An Anlog Bsend Approch for Designing Full-Duplex Rdios Brett Kufmn, Jorm Lilleerg, nd Behnm Azhng Center for Multimedi Communiction, Rice University, Houston, Texs, USA Centre for Wireless Communictions, University of Oulu, Oulu, Finlnd Reness Moile, Oulu, Finlnd Insted of n ctive or pssive leling for the selfinterference cncelltion technique, we cn clssify the cncelltion technique sed on which point long the trnsceiver chin does the cncelltion occur. All of the ove listed techniques re implemented in the nlog RF stge nd re primrily done so in order to not sturte the low noise mplifier. An dditionl reson for focusing on the nlog RF stge of the trnsceiver is the reltive ese in which circuit modifictions nd dditions cn e connected to the existing rdio design. Work in [8] nd [9] demonstrte ctive circuitry tht cn connect to the trnsmitter nd receiver chins respectively. It is due to the ove two resons why the nlog send stge of the trnsceiver hs een lrgely ignored for selfinterference cncelltion. Providing n dditionl stge of nlog cncelltion just efore the nlog-to-digitl converter (ADC) would increse the dynmic rnge of the ADC nd thus provide etter resolution of the desired signl over the self-interference signl. The work in this pper will show tht dding complementry nlog send self-interference cncelltion stge to nlog RF self-interference cncelltion cn significntly improve the totl cncelltion chieved nd help close the gp etween the experimentl implementtions nd the theoreticl expecttions. The reminder of this pper is orgnized s follows. In Section II we define the trnsceiver model nd signl model. In Section III, we chrcterize the self-interference chnnel. Then in Section IV we provide detils of the proposed cncelltion scheme nd quntify its performnce. We then evlute tworxiv:1312.2466v1 [cs.ni] 9 Dec 213 Astrct Recent wireless tested implementtions hve proven tht full-duplex communiction is in fct possile nd cn outperform hlf-duplex systems. Mny of these implementtions modify existing hlf-duplex systems to operte in full-duplex. To relize the full potentil of full-duplex, rdios need to e designed with self-interference in mind. In our work, we use n experimentl setup with ptch ntenn prototype to chrcterize the self-interference chnnel etween two rdios. In doing so, we form n nlyticl model to design nlog send cncelltion techniques. We show tht our cncelltion scheme cn provide up to 1 db improved signl strength, 2.5 ps/hz increse in rte, nd 1 4 improvement in BER s compred to the RF only cncelltion provided y the ptch ntenn. I. INTRODUCTION Wireless full-duplex communiction in which terminl cn simultneously trnsmit nd receive in the sme frequency nd ws first demonstrted in rdr systems [1] s erly s the 194 s. Then in the 198 s, cellulr networks utilized full-duplex in repeters [2] to extend cellulr coverge. Not until recently in 21 ws idirectionl point-to-point fullduplex link, shown in Fig. 1, demonstrted with experimentl testeds [3], [4]. However, insufficient levels of selfinterference cncelltion prevented the expected douling of spectrl efficiency s compred to hlf-duplex communictions from eing chieved. Current self-interference cncelltion techniques cn e clssified into two min techniques: Pssive Suppression nd Active Cncelltion. Pssive techniques ttempt to increse the isoltion etween the trnsmit nd receive ntenns nd re gnostic to the signl chrcteristics of the self-interference signl. A comintion of directionl isoltion, sorptive shielding, nd cross-polriztion in the trnsmit nd receive ntenns ws used in [5]. Another experimentl setup [6] used multiple trnsmit ntenns to crete null point t the receive ntenn. A novel ntenn design in [7] isoltes the trnsmit nd receive strems with two dul-polrized ptch ntenns. Active techniques enle terminl to use the knowledge of it s own self-interference signl to generte cncelltion signl tht cn cn e sutrcted from the received signl. An experimentl setup using the WARP pltform [8] used n extr trnsmit chin to generte n up-converted RF cncelltion signl tht ws then sutrcted from the incoming signl t This work is funded in prt y NSF grnt, y Reness through reserch contrct, nd the Acdemy of Finlnd through the Co-Op grnt. the receive ntenn. A recent work in [9] proposes ctive circuitry tht smples the RF self-interference signl nd uses sinc interpoltion to generte the cncelltion signl. Node Tx Rx Rx r RF r RF Fig. 1. Two-user idirectionl full-duplex link showing self-interference chnnels with dshed rrows nd dt chnnels with solid rrows. Tx Node
terminl full-duplex link in Section V nd then finish with concluding remrks in Section VI. II. SYSTEM MODEL We consider the two-terminl point-to-point full-duplex link shown in Fig. 1 where terminls nd re communicting with ech other using the sme temporl nd frequency resources. Ech terminl hs single trnsmit nd receive ntenn. We now refer to the functionl lock digrm in Fig. 2 s we derive the signl model. We note tht the lock digrm is from the perspective of terminl nd tht everything is identicl for terminl. p PT p e j2 f ct e j2 f ct z RF + LPF crrier frequency f c yielding = x e j2πfct. The signl is then mplified with signl power P T y power mplifier (). At the receiver side, fter down-conversion from f c nd low-pss filtering (LPF), the received send time domin signl r cn e expressed s r = P T h x + P T h x + z, (1) where x is the signl-of-interest trnsmitted over the wireless chnnel h nd x is the self-interference signl trnsmitted over the self-interference chnnel h. The received signl is corrupted y dditive white Gussin noise z CN (, σ 2 z). We note tht the received nlog send signl in (1) is the nlyticl equivlent for the received pssnd signl t crrier frequency f c. Just fter down-conversion nd low-pss filtering, n estimte of the self-interference signl x is dded to the received send signl giving y = r + x, (2) which in turn serves s input to the ADC. Output digitl smples y [k] from the ADC pss through receiver side SRRC yielding symols y which cn e finlly demodulted (DeMod). x p DAC SRRC x [k] h x r y y [k] + ADC SRRC III. SELF-INTERFERENCE MODEL We now provide detils out the RF self-interference chnnel h. We utilize four-lyer ptch ntenn prototype designed y [1] nd similr in design to [11]. The ptch ntenn isoltes the trnsmit nd receive ntenns from ech other in single form fctor. Becuse the isoltion is not perfect, n ttenuted version of the self-interference signl from terminl, the coupling signl, psses thru the ntenn from the trnsmitter side to the receiver side nd mixes with the incoming desired signl from terminl. x y 4 Mod. DeMod. Fig. 2. Functionl lock digrm of full-duplex trnsceiver from the perspective of terminl. At the trnsmitter side, the send signls of ndwidth BW re first modulted (Mod.) into M-K symols x. We ssume verge unit energy symols with E[ x 2 ] = 1 where E[ ] is used to denote the sttisticl expecttion. The symols re then pulse-shped using squre-root-risedcosine (SRRC) filter nd the output digitl smples x [k] serve s input to the Digitl-to-Anlog Converter (DAC). We ssume idel DACs such tht the output send time domin signl x (t) stisfies x [k] = x (t). As our focus will e on the nlog domin, we remove the time nottion t for simplifiction nd simply refer to x (t) s x. We will mintin this ssumption for ll other time domin signls. The nlog send signl x is then up-converted to the Fig. 3. H + RF (f) (db) 45 5 55 2.428 2.438 2.448 f (GHz) Isoltion mesurements for the full-duplex ptch ntenn prototype. The ntenn prototype ws tested inside n Anechoic chmer with n Agilent Network Anlyzer. A rel-time, over-their 2.4 GHz high frequency test signl ws used to mesure
oth the isoltion nd phse effects of the ptch ntenn. We denote the mesured isoltion of the ntenn in the pssnd y H+ RF (f), shown in Fig. 3, nd the phse of the ntenn y H+ RF (f), shown in Fig. 4. The ntenn is optimized for crrier frequency of f c = 2.438 GHz nd mesurements were mde over ndwidth B H = 2 MHz centered t tht frequency. Using those mesurements, we cn nlyticlly express the self-interference chnnel s { H+ RF H (f) = + RF (f) e j HRF + (f), f f c B H 2 (3), elsewhere which is the one-sided FFT of the pssnd chnnel. Using properties of the Fourier trnsform, we cn write H (f) = 1 2 HRF + (f + f c ), (4) which is the FFT of the equivlent send chnnel centered t Hz. The time domin representtion of the selfinterference chnnel cn finlly e written s fter tking the IFFT. Fig. 4. H + RF (f) (rds) π π h = F 1 {H (f)}, (5) 2.428 2.438 2.448 f (GHz) Phse mesurements for the full-duplex ptch ntenn prototype. IV. ANALOG BASEBAND CANCELLATION We now provide the detils for the self-interference cncelltion from the perspective of terminl. A. Chnnel Estimtion In order to form n effective cncelltion signl, the effects of the self-interference need to e estimted. Ech terminl will send trining symols while the other terminl remins silent. Thus using (1), the received signl t terminl during the trining phse is r,tr = P T h x,tr + z, (6) nd is chieved y terminl sending N tr trining symols with terminl silent. Then using Lest Squres chnnel estimtion, terminl cn form n estimte of the chnnel y ĥ = r,trx 1,tr = h + z x 1,tr, (7) where the estimte ĥ consists of the true chnnel corrupted y scled dditive noise. The chnnel estimte cn then e used to form the cncelltion signl x = P T ĥ x which will ttempt to cncel the self-interference signl x during the dt phse. B. Self-Interference Cncelltion Using the cncelltion signl just derived with (2), we cn write y = P T h x + P T (h ĥ)x + z, (8) which is the received nlog send signl t terminl fter cncelltion. We define the unwnted residul self-interference signl t node s y,res P T (h ĥ)x + z, (9) nd notice tht the power of the of the residul E[ y,res 2 ] increses proportionlly with chnnel estimtion error. We introduce two lels, nd, to distinguish etween the two different modes of self-interference cncelltion ville to the full-duplex terminls. We use to denote when the terminls use only the ptch ntenn prototype for pssive RF cncelltion. We then use to denote when the proposed nlog send cncelltion is used in comintion with the ptch ntenn. C. Results We now simulte the performnce of the nlog send cncelltion t terminl using the scheme just descried ove. We will use the Signl-to-Interference-Noise (SINR) rtio s the min metric in order to quntify the strength of the desired signl over the comined self-interference nd noise. If we look t the SINR t terminl Γ = E[ P T h x 2 ] E[ y,res 2, (1) ] we define the strength of the desired signl s P R E[ P T h x 2 ] in order prmeterize it y single vlue. The performnce of the cncelltion scheme ws simulted in Mtl using the experimentl mesurements of the ptch ntenn prototype for the self-interference chnnel. Tle I shows the other relevnt system prmeters. Terminl forms chnnel estimte with N tr trining symols nd then oth terminls exchnge N its of its with ech other. In Fig. 5 the SINR is plotted versus E /N to show the enefit of dding send cncelltion to the RF pssive cncelltion provided y the ptch ntenn. At pproximtely E /N = 1 db, the SINR for the RF scheme egins to sturte while the SINR for the send
Γ (db) TABLE I NETWORK SIMULATION RAMETERS System Prmeters Vlue K Modultion Order (M) 4 Numer of Dt Bits (N its ) 2 Numer of Trining Symols (N tr) 5 Crrier Frequency (f c) 2.438 GHz Smpling Frequency (F s) 2 MHz Chnnel Bndwidth (B H ) 2 MHz Signl Bndwidth (BW ) 1 MHz Terminl s Trnsmit Power (P T ) dbm Received Power from Terminl (P R ) -6 dbm 1 1 2 5 1 15 2 E /N (db) Fig. 5. The signl-to-interference-noise rtio (Γ ) of the desired signl from terminl to the residul self-interference t terminl. Bsend cncelltion improves the SINR s compred to the RF only scheme. scheme continues to increse linerly with E /N. The intersection point of the two curves is explined y the effect of noise on the send cncelltion scheme. In the presence of significnt noise, E /N < 1 in this scenrio, the chnnel estimte will hve high error nd cn ctully cuse more hrm thn good when forming the cncelltion signl. This ffect cn e oserved how the scheme chieves higher SINR thn the scheme. For E /N > 1, the ffects of the dditive noise lessen nd the chnnel estimte cn e used to crete eneficil cncelltion signl. We note tht the vlue of E /N is the sme for oth the trnsmitted dt signls x nd x. In order to quntify the trdeoff etween the RF only cncelltion of the scheme nd the nlog send cnceltion scheme, we define the rtio Λ +B = Γ +B Γ, (11) which clcultes the reltive SINR gin of the scheme over the scheme. In Fig. 6, we plot the SINR gin versus E /N. We cn immeditely see tht the SINR gin is linerly incresing proportionl to the incresing strength of the desired signl when nlog send cncelltion is used in comintion with the RF cncelltion provided y the ptch ntenn. The zero-gin point is the sme E /N = 1 point Λ +B (db) 1 5 5 1 5 1 15 2 E /N (db) Fig. 6. The reltive SINR gin of the RF only cncelltion scheme over the send cncelltion scheme versus E /N. discussed ove. When the dditive noise is too lrge, the RF only scheme outperforms the send scheme y 5 db. However, in low noise situtions, the send scheme relizes gins up to 1 db for the rnge of E /N considered. V. FULL-DUPLEX LINK EVALUATION In the ove section, we quntified the performnce of the nlog send cncelltion scheme in terms of the signl strength of the desired signl. We now quntify the performnce of the point-to-point full-duplex link etween terminls nd with two different metrics. The first performnce metric is the clssicl Shnnon informtion theoretic notion [12] of the chievle rte, R = log 2 (1+Γ ), where the rte is mesured in units of its per second per Hertz (ps/hz). In Fig. 7, we plot the chievle rte t terminl s function of E /N. Becuse the rte is function of the SINR vlue Γ defined nd evluted ove, we see similr trends in the rte of the full-duplex link s were oserved for the SINR. The rte of the RF cncelltion scheme sturtes t out 1.5 ps/hz. The chievle rte of the send scheme is linerly incresing with E /N nd cn chieve up to 4 ps/hz. The second metric we consider is the it error rte (BER) of the its trnsmitted y terminl nd received y terminl. Fig. 8 plots the BER with respect to E /N. It is clerly noticele how the use of send cncelltion improves the link qulity. For E /N = 1 db, we see fctor of 1 improvement in the BER nd t E /N = 2 db, up to 1 4 improvement is oserved. VI. CONCLUSION This pper proposes nd evlutes n nlog send self-interference cncelltion scheme. Rel-time, over-the-ir
4 3 1 1 1 R (ps/hz) 2 1 BER 1 2 1 3 1 4 5 1 15 2 E /N (db) Fig. 7. The chievle rte (R ) t terminl versus E /N. The send scheme chieves lmost 2.5 ps/hz higher rte s compred to the RF scheme. 1 5 5 1 15 2 E /N (db) Fig. 8. The it error rte (BER) t terminl versus E /N. The nlog send cncelltion scheme significntly reduces the error rte s compred to the RF only cncelltion scheme. mesurements of four-lyer RF ptch ntenn prototype were used to chrcterize the RF self-interference chnnel. The chnnel model comined with prcticl trnsceiver model enles us to derive n nlyticl send signl model incorporting the RF self-interference effects. Lest squres chnnel estimtion in the nlog send stge of the receiver is used to estimte the self-interference chnnel nd generte cncelltion signl just prior to the nlog-to-digitl converter. The performnce of the cncelltion scheme ws quntified through the SINR rtio of the desired signl with respect to the residul self-interference signl. The nlog send cncelltion scheme chieves up to 1 db higher SINR thn the RF only cncelltion scheme. We then evlute the performnce of point-to-point full-duple link with the chievle rte nd BER used s metrics. The send scheme is le to chieve up to 2.5 ps/hz improvement in chievle rte s compred to the scheme. A 1 1 1 4 reduction in BER ws chieved y dding nlog send cncelltion to the RF only cncelltion scheme. These initil results provide motivtion for dding nlog send self-interference cncelltion to current systems tht only employ RF self-interference cncelltion. Our proposed send cncelltion scheme is gnostic to the specific RF self-interference chnnel model nd cn e utilized with vrious other chnnel models. In our own extensions of this work, we consider lternte chnnel models in [13] nd provide more in depth nlysis. VII. ACKNOLDGEMENTS REFERENCES [1] F. O Hr nd G. Moore, A high performnce CW receiver using feed thru nulling, Microwve Journl, Sep. 1963. [2] R. Iserg nd W. Lee, Performnce tests of low power cellulr enhncer in prking grge, in IEEE Vehiculr Technology Conference (VTC), 1989. [3] M. Durte nd A. Shrwl, Full-duplex wireless communictions using off-the-shelf rdios: Fesiility nd first results, in Asilomr Conf. on Signls, Systems, nd Comp., Nov. 21. [4] J. Choi, M. Jin, K. Srinivsn, P. Levis, nd S. Ktti, Achieving single chnnel, full duplex wireless communiction, in MOBICOM, Sep. 21. [5] E. Everett, A. Shi, nd A. Shrwl, Pssive self-interference suppression for full-duplex infrstructure nodes, Jn. 213, sumitted to IEEE Trnsctions on Wireless Communiction. [Online]. Aville: http://rxiv.org/s/132.2185 [6] E. Aryfr, M. Khojstepour, K. Sundresn, S. Rngrjn, nd M. Ching, MIDU: enling MIMO full duplex, in MOBICOM, 212. [7] K. Hned, E. Khr, S. Wyne, C. Icheln, nd P. Vinikinen, Mesurement of loop-ck interference chnnels for outdoor-to-indoor fullduplex rdio relys, in Europen Conf. on Antenns nd Propgtion (EuCAP), Apr. 21. [8] M. Durte, C. Dick, nd A. Shrwl, Experiment-driven chrcteriztion of full-duplex wireless systems, IEEE Trns. on Wireless Commun., vol. 11, no. 12, Dec. 212. [9] D. Bhrdi, E. McMilin, nd S. Ktti, Full duplex rdios, in ACM SIGCOMM, Aug. 213. [1] Rice Integrted Systems nd Circuits L (RISC). [Online]. Aville: http://www.ece.rice.edu/ 28/index.html [11] J. Lu, Z. Kui, X. Zhu, nd N. Zhng, A high-isoltion dulpoloriztion microstrip ptch ntenn with qusi-cross-shped coupling slot, IEEE Trns. Antenns Propg., vol. 59, no. 7, Jul. 211. [12] T. M. Cover nd J. A. Thoms, Elements of Informtion Theory. John Wiley & Sons, 1991. [13] B. Kufmn, J. Lilleerg, nd B. Azhng, Anlog send cncelltion for full-duplex: An experiment driven nlysis, sumitted to IEEE Journl on Selected Ares of Communictions (JSAC) Specil Issue on Full-Duplex Communictions nd Networks, Octoer 213. The uthors would like to thnk Dr. Aydin Bkhni nd grdute student reserchers Tulong Yng nd Peiyu Chen. The ptch ntenn prototype ws developed in their l, Rice Integrted Systems nd Circuits (RISC), t Rice University.