Flexible Full-duplex Cognitive Radio Networks by Antenna Reconfiguration

Transcription

1 IEEE/CIC ICCC Symposim on Wireless Commnications Systems Flexible Fll-dplex Cognitive Radio Networks by Antenna Reconfigration Liwei Song Yn Liao and Lingyang Song State Key Laboratory of Advanced Optical Commnication Systems and Networks School of Electrical Engineering and Compter Science Peking University Beijing China. Abstract Fll-dplex radio has been recently proposed to enhance spectrm tilization of secondary networks by a listenand-talk (LA protocol. It allows one antenna of the secondary ser (SU to sense the spectrm band while the other antenna simltaneosly to perform data transmission. De to wireless channel characteristics in this paper we introdce antenna mode selection to the LA protocol to achieve additional spatial diversity. Specifically each SU adaptively selects one antenna for sensing while the other for transmission according to their channel information. wo selection schemes are given: the optimal antenna mode selection maximizes the secondary throghpt performance by comparing the exact secondary throghpt of all antenna modes and a sboptimal method selects the antenna with higher sensing signal to noise ratio (SNR to transmitting SNR ratio for sensing named ratio antenna mode selection (R-AMS. Simlation reslts indicate the throghpt improvement cased by additional spatial diversity with antenna mode selection and the effectiveness of R-AMS. I. INRODUCION Cognitive radio is a promising technology to solve spectrm shortage problem by realizing rese of allocated spectrm when it is not sed by licensed sers. In cognitive radio networks (CRNs secondary sers (SUs are permitted to se the allocated spectrm of primary sers (PUs only if there exist spectrm holes [] []. raditional protocol sed in CRNs is listen-before-talk (LB protocol which is based on half-dplex technology and has been sfficiently analyzed in [3]. In LB protocol SUs have to sense the spectrm for a while and then determine transmission or not according to the sensing reslts. hs it sffers from the problems of shorter transmission time and discontinos transmission for SUs. o resolve the problems faced with LB protocol the listen-and-talk (LA protocol sing fll-dplex (FD technology has been proposed recently [4] []. With contrast to LB protocol in LA protocol each SU is eqipped with two antennas to perform sensing and transmission processes simltaneosly. o realize that SU ses one antenna to do spectrm sensing and the other to transmit or keep silent according to the sensing reslts. However the severe selfinterference from transmitting antenna to sensing antenna is the major challenge for the LA protocol. Eqipped with mltiple antennas SU can choose either antenna for sensing while the other for transmission ths there exist two antenna modes for sensing and transmission processes. However the previos works on LA protocol [4] [] assme the roles of antennas are fixed. In this paper we consider flexible fll-dplex CRNs by antenna reconfigration and introdce antenna mode selection (AMS to the LA protocol which is shown an effective way to achieve diversity gain and improve performance. [6] and [7] show the benefits of AMS to CRNs in nderlay condition where SU and PU are allowed to transmit simltaneosly nder the constraint of interference from SU to PU. While in this paper we analyze AMS in overlay CRN where SU is allowed to se spectrm only when PU is idle. o exploit the spatial diversity SU selects either antenna mode according to the channel state information (CSI of two antennas mainly abot signal to noise ratio (SNR. We simply consider one SU pair in this paper. Or analysis can be easily extended to the scenario with mltiple sers and different channel characteristics. We first discss the optimal antenna mode selection (O- AMS by maximizing throghpt. However it is complex to calclate exact throghpt and to perform. hs in this paper we also propose ratio antenna mode selection (R-AMS as a sboptimal antenna mode selection which selects sensing antenna with higher sensing SNR (SSNR to transmitting SNR (SNR ratio. he R-AMS is proved to redce comptational complexity greatly while still get enogh throghpt improvement. he rest of the paper is organized as follows. Section II describes the system model and the LA protocol. We stdy selection schemes of O-AMS and R-AMS in section III. Section IV derives the analytical throghpt withot AMS and with O-AMS and R-AMS. Simlation reslts are given to show the throghpt improvement introdced by additional spatial diversity with AMS and the effectiveness of R-AMS in Section V. Finally we conclde this paper in Section VI. II. PRELIMINARY A. System Model We consider a CRN consisting of one PU and one FDenabled SU pair denoted by SU and SU respectively. Each SU is eqipped with two antennas (Ant and Ant and SU needs to transmit data to SU. By the LA protocol SU can se one antenna to perform spectrm sensing while the other antenna to perform data transmission to SU simltaneosly when a spectrm hole is detected. Different from the conventional LB protocol in which all the antennas eqipped on one SU are sed for the same prpose i.e. spectrm sensing or data transmission in the LA protocol the two antennas are sed asymmetrically. herefore compared to the conventional LB protocol there exists additional spatial diversity which can be explored by antenna mode selection. he AMS strategies are shown in //$3. IEEE

2 IEEE/CIC ICCC Symposim on Wireless Commnications Systems Fig.. here exists two antenna modes: Ant for sensing while Ant for transmission and Ant for sensing while Ant for transmission. By antenna reconfigration SU can adaptively select either mode according to CSI of its two antennas. Fig.. wo Antenna Modes for Antenna Mode Selection B. LA Protocol In the LA protocol the secondary transmitter SU is allowed to sense the spectrm and transmit data to SU simltaneosly with different antennas. For the sensing process there are two hypothesises abot PU s state when SU is transmitting or not. And we denote by H XY those hypothesises where X Y { } represent the activity of SU and PU respectively. For example the case when PU is absent while SU is transmitting is referred to as H. As shown in Fig. the sensing signal at sensing antenna (Sx can be written as h Pi s P + h I s + H h I s + H y = ( h Pi s P + H H where h Pi (i { } is the sensing channel from PU to Sx and h I is the self-interference channel from transmitting antenna (x to Sx s P and s denote signal of PU and transmitting signal of SU with variance eqal to P respectively and CN ( is the noise. Frthermore we adopt Rayleigh fading channel in or model i.e. h Pi CN ( hi and according to [8] h I s can also be modeled as a Rayleigh distribtion with zero mean and variance χ where χ denotes the self-interference sppression (SIS factor. Energy detection is adopted as the sensing strategy and the process in or model is time-slotted. hen the test statistics can be given by M = N s y(n ( N s n= where y(n denotes the n th sample of sensing signal and N s = f s is the sampling nmber in one slot. SU determines PU s state by comparing the test statistics M with a certain threshold. If M is below the threshold PU is jdged to be absent and x can transmit data otherwise the spectrm is thoght to be bsy and x keeps silent. By ( the sensing signal is different according to SU s activeness and the impact of self-interference only exists in the sensing signal when SU is transmitting. hs SU needs to adaptively change the sensing threshold according to its own activeness to achieve better sensing performance [4]. III. ANENNA MODE SELECION ALGORIHMS In the conventional LA protocol the fnctions of two antennas are fixed e.g. Ant is sed for sensing while Ant for transmission. he neqivalent se of antennas provides the potential to frther exploring the spatial diversity by introdcing proper antenna mode selection schemes to the LA protocol. he antenna mode selection is shown in Fig. : SU can choose either Ant or Ant for sensing according to their wireless channel characteristics and correspondingly the other antenna for transmission. In this section we propose or antenna mode selection schemes. We first present the O-AMS that achieves the maximm throghpt. hen to redce the comptational complexity of O-AMS we propose the R-AMS as a sboptimal antenna mode selection. A. Optimal Antenna Mode Selection he O-AMS compares exact throghpt of those two antenna modes in Fig. and selects the mode with higher throghpt. Let C i (i = denote the throghpt of the antenna mode that Ant i is sed as the Sx while Ant i (i =3 i is sed as x which can be given by C i = R i ( P fi (3 where R i is the channel rate from Ant i of SU to SU and P fi is the false alarm probability for the sensing performance of SU. he concrete expressions of R i and P fi will be given in Section IV. o maximize the throghpt the O-AMS scheme for the Sx i O can be expressed by { when C >C i O = (4 when C <C where Ant io is selected as Sx while Ant io (i O =3 i O is sed as x.

3 IEEE/CIC ICCC Symposim on Wireless Commnications Systems 3 B. Ratio Antenna Mode Selection According to (4 the O-AMS always selects antenna mode with higher throghpt. here are two CSI sitations for AMS. he first sitation is when one antenna say Ant has higher SSNR while the other antenna Ant has higher SNR. hen i O =and SU can get both better sensing performance and better transmitting performance simltaneosly. he second is when one antenna has both higher SSNR and higher SNR in this sitation SU has to compare exact throghpt of two antenna modes and then make an option between better sensing performance and better transmitting performance. o calclate exact throghpt according to ( and (3 SU has to acqire not only SSNR and SNR of its two antennas bt also the interference to noise ratio (INR from x to Sx which is complex. hs it is meaningfl to propose R- AMS as a sboptimal antenna mode selection to redce the comptational complexity. Different from the O-AMS he R-AMS always selects antenna with higher SSNR to SNR ratio for sensing while the other antenna for transmission. We denote by i i the SSNR and the SNR of Ant i respectively. hen the selection scheme of R-AMS can be expressed as when > i R = S when < ( S where SU ses Ant ir for sensing and Ant ir (i R =3 i R for transmission. We also consider the R-AMS in the two CSI sitations. When Ant has higher SSNR while Ant has higher SNR > soi R =and the R-AMS performs the same selection reslt as O-AMS in this sitation. For the latter sitation R-AMS does not need to calclate exact throghpt instead SU only needs to know the SSNR to SNR ratio of its two antennas and chooses the antenna with higher ratio for sensing. herefore the comptational complexity of the R-AMS is greatly redced compared to the O-AMS. IV. PERFORMANCE ANALYSIS In this section we first derive the analytical throghpt performance withot AMS (No-AMS where the fnctions of two antennas on SU are fixed. hen we analyze throghpt performance of O-AMS and R-AMS. he throghpt performance improvement of O-AMS and R-AMS compared to No- AMS is also provided to show the additional spatial diversity with AMS. A. hroghpt withot Antenna Mode Selection Withot AMS the roles of two antennas on SU are fixed ths the throghpt is C N = C i where i = or is fixed. Since the channel in or model is considered to be Rayleigh fading the channel gain is exponentially distribted then SSNR and SNR are also exponential random variables. We first derive the analytical throghpt for certain SSNR and SNR and then analyze the No-AMS throghpt performance according to the probability density fnctions (PDFs of SSNR and SNR. By establishing discrete-time Markov chains to model changes among H XY and approximating test statistics M by a Gassian variable when N s is large enogh [4] gives relationship between a certain miss detection probability (P m and false alarm probability (P fi i = for the sensing performance P fi P fi = Pfi + P fi (6 + I + Si + Ns I ( ( where Pfi = Q Q ( P m + i and Pfi = Q ( Q ( P m (+i +i Ns represent the corresponding false alarm probability when x is transmitting or not respectively in which i = hi P is the SSNR of Ant i and I = χ is the INR from x to Sx. As for the transmission process the received signal at SU can be expressed as [4] r = h Si s + (7 where h Si CN ( hi denotes the transmitting channel from x to SU. hen the channel rate of x can be expressed as R i =log ( + i (8 where i = hi denotes the SNR of Ant i. By sbstitting (6 and (8 into (3 we can get the exact throghpt for certain SSNR and SNR. With (6 and (8 we can find that the channel rate is positively related to the SNR and the false alarm probability is negatively related to the SSNR. hs higher SSNR or SNR leads to better sensing performance or better transmitting performance which is consistent with analysis of selection schemes in Section III. We can also find that the calclation of exact throghpt is complicated especially for the false alarm probability. Now we derive the No-AMS throghpt performance. For simplicity and withot loss of generality we assme the SSNR of two antennas on SU are independent identically distribted (IID with the mean = h P where h denotes the mean of hi. he SNR of two antennas are also assmed to be IID with the mean = h where h is the mean of hi. he cmlative distribtion fnction (CDF of C i can be derived as ( x F Ci (x = e ( i S e ( P fi di. (9 hen we get the analytical No-AMS throghpt performance C N = F Ci (xdx. ( When SSNR is high enogh the sensing performance is nearly perfect. With P fi we can get the asymptotic vale of No-AMS throghpt lim C N = ( ln e E (

4 IEEE/CIC ICCC Symposim on Wireless Commnications Systems 4 where E (x = e t x t dt is the exponential integral fnction of the first order. B. hroghpt of Optimal Antenna Mode Selection As shown in (4 the O-AMS always selects antenna mode with higher throghpt. Becase the SSNR and the SNR of two antennas are IID C i are also IID. hs the CDF of O-AMS throghpt is FC i (x and the analytical throghpt performance of O-AMS can be given by C O = F C i (xdx. ( Frthermore we can get throghpt improvement cased by additional spatial diversity with O-AMS ΔC O = C O C N = F Ci (x( F Ci (x dx. (3 With enogh high SSNR we can also get the asymptotic vales for throghpt performance and improvement with O- AMS scheme lim C O = ( ln e E lim ΔC O = ( ln e E ( ln e E ( ln e E. (4 C. hroghpt of Ratio Antenna Mode Selection We now analyze throghpt performance of R-AMS. Since it is hard to get an analytical reslt by analyzing throghpt directly we get an approximation of R-AMS throghpt by assming the sensing process and the transmission process are independent. We will find that or approximation fits well in the simlation reslts. In order to calclate the average channel rate (R R and the average false alarm probability (P fr we have to derive the PDF of SNR and SSNR selected by R- AMS scheme named by R and R. Let P (A denote the occrrence probability of event A we can get the CDF of R x x + F R (x = e d d x + + x e P ( < d d x + + x e P ( < d d ( where the first integration in the right is the probability that both SNR are less than x the second integral is the probability that <x >xand we choose as SNR since < and the third integral is similar to the second. Solving these integrals we get F R (x = ( x +e x + x E ( x. (6 Similarly we can get CDF of SSNR selected by R-AMS scheme F SR (x = ( x +e x + x E ( x. (7 herefore the PDF of R and R can be derived as the following f R (x = x E ( x f SR (x = x E ( x. S (8 Frthermore the average channel rate by R-AMS scheme can be expressed as R R = log ( + x x E ( x dx = ( ln +( e E ( (9 + ( ln ( + e x x d ln x where the integral ( in the last eqation is so small to be omitted i.e. R R ln +( e E (. And the average false alarm probability of R-AMS can be expressed as P fr = P f (x x E ( x dx S ( g (( + I a g (a+g(( + I a where a = (Q ( P and g(x denotes a fnction m+ N s of x: g(x = P m ( bx x e x(b+x Q(b + x in which b = Q ( P m. he approximations of P fr in ( fits well when INR ( I is below or close to the mean of SSNR (. Finally we have the analytical throghpt performance of R- AMS and its throghpt improvement by exploring additional spatial diversity C R = R R ( P fr ( ΔC R = C R C N. We can also get the asymptotic vales of throghpt and improvement with R-AMS scheme in enogh high SSNR condition lim C R = ( ln +( e E ( lim ΔC R = ( ln e E ( (. V. SIMULAION RESULS In this section to present the throghpt improvement by exploring additional spatial diversity with AMS we provide analytical and simlated throghpt performance of No-AMS O-AMS and R-AMS. We also compare the throghpt of R- AMS with that of O-AMS to prove its effectiveness. he simlation parameters are set as the following. For the sampling nmber the slot dration is.ms and the freqency f s is MHz ths N s =. For the signal power the transmission power of SU and the noise power are set as mw and mw respectively. For the channel information the average

5 IEEE/CIC ICCC Symposim on Wireless Commnications Systems hroghpt C Fig ĥ =4 ĥ =4 No AMS O AMS R AMS SIS factor χ hroghpt verss SIS factor χ hroghpt C Fig No-AMS : χ =.3 No-AMS : χ =.6 No-AMS : Asymptotic O-AMS : χ =.3 O-AMS : χ =.6 O-AMS : Asymptotic R-AMS : χ =.3 R-AMS : χ =.6 R-AMS : Asymptotic Sensing signal power p hroghpt verss sensing signal power P variance of sensing channel h is set to be. Finally the miss detection probability P m is fixed to be.. In Fig. we show throghpt performance verss SIS factor χ with P =.mw. Fig. incldes both analytical reslts (the different lines and simlated reslts (varios types of dots. here are two sets of crves with average variance of transmitting channel h =4 4 respectively. Higher h means higher SNR ths throghpt with h =4is better than that with h = 4. With higher χ self-interference is more severe ths the throghpt decreases with SIS factor. Fig. shows that throghpt performance get greatly improved with O-AMS or R-AMS scheme compared to No-AMS. We can also find that throghpt of R-AMS is fairly close to that of O-AMS: from Fig. the throghpt crves of both are almost overlapping when SIS factor is greater than.4. herefore the R-AMS is proved to be an effective sboptimal selection scheme which not only decreases the comptational complexity bt also gets adeqate throghpt improvement. he fact that the analytical and the simlated reslts of R- AMS fit well shows the accracy of or approximation of C R in Section IV. In Fig. 3 we set average variance of transmitting channel h =4and present throghpt performance verss sensing signal power P. Also two sets of crves are given with χ =.3.6 respectively. he lines denote analytical reslts while the dots are simlated reslts. Frthermore by ( (4 and ( we give corresponding asymptotic throghpt vales of No-AMS O-AMS and R-AMS (horizontal lines. Fig. 3 shows the crves with χ =.6 is lower than corresponding crves with χ =.3 which is consistent with Fig.. Also we can see the throghpt improvement introdced by additional spatial diversity with AMS. Since SSNR is positively proportional to P throghpt increases with the increment of sensing signal power and the gap between two sets decreases to none gradally. When P is large enogh the sensing reslt is considerably reliable and ths two sets of crves reach same corresponding asymptotic vales. VI. CONCLUSIONS In this paper we introdced antenna mode selection into FD cognitive radio networks by antenna reconfigration and proposed R-AMS as a sboptimal antenna mode selection to have lower comptational complexity than the O-AMS and still get fairly throghpt improvement. It was shown by analysis and simlation reslts O-AMS and R-AMS both can get throghpt performance improved by exploring the additional spatial diversity. And we showed that the throghpt of R-AMS is considerably close to that of O-AMS to prove the effectiveness of R-AMS. ACKNOWLEDGEMEN his work was spported in part by the National 973 Project nder Grant 3CB3367 by the National Natral Science Fondation of China nder Grants 64 and U3 and by National Science and echnology Major Project nder Grant 3ZX333. REFERENCES [] J. Mitola and G. Q. Magire Cognitive Radio: Making Software Radios more Personal IEEE Personal Commnications vol. 6 no. 4 pp. 3-8 Ag [] J. Mitola Cognitive Radio-An Integrated Agent Architectre for Software Defined Radio Ph.D. hesis Royal Institte of echnology Sweden May.. [3]. Ycek and H. Arslan A Srvey of Spectrm Sensing Algorithms for Cognitive Radio Applications IEEE Commnications Srveys & torials vol. no. pp. 6-3 Mar. 9. [4] Y. Liao. Wang L. Song and Z. Han Listen-and-alk: Fll-dplex Cognitive Radio Networks in IEEE Global Commnications Conference pp Astin X Dec. 4. [] Y. Liao L. Song Z. Han and Y. Li Fll Dplex Cognitive Radio: A New Design Paradigm for Enhancing Spectrm Usage IEEE Commnication Magazine vol. 3 no. pp May.. [6] J. Zho and J. hompson Single-Antenna Selection for MISO Cognitive Radio in IE Seminar on Cognitive Radio and Software Defined Radios pp. - London Sept. 8. [7] Y. Wang and J. P. Coon Difference Antenna Selection and Power Allocation for Wireless Cognitive Systems IEEE ransactions on Commnications vol. 9 no. pp Dec.. [8] W. Afifi and M. Krnz Exploiting Self-Interference Sppression for Improved Spectrm Awareness/Efficiency in Cognitive Radio Systems in IEEE INFOCOM pp rin Apr. 3.