Active interference cancellation technique for MB-OFDM

Embed Size (px)

Citation preview

  • 8/7/2019 Active interference cancellation technique for MB-OFDM

    1/4

    Active Interference Cancellation Techniquefor MB-OFDM Cognitive RadioHirohisa Yamaguchi

    Texas Instruments, T sukuba Technology Center17 Miyukigaoka Tsukuba Ibaraki, 305-0841, Japan(email: [email protected], tel: 81-29-881-1745)ABSTRACT - Ultra Wideband transmit power (in-band)is regulated to be -41.25 dB dM H z (in the United States),but in the future , it may be flex ibly relaxed sub ject to thecognitive-radio spectrum policy. However, in closeproximity of a protected radio service, transmission at therelaxed pow er migh t inflict an excessive degradatio n on theservice quality. Applying tbe general analog or digitalnotch fdter in the UWB transmitter (baseb and) is thesimplest but not a favorable approach due to the increasedcost and the power consumption of the device. In OFDM,turning off the interfering tones has been studied as thealternative solution, but the inter-carrier interference mayl imit the notch depth to 5-10 dB. This paper discusses anew ap proach that enables the accu rate notch bandwidthand depth control for the general OFDM transmitter. Thetechnique demonstrates the fundam ental advantage of theOFDM-based UWB solution[l] for the future cognitiveratio.

    1. INTRODUCTIONUltra Wideband technology provides a new paradigmfor utilizing the limited frequency resource. Sharing thesame frequencies in close proximity to current andfuture radio services, however, remains to be a

    challenging problem. For the out-of-band (below 3.1GHz mostly) interference, the use of a band-limitingfilter would lower the UWB spillover radiation I O to 20dB below the operative noise level. For the in-band(between 3.1 and 10.6 GH) interference which weexamine in this paper, the use of narrow-band (notch)filters, or an equivalent mechanism is surmised to beessential, or at least necessaty. Such mechanism may berequired in case the upper limit of the UWBtransmission power is relaxed within the vicinity of asusceptible radio receiver. The case directly relates withthe future introduction of the cognitive radio[2].Because UWB is a ubiquitous radio, we expect it toincorporate a coexistence mechanism, which shouldinclude both detection of a nearby victim radio (orcompliance to the regulatoiy rules) and reducing theinterference below the sensitivity threshold; in thispaper, we focus on the latter technology. Because theUWB transmitter is a power-limited radiator, nullifyingthe transmission over an unnecessarily wide bandwidthhurts the throughput; thus for the victim radio, definingthe tightest notch is essential. In the general filterdesign, a rigid band discrimination filter requires a veryhigh sampling rate or a long taps (well over 100 signalsamples) together with the filter coefficients quantizedin 4 bits or more. For example, when the signal

    bandwidth is 1.5 GHz (impulse or DS-type UWB) andthe notch bandwidth is 7 MHz, the digital filter wouldconsume over 600 mW, even in the next-generation 90nm CMOS (and this estimate is very optimistic).We are interested in investigating if the Multi-bandOFDM[l] UWB system, which performs thefundamental signal processing in the frequency domainwould suffer from the similar problem. One aspect ofOFDM which is often pointed out is that the constituenttones can be turned off at the interfering frequencies,creating a spectrum null. In MB-OFDM, each toneoccupies 4.125 MHz, and this bandwidth is reasonablyfine-grained to craft a notch without excessivelysacrificing the system throughput (See Fig.1 for thebaseband M B-OFDM spectrum).

    In OFDM, tones are placed at a regular frequencyinterval to avoid the inter-carrier interference[5],[6](orthogonaIity), The inter-carrier interferencebecomes large in-between the tones, and is generated byall the non-zero tones following the sinc-functioninterpolation. Due to this property, tuming off a singletone does not create a spectral null within its 4.125 MHzband, and its residual in-band power is determined bythe neighboring non-zero tones. For a single tone, thenotch bottom level (the inter-carrier interference power)is typically 5 dB to I O dB below the average signalpower. This may be insufficient (subject to theconditions of bandwidth, distance and sensitivity of thevictim system). In OFDM , a deeper notch is achieved byturning off a larger number of tones. An example isshown for the case of 4 and 16 tones in Fig.2.We will discuss a different solution in the nextsection.

    OFDM spectrum width 12 8 tones x 4.125 MHz = 5 2 8 MHz

    Fig. 1 MB-OFDM baseband spectrum

    34'h European Microwave Conference - Amsterdam, 2004 1105

    mailto:[email protected]:[email protected]
  • 8/7/2019 Active interference cancellation technique for MB-OFDM

    2/4

    One OFOM frame

    Fig. 2 (a) Turning off 4 tones creates 8 dB notchtargeted band position

    Fig. 2(b) Turning off 16 tones creates 15 dB notch11.ACTIVE INTERFERENCE CANCELLATION

    In this and the following sections, we discuss theUWB system based on Multi-Band OFDM[3]. MB-OFDM is an extension of OFDM to the 7.5 GHz-wideUWB band by defining five band-groups, each of whichconsists of three (the highest hand uses two) OFDMsymbols of 128 tones, and applying the time-frequencyinterleaving based on the time-frequency code (TFC). Inthe analysis, we treat the interference in the equivalentbaseband OFDM signal.When the information data is denoted by X@)k=O,. . ,127, the transmitted OFDM signal is given by

    In order to evaluate the interference in-between thetone frequencies, we up-sample (we apply four-timesup-sampling here) the corresponding spectrum, which isgiven by Yp) (1=0, ...4*128-I),

    Combining these two equations, we obtain as therelation between Xan d Y

    (3 ), 127128 i-U= - z X ( k ) P ( l , k )

    r MB-OFDM toneswlhinhe nterference band interferenceband

    Active Interference Cancellation tones(one one at each edge)These tones suppress lhe intederencecaused by the informationdata tones.Fig.3 Definition of the AIC tone position

    where P(n, I) is the transform kemel.Instead of tuming off a large number of tones, wedefine two special tones at the edge of th e interferenceband as shown above, and would prove that these twotones can sufficiently cancel the interference in the band .The tone values can he arbitrarily determined withoutaffecting the information tones due to the orthogonalityrelationship. We call these special tones ActiveInterference Cancellation (AIC) tones. We discussbelow how to compute the AIC tones, and how to createthe notch using the minimum number of tones, or,equivalently, how to m axime the spectrum efficiency.Following the definition of the interfering bandposition, OFD M tones within the hand, and the positionof the AIC tones shown in Fig.3, we take up a specificexample of the (in-band) UWB interference to the RadioAstronomy service of 3260-3267MHz. The tones #85,86 and 87 of the MB-O FDM Band #l co-locate with thisband as shown in the figure. The MB-OFDMinterference to this band is evaluated at four-times (orhigher) finer frequencies denoted by a vector d l . dl( l ) ,d1(4)and dl(8) correspond to the MB-OFDM tones #85,86, 87. We then add two tones on the edge outside thesethree tones, and try to cancel the interference inside theinterference band using the total of five tones. It will beshown later that the AIC tones play the dominant roleand three 'in-band' tones can he simply 'turned o f f . Ithas been found that increasing the number of the AICtones does not significantly improve the cancellationperformance, thus the current solution seems to be nearoptimum.The vector d, is given by

    d, = Pg (4),where P is the kernel defined by (3) an d g is the vectorof the information data tones w ith X(S4) to X(S8) turnedoff (zeroed).In order to cancel the interference within the band, weneed to generate the negative of the interference signalusing the tones X(84) to X(S8). Again using the relation(3) above, setting all the X o zero except for X(S4) toX(88), the equation to be solved is given by

    P,h = - d , ( 5 )where h is th e column vector of (X(S4), ....X(SS)) and P Iis the small kernel derived from P by limiting the index

    1106 34" European Microwave Conference - Amsterdam, 2004

  • 8/7/2019 Active interference cancellation technique for MB-OFDM

    3/4

    according to h an d d, . Thus P I is a 9 x 5 matrix.Here, h is our desired tone values. However, ( 5 )cannot be solved in the straightfonvard way because thematrix P , is not invertible. Hence, instead, we seek theminimization ofe =IIch+d , l12 (61,

    which leads toh = - ( ~ T ~ ) - l ~ T d l-W,d, (7).

    This minimum mean-squared solution is also knownas the M oore-Penrose generalized inverse[4]. It is notedthat the resultant inverse 5 x 9 matrix V i s pre-computable because the interference band location ispre-defined (by the regulatory rules. Also it can bebroadcast by m eans of the inter-device communication ).Now com bining (4) and (7),h is given byh = - W P g - - w , g (81,

    where E is also a pre-computable 5 x 128 matrix111. AIC P ERF ORM ANC E

    For the targeted (Radio Astronomy) 7-M Hz band, wecompare two approaches, both of which use 5 tones forthe notch implementation. By tum ing off all the fivetones, the notch depth is limited to be around 13 dB asshown in Fig.4 (a). Using AIC, a m uch deeper notch ofover 60 dB results. It can be further shown that thenotch depth is determ ined by the quantization accuracyof the AIC coefficients. Here, it can be also shown thatthe optimum tone values are near zero for the threemiddle tones (#85, 86 and 87). Taking thissimplification, we only have to compute the tw o AICtones. Fig. 5 shows the notch depth achieved byquantizing the AIC tones to 2, 3 and 4 bits. The notchdepth vanes in the range of 30 to 40 dB. Hence, in thecase a -80 dBm& fHz notch is desired, 4-bit quantizationis required for the AIC tones (-41.25 dB m/MH z - 40dB= -81.25 dBm/MH z). Fig.6 shows an exam ple of such anotch created by the 2-bit AIC tones.

    tamet band f7MHz)

    cancellation by AICRadio Astronomy band in Japan3260-3267 MHz

    tones, and b) 2-bit AIC tones

    Here, we are interested to compare AIC with thelegacy digital filter technique and this is brieflydiscussed below.Under the same bandwidth constraint, FIR digitalfilters can be easily synthesized ignoring the coefficientquantization. For the Radio Astronomy bandwidth of 7MHz and the transient bandwidth of 4 MHz, it can beshown that a 308-tap FIR filter can create a -63 dB

    ..*...,....,..*

    . . .

    AI C tone quantized in 3 bits(including the sign bit)interference suppression= 35 dBAIC tone quan tized in 4 bits(including the sign bit)Interference suppression= 40 dB

    AI C lone quantized n 2 bits(including the Sign bit)Interferencesuppression= 30 d 6Fig. 5 Quantization of the AIC coefficients and the depth of the created notch. The case of simply tuming offfive tones is shown in the upper half for the reference sake.

    34h European Microwave Conference - Amsterdam, 2004 1IO7

  • 8/7/2019 Active interference cancellation technique for MB-OFDM

    4/4

    24 MHz_j L~~ ~. A: ~ ~ - .- ..,notch. For the notch depth of 24 dB (6 dB w orse thanthe 2-hit AIC in FigS ), the required minimum filter tap- : I: :length is 128 . The filter characteristic is shown below. , , 1

    U QP Q,lFrequency (GHz) ~ ~~Five-tone equivalent equi-ripple 128-tapFIR(suppression depth -24 dB ) Ripple is +-0.6d B over the entire Flg 8 24-MHz notch In a) nine tones are tumed off andb) by AICtonesFig. 7 FIR filter characteristic realized by 128-tap 4-hit

    coefficients. The pass-band and transient bandwidth isequivalent to the cases shown in Fig. 4 , 5 and 6 .The filter entails a ripple of 0.6 dB in the pass-handand this problem becomes more pronounced for the

    shorter filters. Assuming the four-hit quantization forthe filter coefficients (optimistic assumption) and takinginto the symm etry of the general FIR filter structure, thepower consumption of this digital filter (when it isapplied to the base-hand signal of the bandwidth 528MHz) is estimated to be around 300 mW, (0.13 umCMOS) while all that is required for the AICcomputation is less than 2 mW . This directly comesfrom the advantage of the frequency-space signalprocessing.

    For the 7-MHz notch, the total number of the tonesdiscussed in the current paper was five. In practicalapplications, the associated tone reassignment would heeither predefined according to the regulatory rules, orflexibly determined through the inter-devicecommunication. In M B-OFDM, a notch of up to >30MHz (4.125 x 8 = 34 MHz) bandwidth can be createdwithout a serious throughput penalty.Finally, Fig.8 shows an example of the wider (24-MHz) notch. The AIC tones are quantized to tw o bitshere. The AIC concept can be easily extended to meetother notch requirements including m ultiple notches.In light of the ubiquitous radio device that is destinedto coexist with the current and future radio services andoperate in low power (less than 200 mW), theimplementation of the interference suppressionmechanism in low power is an essential feature. Thisleads to our belief that OFDM -based UWB is the key tothe future ubiquitous radio, and the key to open the newtechnology called the cognitive radio[2].

    IV . CONCLUSIONUWB is a new paradigm of the ubiquitous radio, hut

    at the same time, the problem of the interference to thecurrent and future radio services must also he addressed.The AIC technique presented here is based on thebenefits of the frequency-domain signal processing,proposed as MB-OFDM as the worldwide UW Bstandard. The frequency-domain signal processing is theway to go when we consider the performance andcoexistence with the other radio services. MB-OFDM isa solution that achieves both goals.

    ACKNOWLEDGEMENThe author wishes to acknowledge the Multiband

    OFDM Alliance team for the engineering discussion andinsights. He especially thanks Anuj Batra of TexasInstruments for his invaluable comments andsuggestions.

    REFERENCES[ I ] Multiband OFDM Alliance (MBOA) home page(http:i/www.multibandofdm.com)[2] FCC ET Docket No.03-108, Cognitive RadioTechnologies Proceeding (CRTP), December 2003[3] IEEE P802.15-03/26813, Multi-band OFDM Physical

    Layer Proposal for IEEE 802.15.Task Group3a, April2004[4] G.H.Golub and C.F.Van Loan, Matrix Computations, TheJohn Hopkins University Press, 1983[ 5 ] Y. Zhao and S-F Hagpan, lntercarrier InterferenceSelf-C ancellation Scheme for OFDM M obileCommunication System s, IEEE Trans. on Comm.Vo1.49, No.7, July 2001[6 ] S.Diggavi et al. Intercamer Interference in MIMOOFDM, CC2002 0-7802-7400-2

    1108 34 European Microwave Conference - Amsterdam, 2004

    http:///reader/full/i/www.multibandofdm.comhttp:///reader/full/i/www.multibandofdm.com