MIMO Active Interference Cancellation for

MIMO Active Interference Cancellation for Stealth Cognitive Radio in MB-OFDM UWB Hsin-Jui Chou and Jen-Ming Wu Inst. of Communications Engineering National Tsing Hua University Hsinchu, Taiwan, ROC jmwu@ee.nthu.edu.tw ABSTRACT -Cognitive Radio (CR) is considered as a promising approach for efficient utilization of precious radio spectrum resources. In this paper, we present a new approach that combine active interference cancellation (AIC) and vertical Bell Laboratories Layered Space-Time (V-BLAST) techniques to transmit data through a multiple-input and multiple-output (MIMO) system within protected band which has been assigned for primary subscribers. This method allows the transmitter to transmit data in the protected band in a stealth fashion (as low as -120dB to -300dB) so that the primary subscribers will be unaware of the transmission. We show the opportunity of the secondary user to utilize the valuable spectrum of the protected band as well. The simulation shows that BER performance in the protected band is as good as in the unprotected band. We have also simulated the proposed scheme with the co-existence of WiMAX and MB-OFDM UWB devices as the primary user and the secondary user respectively. Index Terms - Cognitive Radio, MIMO, Inter-Carrier Interference Cancellation, VBLAST I. INTRODUCTION Nowadays, the radio electromagnetic spectrum has become a very precious naturalistic resource in the wireless communication. Federal Communications Commission (FCC) in the United States, and its counterparts around the world, allocate a wide range of the radio spectrum of varying channel bandwidths. For examples, AM/FM radio, UHF/VHF television, cell phones, Global Position System (GPS), IEEE802.16e, Ultra-Wide Band (UWB), and so on. Just as wireless services have begun proliferating, there is little radio frequency spectrum left over [1]. However, not every channel in every band is always in use. For example in UWB technique, FCC allocated 7.5 GHz of spectrum in the 3.1 - 10.6 GHz frequency band for unlicensed devices or users who want to access the UWB [2]. However, the spectrum is not utilized as economically as we expect. It is reported that the actual spectrum usage efficiency measure in an urban area in 3-4 GHz is about 0.5%, and moreover drops to 0.3% in 4-5 GHz [3]. In fact, the FCC has determined that, in some locations or at some times of day, 70 percent of the allocated spectrum may be sitting idle. Therefore the solution lies with Cognitive Radios (CR). Cognitive radio defines as a paradigm for wireless communication in which either a network or a wireless node changes its transmission or reception parameters to communicate efficiently avoiding interference with licensed or unlicensed users. Active Interference Cancellation (AIC) technique, provides one of the CR solution [4]. Data in AIC technique is transmitted with some canceling data together, and then it can avoid interference with other user at same frequency band. In this paper, we assume a secondary user with MIMO MB-OFDM UWB system operating in the spectrum of 3.1GHz - 4.6GHz. The primary user of WiMAX service operates at 3.5GHz with 10MHz channel bandwidth. Therefore, with the two services trying to function simultaneously, the narrowband of the primary user becomes a protected band. Conventional CR approach would require the secondary user to provide detect-and-avoidance (DAA) capability so that the second user operation would not interfere with the primary user at the expense giving up using this protected band. In this paper, we seek the possibility that, at the constraint of not interfering with the primary user, the secondary user could still utilize the protected band for its data transmission. It exhibits that the power spectrum is still clean or below the sensitivity level and at the primary user. The receiver of the secondary user employs MIMO avenues, e.g. V-BLAST, to resolve the information. Since power spectrum in protected band is suppressed with respect to the primary user, the primary user receiver will view these signals as noise-like interferences so that the secondary user can secretly transmit data with acceptable bit-error rate (BER) performance and be without polluting protected band. In Section II we overview the SISO AIC technique and signal model. In Section III, we introduce the MIMO AIC. We briefly show that how AIC system avoids interference with the primary user who has already occupied at the frequency band. In Section IV, we discuss about the decoder of MIMO AIC system. We need the special decoder for AIC system. Section V is devoted to BER comparisons and some other simulation. Last, summary and concluding remarks appear in Section VI. II. SISO ACTIVE INTERFERENCE CANCELLATION In this section, we briefly overview the AIC algorithm. Suppose a secondary user tries to send data in a particular frequency band, however, it discovers that there is a primary user has already occupied in a narrow band of three tones conflicting with the secondary user. Customarily, the secondary user could tum off these tones Authorized licensed use limited to: HUNAN UNIVERSITY. Downloaded on April 27, 2009 at 22:52 from IEEE Xplore. Restrictions apply. 12510050 75 subcarrier 25 -300 -+---"T""-...----"T""-...--_-.......-_-.....----.--~ o III. PROPOSED MIMO ACTIVE INTERFERENCE CANCELLATION W opt =-(PI H • PI )-1 PI H • d1. (8) The power in victim band is minimized. It is more efficient than the tum-off method that AIC method sacrifices a few tones to suppress power and does a great job. In finding the AIC signal, we take up-sampling rate of 4 (i.e. r =4). To observe the result more clearly 20 times up-sampling (r =20) is adopted in simulations. For 5-tone AIC, its outcomes as well as a lot of nullifying tones being utilized. For 9-tone AIC, it creates a notch up to 100 --120 dB. We compare AIC method and tum-off method with the same number of tones utilization. Figure 1 tells us that turn-off method suppress about 15-20 dB, but Fig. 2 shows AIC method stripes interferences away for about 120 dB notch. Even 25-tone utilization for tum-off just notch power magnitude around 25 dB. It is still worse than performance of 9-tone AIC as the black line does. ·250 -200 50 I 3-tone AIC I :;. o f'1t\y\~")\1r~'r,;V"{\!yrrVV~r)ifJ\r~~1 -50 Is-tone AIC I E ~ -100 I 7-tone AIC I a.~ I 9-tone AIC I ~ -150 • o a.. where d l is interference extracted from unvictim tones toward victim tones. What we need to do now is to cancel d1 for clean transmission circumstances. Let w represent the X(k)--X(k+q-1) tones, it is the AIC information, which is computed as following: Plw=-dh (6) where PI is the small kernel derived from P by limiting the index according to wand d1. The criterion is minimizing the mean square error, e2, or W opt =arg minlle 2 11 =arg min~P1 · w + d111 2 } (7) w w As a result, Fig. 2. Power spectrum ofdifferent Ale tones use for suppression. We take 3, 5, 7 and 9 tones for comparison. Power Spectrum after Ale In SISO AIC, the secondary user can only use the unvictim frequency band that the primary user doesn't use. The secondary user avoids using the victim band to transmit data. Here we look for the possibility that the secondary user could still utilize the protected band for its data transmission at the constraint of not interfering with the primary user. Let the secondary user be a MIMO OFDM system, which use M antennas to transmit information data. Let the channel between the secondary user's transmitter and the primary user be H' and the channel between the transmitter and the receiver of the 125 3-tone 7-tone -13-tone 25-tone 100 V''YV\ ~V •• 50 75 subcarrier 25 -300 '------__.L--__-"--__--L-__----"---__-----L-..-l o Power spectrum of tum-off -50 -250 50c-----~--.--------,---------,---.j m ~ -100 2 I ~ -150 i -200 Fig. 1 Power spectrum for different turn-off tones, named for 3, 7, 13 and 25 tones for comparison No-l x(n) = "LX(k)'expU21D1kl No) (1) k=O In order to evaluate the interference in-between the tone frequencies, we up-sample the corresponding spectrum, which is given by Y(i), and the up-sampling rate is denoted by r, 1 No-l • n I Y(/) = -"L x(n)· exp( - J2;r--) No n=O No r 1 No-l No-l I (2) = -"L("LX(k). exp(j2;r~(k--))) No n=O k=O No r =_1 IX(k).P(l,k) No k=O where 1=0, ... , (r . No -1) and P(l, k) is the transformation kernel matrix. We first have to compute the interference coming from the secondary user's information data, which might interfere with the protected band that the primary user used. Secondary user tum off tones of victim band signals and see how much interference comes from other tones of the unvictim band signal. g=[X(O) .. X(k-1) 0 ... 0 X(k+q) .. X(No-1)] (3) where g is a vector containing the secondary user's transmit data. Assuming that the primary user occupied q tones in the middle of the frequency band, we turn off q tones in the vector g. And then according to (2), we define the vector d by up-sampling the vector g, d=P'g (4) dl =[d(k· r), ...,d«k +q -1)· r)], (5) • When the information data of OFDM symbol with No subcarriers is denoted by X(k),k = 0, .. .,No-I. The discrete signal can be expressed as : around protected band (which the primary user has already used) to avoid interference. However, the achievements are limited. As Fig. 1 shows, nullifying these 3-tones make about 5--1O-dB notch and nullifying up to 25-tones can merely suppress 25--30dB. As tum-off subcarriers increases, notch depth increases limitedly. The subcarriers involved in the secondary user could be wider than the protected band, so we call these involved subcarriers in the secondary user the victim band. Authorized licensed use limited to: HUNAN UNIVERSITY. Downloaded on April 27, 2009 at 22:52 from IEEE Xplore. Restrictions apply. Fig. 3. The secondary user is a MIMO OFDM system that mIght interfere with the primary. Similarly, the primary user might also confuse the receive part of the seconaary user. Let. Y be the interfering signal that arrives the primary user from the transmitter of the secondary user within the protected band. Y has dimension of No, where No is the number of OFDM subcarriers of the secondary user. Let H' be the estimated channel frequency response between them of dimension No. as well. We have Y=H'.X=[H'l ... H'My[J~]=[Yl+ ...+YM] (9) where each Yk =H'kT.Xk' k=1, ... ,M and Xk is the transmit data stream from the kth antenna of secondary user. We use the first antenna to do the AIC algorithm that cancels the interference caused by other antennas. Therefore, according the same rule to the SISO AIC, we first turn off tones of victim band signal in first transmit antenna and see how much interference comes from other tones of the un-victim band signal, or even from other antennas' data tones. Let Xl =[XI(O) ...X I(k-1) 0 ... 0 XI(k+q) ...XI(No-1)] be the signal of the 1st antenna, where q is the number of tones within the victim band. Then we get some part of received signal in primary user : Yl =H'l X =[Y1(O) ... Y1(k-l) 0 ... 0 ~(k+q) ... ~(No-l)] Let d be the up-sampled signal of Y I with up-sampling rate of r and the interference be d1, that is d=P.(YI+···+YM ) (10) dl =[d(k·r),...,d«k+q-l)·r)] . (11) Let wy =[Y1(k), ... ,Y1(k+q-l)] and W =[X1(k), ... ,Xi(k+q-l)] tones, which is the AIC information. To cancel the interference, the requirement of (12) should be satisfied, PI Wy =-d}, (12) where P I is the small kernel derived from P by limiting the index according to wyand d i . The criterion is minimizing the mean square error, e2, or wy,op' =arg mJnlle2 11 =arg mJn~Pl . wy +d1r} (13) As a result, _ '+ _ (H )-1 H '+ (14) w opt - W Y 'opt H - - PI . PI PI . d . H where (.)+ represent pseudo-inverse matrix operation. At last, after the action of AIC, the transmit data stream becomex=[xl , ••• ,XM ] where Xl = [X1(O) ... X1(k-l), w, X 1(k+Q) ... X 1(No)]T(15)mtfDDRl antfDDam _ ... secondary to primary user interference Secondary User Rx •••••••••~ primary to secondary user interference secondary user ----+ data transmission Ale toors ....~ .. .. .. ... .. .. .. H Secondary User Tx As shown in Fig. 4, we immolate the tones within victim band in first antenna to send AIC signals and then we get the opportunity for the antennas to transmit information data stream in the same frequency band. Therefore, the secondary user can slinkingly transmit data in the frequency band where primary user has used without adding interference to primary user. Later, in the receiver of secondary user, we apply successive interference cancellation technique[7][8] to detect the information from the secondary user transmitter. The block diagram of the proposed scheme is shown in Fig. 5. secondary user be H as shown in Fig. 3. Assuming that the channel information ofR' could be obtained through blind channel estimation [5][6]. Since Rand H' would be different, we can use one of the secondary user's transmit antenna to execute the AIC algorithm based on the channel information H' such that the power spectrum arriving the primary user is minimized at the protected band. Fig. 4. Definition the AIC tones position in the transmit part of secondary user. IV. MIMO RECEIVER with MMSE V-BLAST secondary user tnllSDlit part I~=;·'H ~c Ho.....-Tr_Bfl_Sm_it_ Tx Fig. 5. MIMO AIC structure. The receiver of the secondary user adopts a V-BLAST structure [8] that consists ofM transmitters and N receivers, where M ~ The main idea ofV-BLAST algorithm is to de-multiplex a data stream into M sub-streams, and then each sub-stream is encoded into symbols and taken over by individual transmitters, and finally radiated through its belonged transmit antennas in parallel at the same time and frequency. These transmitters operate co-channel. Each receiver antenna takes over all the transmitted signals in different amplitude and phase variation which are mixed due to the wireless propagation channel, H. Using proper signal processing at the receiver end, these mi~d signals Authorized licensed use limited to: HUNAN UNIVERSITY. Downloaded on April 27, 2009 at 22:52 from IEEE Xplore. Restrictions apply. where (G;)j denotes the jth row of Gj , Uk; stands for the (20j) (20a) (20b) (20c) (20d) (20e) (20f) (20g) (20h) (20i) if-i+l Initialization: i ~ 1 ( H 2)~ HG= H .H+uv ·lm ·H k1 =argmi~l(G) j 11 2 j Recursion: Ukl =(Gi)kl Sk =Uk H ·R. I I I R i+1 = R i - Ski (H)kl Xkl =Q(Skl ) ( H 2 )-1 HG H1 = Hi, .Hi, +G v ·Ii, · Hi, ki+1 = arg min \I(G i+l) j 11 2 j~{kl···kJ V. SIMULATION In the soft-decision we only exchange two formulas' orders of stepf and step g with minor changes. B. Soft Decision MMSE V-BLAST We find that the MMSE V-BLAST with soft decision decoder does not have such problem at decoding AIC tones. In soft decision decoder, after multiplying R with a row vector of G, we obtain the soft-detecting signal. Immediately, we multiply soft-detecting signal with it's corresponding channel vector and then directly subtract to the receive vector R. Afterward the soft-detecting signal quantization is more appropriate to the adopted constellation. We rewrite V-BLAST operating process of (19) for MMSE V-BLAST with soft decision decoder as follows: matrix subtracted the columns kl' k2 , ••• , k i of H, 1\_11 2 is the 2-norm of the vector and Q( . ) denotes the quantization operation appropriate to the constellation in use, Sk j is the soft-detecting signal which is subtracted out from the received signal vector R;, {k1, k2 , ••• , km } is a permutation of the integers 1"'M specifying the order in which components of the transmitted symbol vector X are extracted. Nevertheless, the MMSE V-BLAST decoder algorithm with hard decision is not suitable for MIMO AIC system. The MIMO AIC system, we mentioned in above, applies AIC tones to send the cancelling information that limited the interference to primary user's disturbance. Therefore, data transmitted in AIC tone is not a quantized constellating data (i.e. QPSK or QAM or ... ). In the simulation, we assume the secondary user to be the MB-OFDM UWB device with 128 subcarrier OFDM signal ranging from 3.1GHz to 3.6GHz for the first band of the first group. A primary user of WiMAX device operating at 3.5GHz of lOMHz bandwidth has already occupied the frequency band which located at 86t\ 87th and 88th tone of the secondary user. Then, the secondary user with 4x4 MIMO OFDM simultaneously tries to .send data (19a) (19b) (19c) (19j) (19d) (1ge) (19f) (19g) (19h) ( 19i) k1 =arg~i~l(G) j 11 2 ] Xkl =Q(Skl ) R i+1 = R i - Xkl (H)kl ( H 2 )-1 HG'+1 = Hi, ·Hi, +G v .Ii , . Hi, ki+1 = arg min IICG HI)Jj~{kl···kj } i ~i+l Recursion: U kl =(Gi)kl Sk = Uk H ·R. I I I could be seperated as they are transmitted through a set of independent parallel channels. Let X =[Xl' ... ,XM ] denotes the vector of transmit symbols from M transmit antennas, and R = [R1 ••• RN]T denotes the received vector symbol. Then we can describe the relationship between X and R : R=H·X+v (16) Where H is the N-by-M complex channel matrix with statistically independent entries and v is the complex Gaussian noise vector with zero mean and variance (J"v 2 • Based on the formula (16), we first discuss a detection called Minimum Mean Squared Error (MMSE) linear detection to multiply the received signal vector R with a linear transform matrix G, i.e. G=(HH.H+u v 2 .1 M )-1.HH, (17) ,where superscripts "H" and "-I" represent matrix Hermitian and inverse operations respectively. Then the estimated signal vector in the linear MMSE detection may be represented as X=G.R=G.H·X+G·v (18) A. Hard Decision MMSE V-BLAST Next, we describe the V-BLAST with MMSE receiver. In MMSE V-BLAST algorithm with hard decision, this detection method does not detect the m signals at one run. Instead, it multiply R with a row vector of G replace with the matrix G. This row vector of G which has the best post-detection SNR amongst the all row vectors of G is selected. Then the soft-detecting signal is subtracted out from the received signal vector. At last, we can detect the signal from quantizing the soft detecting signal appropriate to the constellation in use (hard decision). After detecting one of the received signals, we could subtract the detected signal which multiplies with it's corresponding channel vector to the receive vector R, and then start next iteration. This process proceeds until all the signals are detected. The full MMSE V-BLAST with hard decision algorithm can be described compactly as follows : Initialization: i ~ 1 ( H 2 )-1 HG= H ·H+uv ·lm ·H Authorized licensed use limited to: HUNAN UNIVERSITY. Downloaded on April 27, 2009 at 22:52 from IEEE Xplore. Restrictions apply. at same band. Therefore, the secondary user would have to interfere with the primary user at the protected band of 10MHz at 3.5GHz. Let the 84th to 90th tones of first antenna be the AIC tones, which play the canceling signal. And other antennas from secondary user may send QPSK signal. P01Ae!' spectrum 01 in JXimary users recehe part 10:2 .---~--r----r----r----r----r------, 10" 10~ Fig. 6. The 4x4 MIMO AIC system. power spectrum of the transmit signal, which transmIt from the secondary user and received at primary user. Figure 6 shows the power spectrum of interference which caused by the secondary user to the primary user. We can observe that the notch depth is limited to be around -110dB by using AIC technique for 4X4 MIMO OFDM system. Then, Fig. 7 shows the bit error rate after MMSE V-BLAST decoder with hard decision in the received part of secondary user. They are two lines in Fig.7. The first line which limited at 10-2 represent BER which the decoder detect all transmitted data except the 84th -- 90th tones in first transmitter. And the other line then shows the BER that decoder detect only Oth -- 83rd and 91 st -- 127th tones in each transmitter a