated SPRT case, the Wald's choice on thresholds which yield a test satisfying specified false-alarm and misdetection probabilities is not applicable. Alternatively, the thresholds a, b, and γ, and a truncated size M are selected beforehand, either purposefully or randomly, and corresponding P_FA,M and P_MD,M are then computed. This procedure is indicated as block 102 in FIG. 1 and, with more detail, as FIG. 4. These probabilities characterize the effectiveness of a given set of thresholds. If the probabilities are extremely below a desired margin of error, the thresholds should be adjusted in order to decrease the sensing time. If the probabilities are higher than design specifications permit, the thresholds should be adjusted to decrease the probability of error.

If the resulting P_FA,M and P_MD,M do not meet design specifications, the thresholds and truncated size are subsequently adjusted. Such process continues until desirable error probability performance is obtained. In the above process, it is important to efficiently and accurately evaluate false-alarm and misdetection probabilities for prescribed thresholds a, b, and γ, and a truncated size M, as is discussed below.

An exact formulation for false-alarm probability can be derived according to the present principles, and an iterative method is shown to compute misdetection probabilities. In describing these probabilities, the following definitions become useful:

f _{χ t} ^(k)(ξ)=∫_χt^ξdξ_k∫_χt−1^ξkdξ_k−1 . . . ∫_χ1^ξ2dξ₁, k≧1,  (12)

with the initial condition f_χk^(k)(ξ)=1, k=0, where χ₀=Ø and χ_k:=[λ₁, . . . , χ_k−1, χ_k] with 0≦λ₁≦ . . . ≦λ_k. Superscript k and subscript χ_k are used to indicate that f_χk^(k)(ξ) is a k-fold multiple integral with ordered lower limits specified by χ_k. It can be shown that the exact value of f_χt^(k)(ξ) can be obtained recursively. A second helpful integral is defined as:

I⁽⁰⁾:=1, and I⁽ⁿ⁾:=∫_Ω(n) . . . ∫dξ_n, n≧1  (13)

where ξ_n:=[ξ₁, ξ₂, . . . , ξ_n] with 0≦ξ₁≦ξ₂ . . . ≦ξ_n and Ω_(n)={(ξ₁, ξ₂, . . . , ξ_n):0≦ξ₁≦ . . . ≦ξ_n, a_i<ξ_i<b_i, i∈N₁ⁿ}. In particular, when n=1, I⁽¹⁾=∫_a1^b1dξ₁=b₁−a₁. Let c and d denote two positive real numbers with c<d, a_N−1≦c≦b_N, and a_N<d. Then,

${\psi }_{n,c}^N=\{\begin{array}{c}\left[\underset{Q}{\underset{}{b_{n+1},\dots ,b_{n+1}}},\underset{N-Q-n}{\underset{}{a_{Q+n+1},\dots ,a_{N-1},c}}\right],n\in N_0^{N-Q-2}\\ \left[\underset{N-n}{\underset{}{b_{n+1},\dots ,b_{n+1},c}}\right],n\in N_{N-Q-1}^{s-1}\\ b_{n+1}1_{N-n},n\in N_s^{N-2}\end{array}$

where s denotes the integer such that b_s<c≦b_s+1, Q denotes the integer such that a_Q<b₁≦a_Q+, and N≧2. Let A_i be an (N−n)×(N−n−i) matrix defined as A_i=[I_N−i−n|0_{i×(N−i−n)}]^T with i∈N₁^N. Furthermore: ψ_n,c^N−i=ψ_n,c^N·A_i, i∈N₁^N, and a_n1ⁿ²=[a_n1+1, . . . , a_n2], where ψ_n,c^N−i is a (N−i−n)×1 vector and a_n1ⁿ² is a (n₁−n₂)×1 vector. In particular, a_n1ⁿ² is defined as φ if n₁≧n₂.

A third useful integral is defined as:

J _c,d ^(N)(θ):=∫_Yc,d(N) . . . ∫e^−θξNdξ_N  (14)

where θ>0, N≧1, and

Y _c,d ^(N):={(ξ₁, . . . , ξ_N):0≦ξ₁≦ . . . ≦ξ_N, a_i<ξ_i<b_i, i∈N₁^N−1; c<ξ_N<d}

and θ is a positive real number.

Using these integrals, it is possible to formulate a false-alarm probability, shown as block 104 in FIG. 1. Let E_N denote the event that Λ_N≦b and a<Λ_n<b for n∈N₁^N−1 under H₀, where N∈N₁^M−1, and let E_M denote the event that Λ_M≧γ and a<Λ_n<b for n∈N₁^M−1 under H₀. Denote by P_H0(E_N) the probability of the event E_N under H₀, where E_N represents the event that under H₀ the test statistic Λ_N exceeds the upper boundary, where N∈(1,M). The overall false-alarm event is a union of E_N for 1≦N≦M. Recalling that the test procedure given in (5)-(6) is equivalent to that given in (10)-(11), one arrives at

$\begin{array}{cc}{P}_{H_0}\left(E_N\right)=\{\begin{array}{c}{P}_{H_0}\left(a_i<{\xi }_i<b_i,i\in N_1^{N-1};{\xi }_N\ge b_N\right),N\in N_1^{M-1}\\ P_{H_0}\left(a_i<{\xi }_i<b_i,i\in N_1^{M-1};{\xi }_M\ge {\stackrel{\_}{\gamma }}_M\right),N=M\end{array}& \left(15\right)\end{array}$

The false-alarm probability P_FA,M represents the likelihood that the SU will conclude that there is a PU on the channel, despite no such PU actually being present. P_FA,M with truncated size M can be written as P_FA,M=Σ_N=1^MP_H0(E_N). Note that under H₀, v_i is an exponentially distributed RV with rate parameter ½. The probability density function (PDF) of v_i under H₀ is p(v_i|H₀)=e^−vi/2, where v_i>0. Furthermore, the joint PDF of RVs v₁, . . . , v_N is given by

P _{v|H 0} (v₁, . . . , v_N)=2^−Ne^{−Σi=1Nvi/2}, v_i>0,  (16)

where v:=(v₁, . . . , v_N). Due to ξ_N=Σ_i=1^Nv_i, the following transformation between ξ_i and v_i: v₁=ξ₁ arises: v₂=ξ₂−ξ₁, . . . , v_n=ξ_N−ξ_N−1.

By applying this function and equation (16), one arrives at

$\begin{array}{cc}{P}_{\xi |H_0}\left({\xi }_1,\dots ,{\xi }_N\right)=2^{-N}{}^{-\frac{{\xi }_N}{2}},0\le {\xi }_0\le {\xi }_1\le \dots \le {\xi }_N& \left(17\right)\end{array}$

where ξ:=(ξ₁, ξ₂, . . . , ξ_N). According to equations (15), (17), and the definition of Y_bN,∞^(N), one finds

$\begin{array}{cc}\begin{array}{c}{P}_{H_0}\left(E_N\right)={P_H}_0\left(\left({\xi }_1,{\xi }_2,\dots ,{\xi }_N\right)\in Y_{b_N,\infty }^{\left(N\right)}\right)\\ =2^{-N}J_{b_N,\infty }^{\left(N\right)}\left(0.5\right),1\le N<M\end{array}\mathrm{and}& \left(18\right)\\ \begin{array}{c}{P}_{H_0}\left(E_M\right)={P_H}_0\left(\left({\xi }_1,{\xi }_2,\dots ,{\xi }_M\right)\in Y_{{\stackrel{\_}{\gamma }}_M,\infty }^{\left(M\right)}\right)\\ =2^{-N}J_{{\stackrel{\_}{\gamma }}_M,\infty }^{\left(M\right)}\left(0.5\right)\end{array}& \left(19\right)\end{array}$

Taking the above into account, the false-alarm probability, P_FA,M, is given by P_FA,M=Σ_N=1^MP_H0(E_N), where P_H0(E_N) can be recursively computed as

$P_{H_0}\left(E_N\right)=\{\begin{array}{c}{P}_N\frac{b_1b_N^{N-2}}{\left(N-1\right)!},N\in N_1^{P+1}\\ P_N\left[\begin{array}{c}{f}_{a_0^{N-1}}^{\left(N-1\right)}\left(b_{n-1}\right)-\\ I_{\left\{N\ge 3\right\}}\sum _{n=0}^{N-3}P_{H_0}\left(E_{n+1}\right)\times \frac{{\left(b_{N-1}-b_{n+1}\right)}^{N-n-1}2^n{}^{b_{n+1}/2}}{\left(N-n-1\right)!}\end{array}\right],N\in N_{P+2}^{Q+1}\\ P_N\left[f_{a_0^{N-1}}^{\left(N-1\right)}\left(b_{n-1}\right)-\sum _{n=0}^{N-3}f_{{\psi }_{n,a_{N-1}}^{N-1}}^{\left(N-1-n\right)}\left(b_{N+1}\right)\times 2^n{}^{\frac{b_{n+1}}{2}}P_{H_0}\left(E_{n+1}\right)\right],N\in N_{Q+2}^{M-1}\\ 2^{-M}J_{{\stackrel{\_}{\gamma }}_M,\infty }^{\left(M\right)}\left(0.5\right),N=M\end{array}$

where p_N=2^−(N−1)e^−bN/2.

A formulation for the misdetection probability, P_MD,M, is now presented, shown as block 106 in FIG. 1. The misdetection probability represents the likelihood that the SU will incorrectly conclude that there is no PU using the channel. Unlike the false-alarm case, v_i under H₁ is a non-central chi-square RV with two degrees of freedom and non-centrality parameter λ=2|s₁|²/σ_w². Conditioned on λ_i, the PDF of v_i under H₁ is given as

$\begin{array}{cc}p\left(v_i|H_1,{\lambda }_i\right)=\frac{1}{2}{}^{-\left(v_i+{\lambda }_i\right)/2}I_0\left(\sqrt{{\lambda }_iv_i}\right),v_i>0& \left(20\right)\end{array}$

where I₀(•) is the zeroth-order modified Bessel function of the first kind.

To compute the misdetection probability, one must first obtain p(v_i|H₁), as acquiring perfect knowledge of each λ_i is typically infeasible except for constant-modulus primary signals. Alternatively, one can obtain p(v_i|H₁) by applying the Bayesian approach to average over all the possible λ_i. This approach requires knowledge of the exact statistical distribution of the amplitude square of the primary signals, |s_i|². Obtaining such knowledge requires cooperation between the primary and second users. Like energy detection, the SSCT can obviate such a requirement due to the following properties:

(1) For a sufficiently large N, the statistical distribution of Λ_N depends on the mean of λ_i, i=1, . . . , N, irrespective of a specific choice of λ₁, . . . , λ_N. Define ρ_N:=b_N for N∈N₁^M−1 and ρ_M:= γ_M. Let A_N^lN denote the event that a_i<ξ_i<b_i, i∈N₁^lN for some integer l_N∈N₁^N l_N∈N₁^N and let Ã_N^lN denote its counterpart for the constant-modulus case. Let B_N^lN denote the event that ξ_N≧ρ_N, and a_i<ξ_i<b_i, i∈N_lN+1^N, and let {tilde over (B)}_N^lN denote its counterpart in the constant modulus case.(2) Let ε be an arbitrary positive number. If for each N there exists a positive integer l_N∈N₁^N such that

$P_{H_1}\left(A_N^{l_N}\right)\ge 1-\frac{\varepsilon }{3M},P_{H_1}\left({\tilde{A}}_N^{l_N}\right)\ge 1-\frac{\varepsilon }{3M},\mathrm{and}$ $P_{H_1}\left(B_N^{l_N}\right)-P_{H_1}\left({\tilde{B}}_N^{l_N}\right)<\frac{\varepsilon }{3M},\mathrm{then}P_{\mathrm{MD},M}-{\tilde{P}}_{\mathrm{MD},M}\le \varepsilon ,$

where l_N depends on the values of N and ε, and {tilde over (P)}_MD,M denotes the miss-detection probability obtained by assuming constant modulus signals (i.e., when all λ_i are equal).

With these properties, it is reasonable to assume that all λ_i are equal to λ by allowing negligible errors when M is not sufficiently large.

In this case, one can employ an efficient computational method to recursively compute P_MD,M. Defining u_i=v_i− Δ, Λ_N is rewritten as Λ_N=Σ_i=1^Nu_i. Clearly, the PDF of u_i under H₁ may be rewritten as

$p\left(v_i|H_1\right)=\frac{1}{2}{}^{-\left(u_i+\stackrel{\_}{\Delta }+{\lambda }_i\right)}I_0\left(\sqrt{{\lambda }_i\left(u_i+\stackrel{\_}{\Delta }\right)}\right),u_i>-\stackrel{\_}{\Delta }$

Recall that M is the maximum number of samples to observe. Denote Λ_M−k by t_k. Let G_k(t_k) denote the misdetection probability of the SSCT conditioning on that the first (M−k) samples have been observed, the present value t_k= Λ_M−k, and the test statistics have not crossed either boundary in the previous (M−k−1) samples. If ā<t_k< b, an additional sample (the (M−k+1)th sample) is needed. Let u be the next observed value of u_i. The conditional probability G_k(t_k|u) can be readily obtained as

$\begin{array}{cc}{G}_k\left(t_k|u\right)=\{\begin{array}{cc}0& \mathrm{if}u>\stackrel{\_}{b}-t_k\\ 1& \mathrm{if}u<\stackrel{\_}{a}-t_k\\ G_{k-1}\left(t_k+u\right)& \mathrm{if}\stackrel{\_}{a}-t_k<u<\stackrel{\_}{b}-t_k\end{array}& \left(21\right)\end{array}$

Using (21), one can recursively compute G_k(t_k) as

G _k(t_k)=∫_−∞^ā−tkp_H1(u)du+∫ā−t_k^b−tkG_k−1(t_k+u)p_H1(u)du,  (22)

for k=1, . . . , M with the following initial condition:

G ₀(t₀)=0 if t₀≧ γ; G₀(t₀)=1, otherwise.  (23)

Employing the above recursive process, one can obtain G_M(0), which is equal to the misdetection probability, P_MD,M.

Another important quantity in the SSCT is the average sample number (ASN). The number of samples needed to yield a decision is an RV, denoted by N_s. The ASN can be written as

E(N_s)=E_H0(N_s)P_H0+E_H1(N_s)P_H1  (24)

where E_H1(N_s) denotes the ASN conditioned on H_i, and P(H_i) denotes the probability of hypothesis H_i for i=0,1. Since 1≦N_s≦M, one can express E_Hi(N_s) as

$\begin{array}{cc}{E}_{H_i}\left(N_s\right)=\sum _{N=1}^M{\mathrm{NP}}_{H_i}\left(N_s=N\right),i=0,1,& \left(25\right)\end{array}$

where P_Hi(N_s=N) is the conditional probability that the test ends at the N th sample under H_i. Equations (9)-(11) imply that P_Hi(N_s=N) can be obtained as

P _{H i} (N_s=N)^(a)=P_Hi((ξ₁, . . . , ξ_N−1)∈Y_{aN−1,bN−1}^(N−1))−P_Hi((ξ₁, . . . , ξ_N)∈Y_aN,bN^(N)), N∈N₁^M−1  (26)

P _{H i} (N_s=M)^(b)=P_Hi((ξ₁, . . . , ξ_M−1)∈Y_{aM−1,bM−1}^(M−1)),  (27)

where the two terms on the right-hand side of the equality 26 are the probabilities of the events that, under H_i, the test statistic does not cross either of two boundaries at or before samples N−1 and N for N∈N₁^M−1, respectively, and the term on the right-hand side of equality 27 denotes the probability that under H_i (the condition where a PU is using the channel), the test statistic does not cross either boundary at or before samples N−1.

These probabilities can be expressed for each of the hypotheses as

P _{H 0} (N_s=N)=2^−(N−1)J_{aN−1,bN−1}^(N−1)(0.5)−2^−NJ_aN,bN^(N)(0.5)

P _{H 0} (N_s=M)=2^−(M−1)J_{aM−1,bM−1}^(M−1)(0.5)

P _{H 1} (N_s=N)^(c)=P_H1((ξ₁, . . . , ξ_N)∉γ_aN,bN^(N)), −P_H1((ξ₁, . . . , ξ_N−1)∉γ_{aN−1,bN−1}^(N−1))  (28)

P _{H 1} (N_s=M)^(d)=1−P_H1((ξ₁, . . . , ξ_N)∉γ_aN,bN^(N))  (29)

where the two terms on the right-hand side of equation 28 are the probabilities of the events that, under H₁, the test statistic crosses either of the two boundaries at or before samples N and N−1, respectively, and the second term on the right-hand side of equation 29 is the probability that the test statistic crosses either of the two boundaries at or before sample M.

According to equation (23), G_k (t_k) also depends on γ. With a slight abuse of notation, G_k(t_k) is rewritten as G_k(t_k, γ). Let V_k denote the event that the test statistics cross the lower-boundary at or before sample k under H₁, and U_k denote the event that the test statistics do not cross the upper-boundary at or before sample k under H₁. It is not hard to see P_H1(V_k)=G_k(0,ā) and P_H1(U_k)=G_k(0, b). One can now obtain

P _{H 1} ((ξ₁, . . . , ξ_N)∉Y_aN,bN^(N))=G_N(0,ā)+1−G_N(0, b),  (31)

where G_N(t,ā) and G_N(t, b) can be recursively obtained by applying (22). After obtaining P_H0(N_s=N) and P_H1(N_s=N), one can readily compute E(N_s) from (24) and (25).

As can be seen from the above, the SSCT scheme provides advantages over the prior art in that: 1) the test statistic is simple; 2) it does not need deterministic knowledge about the primary signals; 3) it can substantially reduce sensing time while maintaining a comparable detection performance as compared with energy detection; 4) and it offers desirable flexibility in striking the trade-off between detection performance and sensing time when SNR_a mismatches with SNR_m.

Referring now to FIG. 2, a method is shown for determining whether a given spectrum is occupied. A received signal is sampled at block 202. This produces a value representing the strength of the signal at the time of sampling. The sample is then squared at block 204. This squared sample value is used to update the test statistic at block 206. As noted above, an exemplary test statistic according to the present principles is

${\Lambda }_N=\sum _{i=1}^N\left({r_i}^2-\Delta \right),$

where Δ is a constant and N represents the number of samples collected so far. The test statistic is then evaluated at block 208 to determine whether it exceeds the upper-boundary b or falls below the lower-boundary a, such that a determination is made regarding whether the spectrum is occupied. If a threshold is exceeded, the procedure ends with the information that the SU may or may not use the spectrum. Otherwise, the number of samples received is determined at block 210. If no decision has been made when that number reaches a maximum number of samples (described above as the quantity M), then the sensing stops and makes a decision by comparing the final test statistic Λ_M with γ.

Referring now to FIG. 3, a system is shown to make determinations regarding whether a spectrum is occupied. An incoming signal reaches filter 302, which removes out-of-band noise from the signal. An analog-to-digital converter (ADC) 304 then samples the signal, converting it from a continuous-time waveform to discrete-time signals. These discrete-time signals then pass through a squaring module 306, which outputs the square of the magnitude of its input. The squared samples pass through an update module 308, which updates the test statistic as described above. The update module 308 outputs the test statistic, which is then used by test module 310. The test module 310 determines whether the test statistic has exceeded a given threshold. The test module 310 also determines whether the number of samples used has reached the maximum allowable sample size M. If a threshold has been exceeded, the test module sends a signal to filter 302 indicating that the filter 302 should stop sensing the incoming signal. The test module then outputs the end result. The thresholds may be set and adjusted according to the procedure set forth below prior to beginning detection at threshold adjustment module 312.

The present principles involve the use of several thresholds. If the test statistic is below a threshold “a”, then the present principles arrive at a determination that the spectrum is unoccupied. If the test statistic is above a second threshold “b”, the present principles determine that the spectrum is occupied. A third threshold “γ” may be used to provide a determination for a final test statistic. A fourth threshold, “M,” is selected as the maximum allowable sensing time.

The thresholds “a,” “b,” and “γ” can be determined based on system design specifications. The values of these thresholds determine the probabilities for false-alarm and for misdetection. Referring now to FIG. 4, a technique for selecting the thresholds is shown. Block 402 begins by making initial guesses for the thresholds. The thresholds should obey a<0, b>0, and a≦γ≦b. Block 404 computes the false alarm and misdetection probabilities using, for example, the formulations described above. Block 406 determines whether the probabilities meet design specifications. If the obtained probabilities are much smaller than a set of target probabilities, the thresholds should be adjusted to improve the sensing time. On the other hand, if the obtained probabilities are larger than a set of target probabilities, the thresholds need to be adjusted to ensure that the target probability is satisfied. If the design specifications are not met, block 408 adjusts the thresholds and returns to block 404. If the design specifications are met, block 406 terminates and outputs the thresholds.

The thresholds may be adjusted on a trial-and-error basis. After a set of thresholds has been generated and the false-alarm and misdetection probabilities have been calculated, if the misdetection probability is larger than design specifications, the value of threshold a may be decreased, and vice versa. If the false alarm probability is larger than the design specifications, the value of threshold b may be increased. The difference between the two thresholds (b−a) is also considered.

In the above described sensing scheme, the test statistics Λ_N is compared with two predetermined thresholds a and b every received signal sample. This may be difficult or even infeasible in practice especially when the SNR is low and/or the sampling rate is high. To overcome this shortcoming, we propose an extension of the SSCT, simply called the block-wise SSCT (B-SSCT).

In the B-SSCT, the received signal samples are first parsed into a block of length L and threshold comparisons are performed at the end of each block, as showed in FIG. 6. The test static at the pth block is computed as follows:

Ξ_q=Σ_i=1^q×L(|r_i|²−Δ)  (32)

where Δ is a predetermined constant. Similar to ones described for equations (4)-(6) above, the test procedure is described as follows:

Reject H₀:

if Ξ_q≧b_B and q≦Q−1 or if Ξ_Q≧c_B;  (33)

Accept H₀:

if Ξq<a_B and q≦Q−1 or if Ξ_Q<c_B;  (34)

Continue Sensing:

if Ξ_q∈(a_B,b_B) and q<Q.  (35)

where a_B, b_B and c_B are three predetermined thresholds with a_B<0, b_B>0, and a_B<c_B<b_B, and Q is the truncated block number of the test. It is clear from (32) that the truncate sample number of the B-SSCT is Q×L, and the threshold comparison is only performed every L samples.

The false-alarm probability denoted by P_FA and the miss-detection probability denoted by P_MD can be evaluated by using a similar procedure described above for SSCT. In the following, we briefly describe the procedure to evaluate false-alarm and miss-detection probabilities, which illustrates the differences between the evaluation procedure for the B-SSCT and the one described above for SSCT. Since evaluating P_FA and P_MD follows the same procedure, we only show how to compute P_MD in detail.

Defining u_s=τ_l=1^L(|r_(s−1)L+l|²−Δ), we can rewrite (32) as

${\Xi }_q=\sum _{s=1}^qu_s$

Based on the central limit theorem (CLT), the distributions of u_s under hypotheses H₀ and H₁ can be approximated as

$\hspace{1em}\{\begin{array}{c}{u}_s|H_0\sim N\left(L\left({\sigma }_w^2-\Delta \right),L{\sigma }_w^4\right)\\ u_s|H_1\sim N\left(L\left(\left(1+{\mathrm{SNR}}_m\right){\sigma }_w^2-\Delta \right),L\left(1+2{\mathrm{SNR}}_m\right){\sigma }_w^4\right)\end{array}$

Hence, we can write the PDFs of u_s under H₀ can be written as

$p_{H_0}\left(u_s\right)=\frac{1}{\sqrt{2\pi L{\sigma }_w^4}}\exp \left(-\frac{{\left(u_s-L\left({\sigma }_w^2-\Delta \right)\right)}^2}{2L{\sigma }_w^2}\right)$

and the PDF of u_s under H₁ can be written as

$p_{H_1}\left(u_s\right)=\frac{{\left(1+2{\mathrm{SNR}}_m\right)}^{-1/2}}{\sqrt{2\pi L{\sigma }_w^4}}\exp \left(-\frac{{\left(u_s-L\left(\left(1+{\mathrm{SNR}}_m\right){\sigma }_w^2-\Delta \right)\right)}^2}{2\left(1+2{\mathrm{SNR}}_m\right)L{\sigma }_w^2}\right)$

To compute P_MD, we make corresponding changes on upper- and lower-limits and replace P_H1(u) with P_H1(u_s) in (21)-(23). Similarly, P_FA can be obtained by replacing P_H1(u) with P_H0(u_s) and making certain changes in (21)-(23).

The thresholds a_B, b_B and c_B may be adjusted on a trial-and-error basis, as described above.

The ASN of the B-SSCT can be evaluated by using the similar approaches described above for SSCT.

Having described preferred embodiments of a system and method (which are intended to be illustrative and not limiting), it is noted that modifications and variations can be made by persons skilled in the art in light of the above teachings. It is therefore to be understood that changes may be made in the particular embodiments disclosed which are within the scope of the invention as outlined by the appended claims. Having thus described aspects of the invention, with the details and particularity required by the patent laws, what is claimed and desired protected by Letters Patent is set forth in the appended claims.

Claims

1. A method for determining whether a given electromagnetic frequency is in use, comprising the steps of: applying a transformation to an amplitude of received samples;adjusting the transformed samples by a constant based on a minimum detection signal-to-noise ratio;combining the adjusted samples to produce a test statistic; andusing a processor to make a determination regarding if the frequency is in use based on the test statistic exceeded or falling below a threshold, said test statistic being based on Ξq=Σi=1q×L(|ri|2−Δ), where q is the block index, ri is the ith received sample, and Δ is the constant.

2. The method of claim 1, further comprising the steps of: adjusting thresholds to meet design specifications; andcalculating false alarm and misdetection probabilities to determine whether a particular set of thresholds meets design specifications, said false-alarm and miss-detection probabilities being calculated recursively.

3. A system for determining whether a given electromagnetic frequency is in use, comprising: a transformation module configured to transform an amplitude of received samples;an adjustment module configured to adjust the transformed samples by a constant based on a minimum detection signal-to-noise ratio and to combine the adjusted samples to produce a test statistic; anda test module configured for making a determination using a processor as to whether the frequency is in use based on the test statistic exceeding or falling below a threshold, said test statistic being based on Ξq=Σi=1q×L(|ri|2−Δ), where q is the block index, ri is the ith received sample, and Δ is the constant.